* [PATCHv5 00/14] remoteproc: updates for omap remoteproc support
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
Hi,
This version addresses review comments received for v4 of the series.
Mostly cosmetic changes, the most important changes here:
- minor dt binding updates and any driver changes induced by this
- reset handling fixes to avoid asserting/deasserting same reset twice,
this appeared to still produce some timeouts in some cases; this was
masked earlier but became visible with one error handling fix
All changes are listed under individual patches.
-Tero
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCHv5 01/14] dt-bindings: remoteproc: Add OMAP remoteproc bindings
2020-01-16 13:53 ` Tero Kristo
(?)
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Rob Herring,
devicetree, Tero Kristo
From: Suman Anna <s-anna@ti.com>
Add the device tree bindings document for the IPU and DSP
remote processor devices on OMAP4+ SoCs.
Cc: Rob Herring <robh@kernel.org>
Cc: devicetree@vger.kernel.org
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to schema]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- enforced ordering of reg-names
- removed ti,bootreg-shift, instead added this under ti,bootreg
property
- renamed ti,autosuspend-delay to ti,autosuspend-delay-ms
- fixed required property handling
- fixed the omap5 ipu example firmware name extension
.../remoteproc/ti,omap-remoteproc.yaml | 322 ++++++++++++++++++
1 file changed, 322 insertions(+)
create mode 100644 Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
diff --git a/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml b/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
new file mode 100644
index 000000000000..b910aac7895e
--- /dev/null
+++ b/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
@@ -0,0 +1,322 @@
+# SPDX-License-Identifier: (GPL-2.0-only or BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/remoteproc/ti,omap-remoteproc.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: OMAP4+ Remoteproc Devices
+
+maintainers:
+ - Suman Anna <s-anna@ti.com>
+
+description:
+ The OMAP family of SoCs usually have one or more slave processor sub-systems
+ that are used to offload some of the processor-intensive tasks, or to manage
+ other hardware accelerators, for achieving various system level goals.
+
+ The processor cores in the sub-system are usually behind an IOMMU, and may
+ contain additional sub-modules like Internal RAM and/or ROMs, L1 and/or L2
+ caches, an Interrupt Controller, a Cache Controller etc.
+
+ The OMAP SoCs usually have a DSP processor sub-system and/or an IPU processor
+ sub-system. The DSP processor sub-system can contain any of the TI's C64x,
+ C66x or C67x family of DSP cores as the main execution unit. The IPU processor
+ sub-system usually contains either a Dual-Core Cortex-M3 or Dual-Core
+ Cortex-M4 processors.
+
+ Each remote processor sub-system is represented as a single DT node. Each node
+ has a number of required or optional properties that enable the OS running on
+ the host processor (MPU) to perform the device management of the remote
+ processor and to communicate with the remote processor. The various properties
+ can be classified as constant or variable. The constant properties are
+ dictated by the SoC and does not change from one board to another having the
+ same SoC. Examples of constant properties include 'iommus', 'reg'. The
+ variable properties are dictated by the system integration aspects such as
+ memory on the board, or configuration used within the corresponding firmware
+ image. Examples of variable properties include 'mboxes', 'memory-region',
+ 'timers', 'watchdog-timers' etc.
+
+properties:
+ compatible:
+ enum:
+ - ti,omap4-dsp
+ - ti,omap5-dsp
+ - ti,dra7-dsp
+ - ti,omap4-ipu
+ - ti,omap5-ipu
+ - ti,dra7-ipu
+
+ iommus:
+ minItems: 1
+ maxItems: 2
+ description: |
+ phandles to OMAP IOMMU nodes, that need to be programmed
+ for this remote processor to access any external RAM memory or
+ other peripheral device address spaces. This property usually
+ has only a single phandle. Multiple phandles are used only in
+ cases where the sub-system has different ports for different
+ sub-modules within the processor sub-system (eg: DRA7 DSPs),
+ and need the same programming in both the MMUs.
+
+ mboxes:
+ minItems: 1
+ maxItems: 2
+ description: |
+ OMAP Mailbox specifier denoting the sub-mailbox, to be used for
+ communication with the remote processor. The specifier format is
+ as per the bindings,
+ Documentation/devicetree/bindings/mailbox/omap-mailbox.txt
+ This property should match with the sub-mailbox node used in
+ the firmware image.
+
+ clocks:
+ description: |
+ Main functional clock for the remote processor
+
+ resets:
+ description: |
+ Reset handles for the remote processor
+
+ memory-region:
+ $ref: /schemas/types.yaml#/definitions/phandle
+ description: |
+ phandle to the reserved memory node to be associated
+ with the remoteproc device. The reserved memory node
+ can be a CMA memory node, and should be defined as
+ per the bindings,
+ Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
+
+ firmware-name:
+ description: |
+ Default name of the firmware to load to the remote processor.
+
+# Optional properties:
+# --------------------
+# Some of these properties are mandatory on some SoCs, and some are optional
+# depending on the configuration of the firmware image to be executed on the
+# remote processor. The conditions are mentioned for each property.
+#
+# The following are the optional properties:
+
+ reg:
+ description: |
+ Address space for any remoteproc memories present on
+ the SoC. Should contain an entry for each value in
+ 'reg-names'. These are mandatory for all DSP and IPU
+ processors that have them (OMAP4/OMAP5 DSPs do not have
+ any RAMs)
+
+ reg-names:
+ description: |
+ Required names for each of the address spaces defined in
+ the 'reg' property. Expects the names from the following
+ list, in the specified order, each representing the corresponding
+ internal RAM memory region.
+ minItems: 1
+ maxItems: 3
+ items:
+ - const: l2ram
+ - const: l1pram
+ - const: l1dram
+
+ ti,bootreg:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ Should be a triple of the phandle to the System Control
+ Configuration region that contains the boot address
+ register, the register offset of the boot address
+ register within the System Control module, and the bit
+ shift within the register. This property is required for
+ all the DSP instances on OMAP4, OMAP5 and DRA7xx SoCs.
+
+ ti,autosuspend-delay-ms:
+ description: |
+ Custom autosuspend delay for the remoteproc in milliseconds.
+
+ ti,timers:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ One or more phandles to OMAP DMTimer nodes, that serve
+ as System/Tick timers for the OS running on the remote
+ processors. This will usually be a single timer if the
+ processor sub-system is running in SMP mode, or one per
+ core in the processor sub-system. This can also be used
+ to reserve specific timers to be dedicated to the
+ remote processors.
+
+ This property is mandatory on remote processors requiring
+ external tick wakeup, and to support Power Management
+ features. The timers to be used should match with the
+ timers used in the firmware image.
+
+ ti,watchdog-timers:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ One or more phandles to OMAP DMTimer nodes, used to
+ serve as Watchdog timers for the processor cores. This
+ will usually be one per executing processor core, even
+ if the processor sub-system is running a SMP OS.
+
+ The timers to be used should match with the watchdog
+ timers used in the firmware image.
+
+if:
+ properties:
+ compatible:
+ enum:
+ - ti,dra7-dsp
+then:
+ properties:
+ reg:
+ minItems: 3
+ maxItems: 3
+ required:
+ - reg
+ - reg-names
+ - ti,bootreg
+
+else:
+ if:
+ properties:
+ compatible:
+ enum:
+ - ti,omap4-ipu
+ - ti,omap5-ipu
+ - ti,dra7-ipu
+ then:
+ properties:
+ reg:
+ minItems: 1
+ maxItems: 1
+ ti,bootreg: false
+ required:
+ - reg
+ - reg-names
+
+ else:
+ properties:
+ reg: false
+ required:
+ - ti,bootreg
+
+required:
+ - compatible
+ - iommus
+ - mboxes
+ - memory-region
+ - clocks
+ - resets
+ - firmware-name
+
+additionalProperties: false
+
+examples:
+ - |
+
+ //Example 1: OMAP4 DSP
+
+ /* DSP Reserved Memory node */
+ #include <dt-bindings/clock/omap4.h>
+ reserved-memory {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ dsp_memory_region: dsp-memory@98000000 {
+ compatible = "shared-dma-pool";
+ reg = <0x98000000 0x800000>;
+ reusable;
+ };
+ };
+
+ /* DSP node */
+ ocp {
+ dsp: dsp {
+ compatible = "ti,omap4-dsp";
+ ti,bootreg = <&scm_conf 0x304 0>;
+ iommus = <&mmu_dsp>;
+ mboxes = <&mailbox &mbox_dsp>;
+ memory-region = <&dsp_memory_region>;
+ ti,timers = <&timer5>;
+ ti,watchdog-timers = <&timer6>;
+ clocks = <&tesla_clkctrl OMAP4_DSP_CLKCTRL 0>;
+ resets = <&prm_tesla 0>, <&prm_tesla 1>;
+ firmware-name = "omap4-dsp-fw.xe64T";
+ };
+ };
+
+ - |+
+
+ //Example 2: OMAP5 IPU
+
+ /* IPU Reserved Memory node */
+ #include <dt-bindings/clock/omap5.h>
+ reserved-memory {
+ #address-cells = <2>;
+ #size-cells = <2>;
+
+ ipu_memory_region: ipu-memory@95800000 {
+ compatible = "shared-dma-pool";
+ reg = <0 0x95800000 0 0x3800000>;
+ reusable;
+ };
+ };
+
+ /* IPU node */
+ ocp {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ ipu: ipu@55020000 {
+ compatible = "ti,omap5-ipu";
+ reg = <0x55020000 0x10000>;
+ reg-names = "l2ram";
+ iommus = <&mmu_ipu>;
+ mboxes = <&mailbox &mbox_ipu>;
+ memory-region = <&ipu_memory_region>;
+ ti,timers = <&timer3>, <&timer4>;
+ ti,watchdog-timers = <&timer9>, <&timer11>;
+ clocks = <&ipu_clkctrl OMAP5_MMU_IPU_CLKCTRL 0>;
+ resets = <&prm_core 2>;
+ firmware-name = "omap5-ipu-fw.xem4";
+ };
+ };
+
+ - |+
+
+ //Example 3: DRA7xx/AM57xx DSP
+
+ /* DSP1 Reserved Memory node */
+ #include <dt-bindings/clock/dra7.h>
+ reserved-memory {
+ #address-cells = <2>;
+ #size-cells = <2>;
+
+ dsp1_memory_region: dsp1-memory@99000000 {
+ compatible = "shared-dma-pool";
+ reg = <0x0 0x99000000 0x0 0x4000000>;
+ reusable;
+ };
+ };
+
+ /* DSP1 node */
+ ocp {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ dsp1: dsp@40800000 {
+ compatible = "ti,dra7-dsp";
+ reg = <0x40800000 0x48000>,
+ <0x40e00000 0x8000>,
+ <0x40f00000 0x8000>;
+ reg-names = "l2ram", "l1pram", "l1dram";
+ ti,bootreg = <&scm_conf 0x55c 0>;
+ iommus = <&mmu0_dsp1>, <&mmu1_dsp1>;
+ mboxes = <&mailbox5 &mbox_dsp1_ipc3x>;
+ memory-region = <&dsp1_memory_region>;
+ ti,timers = <&timer5>;
+ ti,watchdog-timers = <&timer10>;
+ resets = <&prm_dsp1 0>;
+ clocks = <&dsp1_clkctrl DRA7_DSP1_MMU0_DSP1_CLKCTRL 0>;
+ firmware-name = "dra7-dsp1-fw.xe66";
+ };
+ };
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 01/14] dt-bindings: remoteproc: Add OMAP remoteproc bindings
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson-QSEj5FYQhm4dnm+yROfE0A,
ohad-Ix1uc/W3ht7QT0dZR+AlfA,
linux-remoteproc-u79uwXL29TY76Z2rM5mHXA
Cc: linux-kernel-u79uwXL29TY76Z2rM5mHXA,
mathieu.poirier-QSEj5FYQhm4dnm+yROfE0A, s-anna-l0cyMroinI0,
linux-omap-u79uwXL29TY76Z2rM5mHXA, Rob Herring,
devicetree-u79uwXL29TY76Z2rM5mHXA, Tero Kristo
From: Suman Anna <s-anna-l0cyMroinI0@public.gmane.org>
Add the device tree bindings document for the IPU and DSP
remote processor devices on OMAP4+ SoCs.
Cc: Rob Herring <robh-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
Cc: devicetree-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Signed-off-by: Suman Anna <s-anna-l0cyMroinI0@public.gmane.org>
[t-kristo-l0cyMroinI0@public.gmane.org: converted to schema]
Signed-off-by: Tero Kristo <t-kristo-l0cyMroinI0@public.gmane.org>
---
v5:
- enforced ordering of reg-names
- removed ti,bootreg-shift, instead added this under ti,bootreg
property
- renamed ti,autosuspend-delay to ti,autosuspend-delay-ms
- fixed required property handling
- fixed the omap5 ipu example firmware name extension
.../remoteproc/ti,omap-remoteproc.yaml | 322 ++++++++++++++++++
1 file changed, 322 insertions(+)
create mode 100644 Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
diff --git a/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml b/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
new file mode 100644
index 000000000000..b910aac7895e
--- /dev/null
+++ b/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
@@ -0,0 +1,322 @@
+# SPDX-License-Identifier: (GPL-2.0-only or BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/remoteproc/ti,omap-remoteproc.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: OMAP4+ Remoteproc Devices
+
+maintainers:
+ - Suman Anna <s-anna-l0cyMroinI0@public.gmane.org>
+
+description:
+ The OMAP family of SoCs usually have one or more slave processor sub-systems
+ that are used to offload some of the processor-intensive tasks, or to manage
+ other hardware accelerators, for achieving various system level goals.
+
+ The processor cores in the sub-system are usually behind an IOMMU, and may
+ contain additional sub-modules like Internal RAM and/or ROMs, L1 and/or L2
+ caches, an Interrupt Controller, a Cache Controller etc.
+
+ The OMAP SoCs usually have a DSP processor sub-system and/or an IPU processor
+ sub-system. The DSP processor sub-system can contain any of the TI's C64x,
+ C66x or C67x family of DSP cores as the main execution unit. The IPU processor
+ sub-system usually contains either a Dual-Core Cortex-M3 or Dual-Core
+ Cortex-M4 processors.
+
+ Each remote processor sub-system is represented as a single DT node. Each node
+ has a number of required or optional properties that enable the OS running on
+ the host processor (MPU) to perform the device management of the remote
+ processor and to communicate with the remote processor. The various properties
+ can be classified as constant or variable. The constant properties are
+ dictated by the SoC and does not change from one board to another having the
+ same SoC. Examples of constant properties include 'iommus', 'reg'. The
+ variable properties are dictated by the system integration aspects such as
+ memory on the board, or configuration used within the corresponding firmware
+ image. Examples of variable properties include 'mboxes', 'memory-region',
+ 'timers', 'watchdog-timers' etc.
+
+properties:
+ compatible:
+ enum:
+ - ti,omap4-dsp
+ - ti,omap5-dsp
+ - ti,dra7-dsp
+ - ti,omap4-ipu
+ - ti,omap5-ipu
+ - ti,dra7-ipu
+
+ iommus:
+ minItems: 1
+ maxItems: 2
+ description: |
+ phandles to OMAP IOMMU nodes, that need to be programmed
+ for this remote processor to access any external RAM memory or
+ other peripheral device address spaces. This property usually
+ has only a single phandle. Multiple phandles are used only in
+ cases where the sub-system has different ports for different
+ sub-modules within the processor sub-system (eg: DRA7 DSPs),
+ and need the same programming in both the MMUs.
+
+ mboxes:
+ minItems: 1
+ maxItems: 2
+ description: |
+ OMAP Mailbox specifier denoting the sub-mailbox, to be used for
+ communication with the remote processor. The specifier format is
+ as per the bindings,
+ Documentation/devicetree/bindings/mailbox/omap-mailbox.txt
+ This property should match with the sub-mailbox node used in
+ the firmware image.
+
+ clocks:
+ description: |
+ Main functional clock for the remote processor
+
+ resets:
+ description: |
+ Reset handles for the remote processor
+
+ memory-region:
+ $ref: /schemas/types.yaml#/definitions/phandle
+ description: |
+ phandle to the reserved memory node to be associated
+ with the remoteproc device. The reserved memory node
+ can be a CMA memory node, and should be defined as
+ per the bindings,
+ Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
+
+ firmware-name:
+ description: |
+ Default name of the firmware to load to the remote processor.
+
+# Optional properties:
+# --------------------
+# Some of these properties are mandatory on some SoCs, and some are optional
+# depending on the configuration of the firmware image to be executed on the
+# remote processor. The conditions are mentioned for each property.
+#
+# The following are the optional properties:
+
+ reg:
+ description: |
+ Address space for any remoteproc memories present on
+ the SoC. Should contain an entry for each value in
+ 'reg-names'. These are mandatory for all DSP and IPU
+ processors that have them (OMAP4/OMAP5 DSPs do not have
+ any RAMs)
+
+ reg-names:
+ description: |
+ Required names for each of the address spaces defined in
+ the 'reg' property. Expects the names from the following
+ list, in the specified order, each representing the corresponding
+ internal RAM memory region.
+ minItems: 1
+ maxItems: 3
+ items:
+ - const: l2ram
+ - const: l1pram
+ - const: l1dram
+
+ ti,bootreg:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ Should be a triple of the phandle to the System Control
+ Configuration region that contains the boot address
+ register, the register offset of the boot address
+ register within the System Control module, and the bit
+ shift within the register. This property is required for
+ all the DSP instances on OMAP4, OMAP5 and DRA7xx SoCs.
+
+ ti,autosuspend-delay-ms:
+ description: |
+ Custom autosuspend delay for the remoteproc in milliseconds.
+
+ ti,timers:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ One or more phandles to OMAP DMTimer nodes, that serve
+ as System/Tick timers for the OS running on the remote
+ processors. This will usually be a single timer if the
+ processor sub-system is running in SMP mode, or one per
+ core in the processor sub-system. This can also be used
+ to reserve specific timers to be dedicated to the
+ remote processors.
+
+ This property is mandatory on remote processors requiring
+ external tick wakeup, and to support Power Management
+ features. The timers to be used should match with the
+ timers used in the firmware image.
+
+ ti,watchdog-timers:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ One or more phandles to OMAP DMTimer nodes, used to
+ serve as Watchdog timers for the processor cores. This
+ will usually be one per executing processor core, even
+ if the processor sub-system is running a SMP OS.
+
+ The timers to be used should match with the watchdog
+ timers used in the firmware image.
+
+if:
+ properties:
+ compatible:
+ enum:
+ - ti,dra7-dsp
+then:
+ properties:
+ reg:
+ minItems: 3
+ maxItems: 3
+ required:
+ - reg
+ - reg-names
+ - ti,bootreg
+
+else:
+ if:
+ properties:
+ compatible:
+ enum:
+ - ti,omap4-ipu
+ - ti,omap5-ipu
+ - ti,dra7-ipu
+ then:
+ properties:
+ reg:
+ minItems: 1
+ maxItems: 1
+ ti,bootreg: false
+ required:
+ - reg
+ - reg-names
+
+ else:
+ properties:
+ reg: false
+ required:
+ - ti,bootreg
+
+required:
+ - compatible
+ - iommus
+ - mboxes
+ - memory-region
+ - clocks
+ - resets
+ - firmware-name
+
+additionalProperties: false
+
+examples:
+ - |
+
+ //Example 1: OMAP4 DSP
+
+ /* DSP Reserved Memory node */
+ #include <dt-bindings/clock/omap4.h>
+ reserved-memory {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ dsp_memory_region: dsp-memory@98000000 {
+ compatible = "shared-dma-pool";
+ reg = <0x98000000 0x800000>;
+ reusable;
+ };
+ };
+
+ /* DSP node */
+ ocp {
+ dsp: dsp {
+ compatible = "ti,omap4-dsp";
+ ti,bootreg = <&scm_conf 0x304 0>;
+ iommus = <&mmu_dsp>;
+ mboxes = <&mailbox &mbox_dsp>;
+ memory-region = <&dsp_memory_region>;
+ ti,timers = <&timer5>;
+ ti,watchdog-timers = <&timer6>;
+ clocks = <&tesla_clkctrl OMAP4_DSP_CLKCTRL 0>;
+ resets = <&prm_tesla 0>, <&prm_tesla 1>;
+ firmware-name = "omap4-dsp-fw.xe64T";
+ };
+ };
+
+ - |+
+
+ //Example 2: OMAP5 IPU
+
+ /* IPU Reserved Memory node */
+ #include <dt-bindings/clock/omap5.h>
+ reserved-memory {
+ #address-cells = <2>;
+ #size-cells = <2>;
+
+ ipu_memory_region: ipu-memory@95800000 {
+ compatible = "shared-dma-pool";
+ reg = <0 0x95800000 0 0x3800000>;
+ reusable;
+ };
+ };
+
+ /* IPU node */
+ ocp {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ ipu: ipu@55020000 {
+ compatible = "ti,omap5-ipu";
+ reg = <0x55020000 0x10000>;
+ reg-names = "l2ram";
+ iommus = <&mmu_ipu>;
+ mboxes = <&mailbox &mbox_ipu>;
+ memory-region = <&ipu_memory_region>;
+ ti,timers = <&timer3>, <&timer4>;
+ ti,watchdog-timers = <&timer9>, <&timer11>;
+ clocks = <&ipu_clkctrl OMAP5_MMU_IPU_CLKCTRL 0>;
+ resets = <&prm_core 2>;
+ firmware-name = "omap5-ipu-fw.xem4";
+ };
+ };
+
+ - |+
+
+ //Example 3: DRA7xx/AM57xx DSP
+
+ /* DSP1 Reserved Memory node */
+ #include <dt-bindings/clock/dra7.h>
+ reserved-memory {
+ #address-cells = <2>;
+ #size-cells = <2>;
+
+ dsp1_memory_region: dsp1-memory@99000000 {
+ compatible = "shared-dma-pool";
+ reg = <0x0 0x99000000 0x0 0x4000000>;
+ reusable;
+ };
+ };
+
+ /* DSP1 node */
+ ocp {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ dsp1: dsp@40800000 {
+ compatible = "ti,dra7-dsp";
+ reg = <0x40800000 0x48000>,
+ <0x40e00000 0x8000>,
+ <0x40f00000 0x8000>;
+ reg-names = "l2ram", "l1pram", "l1dram";
+ ti,bootreg = <&scm_conf 0x55c 0>;
+ iommus = <&mmu0_dsp1>, <&mmu1_dsp1>;
+ mboxes = <&mailbox5 &mbox_dsp1_ipc3x>;
+ memory-region = <&dsp1_memory_region>;
+ ti,timers = <&timer5>;
+ ti,watchdog-timers = <&timer10>;
+ resets = <&prm_dsp1 0>;
+ clocks = <&dsp1_clkctrl DRA7_DSP1_MMU0_DSP1_CLKCTRL 0>;
+ firmware-name = "dra7-dsp1-fw.xe66";
+ };
+ };
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 01/14] dt-bindings: remoteproc: Add OMAP remoteproc bindings
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Rob Herring,
devicetree, Tero Kristo
From: Suman Anna <s-anna@ti.com>
Add the device tree bindings document for the IPU and DSP
remote processor devices on OMAP4+ SoCs.
Cc: Rob Herring <robh@kernel.org>
Cc: devicetree@vger.kernel.org
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to schema]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- enforced ordering of reg-names
- removed ti,bootreg-shift, instead added this under ti,bootreg
property
- renamed ti,autosuspend-delay to ti,autosuspend-delay-ms
- fixed required property handling
- fixed the omap5 ipu example firmware name extension
.../remoteproc/ti,omap-remoteproc.yaml | 322 ++++++++++++++++++
1 file changed, 322 insertions(+)
create mode 100644 Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
diff --git a/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml b/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
new file mode 100644
index 000000000000..b910aac7895e
--- /dev/null
+++ b/Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
@@ -0,0 +1,322 @@
+# SPDX-License-Identifier: (GPL-2.0-only or BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/remoteproc/ti,omap-remoteproc.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: OMAP4+ Remoteproc Devices
+
+maintainers:
+ - Suman Anna <s-anna@ti.com>
+
+description:
+ The OMAP family of SoCs usually have one or more slave processor sub-systems
+ that are used to offload some of the processor-intensive tasks, or to manage
+ other hardware accelerators, for achieving various system level goals.
+
+ The processor cores in the sub-system are usually behind an IOMMU, and may
+ contain additional sub-modules like Internal RAM and/or ROMs, L1 and/or L2
+ caches, an Interrupt Controller, a Cache Controller etc.
+
+ The OMAP SoCs usually have a DSP processor sub-system and/or an IPU processor
+ sub-system. The DSP processor sub-system can contain any of the TI's C64x,
+ C66x or C67x family of DSP cores as the main execution unit. The IPU processor
+ sub-system usually contains either a Dual-Core Cortex-M3 or Dual-Core
+ Cortex-M4 processors.
+
+ Each remote processor sub-system is represented as a single DT node. Each node
+ has a number of required or optional properties that enable the OS running on
+ the host processor (MPU) to perform the device management of the remote
+ processor and to communicate with the remote processor. The various properties
+ can be classified as constant or variable. The constant properties are
+ dictated by the SoC and does not change from one board to another having the
+ same SoC. Examples of constant properties include 'iommus', 'reg'. The
+ variable properties are dictated by the system integration aspects such as
+ memory on the board, or configuration used within the corresponding firmware
+ image. Examples of variable properties include 'mboxes', 'memory-region',
+ 'timers', 'watchdog-timers' etc.
+
+properties:
+ compatible:
+ enum:
+ - ti,omap4-dsp
+ - ti,omap5-dsp
+ - ti,dra7-dsp
+ - ti,omap4-ipu
+ - ti,omap5-ipu
+ - ti,dra7-ipu
+
+ iommus:
+ minItems: 1
+ maxItems: 2
+ description: |
+ phandles to OMAP IOMMU nodes, that need to be programmed
+ for this remote processor to access any external RAM memory or
+ other peripheral device address spaces. This property usually
+ has only a single phandle. Multiple phandles are used only in
+ cases where the sub-system has different ports for different
+ sub-modules within the processor sub-system (eg: DRA7 DSPs),
+ and need the same programming in both the MMUs.
+
+ mboxes:
+ minItems: 1
+ maxItems: 2
+ description: |
+ OMAP Mailbox specifier denoting the sub-mailbox, to be used for
+ communication with the remote processor. The specifier format is
+ as per the bindings,
+ Documentation/devicetree/bindings/mailbox/omap-mailbox.txt
+ This property should match with the sub-mailbox node used in
+ the firmware image.
+
+ clocks:
+ description: |
+ Main functional clock for the remote processor
+
+ resets:
+ description: |
+ Reset handles for the remote processor
+
+ memory-region:
+ $ref: /schemas/types.yaml#/definitions/phandle
+ description: |
+ phandle to the reserved memory node to be associated
+ with the remoteproc device. The reserved memory node
+ can be a CMA memory node, and should be defined as
+ per the bindings,
+ Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
+
+ firmware-name:
+ description: |
+ Default name of the firmware to load to the remote processor.
+
+# Optional properties:
+# --------------------
+# Some of these properties are mandatory on some SoCs, and some are optional
+# depending on the configuration of the firmware image to be executed on the
+# remote processor. The conditions are mentioned for each property.
+#
+# The following are the optional properties:
+
+ reg:
+ description: |
+ Address space for any remoteproc memories present on
+ the SoC. Should contain an entry for each value in
+ 'reg-names'. These are mandatory for all DSP and IPU
+ processors that have them (OMAP4/OMAP5 DSPs do not have
+ any RAMs)
+
+ reg-names:
+ description: |
+ Required names for each of the address spaces defined in
+ the 'reg' property. Expects the names from the following
+ list, in the specified order, each representing the corresponding
+ internal RAM memory region.
+ minItems: 1
+ maxItems: 3
+ items:
+ - const: l2ram
+ - const: l1pram
+ - const: l1dram
+
+ ti,bootreg:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ Should be a triple of the phandle to the System Control
+ Configuration region that contains the boot address
+ register, the register offset of the boot address
+ register within the System Control module, and the bit
+ shift within the register. This property is required for
+ all the DSP instances on OMAP4, OMAP5 and DRA7xx SoCs.
+
+ ti,autosuspend-delay-ms:
+ description: |
+ Custom autosuspend delay for the remoteproc in milliseconds.
+
+ ti,timers:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ One or more phandles to OMAP DMTimer nodes, that serve
+ as System/Tick timers for the OS running on the remote
+ processors. This will usually be a single timer if the
+ processor sub-system is running in SMP mode, or one per
+ core in the processor sub-system. This can also be used
+ to reserve specific timers to be dedicated to the
+ remote processors.
+
+ This property is mandatory on remote processors requiring
+ external tick wakeup, and to support Power Management
+ features. The timers to be used should match with the
+ timers used in the firmware image.
+
+ ti,watchdog-timers:
+ $ref: /schemas/types.yaml#/definitions/phandle-array
+ description: |
+ One or more phandles to OMAP DMTimer nodes, used to
+ serve as Watchdog timers for the processor cores. This
+ will usually be one per executing processor core, even
+ if the processor sub-system is running a SMP OS.
+
+ The timers to be used should match with the watchdog
+ timers used in the firmware image.
+
+if:
+ properties:
+ compatible:
+ enum:
+ - ti,dra7-dsp
+then:
+ properties:
+ reg:
+ minItems: 3
+ maxItems: 3
+ required:
+ - reg
+ - reg-names
+ - ti,bootreg
+
+else:
+ if:
+ properties:
+ compatible:
+ enum:
+ - ti,omap4-ipu
+ - ti,omap5-ipu
+ - ti,dra7-ipu
+ then:
+ properties:
+ reg:
+ minItems: 1
+ maxItems: 1
+ ti,bootreg: false
+ required:
+ - reg
+ - reg-names
+
+ else:
+ properties:
+ reg: false
+ required:
+ - ti,bootreg
+
+required:
+ - compatible
+ - iommus
+ - mboxes
+ - memory-region
+ - clocks
+ - resets
+ - firmware-name
+
+additionalProperties: false
+
+examples:
+ - |
+
+ //Example 1: OMAP4 DSP
+
+ /* DSP Reserved Memory node */
+ #include <dt-bindings/clock/omap4.h>
+ reserved-memory {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ dsp_memory_region: dsp-memory@98000000 {
+ compatible = "shared-dma-pool";
+ reg = <0x98000000 0x800000>;
+ reusable;
+ };
+ };
+
+ /* DSP node */
+ ocp {
+ dsp: dsp {
+ compatible = "ti,omap4-dsp";
+ ti,bootreg = <&scm_conf 0x304 0>;
+ iommus = <&mmu_dsp>;
+ mboxes = <&mailbox &mbox_dsp>;
+ memory-region = <&dsp_memory_region>;
+ ti,timers = <&timer5>;
+ ti,watchdog-timers = <&timer6>;
+ clocks = <&tesla_clkctrl OMAP4_DSP_CLKCTRL 0>;
+ resets = <&prm_tesla 0>, <&prm_tesla 1>;
+ firmware-name = "omap4-dsp-fw.xe64T";
+ };
+ };
+
+ - |+
+
+ //Example 2: OMAP5 IPU
+
+ /* IPU Reserved Memory node */
+ #include <dt-bindings/clock/omap5.h>
+ reserved-memory {
+ #address-cells = <2>;
+ #size-cells = <2>;
+
+ ipu_memory_region: ipu-memory@95800000 {
+ compatible = "shared-dma-pool";
+ reg = <0 0x95800000 0 0x3800000>;
+ reusable;
+ };
+ };
+
+ /* IPU node */
+ ocp {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ ipu: ipu@55020000 {
+ compatible = "ti,omap5-ipu";
+ reg = <0x55020000 0x10000>;
+ reg-names = "l2ram";
+ iommus = <&mmu_ipu>;
+ mboxes = <&mailbox &mbox_ipu>;
+ memory-region = <&ipu_memory_region>;
+ ti,timers = <&timer3>, <&timer4>;
+ ti,watchdog-timers = <&timer9>, <&timer11>;
+ clocks = <&ipu_clkctrl OMAP5_MMU_IPU_CLKCTRL 0>;
+ resets = <&prm_core 2>;
+ firmware-name = "omap5-ipu-fw.xem4";
+ };
+ };
+
+ - |+
+
+ //Example 3: DRA7xx/AM57xx DSP
+
+ /* DSP1 Reserved Memory node */
+ #include <dt-bindings/clock/dra7.h>
+ reserved-memory {
+ #address-cells = <2>;
+ #size-cells = <2>;
+
+ dsp1_memory_region: dsp1-memory@99000000 {
+ compatible = "shared-dma-pool";
+ reg = <0x0 0x99000000 0x0 0x4000000>;
+ reusable;
+ };
+ };
+
+ /* DSP1 node */
+ ocp {
+ #address-cells = <1>;
+ #size-cells = <1>;
+
+ dsp1: dsp@40800000 {
+ compatible = "ti,dra7-dsp";
+ reg = <0x40800000 0x48000>,
+ <0x40e00000 0x8000>,
+ <0x40f00000 0x8000>;
+ reg-names = "l2ram", "l1pram", "l1dram";
+ ti,bootreg = <&scm_conf 0x55c 0>;
+ iommus = <&mmu0_dsp1>, <&mmu1_dsp1>;
+ mboxes = <&mailbox5 &mbox_dsp1_ipc3x>;
+ memory-region = <&dsp1_memory_region>;
+ ti,timers = <&timer5>;
+ ti,watchdog-timers = <&timer10>;
+ resets = <&prm_dsp1 0>;
+ clocks = <&dsp1_clkctrl DRA7_DSP1_MMU0_DSP1_CLKCTRL 0>;
+ firmware-name = "dra7-dsp1-fw.xe66";
+ };
+ };
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* Re: [PATCHv5 01/14] dt-bindings: remoteproc: Add OMAP remoteproc bindings
2020-01-16 13:53 ` Tero Kristo
(?)
@ 2020-01-22 2:34 ` Rob Herring
-1 siblings, 0 replies; 62+ messages in thread
From: Rob Herring @ 2020-01-22 2:34 UTC (permalink / raw)
To: Tero Kristo
Cc: bjorn.andersson, ohad, linux-remoteproc, linux-kernel,
mathieu.poirier, s-anna, linux-omap, devicetree
On Thu, 16 Jan 2020 15:53:19 +0200, Tero Kristo wrote:
> From: Suman Anna <s-anna@ti.com>
>
> Add the device tree bindings document for the IPU and DSP
> remote processor devices on OMAP4+ SoCs.
>
> Cc: Rob Herring <robh@kernel.org>
> Cc: devicetree@vger.kernel.org
> Signed-off-by: Suman Anna <s-anna@ti.com>
> [t-kristo@ti.com: converted to schema]
> Signed-off-by: Tero Kristo <t-kristo@ti.com>
> ---
> v5:
> - enforced ordering of reg-names
> - removed ti,bootreg-shift, instead added this under ti,bootreg
> property
> - renamed ti,autosuspend-delay to ti,autosuspend-delay-ms
> - fixed required property handling
> - fixed the omap5 ipu example firmware name extension
>
> .../remoteproc/ti,omap-remoteproc.yaml | 322 ++++++++++++++++++
> 1 file changed, 322 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
>
Reviewed-by: Rob Herring <robh@kernel.org>
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCHv5 01/14] dt-bindings: remoteproc: Add OMAP remoteproc bindings
@ 2020-01-22 2:34 ` Rob Herring
0 siblings, 0 replies; 62+ messages in thread
From: Rob Herring @ 2020-01-22 2:34 UTC (permalink / raw)
Cc: bjorn.andersson-QSEj5FYQhm4dnm+yROfE0A,
ohad-Ix1uc/W3ht7QT0dZR+AlfA,
linux-remoteproc-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
mathieu.poirier-QSEj5FYQhm4dnm+yROfE0A, s-anna-l0cyMroinI0,
linux-omap-u79uwXL29TY76Z2rM5mHXA,
devicetree-u79uwXL29TY76Z2rM5mHXA, Tero Kristo
On Thu, 16 Jan 2020 15:53:19 +0200, Tero Kristo wrote:
> From: Suman Anna <s-anna-l0cyMroinI0@public.gmane.org>
>
> Add the device tree bindings document for the IPU and DSP
> remote processor devices on OMAP4+ SoCs.
>
> Cc: Rob Herring <robh-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
> Cc: devicetree-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> Signed-off-by: Suman Anna <s-anna-l0cyMroinI0@public.gmane.org>
> [t-kristo-l0cyMroinI0@public.gmane.org: converted to schema]
> Signed-off-by: Tero Kristo <t-kristo-l0cyMroinI0@public.gmane.org>
> ---
> v5:
> - enforced ordering of reg-names
> - removed ti,bootreg-shift, instead added this under ti,bootreg
> property
> - renamed ti,autosuspend-delay to ti,autosuspend-delay-ms
> - fixed required property handling
> - fixed the omap5 ipu example firmware name extension
>
> .../remoteproc/ti,omap-remoteproc.yaml | 322 ++++++++++++++++++
> 1 file changed, 322 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
>
Reviewed-by: Rob Herring <robh-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
^ permalink raw reply [flat|nested] 62+ messages in thread
* Re: [PATCHv5 01/14] dt-bindings: remoteproc: Add OMAP remoteproc bindings
@ 2020-01-22 2:34 ` Rob Herring
0 siblings, 0 replies; 62+ messages in thread
From: Rob Herring @ 2020-01-22 2:34 UTC (permalink / raw)
To: Tero Kristo
Cc: bjorn.andersson, ohad, linux-remoteproc, linux-kernel,
mathieu.poirier, s-anna, linux-omap, devicetree, Tero Kristo
On Thu, 16 Jan 2020 15:53:19 +0200, Tero Kristo wrote:
> From: Suman Anna <s-anna@ti.com>
>
> Add the device tree bindings document for the IPU and DSP
> remote processor devices on OMAP4+ SoCs.
>
> Cc: Rob Herring <robh@kernel.org>
> Cc: devicetree@vger.kernel.org
> Signed-off-by: Suman Anna <s-anna@ti.com>
> [t-kristo@ti.com: converted to schema]
> Signed-off-by: Tero Kristo <t-kristo@ti.com>
> ---
> v5:
> - enforced ordering of reg-names
> - removed ti,bootreg-shift, instead added this under ti,bootreg
> property
> - renamed ti,autosuspend-delay to ti,autosuspend-delay-ms
> - fixed required property handling
> - fixed the omap5 ipu example firmware name extension
>
> .../remoteproc/ti,omap-remoteproc.yaml | 322 ++++++++++++++++++
> 1 file changed, 322 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/remoteproc/ti,omap-remoteproc.yaml
>
Reviewed-by: Rob Herring <robh@kernel.org>
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCHv5 02/14] remoteproc/omap: Add device tree support
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tony Lindgren,
Tero Kristo
From: Suman Anna <s-anna@ti.com>
OMAP4+ SoCs support device tree boot only. The OMAP remoteproc
driver is enhanced to support remoteproc devices created through
Device Tree, support for legacy platform devices has been
deprecated. The current DT support handles the IPU and DSP
processor subsystems on OMAP4 and OMAP5 SoCs.
The OMAP remoteproc driver relies on the ti-sysc, reset, and
syscon layers for performing clock, reset and boot vector
management (DSP remoteprocs only) of the devices, but some of
these are limited only to the machine-specific layers
in arch/arm. The dependency against control module API for boot
vector management of the DSP remoteprocs has now been removed
with added logic to parse the boot register from the DT node
and program it appropriately directly within the driver.
The OMAP remoteproc driver expects the firmware names to be
provided via device tree entries (firmware-name.) These are used
to load the proper firmware during boot of the remote processor.
Cc: Tony Lindgren <tony@atomide.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to use ti-sysc framework]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 177 ++++++++++++++++++++++++---
1 file changed, 160 insertions(+), 17 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 6398194075aa..6c603c2cd895 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -2,7 +2,7 @@
/*
* OMAP Remote Processor driver
*
- * Copyright (C) 2011 Texas Instruments, Inc.
+ * Copyright (C) 2011-2020 Texas Instruments Incorporated - http://www.ti.com/
* Copyright (C) 2011 Google, Inc.
*
* Ohad Ben-Cohen <ohad@wizery.com>
@@ -16,27 +16,51 @@
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/err.h>
+#include <linux/of_device.h>
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
#include <linux/omap-mailbox.h>
-
-#include <linux/platform_data/remoteproc-omap.h>
+#include <linux/regmap.h>
+#include <linux/mfd/syscon.h>
+#include <linux/reset.h>
#include "omap_remoteproc.h"
#include "remoteproc_internal.h"
+/**
+ * struct omap_rproc_boot_data - boot data structure for the DSP omap rprocs
+ * @syscon: regmap handle for the system control configuration module
+ * @boot_reg: boot register offset within the @syscon regmap
+ */
+struct omap_rproc_boot_data {
+ struct regmap *syscon;
+ unsigned int boot_reg;
+};
+
/**
* struct omap_rproc - omap remote processor state
* @mbox: mailbox channel handle
* @client: mailbox client to request the mailbox channel
+ * @boot_data: boot data structure for setting processor boot address
* @rproc: rproc handle
+ * @reset: reset handle
*/
struct omap_rproc {
struct mbox_chan *mbox;
struct mbox_client client;
+ struct omap_rproc_boot_data *boot_data;
struct rproc *rproc;
+ struct reset_control *reset;
+};
+
+/**
+ * struct omap_rproc_dev_data - device data for the omap remote processor
+ * @device_name: device name of the remote processor
+ */
+struct omap_rproc_dev_data {
+ const char *device_name;
};
/**
@@ -92,6 +116,21 @@ static void omap_rproc_kick(struct rproc *rproc, int vqid)
ret);
}
+/**
+ * omap_rproc_write_dsp_boot_addr - set boot address for a DSP remote processor
+ * @rproc: handle of a remote processor
+ *
+ * Set boot address for a supported DSP remote processor.
+ */
+static void omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
+{
+ struct omap_rproc *oproc = rproc->priv;
+ struct omap_rproc_boot_data *bdata = oproc->boot_data;
+ u32 offset = bdata->boot_reg;
+
+ regmap_write(bdata->syscon, offset, rproc->bootaddr);
+}
+
/*
* Power up the remote processor.
*
@@ -103,13 +142,11 @@ static int omap_rproc_start(struct rproc *rproc)
{
struct omap_rproc *oproc = rproc->priv;
struct device *dev = rproc->dev.parent;
- struct platform_device *pdev = to_platform_device(dev);
- struct omap_rproc_pdata *pdata = pdev->dev.platform_data;
int ret;
struct mbox_client *client = &oproc->client;
- if (pdata->set_bootaddr)
- pdata->set_bootaddr(rproc->bootaddr);
+ if (oproc->boot_data)
+ omap_rproc_write_dsp_boot_addr(rproc);
client->dev = dev;
client->tx_done = NULL;
@@ -117,7 +154,7 @@ static int omap_rproc_start(struct rproc *rproc)
client->tx_block = false;
client->knows_txdone = false;
- oproc->mbox = omap_mbox_request_channel(client, pdata->mbox_name);
+ oproc->mbox = mbox_request_channel(client, 0);
if (IS_ERR(oproc->mbox)) {
ret = -EBUSY;
dev_err(dev, "mbox_request_channel failed: %ld\n",
@@ -138,9 +175,9 @@ static int omap_rproc_start(struct rproc *rproc)
goto put_mbox;
}
- ret = pdata->device_enable(pdev);
+ ret = reset_control_deassert(oproc->reset);
if (ret) {
- dev_err(dev, "omap_device_enable failed: %d\n", ret);
+ dev_err(dev, "reset control deassert failed: %d\n", ret);
goto put_mbox;
}
@@ -154,13 +191,10 @@ static int omap_rproc_start(struct rproc *rproc)
/* power off the remote processor */
static int omap_rproc_stop(struct rproc *rproc)
{
- struct device *dev = rproc->dev.parent;
- struct platform_device *pdev = to_platform_device(dev);
- struct omap_rproc_pdata *pdata = pdev->dev.platform_data;
struct omap_rproc *oproc = rproc->priv;
int ret;
- ret = pdata->device_shutdown(pdev);
+ ret = reset_control_assert(oproc->reset);
if (ret)
return ret;
@@ -175,12 +209,115 @@ static const struct rproc_ops omap_rproc_ops = {
.kick = omap_rproc_kick,
};
+static const struct omap_rproc_dev_data omap4_dsp_dev_data = {
+ .device_name = "dsp",
+};
+
+static const struct omap_rproc_dev_data omap4_ipu_dev_data = {
+ .device_name = "ipu",
+};
+
+static const struct omap_rproc_dev_data omap5_dsp_dev_data = {
+ .device_name = "dsp",
+};
+
+static const struct omap_rproc_dev_data omap5_ipu_dev_data = {
+ .device_name = "ipu",
+};
+
+static const struct of_device_id omap_rproc_of_match[] = {
+ {
+ .compatible = "ti,omap4-dsp",
+ .data = &omap4_dsp_dev_data,
+ },
+ {
+ .compatible = "ti,omap4-ipu",
+ .data = &omap4_ipu_dev_data,
+ },
+ {
+ .compatible = "ti,omap5-dsp",
+ .data = &omap5_dsp_dev_data,
+ },
+ {
+ .compatible = "ti,omap5-ipu",
+ .data = &omap5_ipu_dev_data,
+ },
+ {
+ /* end */
+ },
+};
+MODULE_DEVICE_TABLE(of, omap_rproc_of_match);
+
+static const char *omap_rproc_get_firmware(struct platform_device *pdev)
+{
+ const char *fw_name;
+ int ret;
+
+ ret = of_property_read_string(pdev->dev.of_node, "firmware-name",
+ &fw_name);
+ if (ret)
+ return ERR_PTR(ret);
+
+ return fw_name;
+}
+
+static int omap_rproc_get_boot_data(struct platform_device *pdev,
+ struct rproc *rproc)
+{
+ struct device_node *np = pdev->dev.of_node;
+ struct omap_rproc *oproc = rproc->priv;
+ const struct omap_rproc_dev_data *data;
+ int ret;
+
+ data = of_device_get_match_data(&pdev->dev);
+ if (!data)
+ return -ENODEV;
+
+ if (!of_property_read_bool(np, "ti,bootreg"))
+ return 0;
+
+ oproc->boot_data = devm_kzalloc(&pdev->dev, sizeof(*oproc->boot_data),
+ GFP_KERNEL);
+ if (!oproc->boot_data)
+ return -ENOMEM;
+
+ oproc->boot_data->syscon =
+ syscon_regmap_lookup_by_phandle(np, "ti,bootreg");
+ if (IS_ERR(oproc->boot_data->syscon)) {
+ ret = PTR_ERR(oproc->boot_data->syscon);
+ return ret;
+ }
+
+ if (of_property_read_u32_index(np, "ti,bootreg", 1,
+ &oproc->boot_data->boot_reg)) {
+ dev_err(&pdev->dev, "couldn't get the boot register\n");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
static int omap_rproc_probe(struct platform_device *pdev)
{
- struct omap_rproc_pdata *pdata = pdev->dev.platform_data;
+ struct device_node *np = pdev->dev.of_node;
struct omap_rproc *oproc;
struct rproc *rproc;
+ const char *firmware;
int ret;
+ struct reset_control *reset;
+
+ if (!np) {
+ dev_err(&pdev->dev, "only DT-based devices are supported\n");
+ return -ENODEV;
+ }
+
+ reset = devm_reset_control_array_get_exclusive(&pdev->dev);
+ if (IS_ERR(reset))
+ return PTR_ERR(reset);
+
+ firmware = omap_rproc_get_firmware(pdev);
+ if (IS_ERR(firmware))
+ return PTR_ERR(firmware);
ret = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
if (ret) {
@@ -188,16 +325,21 @@ static int omap_rproc_probe(struct platform_device *pdev)
return ret;
}
- rproc = rproc_alloc(&pdev->dev, pdata->name, &omap_rproc_ops,
- pdata->firmware, sizeof(*oproc));
+ rproc = rproc_alloc(&pdev->dev, dev_name(&pdev->dev), &omap_rproc_ops,
+ firmware, sizeof(*oproc));
if (!rproc)
return -ENOMEM;
oproc = rproc->priv;
oproc->rproc = rproc;
+ oproc->reset = reset;
/* All existing OMAP IPU and DSP processors have an MMU */
rproc->has_iommu = true;
+ ret = omap_rproc_get_boot_data(pdev, rproc);
+ if (ret)
+ goto free_rproc;
+
platform_set_drvdata(pdev, rproc);
ret = rproc_add(rproc);
@@ -226,6 +368,7 @@ static struct platform_driver omap_rproc_driver = {
.remove = omap_rproc_remove,
.driver = {
.name = "omap-rproc",
+ .of_match_table = omap_rproc_of_match,
},
};
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 02/14] remoteproc/omap: Add device tree support
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tony Lindgren,
Tero Kristo
From: Suman Anna <s-anna@ti.com>
OMAP4+ SoCs support device tree boot only. The OMAP remoteproc
driver is enhanced to support remoteproc devices created through
Device Tree, support for legacy platform devices has been
deprecated. The current DT support handles the IPU and DSP
processor subsystems on OMAP4 and OMAP5 SoCs.
The OMAP remoteproc driver relies on the ti-sysc, reset, and
syscon layers for performing clock, reset and boot vector
management (DSP remoteprocs only) of the devices, but some of
these are limited only to the machine-specific layers
in arch/arm. The dependency against control module API for boot
vector management of the DSP remoteprocs has now been removed
with added logic to parse the boot register from the DT node
and program it appropriately directly within the driver.
The OMAP remoteproc driver expects the firmware names to be
provided via device tree entries (firmware-name.) These are used
to load the proper firmware during boot of the remote processor.
Cc: Tony Lindgren <tony@atomide.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to use ti-sysc framework]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 177 ++++++++++++++++++++++++---
1 file changed, 160 insertions(+), 17 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 6398194075aa..6c603c2cd895 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -2,7 +2,7 @@
/*
* OMAP Remote Processor driver
*
- * Copyright (C) 2011 Texas Instruments, Inc.
+ * Copyright (C) 2011-2020 Texas Instruments Incorporated - http://www.ti.com/
* Copyright (C) 2011 Google, Inc.
*
* Ohad Ben-Cohen <ohad@wizery.com>
@@ -16,27 +16,51 @@
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/err.h>
+#include <linux/of_device.h>
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
#include <linux/omap-mailbox.h>
-
-#include <linux/platform_data/remoteproc-omap.h>
+#include <linux/regmap.h>
+#include <linux/mfd/syscon.h>
+#include <linux/reset.h>
#include "omap_remoteproc.h"
#include "remoteproc_internal.h"
+/**
+ * struct omap_rproc_boot_data - boot data structure for the DSP omap rprocs
+ * @syscon: regmap handle for the system control configuration module
+ * @boot_reg: boot register offset within the @syscon regmap
+ */
+struct omap_rproc_boot_data {
+ struct regmap *syscon;
+ unsigned int boot_reg;
+};
+
/**
* struct omap_rproc - omap remote processor state
* @mbox: mailbox channel handle
* @client: mailbox client to request the mailbox channel
+ * @boot_data: boot data structure for setting processor boot address
* @rproc: rproc handle
+ * @reset: reset handle
*/
struct omap_rproc {
struct mbox_chan *mbox;
struct mbox_client client;
+ struct omap_rproc_boot_data *boot_data;
struct rproc *rproc;
+ struct reset_control *reset;
+};
+
+/**
+ * struct omap_rproc_dev_data - device data for the omap remote processor
+ * @device_name: device name of the remote processor
+ */
+struct omap_rproc_dev_data {
+ const char *device_name;
};
/**
@@ -92,6 +116,21 @@ static void omap_rproc_kick(struct rproc *rproc, int vqid)
ret);
}
+/**
+ * omap_rproc_write_dsp_boot_addr - set boot address for a DSP remote processor
+ * @rproc: handle of a remote processor
+ *
+ * Set boot address for a supported DSP remote processor.
+ */
+static void omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
+{
+ struct omap_rproc *oproc = rproc->priv;
+ struct omap_rproc_boot_data *bdata = oproc->boot_data;
+ u32 offset = bdata->boot_reg;
+
+ regmap_write(bdata->syscon, offset, rproc->bootaddr);
+}
+
/*
* Power up the remote processor.
*
@@ -103,13 +142,11 @@ static int omap_rproc_start(struct rproc *rproc)
{
struct omap_rproc *oproc = rproc->priv;
struct device *dev = rproc->dev.parent;
- struct platform_device *pdev = to_platform_device(dev);
- struct omap_rproc_pdata *pdata = pdev->dev.platform_data;
int ret;
struct mbox_client *client = &oproc->client;
- if (pdata->set_bootaddr)
- pdata->set_bootaddr(rproc->bootaddr);
+ if (oproc->boot_data)
+ omap_rproc_write_dsp_boot_addr(rproc);
client->dev = dev;
client->tx_done = NULL;
@@ -117,7 +154,7 @@ static int omap_rproc_start(struct rproc *rproc)
client->tx_block = false;
client->knows_txdone = false;
- oproc->mbox = omap_mbox_request_channel(client, pdata->mbox_name);
+ oproc->mbox = mbox_request_channel(client, 0);
if (IS_ERR(oproc->mbox)) {
ret = -EBUSY;
dev_err(dev, "mbox_request_channel failed: %ld\n",
@@ -138,9 +175,9 @@ static int omap_rproc_start(struct rproc *rproc)
goto put_mbox;
}
- ret = pdata->device_enable(pdev);
+ ret = reset_control_deassert(oproc->reset);
if (ret) {
- dev_err(dev, "omap_device_enable failed: %d\n", ret);
+ dev_err(dev, "reset control deassert failed: %d\n", ret);
goto put_mbox;
}
@@ -154,13 +191,10 @@ static int omap_rproc_start(struct rproc *rproc)
/* power off the remote processor */
static int omap_rproc_stop(struct rproc *rproc)
{
- struct device *dev = rproc->dev.parent;
- struct platform_device *pdev = to_platform_device(dev);
- struct omap_rproc_pdata *pdata = pdev->dev.platform_data;
struct omap_rproc *oproc = rproc->priv;
int ret;
- ret = pdata->device_shutdown(pdev);
+ ret = reset_control_assert(oproc->reset);
if (ret)
return ret;
@@ -175,12 +209,115 @@ static const struct rproc_ops omap_rproc_ops = {
.kick = omap_rproc_kick,
};
+static const struct omap_rproc_dev_data omap4_dsp_dev_data = {
+ .device_name = "dsp",
+};
+
+static const struct omap_rproc_dev_data omap4_ipu_dev_data = {
+ .device_name = "ipu",
+};
+
+static const struct omap_rproc_dev_data omap5_dsp_dev_data = {
+ .device_name = "dsp",
+};
+
+static const struct omap_rproc_dev_data omap5_ipu_dev_data = {
+ .device_name = "ipu",
+};
+
+static const struct of_device_id omap_rproc_of_match[] = {
+ {
+ .compatible = "ti,omap4-dsp",
+ .data = &omap4_dsp_dev_data,
+ },
+ {
+ .compatible = "ti,omap4-ipu",
+ .data = &omap4_ipu_dev_data,
+ },
+ {
+ .compatible = "ti,omap5-dsp",
+ .data = &omap5_dsp_dev_data,
+ },
+ {
+ .compatible = "ti,omap5-ipu",
+ .data = &omap5_ipu_dev_data,
+ },
+ {
+ /* end */
+ },
+};
+MODULE_DEVICE_TABLE(of, omap_rproc_of_match);
+
+static const char *omap_rproc_get_firmware(struct platform_device *pdev)
+{
+ const char *fw_name;
+ int ret;
+
+ ret = of_property_read_string(pdev->dev.of_node, "firmware-name",
+ &fw_name);
+ if (ret)
+ return ERR_PTR(ret);
+
+ return fw_name;
+}
+
+static int omap_rproc_get_boot_data(struct platform_device *pdev,
+ struct rproc *rproc)
+{
+ struct device_node *np = pdev->dev.of_node;
+ struct omap_rproc *oproc = rproc->priv;
+ const struct omap_rproc_dev_data *data;
+ int ret;
+
+ data = of_device_get_match_data(&pdev->dev);
+ if (!data)
+ return -ENODEV;
+
+ if (!of_property_read_bool(np, "ti,bootreg"))
+ return 0;
+
+ oproc->boot_data = devm_kzalloc(&pdev->dev, sizeof(*oproc->boot_data),
+ GFP_KERNEL);
+ if (!oproc->boot_data)
+ return -ENOMEM;
+
+ oproc->boot_data->syscon =
+ syscon_regmap_lookup_by_phandle(np, "ti,bootreg");
+ if (IS_ERR(oproc->boot_data->syscon)) {
+ ret = PTR_ERR(oproc->boot_data->syscon);
+ return ret;
+ }
+
+ if (of_property_read_u32_index(np, "ti,bootreg", 1,
+ &oproc->boot_data->boot_reg)) {
+ dev_err(&pdev->dev, "couldn't get the boot register\n");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
static int omap_rproc_probe(struct platform_device *pdev)
{
- struct omap_rproc_pdata *pdata = pdev->dev.platform_data;
+ struct device_node *np = pdev->dev.of_node;
struct omap_rproc *oproc;
struct rproc *rproc;
+ const char *firmware;
int ret;
+ struct reset_control *reset;
+
+ if (!np) {
+ dev_err(&pdev->dev, "only DT-based devices are supported\n");
+ return -ENODEV;
+ }
+
+ reset = devm_reset_control_array_get_exclusive(&pdev->dev);
+ if (IS_ERR(reset))
+ return PTR_ERR(reset);
+
+ firmware = omap_rproc_get_firmware(pdev);
+ if (IS_ERR(firmware))
+ return PTR_ERR(firmware);
ret = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
if (ret) {
@@ -188,16 +325,21 @@ static int omap_rproc_probe(struct platform_device *pdev)
return ret;
}
- rproc = rproc_alloc(&pdev->dev, pdata->name, &omap_rproc_ops,
- pdata->firmware, sizeof(*oproc));
+ rproc = rproc_alloc(&pdev->dev, dev_name(&pdev->dev), &omap_rproc_ops,
+ firmware, sizeof(*oproc));
if (!rproc)
return -ENOMEM;
oproc = rproc->priv;
oproc->rproc = rproc;
+ oproc->reset = reset;
/* All existing OMAP IPU and DSP processors have an MMU */
rproc->has_iommu = true;
+ ret = omap_rproc_get_boot_data(pdev, rproc);
+ if (ret)
+ goto free_rproc;
+
platform_set_drvdata(pdev, rproc);
ret = rproc_add(rproc);
@@ -226,6 +368,7 @@ static struct platform_driver omap_rproc_driver = {
.remove = omap_rproc_remove,
.driver = {
.name = "omap-rproc",
+ .of_match_table = omap_rproc_of_match,
},
};
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 03/14] remoteproc/omap: Add a sanity check for DSP boot address alignment
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The DSP remote processors on OMAP SoCs require a boot register to
be programmed with a boot address, and this boot address needs to
be on a 1KB boundary. The current code is simply masking the boot
address appropriately without performing any sanity checks before
releasing the resets. An unaligned boot address results in an
undefined execution behavior and can result in various bus errors
like MMU Faults or L3 NoC errors. Such errors are hard to debug and
can be easily avoided by adding a sanity check for the alignment
before booting a DSP remote processor.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 18 +++++++++++++++---
1 file changed, 15 insertions(+), 3 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 6c603c2cd895..2935a8edb3a6 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -122,13 +122,22 @@ static void omap_rproc_kick(struct rproc *rproc, int vqid)
*
* Set boot address for a supported DSP remote processor.
*/
-static void omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
+static int omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
{
+ struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
struct omap_rproc_boot_data *bdata = oproc->boot_data;
u32 offset = bdata->boot_reg;
+ if (rproc->bootaddr & (SZ_1K - 1)) {
+ dev_err(dev, "invalid boot address 0x%x, must be aligned on a 1KB boundary\n",
+ rproc->bootaddr);
+ return -EINVAL;
+ }
+
regmap_write(bdata->syscon, offset, rproc->bootaddr);
+
+ return 0;
}
/*
@@ -145,8 +154,11 @@ static int omap_rproc_start(struct rproc *rproc)
int ret;
struct mbox_client *client = &oproc->client;
- if (oproc->boot_data)
- omap_rproc_write_dsp_boot_addr(rproc);
+ if (oproc->boot_data) {
+ ret = omap_rproc_write_dsp_boot_addr(rproc);
+ if (ret)
+ return ret;
+ }
client->dev = dev;
client->tx_done = NULL;
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 03/14] remoteproc/omap: Add a sanity check for DSP boot address alignment
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The DSP remote processors on OMAP SoCs require a boot register to
be programmed with a boot address, and this boot address needs to
be on a 1KB boundary. The current code is simply masking the boot
address appropriately without performing any sanity checks before
releasing the resets. An unaligned boot address results in an
undefined execution behavior and can result in various bus errors
like MMU Faults or L3 NoC errors. Such errors are hard to debug and
can be easily avoided by adding a sanity check for the alignment
before booting a DSP remote processor.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 18 +++++++++++++++---
1 file changed, 15 insertions(+), 3 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 6c603c2cd895..2935a8edb3a6 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -122,13 +122,22 @@ static void omap_rproc_kick(struct rproc *rproc, int vqid)
*
* Set boot address for a supported DSP remote processor.
*/
-static void omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
+static int omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
{
+ struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
struct omap_rproc_boot_data *bdata = oproc->boot_data;
u32 offset = bdata->boot_reg;
+ if (rproc->bootaddr & (SZ_1K - 1)) {
+ dev_err(dev, "invalid boot address 0x%x, must be aligned on a 1KB boundary\n",
+ rproc->bootaddr);
+ return -EINVAL;
+ }
+
regmap_write(bdata->syscon, offset, rproc->bootaddr);
+
+ return 0;
}
/*
@@ -145,8 +154,11 @@ static int omap_rproc_start(struct rproc *rproc)
int ret;
struct mbox_client *client = &oproc->client;
- if (oproc->boot_data)
- omap_rproc_write_dsp_boot_addr(rproc);
+ if (oproc->boot_data) {
+ ret = omap_rproc_write_dsp_boot_addr(rproc);
+ if (ret)
+ return ret;
+ }
client->dev = dev;
client->tx_done = NULL;
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 04/14] remoteproc/omap: Add support to parse internal memories from DT
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The OMAP remoteproc driver has been enhanced to parse and store
the kernel mappings for different internal RAM memories that may
be present within each remote processor IP subsystem. Different
devices have varying memories present on current SoCs. The current
support handles the L2RAM for all IPU devices on OMAP4+ SoCs. The
DSPs on OMAP4/OMAP5 only have Unicaches and do not have any L1 or
L2 RAM memories.
IPUs are expected to have the L2RAM at a fixed device address of
0x20000000, based on the current limitations on Attribute MMU
configurations.
NOTE:
The current logic doesn't handle the parsing of memories for DRA7
remoteproc devices, and will be added alongside the DRA7 support.
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo: converted to parse mem names / device addresses from pdata]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 89 ++++++++++++++++++++++++++++
1 file changed, 89 insertions(+)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 2935a8edb3a6..a99e542da364 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -39,11 +39,27 @@ struct omap_rproc_boot_data {
unsigned int boot_reg;
};
+/**
+ * struct omap_rproc_mem - internal memory structure
+ * @cpu_addr: MPU virtual address of the memory region
+ * @bus_addr: bus address used to access the memory region
+ * @dev_addr: device address of the memory region from DSP view
+ * @size: size of the memory region
+ */
+struct omap_rproc_mem {
+ void __iomem *cpu_addr;
+ phys_addr_t bus_addr;
+ u32 dev_addr;
+ size_t size;
+};
+
/**
* struct omap_rproc - omap remote processor state
* @mbox: mailbox channel handle
* @client: mailbox client to request the mailbox channel
* @boot_data: boot data structure for setting processor boot address
+ * @mem: internal memory regions data
+ * @num_mems: number of internal memory regions
* @rproc: rproc handle
* @reset: reset handle
*/
@@ -51,16 +67,30 @@ struct omap_rproc {
struct mbox_chan *mbox;
struct mbox_client client;
struct omap_rproc_boot_data *boot_data;
+ struct omap_rproc_mem *mem;
+ int num_mems;
struct rproc *rproc;
struct reset_control *reset;
};
+/**
+ * struct omap_rproc_mem_data - memory definitions for an omap remote processor
+ * @mem_name: name for this memory entry
+ * @dev_addr: device address for the memory entry
+ */
+struct omap_rproc_mem_data {
+ const char *name;
+ const u32 dev_addr;
+};
+
/**
* struct omap_rproc_dev_data - device data for the omap remote processor
* @device_name: device name of the remote processor
+ * @mems: memory definitions for this remote processor
*/
struct omap_rproc_dev_data {
const char *device_name;
+ const struct omap_rproc_mem_data *mems;
};
/**
@@ -221,12 +251,18 @@ static const struct rproc_ops omap_rproc_ops = {
.kick = omap_rproc_kick,
};
+static const struct omap_rproc_mem_data ipu_mems[] = {
+ { .name = "l2ram", .dev_addr = 0x20000000 },
+ { },
+};
+
static const struct omap_rproc_dev_data omap4_dsp_dev_data = {
.device_name = "dsp",
};
static const struct omap_rproc_dev_data omap4_ipu_dev_data = {
.device_name = "ipu",
+ .mems = ipu_mems,
};
static const struct omap_rproc_dev_data omap5_dsp_dev_data = {
@@ -235,6 +271,7 @@ static const struct omap_rproc_dev_data omap5_dsp_dev_data = {
static const struct omap_rproc_dev_data omap5_ipu_dev_data = {
.device_name = "ipu",
+ .mems = ipu_mems,
};
static const struct of_device_id omap_rproc_of_match[] = {
@@ -309,6 +346,54 @@ static int omap_rproc_get_boot_data(struct platform_device *pdev,
return 0;
}
+static int omap_rproc_of_get_internal_memories(struct platform_device *pdev,
+ struct rproc *rproc)
+{
+ struct omap_rproc *oproc = rproc->priv;
+ struct device *dev = &pdev->dev;
+ const struct omap_rproc_dev_data *data;
+ struct resource *res;
+ int num_mems;
+ int i;
+
+ data = of_device_get_match_data(&pdev->dev);
+ if (!data)
+ return -ENODEV;
+
+ if (!data->mems)
+ return 0;
+
+ for (num_mems = 0; data->mems[num_mems].name; num_mems++)
+ ;
+
+ oproc->mem = devm_kcalloc(dev, num_mems, sizeof(*oproc->mem),
+ GFP_KERNEL);
+ if (!oproc->mem)
+ return -ENOMEM;
+
+ for (i = 0; i < num_mems; i++) {
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+ data->mems[i].name);
+ oproc->mem[i].cpu_addr = devm_ioremap_resource(dev, res);
+ if (IS_ERR(oproc->mem[i].cpu_addr)) {
+ dev_err(dev, "failed to parse and map %s memory\n",
+ data->mems[i].name);
+ return PTR_ERR(oproc->mem[i].cpu_addr);
+ }
+ oproc->mem[i].bus_addr = res->start;
+ oproc->mem[i].dev_addr = data->mems[i].dev_addr;
+ oproc->mem[i].size = resource_size(res);
+
+ dev_dbg(dev, "memory %8s: bus addr %pa size 0x%x va %p da 0x%x\n",
+ data->mems[i].name, &oproc->mem[i].bus_addr,
+ oproc->mem[i].size, oproc->mem[i].cpu_addr,
+ oproc->mem[i].dev_addr);
+ }
+ oproc->num_mems = num_mems;
+
+ return 0;
+}
+
static int omap_rproc_probe(struct platform_device *pdev)
{
struct device_node *np = pdev->dev.of_node;
@@ -348,6 +433,10 @@ static int omap_rproc_probe(struct platform_device *pdev)
/* All existing OMAP IPU and DSP processors have an MMU */
rproc->has_iommu = true;
+ ret = omap_rproc_of_get_internal_memories(pdev, rproc);
+ if (ret)
+ goto free_rproc;
+
ret = omap_rproc_get_boot_data(pdev, rproc);
if (ret)
goto free_rproc;
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 04/14] remoteproc/omap: Add support to parse internal memories from DT
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The OMAP remoteproc driver has been enhanced to parse and store
the kernel mappings for different internal RAM memories that may
be present within each remote processor IP subsystem. Different
devices have varying memories present on current SoCs. The current
support handles the L2RAM for all IPU devices on OMAP4+ SoCs. The
DSPs on OMAP4/OMAP5 only have Unicaches and do not have any L1 or
L2 RAM memories.
IPUs are expected to have the L2RAM at a fixed device address of
0x20000000, based on the current limitations on Attribute MMU
configurations.
NOTE:
The current logic doesn't handle the parsing of memories for DRA7
remoteproc devices, and will be added alongside the DRA7 support.
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo: converted to parse mem names / device addresses from pdata]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 89 ++++++++++++++++++++++++++++
1 file changed, 89 insertions(+)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 2935a8edb3a6..a99e542da364 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -39,11 +39,27 @@ struct omap_rproc_boot_data {
unsigned int boot_reg;
};
+/**
+ * struct omap_rproc_mem - internal memory structure
+ * @cpu_addr: MPU virtual address of the memory region
+ * @bus_addr: bus address used to access the memory region
+ * @dev_addr: device address of the memory region from DSP view
+ * @size: size of the memory region
+ */
+struct omap_rproc_mem {
+ void __iomem *cpu_addr;
+ phys_addr_t bus_addr;
+ u32 dev_addr;
+ size_t size;
+};
+
/**
* struct omap_rproc - omap remote processor state
* @mbox: mailbox channel handle
* @client: mailbox client to request the mailbox channel
* @boot_data: boot data structure for setting processor boot address
+ * @mem: internal memory regions data
+ * @num_mems: number of internal memory regions
* @rproc: rproc handle
* @reset: reset handle
*/
@@ -51,16 +67,30 @@ struct omap_rproc {
struct mbox_chan *mbox;
struct mbox_client client;
struct omap_rproc_boot_data *boot_data;
+ struct omap_rproc_mem *mem;
+ int num_mems;
struct rproc *rproc;
struct reset_control *reset;
};
+/**
+ * struct omap_rproc_mem_data - memory definitions for an omap remote processor
+ * @mem_name: name for this memory entry
+ * @dev_addr: device address for the memory entry
+ */
+struct omap_rproc_mem_data {
+ const char *name;
+ const u32 dev_addr;
+};
+
/**
* struct omap_rproc_dev_data - device data for the omap remote processor
* @device_name: device name of the remote processor
+ * @mems: memory definitions for this remote processor
*/
struct omap_rproc_dev_data {
const char *device_name;
+ const struct omap_rproc_mem_data *mems;
};
/**
@@ -221,12 +251,18 @@ static const struct rproc_ops omap_rproc_ops = {
.kick = omap_rproc_kick,
};
+static const struct omap_rproc_mem_data ipu_mems[] = {
+ { .name = "l2ram", .dev_addr = 0x20000000 },
+ { },
+};
+
static const struct omap_rproc_dev_data omap4_dsp_dev_data = {
.device_name = "dsp",
};
static const struct omap_rproc_dev_data omap4_ipu_dev_data = {
.device_name = "ipu",
+ .mems = ipu_mems,
};
static const struct omap_rproc_dev_data omap5_dsp_dev_data = {
@@ -235,6 +271,7 @@ static const struct omap_rproc_dev_data omap5_dsp_dev_data = {
static const struct omap_rproc_dev_data omap5_ipu_dev_data = {
.device_name = "ipu",
+ .mems = ipu_mems,
};
static const struct of_device_id omap_rproc_of_match[] = {
@@ -309,6 +346,54 @@ static int omap_rproc_get_boot_data(struct platform_device *pdev,
return 0;
}
+static int omap_rproc_of_get_internal_memories(struct platform_device *pdev,
+ struct rproc *rproc)
+{
+ struct omap_rproc *oproc = rproc->priv;
+ struct device *dev = &pdev->dev;
+ const struct omap_rproc_dev_data *data;
+ struct resource *res;
+ int num_mems;
+ int i;
+
+ data = of_device_get_match_data(&pdev->dev);
+ if (!data)
+ return -ENODEV;
+
+ if (!data->mems)
+ return 0;
+
+ for (num_mems = 0; data->mems[num_mems].name; num_mems++)
+ ;
+
+ oproc->mem = devm_kcalloc(dev, num_mems, sizeof(*oproc->mem),
+ GFP_KERNEL);
+ if (!oproc->mem)
+ return -ENOMEM;
+
+ for (i = 0; i < num_mems; i++) {
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+ data->mems[i].name);
+ oproc->mem[i].cpu_addr = devm_ioremap_resource(dev, res);
+ if (IS_ERR(oproc->mem[i].cpu_addr)) {
+ dev_err(dev, "failed to parse and map %s memory\n",
+ data->mems[i].name);
+ return PTR_ERR(oproc->mem[i].cpu_addr);
+ }
+ oproc->mem[i].bus_addr = res->start;
+ oproc->mem[i].dev_addr = data->mems[i].dev_addr;
+ oproc->mem[i].size = resource_size(res);
+
+ dev_dbg(dev, "memory %8s: bus addr %pa size 0x%x va %p da 0x%x\n",
+ data->mems[i].name, &oproc->mem[i].bus_addr,
+ oproc->mem[i].size, oproc->mem[i].cpu_addr,
+ oproc->mem[i].dev_addr);
+ }
+ oproc->num_mems = num_mems;
+
+ return 0;
+}
+
static int omap_rproc_probe(struct platform_device *pdev)
{
struct device_node *np = pdev->dev.of_node;
@@ -348,6 +433,10 @@ static int omap_rproc_probe(struct platform_device *pdev)
/* All existing OMAP IPU and DSP processors have an MMU */
rproc->has_iommu = true;
+ ret = omap_rproc_of_get_internal_memories(pdev, rproc);
+ if (ret)
+ goto free_rproc;
+
ret = omap_rproc_get_boot_data(pdev, rproc);
if (ret)
goto free_rproc;
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 05/14] remoteproc/omap: Add the rproc ops .da_to_va() implementation
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
An implementation for the rproc ops .da_to_va() has been added
that provides the address translation between device addresses
to kernel virtual addresses for internal RAMs present on that
particular remote processor device. The implementation provides
the translations based on the addresses parsed and stored during
the probe.
This ops gets invoked by the exported rproc_da_to_va() function
and allows the remoteproc core's ELF loader to be able to load
program data directly into the internal memories.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 39 ++++++++++++++++++++++++++++
1 file changed, 39 insertions(+)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index a99e542da364..0846839b2c97 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -245,10 +245,49 @@ static int omap_rproc_stop(struct rproc *rproc)
return 0;
}
+/**
+ * omap_rproc_da_to_va() - internal memory translation helper
+ * @rproc: remote processor to apply the address translation for
+ * @da: device address to translate
+ * @len: length of the memory buffer
+ *
+ * Custom function implementing the rproc .da_to_va ops to provide address
+ * translation (device address to kernel virtual address) for internal RAMs
+ * present in a DSP or IPU device). The translated addresses can be used
+ * either by the remoteproc core for loading, or by any rpmsg bus drivers.
+ * Returns the translated virtual address in kernel memory space, or NULL
+ * in failure.
+ */
+static void *omap_rproc_da_to_va(struct rproc *rproc, u64 da, int len)
+{
+ struct omap_rproc *oproc = rproc->priv;
+ int i;
+ u32 offset;
+
+ if (len <= 0)
+ return NULL;
+
+ if (!oproc->num_mems)
+ return NULL;
+
+ for (i = 0; i < oproc->num_mems; i++) {
+ if (da >= oproc->mem[i].dev_addr && da + len <=
+ oproc->mem[i].dev_addr + oproc->mem[i].size) {
+ offset = da - oproc->mem[i].dev_addr;
+ /* __force to make sparse happy with type conversion */
+ return (__force void *)(oproc->mem[i].cpu_addr +
+ offset);
+ }
+ }
+
+ return NULL;
+}
+
static const struct rproc_ops omap_rproc_ops = {
.start = omap_rproc_start,
.stop = omap_rproc_stop,
.kick = omap_rproc_kick,
+ .da_to_va = omap_rproc_da_to_va,
};
static const struct omap_rproc_mem_data ipu_mems[] = {
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 05/14] remoteproc/omap: Add the rproc ops .da_to_va() implementation
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
An implementation for the rproc ops .da_to_va() has been added
that provides the address translation between device addresses
to kernel virtual addresses for internal RAMs present on that
particular remote processor device. The implementation provides
the translations based on the addresses parsed and stored during
the probe.
This ops gets invoked by the exported rproc_da_to_va() function
and allows the remoteproc core's ELF loader to be able to load
program data directly into the internal memories.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 39 ++++++++++++++++++++++++++++
1 file changed, 39 insertions(+)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index a99e542da364..0846839b2c97 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -245,10 +245,49 @@ static int omap_rproc_stop(struct rproc *rproc)
return 0;
}
+/**
+ * omap_rproc_da_to_va() - internal memory translation helper
+ * @rproc: remote processor to apply the address translation for
+ * @da: device address to translate
+ * @len: length of the memory buffer
+ *
+ * Custom function implementing the rproc .da_to_va ops to provide address
+ * translation (device address to kernel virtual address) for internal RAMs
+ * present in a DSP or IPU device). The translated addresses can be used
+ * either by the remoteproc core for loading, or by any rpmsg bus drivers.
+ * Returns the translated virtual address in kernel memory space, or NULL
+ * in failure.
+ */
+static void *omap_rproc_da_to_va(struct rproc *rproc, u64 da, int len)
+{
+ struct omap_rproc *oproc = rproc->priv;
+ int i;
+ u32 offset;
+
+ if (len <= 0)
+ return NULL;
+
+ if (!oproc->num_mems)
+ return NULL;
+
+ for (i = 0; i < oproc->num_mems; i++) {
+ if (da >= oproc->mem[i].dev_addr && da + len <=
+ oproc->mem[i].dev_addr + oproc->mem[i].size) {
+ offset = da - oproc->mem[i].dev_addr;
+ /* __force to make sparse happy with type conversion */
+ return (__force void *)(oproc->mem[i].cpu_addr +
+ offset);
+ }
+ }
+
+ return NULL;
+}
+
static const struct rproc_ops omap_rproc_ops = {
.start = omap_rproc_start,
.stop = omap_rproc_stop,
.kick = omap_rproc_kick,
+ .da_to_va = omap_rproc_da_to_va,
};
static const struct omap_rproc_mem_data ipu_mems[] = {
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The reserved memory nodes are not assigned to platform devices by
default in the driver core to avoid the lookup for every platform
device and incur a penalty as the real users are expected to be
only a few devices.
OMAP remoteproc devices fall into the above category and the OMAP
remoteproc driver _requires_ specific CMA pools to be assigned
for each device at the moment to align on the location of the
vrings and vring buffers in the RTOS-side firmware images. So,
use the of_reserved_mem_device_init/release() API appropriately
to assign the corresponding reserved memory region to the OMAP
remoteproc device. Note that only one region per device is
allowed by the framework.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 0846839b2c97..194303b860b2 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -17,6 +17,7 @@
#include <linux/module.h>
#include <linux/err.h>
#include <linux/of_device.h>
+#include <linux/of_reserved_mem.h>
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
@@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev)
if (ret)
goto free_rproc;
+ ret = of_reserved_mem_device_init(&pdev->dev);
+ if (ret) {
+ dev_err(&pdev->dev, "device does not have specific CMA pool\n");
+ goto free_rproc;
+ }
+
platform_set_drvdata(pdev, rproc);
ret = rproc_add(rproc);
if (ret)
- goto free_rproc;
+ goto release_mem;
return 0;
+release_mem:
+ of_reserved_mem_device_release(&pdev->dev);
free_rproc:
rproc_free(rproc);
return ret;
@@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev)
rproc_del(rproc);
rproc_free(rproc);
+ of_reserved_mem_device_release(&pdev->dev);
return 0;
}
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The reserved memory nodes are not assigned to platform devices by
default in the driver core to avoid the lookup for every platform
device and incur a penalty as the real users are expected to be
only a few devices.
OMAP remoteproc devices fall into the above category and the OMAP
remoteproc driver _requires_ specific CMA pools to be assigned
for each device at the moment to align on the location of the
vrings and vring buffers in the RTOS-side firmware images. So,
use the of_reserved_mem_device_init/release() API appropriately
to assign the corresponding reserved memory region to the OMAP
remoteproc device. Note that only one region per device is
allowed by the framework.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 0846839b2c97..194303b860b2 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -17,6 +17,7 @@
#include <linux/module.h>
#include <linux/err.h>
#include <linux/of_device.h>
+#include <linux/of_reserved_mem.h>
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
@@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev)
if (ret)
goto free_rproc;
+ ret = of_reserved_mem_device_init(&pdev->dev);
+ if (ret) {
+ dev_err(&pdev->dev, "device does not have specific CMA pool\n");
+ goto free_rproc;
+ }
+
platform_set_drvdata(pdev, rproc);
ret = rproc_add(rproc);
if (ret)
- goto free_rproc;
+ goto release_mem;
return 0;
+release_mem:
+ of_reserved_mem_device_release(&pdev->dev);
free_rproc:
rproc_free(rproc);
return ret;
@@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev)
rproc_del(rproc);
rproc_free(rproc);
+ of_reserved_mem_device_release(&pdev->dev);
return 0;
}
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-30 18:11 ` Andrew F. Davis
-1 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 18:11 UTC (permalink / raw)
To: Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 1/16/20 8:53 AM, Tero Kristo wrote:
> From: Suman Anna <s-anna@ti.com>
>
> The reserved memory nodes are not assigned to platform devices by
> default in the driver core to avoid the lookup for every platform
> device and incur a penalty as the real users are expected to be
> only a few devices.
>
> OMAP remoteproc devices fall into the above category and the OMAP
> remoteproc driver _requires_ specific CMA pools to be assigned
> for each device at the moment to align on the location of the
> vrings and vring buffers in the RTOS-side firmware images. So,
Same comment as before, this is a firmware issue for only some firmwares
that do not handle being assigned vring locations correctly and instead
hard-code them.
This is not a requirement of the remote processor itself and so it
should not fail to probe if a specific memory carveout isn't given.
Andrew
> use the of_reserved_mem_device_init/release() API appropriately
> to assign the corresponding reserved memory region to the OMAP
> remoteproc device. Note that only one region per device is
> allowed by the framework.
>
> Signed-off-by: Suman Anna <s-anna@ti.com>
> Signed-off-by: Tero Kristo <t-kristo@ti.com>
> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
> ---
> v5: no changes
>
> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
> index 0846839b2c97..194303b860b2 100644
> --- a/drivers/remoteproc/omap_remoteproc.c
> +++ b/drivers/remoteproc/omap_remoteproc.c
> @@ -17,6 +17,7 @@
> #include <linux/module.h>
> #include <linux/err.h>
> #include <linux/of_device.h>
> +#include <linux/of_reserved_mem.h>
> #include <linux/platform_device.h>
> #include <linux/dma-mapping.h>
> #include <linux/remoteproc.h>
> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev)
> if (ret)
> goto free_rproc;
>
> + ret = of_reserved_mem_device_init(&pdev->dev);
> + if (ret) {
> + dev_err(&pdev->dev, "device does not have specific CMA pool\n");
> + goto free_rproc;
> + }
> +
> platform_set_drvdata(pdev, rproc);
>
> ret = rproc_add(rproc);
> if (ret)
> - goto free_rproc;
> + goto release_mem;
>
> return 0;
>
> +release_mem:
> + of_reserved_mem_device_release(&pdev->dev);
> free_rproc:
> rproc_free(rproc);
> return ret;
> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev)
>
> rproc_del(rproc);
> rproc_free(rproc);
> + of_reserved_mem_device_release(&pdev->dev);
>
> return 0;
> }
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 18:11 ` Andrew F. Davis
0 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 18:11 UTC (permalink / raw)
To: Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 1/16/20 8:53 AM, Tero Kristo wrote:
> From: Suman Anna <s-anna@ti.com>
>
> The reserved memory nodes are not assigned to platform devices by
> default in the driver core to avoid the lookup for every platform
> device and incur a penalty as the real users are expected to be
> only a few devices.
>
> OMAP remoteproc devices fall into the above category and the OMAP
> remoteproc driver _requires_ specific CMA pools to be assigned
> for each device at the moment to align on the location of the
> vrings and vring buffers in the RTOS-side firmware images. So,
Same comment as before, this is a firmware issue for only some firmwares
that do not handle being assigned vring locations correctly and instead
hard-code them.
This is not a requirement of the remote processor itself and so it
should not fail to probe if a specific memory carveout isn't given.
Andrew
> use the of_reserved_mem_device_init/release() API appropriately
> to assign the corresponding reserved memory region to the OMAP
> remoteproc device. Note that only one region per device is
> allowed by the framework.
>
> Signed-off-by: Suman Anna <s-anna@ti.com>
> Signed-off-by: Tero Kristo <t-kristo@ti.com>
> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
> ---
> v5: no changes
>
> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
> index 0846839b2c97..194303b860b2 100644
> --- a/drivers/remoteproc/omap_remoteproc.c
> +++ b/drivers/remoteproc/omap_remoteproc.c
> @@ -17,6 +17,7 @@
> #include <linux/module.h>
> #include <linux/err.h>
> #include <linux/of_device.h>
> +#include <linux/of_reserved_mem.h>
> #include <linux/platform_device.h>
> #include <linux/dma-mapping.h>
> #include <linux/remoteproc.h>
> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev)
> if (ret)
> goto free_rproc;
>
> + ret = of_reserved_mem_device_init(&pdev->dev);
> + if (ret) {
> + dev_err(&pdev->dev, "device does not have specific CMA pool\n");
> + goto free_rproc;
> + }
> +
> platform_set_drvdata(pdev, rproc);
>
> ret = rproc_add(rproc);
> if (ret)
> - goto free_rproc;
> + goto release_mem;
>
> return 0;
>
> +release_mem:
> + of_reserved_mem_device_release(&pdev->dev);
> free_rproc:
> rproc_free(rproc);
> return ret;
> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev)
>
> rproc_del(rproc);
> rproc_free(rproc);
> + of_reserved_mem_device_release(&pdev->dev);
>
> return 0;
> }
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 18:11 ` Andrew F. Davis
@ 2020-01-30 19:18 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-30 19:18 UTC (permalink / raw)
To: Andrew F. Davis, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 30/01/2020 20:11, Andrew F. Davis wrote:
> On 1/16/20 8:53 AM, Tero Kristo wrote:
>> From: Suman Anna <s-anna@ti.com>
>>
>> The reserved memory nodes are not assigned to platform devices by
>> default in the driver core to avoid the lookup for every platform
>> device and incur a penalty as the real users are expected to be
>> only a few devices.
>>
>> OMAP remoteproc devices fall into the above category and the OMAP
>> remoteproc driver _requires_ specific CMA pools to be assigned
>> for each device at the moment to align on the location of the
>> vrings and vring buffers in the RTOS-side firmware images. So,
>
>
> Same comment as before, this is a firmware issue for only some firmwares
> that do not handle being assigned vring locations correctly and instead
> hard-code them.
I believe we discussed this topic in length in previous version but
there was no conclusion on it.
The commit desc might be a bit misleading, we are not actually forced to
use specific CMA buffers, as we use IOMMU to map these to device
addresses. For example IPU1/IPU2 use internally exact same memory
addresses, iommu is used to map these to specific CMA buffer.
CMA buffers are mostly used so that we get aligned large chunk of memory
which can be mapped properly with the limited IOMMU OMAP family of chips
have. Not sure if there is any sane way to get this done in any other
manner.
-Tero
>
> This is not a requirement of the remote processor itself and so it
> should not fail to probe if a specific memory carveout isn't given.
>
>
>> use the of_reserved_mem_device_init/release() API appropriately
>> to assign the corresponding reserved memory region to the OMAP
>> remoteproc device. Note that only one region per device is
>> allowed by the framework.
>>
>> Signed-off-by: Suman Anna <s-anna@ti.com>
>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>> ---
>> v5: no changes
>>
>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
>> index 0846839b2c97..194303b860b2 100644
>> --- a/drivers/remoteproc/omap_remoteproc.c
>> +++ b/drivers/remoteproc/omap_remoteproc.c
>> @@ -17,6 +17,7 @@
>> #include <linux/module.h>
>> #include <linux/err.h>
>> #include <linux/of_device.h>
>> +#include <linux/of_reserved_mem.h>
>> #include <linux/platform_device.h>
>> #include <linux/dma-mapping.h>
>> #include <linux/remoteproc.h>
>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev)
>> if (ret)
>> goto free_rproc;
>>
>> + ret = of_reserved_mem_device_init(&pdev->dev);
>> + if (ret) {
>> + dev_err(&pdev->dev, "device does not have specific CMA pool\n");
>> + goto free_rproc;
>> + }
>> +
>> platform_set_drvdata(pdev, rproc);
>>
>> ret = rproc_add(rproc);
>> if (ret)
>> - goto free_rproc;
>> + goto release_mem;
>>
>> return 0;
>>
>> +release_mem:
>> + of_reserved_mem_device_release(&pdev->dev);
>> free_rproc:
>> rproc_free(rproc);
>> return ret;
>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev)
>>
>> rproc_del(rproc);
>> rproc_free(rproc);
>> + of_reserved_mem_device_release(&pdev->dev);
>>
>> return 0;
>> }
>>
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 19:18 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-30 19:18 UTC (permalink / raw)
To: Andrew F. Davis, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 30/01/2020 20:11, Andrew F. Davis wrote:
> On 1/16/20 8:53 AM, Tero Kristo wrote:
>> From: Suman Anna <s-anna@ti.com>
>>
>> The reserved memory nodes are not assigned to platform devices by
>> default in the driver core to avoid the lookup for every platform
>> device and incur a penalty as the real users are expected to be
>> only a few devices.
>>
>> OMAP remoteproc devices fall into the above category and the OMAP
>> remoteproc driver _requires_ specific CMA pools to be assigned
>> for each device at the moment to align on the location of the
>> vrings and vring buffers in the RTOS-side firmware images. So,
>
>
> Same comment as before, this is a firmware issue for only some firmwares
> that do not handle being assigned vring locations correctly and instead
> hard-code them.
I believe we discussed this topic in length in previous version but
there was no conclusion on it.
The commit desc might be a bit misleading, we are not actually forced to
use specific CMA buffers, as we use IOMMU to map these to device
addresses. For example IPU1/IPU2 use internally exact same memory
addresses, iommu is used to map these to specific CMA buffer.
CMA buffers are mostly used so that we get aligned large chunk of memory
which can be mapped properly with the limited IOMMU OMAP family of chips
have. Not sure if there is any sane way to get this done in any other
manner.
-Tero
>
> This is not a requirement of the remote processor itself and so it
> should not fail to probe if a specific memory carveout isn't given.
>
>
>> use the of_reserved_mem_device_init/release() API appropriately
>> to assign the corresponding reserved memory region to the OMAP
>> remoteproc device. Note that only one region per device is
>> allowed by the framework.
>>
>> Signed-off-by: Suman Anna <s-anna@ti.com>
>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>> ---
>> v5: no changes
>>
>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
>> index 0846839b2c97..194303b860b2 100644
>> --- a/drivers/remoteproc/omap_remoteproc.c
>> +++ b/drivers/remoteproc/omap_remoteproc.c
>> @@ -17,6 +17,7 @@
>> #include <linux/module.h>
>> #include <linux/err.h>
>> #include <linux/of_device.h>
>> +#include <linux/of_reserved_mem.h>
>> #include <linux/platform_device.h>
>> #include <linux/dma-mapping.h>
>> #include <linux/remoteproc.h>
>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev)
>> if (ret)
>> goto free_rproc;
>>
>> + ret = of_reserved_mem_device_init(&pdev->dev);
>> + if (ret) {
>> + dev_err(&pdev->dev, "device does not have specific CMA pool\n");
>> + goto free_rproc;
>> + }
>> +
>> platform_set_drvdata(pdev, rproc);
>>
>> ret = rproc_add(rproc);
>> if (ret)
>> - goto free_rproc;
>> + goto release_mem;
>>
>> return 0;
>>
>> +release_mem:
>> + of_reserved_mem_device_release(&pdev->dev);
>> free_rproc:
>> rproc_free(rproc);
>> return ret;
>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev)
>>
>> rproc_del(rproc);
>> rproc_free(rproc);
>> + of_reserved_mem_device_release(&pdev->dev);
>>
>> return 0;
>> }
>>
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 19:18 ` Tero Kristo
@ 2020-01-30 19:20 ` Andrew F. Davis
-1 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 19:20 UTC (permalink / raw)
To: Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 1/30/20 2:18 PM, Tero Kristo wrote:
> On 30/01/2020 20:11, Andrew F. Davis wrote:
>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>> From: Suman Anna <s-anna@ti.com>
>>>
>>> The reserved memory nodes are not assigned to platform devices by
>>> default in the driver core to avoid the lookup for every platform
>>> device and incur a penalty as the real users are expected to be
>>> only a few devices.
>>>
>>> OMAP remoteproc devices fall into the above category and the OMAP
>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>> for each device at the moment to align on the location of the
>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>
>>
>> Same comment as before, this is a firmware issue for only some firmwares
>> that do not handle being assigned vring locations correctly and instead
>> hard-code them.
>
> I believe we discussed this topic in length in previous version but
> there was no conclusion on it.
>
> The commit desc might be a bit misleading, we are not actually forced to
> use specific CMA buffers, as we use IOMMU to map these to device
> addresses. For example IPU1/IPU2 use internally exact same memory
> addresses, iommu is used to map these to specific CMA buffer.
>
> CMA buffers are mostly used so that we get aligned large chunk of memory
> which can be mapped properly with the limited IOMMU OMAP family of chips
> have. Not sure if there is any sane way to get this done in any other
> manner.
>
Why not use the default CMA area?
Andrew
> -Tero
>
>>
>> This is not a requirement of the remote processor itself and so it
>> should not fail to probe if a specific memory carveout isn't given.
>>
>>
>>> use the of_reserved_mem_device_init/release() API appropriately
>>> to assign the corresponding reserved memory region to the OMAP
>>> remoteproc device. Note that only one region per device is
>>> allowed by the framework.
>>>
>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>> ---
>>> v5: no changes
>>>
>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>> b/drivers/remoteproc/omap_remoteproc.c
>>> index 0846839b2c97..194303b860b2 100644
>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>> @@ -17,6 +17,7 @@
>>> #include <linux/module.h>
>>> #include <linux/err.h>
>>> #include <linux/of_device.h>
>>> +#include <linux/of_reserved_mem.h>
>>> #include <linux/platform_device.h>
>>> #include <linux/dma-mapping.h>
>>> #include <linux/remoteproc.h>
>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>> platform_device *pdev)
>>> if (ret)
>>> goto free_rproc;
>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>> + if (ret) {
>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>> pool\n");
>>> + goto free_rproc;
>>> + }
>>> +
>>> platform_set_drvdata(pdev, rproc);
>>> ret = rproc_add(rproc);
>>> if (ret)
>>> - goto free_rproc;
>>> + goto release_mem;
>>> return 0;
>>> +release_mem:
>>> + of_reserved_mem_device_release(&pdev->dev);
>>> free_rproc:
>>> rproc_free(rproc);
>>> return ret;
>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>> platform_device *pdev)
>>> rproc_del(rproc);
>>> rproc_free(rproc);
>>> + of_reserved_mem_device_release(&pdev->dev);
>>> return 0;
>>> }
>>>
>
> --
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 19:20 ` Andrew F. Davis
0 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 19:20 UTC (permalink / raw)
To: Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 1/30/20 2:18 PM, Tero Kristo wrote:
> On 30/01/2020 20:11, Andrew F. Davis wrote:
>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>> From: Suman Anna <s-anna@ti.com>
>>>
>>> The reserved memory nodes are not assigned to platform devices by
>>> default in the driver core to avoid the lookup for every platform
>>> device and incur a penalty as the real users are expected to be
>>> only a few devices.
>>>
>>> OMAP remoteproc devices fall into the above category and the OMAP
>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>> for each device at the moment to align on the location of the
>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>
>>
>> Same comment as before, this is a firmware issue for only some firmwares
>> that do not handle being assigned vring locations correctly and instead
>> hard-code them.
>
> I believe we discussed this topic in length in previous version but
> there was no conclusion on it.
>
> The commit desc might be a bit misleading, we are not actually forced to
> use specific CMA buffers, as we use IOMMU to map these to device
> addresses. For example IPU1/IPU2 use internally exact same memory
> addresses, iommu is used to map these to specific CMA buffer.
>
> CMA buffers are mostly used so that we get aligned large chunk of memory
> which can be mapped properly with the limited IOMMU OMAP family of chips
> have. Not sure if there is any sane way to get this done in any other
> manner.
>
Why not use the default CMA area?
Andrew
> -Tero
>
>>
>> This is not a requirement of the remote processor itself and so it
>> should not fail to probe if a specific memory carveout isn't given.
>>
>>
>>> use the of_reserved_mem_device_init/release() API appropriately
>>> to assign the corresponding reserved memory region to the OMAP
>>> remoteproc device. Note that only one region per device is
>>> allowed by the framework.
>>>
>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>> ---
>>> v5: no changes
>>>
>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>> b/drivers/remoteproc/omap_remoteproc.c
>>> index 0846839b2c97..194303b860b2 100644
>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>> @@ -17,6 +17,7 @@
>>> #include <linux/module.h>
>>> #include <linux/err.h>
>>> #include <linux/of_device.h>
>>> +#include <linux/of_reserved_mem.h>
>>> #include <linux/platform_device.h>
>>> #include <linux/dma-mapping.h>
>>> #include <linux/remoteproc.h>
>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>> platform_device *pdev)
>>> if (ret)
>>> goto free_rproc;
>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>> + if (ret) {
>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>> pool\n");
>>> + goto free_rproc;
>>> + }
>>> +
>>> platform_set_drvdata(pdev, rproc);
>>> ret = rproc_add(rproc);
>>> if (ret)
>>> - goto free_rproc;
>>> + goto release_mem;
>>> return 0;
>>> +release_mem:
>>> + of_reserved_mem_device_release(&pdev->dev);
>>> free_rproc:
>>> rproc_free(rproc);
>>> return ret;
>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>> platform_device *pdev)
>>> rproc_del(rproc);
>>> rproc_free(rproc);
>>> + of_reserved_mem_device_release(&pdev->dev);
>>> return 0;
>>> }
>>>
>
> --
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 19:20 ` Andrew F. Davis
@ 2020-01-30 19:42 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-30 19:42 UTC (permalink / raw)
To: Andrew F. Davis, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 30/01/2020 21:20, Andrew F. Davis wrote:
> On 1/30/20 2:18 PM, Tero Kristo wrote:
>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>> From: Suman Anna <s-anna@ti.com>
>>>>
>>>> The reserved memory nodes are not assigned to platform devices by
>>>> default in the driver core to avoid the lookup for every platform
>>>> device and incur a penalty as the real users are expected to be
>>>> only a few devices.
>>>>
>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>> for each device at the moment to align on the location of the
>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>
>>>
>>> Same comment as before, this is a firmware issue for only some firmwares
>>> that do not handle being assigned vring locations correctly and instead
>>> hard-code them.
>>
>> I believe we discussed this topic in length in previous version but
>> there was no conclusion on it.
>>
>> The commit desc might be a bit misleading, we are not actually forced to
>> use specific CMA buffers, as we use IOMMU to map these to device
>> addresses. For example IPU1/IPU2 use internally exact same memory
>> addresses, iommu is used to map these to specific CMA buffer.
>>
>> CMA buffers are mostly used so that we get aligned large chunk of memory
>> which can be mapped properly with the limited IOMMU OMAP family of chips
>> have. Not sure if there is any sane way to get this done in any other
>> manner.
>>
>
>
> Why not use the default CMA area?
I think using default CMA area getting the actual memory block is not
guaranteed and might fail. There are other users for the memory, and it
might get fragmented at the very late phase we are grabbing the memory
(omap remoteproc driver probe time.) Some chunks we need are pretty large.
I believe I could experiment with this a bit though and see, or Suman
could maybe provide feedback why this was designed initially like this
and why this would not be a good idea.
-Tero
>
> Andrew
>
>
>> -Tero
>>
>>>
>>> This is not a requirement of the remote processor itself and so it
>>> should not fail to probe if a specific memory carveout isn't given.
>>>
>>>
>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>> to assign the corresponding reserved memory region to the OMAP
>>>> remoteproc device. Note that only one region per device is
>>>> allowed by the framework.
>>>>
>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>> ---
>>>> v5: no changes
>>>>
>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>> index 0846839b2c97..194303b860b2 100644
>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>> @@ -17,6 +17,7 @@
>>>> #include <linux/module.h>
>>>> #include <linux/err.h>
>>>> #include <linux/of_device.h>
>>>> +#include <linux/of_reserved_mem.h>
>>>> #include <linux/platform_device.h>
>>>> #include <linux/dma-mapping.h>
>>>> #include <linux/remoteproc.h>
>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>> platform_device *pdev)
>>>> if (ret)
>>>> goto free_rproc;
>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>> + if (ret) {
>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>> pool\n");
>>>> + goto free_rproc;
>>>> + }
>>>> +
>>>> platform_set_drvdata(pdev, rproc);
>>>> ret = rproc_add(rproc);
>>>> if (ret)
>>>> - goto free_rproc;
>>>> + goto release_mem;
>>>> return 0;
>>>> +release_mem:
>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>> free_rproc:
>>>> rproc_free(rproc);
>>>> return ret;
>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>> platform_device *pdev)
>>>> rproc_del(rproc);
>>>> rproc_free(rproc);
>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>> return 0;
>>>> }
>>>>
>>
>> --
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 19:42 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-30 19:42 UTC (permalink / raw)
To: Andrew F. Davis, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap
On 30/01/2020 21:20, Andrew F. Davis wrote:
> On 1/30/20 2:18 PM, Tero Kristo wrote:
>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>> From: Suman Anna <s-anna@ti.com>
>>>>
>>>> The reserved memory nodes are not assigned to platform devices by
>>>> default in the driver core to avoid the lookup for every platform
>>>> device and incur a penalty as the real users are expected to be
>>>> only a few devices.
>>>>
>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>> for each device at the moment to align on the location of the
>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>
>>>
>>> Same comment as before, this is a firmware issue for only some firmwares
>>> that do not handle being assigned vring locations correctly and instead
>>> hard-code them.
>>
>> I believe we discussed this topic in length in previous version but
>> there was no conclusion on it.
>>
>> The commit desc might be a bit misleading, we are not actually forced to
>> use specific CMA buffers, as we use IOMMU to map these to device
>> addresses. For example IPU1/IPU2 use internally exact same memory
>> addresses, iommu is used to map these to specific CMA buffer.
>>
>> CMA buffers are mostly used so that we get aligned large chunk of memory
>> which can be mapped properly with the limited IOMMU OMAP family of chips
>> have. Not sure if there is any sane way to get this done in any other
>> manner.
>>
>
>
> Why not use the default CMA area?
I think using default CMA area getting the actual memory block is not
guaranteed and might fail. There are other users for the memory, and it
might get fragmented at the very late phase we are grabbing the memory
(omap remoteproc driver probe time.) Some chunks we need are pretty large.
I believe I could experiment with this a bit though and see, or Suman
could maybe provide feedback why this was designed initially like this
and why this would not be a good idea.
-Tero
>
> Andrew
>
>
>> -Tero
>>
>>>
>>> This is not a requirement of the remote processor itself and so it
>>> should not fail to probe if a specific memory carveout isn't given.
>>>
>>>
>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>> to assign the corresponding reserved memory region to the OMAP
>>>> remoteproc device. Note that only one region per device is
>>>> allowed by the framework.
>>>>
>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>> ---
>>>> v5: no changes
>>>>
>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>> index 0846839b2c97..194303b860b2 100644
>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>> @@ -17,6 +17,7 @@
>>>> #include <linux/module.h>
>>>> #include <linux/err.h>
>>>> #include <linux/of_device.h>
>>>> +#include <linux/of_reserved_mem.h>
>>>> #include <linux/platform_device.h>
>>>> #include <linux/dma-mapping.h>
>>>> #include <linux/remoteproc.h>
>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>> platform_device *pdev)
>>>> if (ret)
>>>> goto free_rproc;
>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>> + if (ret) {
>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>> pool\n");
>>>> + goto free_rproc;
>>>> + }
>>>> +
>>>> platform_set_drvdata(pdev, rproc);
>>>> ret = rproc_add(rproc);
>>>> if (ret)
>>>> - goto free_rproc;
>>>> + goto release_mem;
>>>> return 0;
>>>> +release_mem:
>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>> free_rproc:
>>>> rproc_free(rproc);
>>>> return ret;
>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>> platform_device *pdev)
>>>> rproc_del(rproc);
>>>> rproc_free(rproc);
>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>> return 0;
>>>> }
>>>>
>>
>> --
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 19:42 ` Tero Kristo
@ 2020-01-30 19:55 ` Suman Anna
-1 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 19:55 UTC (permalink / raw)
To: Tero Kristo, Andrew F. Davis, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 1:42 PM, Tero Kristo wrote:
> On 30/01/2020 21:20, Andrew F. Davis wrote:
>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>
>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>> default in the driver core to avoid the lookup for every platform
>>>>> device and incur a penalty as the real users are expected to be
>>>>> only a few devices.
>>>>>
>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>> for each device at the moment to align on the location of the
>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>
>>>>
>>>> Same comment as before, this is a firmware issue for only some
>>>> firmwares
>>>> that do not handle being assigned vring locations correctly and instead
>>>> hard-code them.
As for this statement, this can do with some updating. Post 4.20,
because of the lazy allocation scheme used for carveouts including the
vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
have to wait for the vdev synchronization to happen.
>>>
>>> I believe we discussed this topic in length in previous version but
>>> there was no conclusion on it.
>>>
>>> The commit desc might be a bit misleading, we are not actually forced to
>>> use specific CMA buffers, as we use IOMMU to map these to device
>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>> addresses, iommu is used to map these to specific CMA buffer.
>>>
>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>> have. Not sure if there is any sane way to get this done in any other
>>> manner.
>>>
>>
>>
>> Why not use the default CMA area?
>
> I think using default CMA area getting the actual memory block is not
> guaranteed and might fail. There are other users for the memory, and it
> might get fragmented at the very late phase we are grabbing the memory
> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>
> I believe I could experiment with this a bit though and see, or Suman
> could maybe provide feedback why this was designed initially like this
> and why this would not be a good idea.
I have given some explanation on this on v4 as well, but if it is not
clear, there are restrictions with using default CMA. Default CMA has
switched to be assigned from the top of the memory (higher addresses,
since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
32-bits. So, we cannot blindly use the default CMA pool, and this will
definitely not work on boards > 2 GB RAM. And, if you want to add in any
firewall capability, then specific physical addresses becomes mandatory.
regards
Suman
>
> -Tero
>
>>
>> Andrew
>>
>>
>>> -Tero
>>>
>>>>
>>>> This is not a requirement of the remote processor itself and so it
>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>
>>>>
>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>> remoteproc device. Note that only one region per device is
>>>>> allowed by the framework.
>>>>>
>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>> ---
>>>>> v5: no changes
>>>>>
>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>> index 0846839b2c97..194303b860b2 100644
>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>> @@ -17,6 +17,7 @@
>>>>> #include <linux/module.h>
>>>>> #include <linux/err.h>
>>>>> #include <linux/of_device.h>
>>>>> +#include <linux/of_reserved_mem.h>
>>>>> #include <linux/platform_device.h>
>>>>> #include <linux/dma-mapping.h>
>>>>> #include <linux/remoteproc.h>
>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>> platform_device *pdev)
>>>>> if (ret)
>>>>> goto free_rproc;
>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>> + if (ret) {
>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>> pool\n");
>>>>> + goto free_rproc;
>>>>> + }
>>>>> +
>>>>> platform_set_drvdata(pdev, rproc);
>>>>> ret = rproc_add(rproc);
>>>>> if (ret)
>>>>> - goto free_rproc;
>>>>> + goto release_mem;
>>>>> return 0;
>>>>> +release_mem:
>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>> free_rproc:
>>>>> rproc_free(rproc);
>>>>> return ret;
>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>> platform_device *pdev)
>>>>> rproc_del(rproc);
>>>>> rproc_free(rproc);
>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>> return 0;
>>>>> }
>>>>>
>>>
>>> --
>
> --
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 19:55 ` Suman Anna
0 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 19:55 UTC (permalink / raw)
To: Tero Kristo, Andrew F. Davis, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 1:42 PM, Tero Kristo wrote:
> On 30/01/2020 21:20, Andrew F. Davis wrote:
>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>
>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>> default in the driver core to avoid the lookup for every platform
>>>>> device and incur a penalty as the real users are expected to be
>>>>> only a few devices.
>>>>>
>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>> for each device at the moment to align on the location of the
>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>
>>>>
>>>> Same comment as before, this is a firmware issue for only some
>>>> firmwares
>>>> that do not handle being assigned vring locations correctly and instead
>>>> hard-code them.
As for this statement, this can do with some updating. Post 4.20,
because of the lazy allocation scheme used for carveouts including the
vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
have to wait for the vdev synchronization to happen.
>>>
>>> I believe we discussed this topic in length in previous version but
>>> there was no conclusion on it.
>>>
>>> The commit desc might be a bit misleading, we are not actually forced to
>>> use specific CMA buffers, as we use IOMMU to map these to device
>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>> addresses, iommu is used to map these to specific CMA buffer.
>>>
>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>> have. Not sure if there is any sane way to get this done in any other
>>> manner.
>>>
>>
>>
>> Why not use the default CMA area?
>
> I think using default CMA area getting the actual memory block is not
> guaranteed and might fail. There are other users for the memory, and it
> might get fragmented at the very late phase we are grabbing the memory
> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>
> I believe I could experiment with this a bit though and see, or Suman
> could maybe provide feedback why this was designed initially like this
> and why this would not be a good idea.
I have given some explanation on this on v4 as well, but if it is not
clear, there are restrictions with using default CMA. Default CMA has
switched to be assigned from the top of the memory (higher addresses,
since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
32-bits. So, we cannot blindly use the default CMA pool, and this will
definitely not work on boards > 2 GB RAM. And, if you want to add in any
firewall capability, then specific physical addresses becomes mandatory.
regards
Suman
>
> -Tero
>
>>
>> Andrew
>>
>>
>>> -Tero
>>>
>>>>
>>>> This is not a requirement of the remote processor itself and so it
>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>
>>>>
>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>> remoteproc device. Note that only one region per device is
>>>>> allowed by the framework.
>>>>>
>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>> ---
>>>>> v5: no changes
>>>>>
>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>> index 0846839b2c97..194303b860b2 100644
>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>> @@ -17,6 +17,7 @@
>>>>> #include <linux/module.h>
>>>>> #include <linux/err.h>
>>>>> #include <linux/of_device.h>
>>>>> +#include <linux/of_reserved_mem.h>
>>>>> #include <linux/platform_device.h>
>>>>> #include <linux/dma-mapping.h>
>>>>> #include <linux/remoteproc.h>
>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>> platform_device *pdev)
>>>>> if (ret)
>>>>> goto free_rproc;
>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>> + if (ret) {
>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>> pool\n");
>>>>> + goto free_rproc;
>>>>> + }
>>>>> +
>>>>> platform_set_drvdata(pdev, rproc);
>>>>> ret = rproc_add(rproc);
>>>>> if (ret)
>>>>> - goto free_rproc;
>>>>> + goto release_mem;
>>>>> return 0;
>>>>> +release_mem:
>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>> free_rproc:
>>>>> rproc_free(rproc);
>>>>> return ret;
>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>> platform_device *pdev)
>>>>> rproc_del(rproc);
>>>>> rproc_free(rproc);
>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>> return 0;
>>>>> }
>>>>>
>>>
>>> --
>
> --
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 19:55 ` Suman Anna
@ 2020-01-30 20:22 ` Andrew F. Davis
-1 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 20:22 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 2:55 PM, Suman Anna wrote:
> On 1/30/20 1:42 PM, Tero Kristo wrote:
>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>
>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>> device and incur a penalty as the real users are expected to be
>>>>>> only a few devices.
>>>>>>
>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>> for each device at the moment to align on the location of the
>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>
>>>>>
>>>>> Same comment as before, this is a firmware issue for only some
>>>>> firmwares
>>>>> that do not handle being assigned vring locations correctly and instead
>>>>> hard-code them.
>
> As for this statement, this can do with some updating. Post 4.20,
> because of the lazy allocation scheme used for carveouts including the
> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
> have to wait for the vdev synchronization to happen.
>
>>>>
>>>> I believe we discussed this topic in length in previous version but
>>>> there was no conclusion on it.
>>>>
>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>
>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>> have. Not sure if there is any sane way to get this done in any other
>>>> manner.
>>>>
>>>
>>>
>>> Why not use the default CMA area?
>>
>> I think using default CMA area getting the actual memory block is not
>> guaranteed and might fail. There are other users for the memory, and it
>> might get fragmented at the very late phase we are grabbing the memory
>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>
>> I believe I could experiment with this a bit though and see, or Suman
>> could maybe provide feedback why this was designed initially like this
>> and why this would not be a good idea.
>
> I have given some explanation on this on v4 as well, but if it is not
> clear, there are restrictions with using default CMA. Default CMA has
> switched to be assigned from the top of the memory (higher addresses,
> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
> 32-bits. So, we cannot blindly use the default CMA pool, and this will
> definitely not work on boards > 2 GB RAM. And, if you want to add in any
> firewall capability, then specific physical addresses becomes mandatory.
>
If you need 32bit range allocations then
dma_set_mask(dev, DMA_BIT_MASK(32));
I'm not saying don't have support for carveouts, just make them
optional, keystone_remoteproc.c does this:
if (of_reserved_mem_device_init(dev))
dev_warn(dev, "device does not have specific CMA pool\n");
There doesn't even needs to be a warning but that is up to you.
Andrew
> regards
> Suman
>
>>
>> -Tero
>>
>>>
>>> Andrew
>>>
>>>
>>>> -Tero
>>>>
>>>>>
>>>>> This is not a requirement of the remote processor itself and so it
>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>
>>>>>
>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>> remoteproc device. Note that only one region per device is
>>>>>> allowed by the framework.
>>>>>>
>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>> ---
>>>>>> v5: no changes
>>>>>>
>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>> @@ -17,6 +17,7 @@
>>>>>> #include <linux/module.h>
>>>>>> #include <linux/err.h>
>>>>>> #include <linux/of_device.h>
>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>> #include <linux/platform_device.h>
>>>>>> #include <linux/dma-mapping.h>
>>>>>> #include <linux/remoteproc.h>
>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>> platform_device *pdev)
>>>>>> if (ret)
>>>>>> goto free_rproc;
>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>> + if (ret) {
>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>> pool\n");
>>>>>> + goto free_rproc;
>>>>>> + }
>>>>>> +
>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>> ret = rproc_add(rproc);
>>>>>> if (ret)
>>>>>> - goto free_rproc;
>>>>>> + goto release_mem;
>>>>>> return 0;
>>>>>> +release_mem:
>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>> free_rproc:
>>>>>> rproc_free(rproc);
>>>>>> return ret;
>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>> platform_device *pdev)
>>>>>> rproc_del(rproc);
>>>>>> rproc_free(rproc);
>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>> return 0;
>>>>>> }
>>>>>>
>>>>
>>>> --
>>
>> --
>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 20:22 ` Andrew F. Davis
0 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 20:22 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 2:55 PM, Suman Anna wrote:
> On 1/30/20 1:42 PM, Tero Kristo wrote:
>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>
>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>> device and incur a penalty as the real users are expected to be
>>>>>> only a few devices.
>>>>>>
>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>> for each device at the moment to align on the location of the
>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>
>>>>>
>>>>> Same comment as before, this is a firmware issue for only some
>>>>> firmwares
>>>>> that do not handle being assigned vring locations correctly and instead
>>>>> hard-code them.
>
> As for this statement, this can do with some updating. Post 4.20,
> because of the lazy allocation scheme used for carveouts including the
> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
> have to wait for the vdev synchronization to happen.
>
>>>>
>>>> I believe we discussed this topic in length in previous version but
>>>> there was no conclusion on it.
>>>>
>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>
>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>> have. Not sure if there is any sane way to get this done in any other
>>>> manner.
>>>>
>>>
>>>
>>> Why not use the default CMA area?
>>
>> I think using default CMA area getting the actual memory block is not
>> guaranteed and might fail. There are other users for the memory, and it
>> might get fragmented at the very late phase we are grabbing the memory
>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>
>> I believe I could experiment with this a bit though and see, or Suman
>> could maybe provide feedback why this was designed initially like this
>> and why this would not be a good idea.
>
> I have given some explanation on this on v4 as well, but if it is not
> clear, there are restrictions with using default CMA. Default CMA has
> switched to be assigned from the top of the memory (higher addresses,
> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
> 32-bits. So, we cannot blindly use the default CMA pool, and this will
> definitely not work on boards > 2 GB RAM. And, if you want to add in any
> firewall capability, then specific physical addresses becomes mandatory.
>
If you need 32bit range allocations then
dma_set_mask(dev, DMA_BIT_MASK(32));
I'm not saying don't have support for carveouts, just make them
optional, keystone_remoteproc.c does this:
if (of_reserved_mem_device_init(dev))
dev_warn(dev, "device does not have specific CMA pool\n");
There doesn't even needs to be a warning but that is up to you.
Andrew
> regards
> Suman
>
>>
>> -Tero
>>
>>>
>>> Andrew
>>>
>>>
>>>> -Tero
>>>>
>>>>>
>>>>> This is not a requirement of the remote processor itself and so it
>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>
>>>>>
>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>> remoteproc device. Note that only one region per device is
>>>>>> allowed by the framework.
>>>>>>
>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>> ---
>>>>>> v5: no changes
>>>>>>
>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>> @@ -17,6 +17,7 @@
>>>>>> #include <linux/module.h>
>>>>>> #include <linux/err.h>
>>>>>> #include <linux/of_device.h>
>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>> #include <linux/platform_device.h>
>>>>>> #include <linux/dma-mapping.h>
>>>>>> #include <linux/remoteproc.h>
>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>> platform_device *pdev)
>>>>>> if (ret)
>>>>>> goto free_rproc;
>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>> + if (ret) {
>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>> pool\n");
>>>>>> + goto free_rproc;
>>>>>> + }
>>>>>> +
>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>> ret = rproc_add(rproc);
>>>>>> if (ret)
>>>>>> - goto free_rproc;
>>>>>> + goto release_mem;
>>>>>> return 0;
>>>>>> +release_mem:
>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>> free_rproc:
>>>>>> rproc_free(rproc);
>>>>>> return ret;
>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>> platform_device *pdev)
>>>>>> rproc_del(rproc);
>>>>>> rproc_free(rproc);
>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>> return 0;
>>>>>> }
>>>>>>
>>>>
>>>> --
>>
>> --
>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 20:22 ` Andrew F. Davis
@ 2020-01-30 20:39 ` Suman Anna
-1 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 20:39 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 2:22 PM, Andrew F. Davis wrote:
> On 1/30/20 2:55 PM, Suman Anna wrote:
>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>
>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>> only a few devices.
>>>>>>>
>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>> for each device at the moment to align on the location of the
>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>
>>>>>>
>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>> firmwares
>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>> hard-code them.
>>
>> As for this statement, this can do with some updating. Post 4.20,
>> because of the lazy allocation scheme used for carveouts including the
>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>> have to wait for the vdev synchronization to happen.
>>
>>>>>
>>>>> I believe we discussed this topic in length in previous version but
>>>>> there was no conclusion on it.
>>>>>
>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>
>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>> manner.
>>>>>
>>>>
>>>>
>>>> Why not use the default CMA area?
>>>
>>> I think using default CMA area getting the actual memory block is not
>>> guaranteed and might fail. There are other users for the memory, and it
>>> might get fragmented at the very late phase we are grabbing the memory
>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>
>>> I believe I could experiment with this a bit though and see, or Suman
>>> could maybe provide feedback why this was designed initially like this
>>> and why this would not be a good idea.
>>
>> I have given some explanation on this on v4 as well, but if it is not
>> clear, there are restrictions with using default CMA. Default CMA has
>> switched to be assigned from the top of the memory (higher addresses,
>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>> firewall capability, then specific physical addresses becomes mandatory.
>>
>
>
> If you need 32bit range allocations then
> dma_set_mask(dev, DMA_BIT_MASK(32));
>
> I'm not saying don't have support for carveouts, just make them
> optional, keystone_remoteproc.c does this:
>
> if (of_reserved_mem_device_init(dev))
> dev_warn(dev, "device does not have specific CMA pool\n");
>
> There doesn't even needs to be a warning but that is up to you.
It is not exactly an apples to apples comparison. K2s do not have MMUs,
and most of our firmware images on K2 are actually running out of the
DSP internal memory.
regards
Suman
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> -Tero
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> -Tero
>>>>>
>>>>>>
>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>
>>>>>>
>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>> allowed by the framework.
>>>>>>>
>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>> ---
>>>>>>> v5: no changes
>>>>>>>
>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>> @@ -17,6 +17,7 @@
>>>>>>> #include <linux/module.h>
>>>>>>> #include <linux/err.h>
>>>>>>> #include <linux/of_device.h>
>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>> #include <linux/platform_device.h>
>>>>>>> #include <linux/dma-mapping.h>
>>>>>>> #include <linux/remoteproc.h>
>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>> platform_device *pdev)
>>>>>>> if (ret)
>>>>>>> goto free_rproc;
>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>> + if (ret) {
>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>> pool\n");
>>>>>>> + goto free_rproc;
>>>>>>> + }
>>>>>>> +
>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>> ret = rproc_add(rproc);
>>>>>>> if (ret)
>>>>>>> - goto free_rproc;
>>>>>>> + goto release_mem;
>>>>>>> return 0;
>>>>>>> +release_mem:
>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>> free_rproc:
>>>>>>> rproc_free(rproc);
>>>>>>> return ret;
>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>> platform_device *pdev)
>>>>>>> rproc_del(rproc);
>>>>>>> rproc_free(rproc);
>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>> return 0;
>>>>>>> }
>>>>>>>
>>>>>
>>>>> --
>>>
>>> --
>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 20:39 ` Suman Anna
0 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 20:39 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 2:22 PM, Andrew F. Davis wrote:
> On 1/30/20 2:55 PM, Suman Anna wrote:
>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>
>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>> only a few devices.
>>>>>>>
>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>> for each device at the moment to align on the location of the
>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>
>>>>>>
>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>> firmwares
>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>> hard-code them.
>>
>> As for this statement, this can do with some updating. Post 4.20,
>> because of the lazy allocation scheme used for carveouts including the
>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>> have to wait for the vdev synchronization to happen.
>>
>>>>>
>>>>> I believe we discussed this topic in length in previous version but
>>>>> there was no conclusion on it.
>>>>>
>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>
>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>> manner.
>>>>>
>>>>
>>>>
>>>> Why not use the default CMA area?
>>>
>>> I think using default CMA area getting the actual memory block is not
>>> guaranteed and might fail. There are other users for the memory, and it
>>> might get fragmented at the very late phase we are grabbing the memory
>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>
>>> I believe I could experiment with this a bit though and see, or Suman
>>> could maybe provide feedback why this was designed initially like this
>>> and why this would not be a good idea.
>>
>> I have given some explanation on this on v4 as well, but if it is not
>> clear, there are restrictions with using default CMA. Default CMA has
>> switched to be assigned from the top of the memory (higher addresses,
>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>> firewall capability, then specific physical addresses becomes mandatory.
>>
>
>
> If you need 32bit range allocations then
> dma_set_mask(dev, DMA_BIT_MASK(32));
>
> I'm not saying don't have support for carveouts, just make them
> optional, keystone_remoteproc.c does this:
>
> if (of_reserved_mem_device_init(dev))
> dev_warn(dev, "device does not have specific CMA pool\n");
>
> There doesn't even needs to be a warning but that is up to you.
It is not exactly an apples to apples comparison. K2s do not have MMUs,
and most of our firmware images on K2 are actually running out of the
DSP internal memory.
regards
Suman
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> -Tero
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> -Tero
>>>>>
>>>>>>
>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>
>>>>>>
>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>> allowed by the framework.
>>>>>>>
>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>> ---
>>>>>>> v5: no changes
>>>>>>>
>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>> @@ -17,6 +17,7 @@
>>>>>>> #include <linux/module.h>
>>>>>>> #include <linux/err.h>
>>>>>>> #include <linux/of_device.h>
>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>> #include <linux/platform_device.h>
>>>>>>> #include <linux/dma-mapping.h>
>>>>>>> #include <linux/remoteproc.h>
>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>> platform_device *pdev)
>>>>>>> if (ret)
>>>>>>> goto free_rproc;
>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>> + if (ret) {
>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>> pool\n");
>>>>>>> + goto free_rproc;
>>>>>>> + }
>>>>>>> +
>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>> ret = rproc_add(rproc);
>>>>>>> if (ret)
>>>>>>> - goto free_rproc;
>>>>>>> + goto release_mem;
>>>>>>> return 0;
>>>>>>> +release_mem:
>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>> free_rproc:
>>>>>>> rproc_free(rproc);
>>>>>>> return ret;
>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>> platform_device *pdev)
>>>>>>> rproc_del(rproc);
>>>>>>> rproc_free(rproc);
>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>> return 0;
>>>>>>> }
>>>>>>>
>>>>>
>>>>> --
>>>
>>> --
>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 20:39 ` Suman Anna
@ 2020-01-30 21:19 ` Andrew F. Davis
-1 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 21:19 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:39 PM, Suman Anna wrote:
> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>
>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>> only a few devices.
>>>>>>>>
>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>
>>>>>>>
>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>> firmwares
>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>> hard-code them.
>>>
>>> As for this statement, this can do with some updating. Post 4.20,
>>> because of the lazy allocation scheme used for carveouts including the
>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>> have to wait for the vdev synchronization to happen.
>>>
>>>>>>
>>>>>> I believe we discussed this topic in length in previous version but
>>>>>> there was no conclusion on it.
>>>>>>
>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>
>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>> manner.
>>>>>>
>>>>>
>>>>>
>>>>> Why not use the default CMA area?
>>>>
>>>> I think using default CMA area getting the actual memory block is not
>>>> guaranteed and might fail. There are other users for the memory, and it
>>>> might get fragmented at the very late phase we are grabbing the memory
>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>
>>>> I believe I could experiment with this a bit though and see, or Suman
>>>> could maybe provide feedback why this was designed initially like this
>>>> and why this would not be a good idea.
>>>
>>> I have given some explanation on this on v4 as well, but if it is not
>>> clear, there are restrictions with using default CMA. Default CMA has
>>> switched to be assigned from the top of the memory (higher addresses,
>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>> firewall capability, then specific physical addresses becomes mandatory.
>>>
>>
>>
>> If you need 32bit range allocations then
>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>
>> I'm not saying don't have support for carveouts, just make them
>> optional, keystone_remoteproc.c does this:
>>
>> if (of_reserved_mem_device_init(dev))
>> dev_warn(dev, "device does not have specific CMA pool\n");
>>
>> There doesn't even needs to be a warning but that is up to you.
>
> It is not exactly an apples to apples comparison. K2s do not have MMUs,
> and most of our firmware images on K2 are actually running out of the
> DSP internal memory.
>
So again we circle back to it being a firmware issue, if K2 can get away
without needing carveouts and it doesn't even have an MMU then certainly
OMAP/DRA7x class devices can handle it even better given they *do* have
an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
which case we are still just hacking around the problem here with
mandatory specific address memory carveouts.
Andrew
> regards
> Suman
>
>>
>> Andrew
>>
>>
>>> regards
>>> Suman
>>>
>>>>
>>>> -Tero
>>>>
>>>>>
>>>>> Andrew
>>>>>
>>>>>
>>>>>> -Tero
>>>>>>
>>>>>>>
>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>
>>>>>>>
>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>> allowed by the framework.
>>>>>>>>
>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>> ---
>>>>>>>> v5: no changes
>>>>>>>>
>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>> #include <linux/module.h>
>>>>>>>> #include <linux/err.h>
>>>>>>>> #include <linux/of_device.h>
>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>> #include <linux/platform_device.h>
>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>> platform_device *pdev)
>>>>>>>> if (ret)
>>>>>>>> goto free_rproc;
>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>> + if (ret) {
>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>> pool\n");
>>>>>>>> + goto free_rproc;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>> ret = rproc_add(rproc);
>>>>>>>> if (ret)
>>>>>>>> - goto free_rproc;
>>>>>>>> + goto release_mem;
>>>>>>>> return 0;
>>>>>>>> +release_mem:
>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>> free_rproc:
>>>>>>>> rproc_free(rproc);
>>>>>>>> return ret;
>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>> platform_device *pdev)
>>>>>>>> rproc_del(rproc);
>>>>>>>> rproc_free(rproc);
>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>> return 0;
>>>>>>>> }
>>>>>>>>
>>>>>>
>>>>>> --
>>>>
>>>> --
>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 21:19 ` Andrew F. Davis
0 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 21:19 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:39 PM, Suman Anna wrote:
> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>
>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>> only a few devices.
>>>>>>>>
>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>
>>>>>>>
>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>> firmwares
>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>> hard-code them.
>>>
>>> As for this statement, this can do with some updating. Post 4.20,
>>> because of the lazy allocation scheme used for carveouts including the
>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>> have to wait for the vdev synchronization to happen.
>>>
>>>>>>
>>>>>> I believe we discussed this topic in length in previous version but
>>>>>> there was no conclusion on it.
>>>>>>
>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>
>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>> manner.
>>>>>>
>>>>>
>>>>>
>>>>> Why not use the default CMA area?
>>>>
>>>> I think using default CMA area getting the actual memory block is not
>>>> guaranteed and might fail. There are other users for the memory, and it
>>>> might get fragmented at the very late phase we are grabbing the memory
>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>
>>>> I believe I could experiment with this a bit though and see, or Suman
>>>> could maybe provide feedback why this was designed initially like this
>>>> and why this would not be a good idea.
>>>
>>> I have given some explanation on this on v4 as well, but if it is not
>>> clear, there are restrictions with using default CMA. Default CMA has
>>> switched to be assigned from the top of the memory (higher addresses,
>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>> firewall capability, then specific physical addresses becomes mandatory.
>>>
>>
>>
>> If you need 32bit range allocations then
>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>
>> I'm not saying don't have support for carveouts, just make them
>> optional, keystone_remoteproc.c does this:
>>
>> if (of_reserved_mem_device_init(dev))
>> dev_warn(dev, "device does not have specific CMA pool\n");
>>
>> There doesn't even needs to be a warning but that is up to you.
>
> It is not exactly an apples to apples comparison. K2s do not have MMUs,
> and most of our firmware images on K2 are actually running out of the
> DSP internal memory.
>
So again we circle back to it being a firmware issue, if K2 can get away
without needing carveouts and it doesn't even have an MMU then certainly
OMAP/DRA7x class devices can handle it even better given they *do* have
an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
which case we are still just hacking around the problem here with
mandatory specific address memory carveouts.
Andrew
> regards
> Suman
>
>>
>> Andrew
>>
>>
>>> regards
>>> Suman
>>>
>>>>
>>>> -Tero
>>>>
>>>>>
>>>>> Andrew
>>>>>
>>>>>
>>>>>> -Tero
>>>>>>
>>>>>>>
>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>
>>>>>>>
>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>> allowed by the framework.
>>>>>>>>
>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>> ---
>>>>>>>> v5: no changes
>>>>>>>>
>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>> #include <linux/module.h>
>>>>>>>> #include <linux/err.h>
>>>>>>>> #include <linux/of_device.h>
>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>> #include <linux/platform_device.h>
>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>> platform_device *pdev)
>>>>>>>> if (ret)
>>>>>>>> goto free_rproc;
>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>> + if (ret) {
>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>> pool\n");
>>>>>>>> + goto free_rproc;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>> ret = rproc_add(rproc);
>>>>>>>> if (ret)
>>>>>>>> - goto free_rproc;
>>>>>>>> + goto release_mem;
>>>>>>>> return 0;
>>>>>>>> +release_mem:
>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>> free_rproc:
>>>>>>>> rproc_free(rproc);
>>>>>>>> return ret;
>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>> platform_device *pdev)
>>>>>>>> rproc_del(rproc);
>>>>>>>> rproc_free(rproc);
>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>> return 0;
>>>>>>>> }
>>>>>>>>
>>>>>>
>>>>>> --
>>>>
>>>> --
>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 21:19 ` Andrew F. Davis
@ 2020-01-30 21:39 ` Suman Anna
-1 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 21:39 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:19 PM, Andrew F. Davis wrote:
> On 1/30/20 3:39 PM, Suman Anna wrote:
>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>
>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>> only a few devices.
>>>>>>>>>
>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>
>>>>>>>>
>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>> firmwares
>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>> hard-code them.
>>>>
>>>> As for this statement, this can do with some updating. Post 4.20,
>>>> because of the lazy allocation scheme used for carveouts including the
>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>> have to wait for the vdev synchronization to happen.
>>>>
>>>>>>>
>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>> there was no conclusion on it.
>>>>>>>
>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>
>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>> manner.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> Why not use the default CMA area?
>>>>>
>>>>> I think using default CMA area getting the actual memory block is not
>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>
>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>> could maybe provide feedback why this was designed initially like this
>>>>> and why this would not be a good idea.
>>>>
>>>> I have given some explanation on this on v4 as well, but if it is not
>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>> switched to be assigned from the top of the memory (higher addresses,
>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>
>>>
>>>
>>> If you need 32bit range allocations then
>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>
>>> I'm not saying don't have support for carveouts, just make them
>>> optional, keystone_remoteproc.c does this:
>>>
>>> if (of_reserved_mem_device_init(dev))
>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>
>>> There doesn't even needs to be a warning but that is up to you.
>>
>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>> and most of our firmware images on K2 are actually running out of the
>> DSP internal memory.
>>
>
>
> So again we circle back to it being a firmware issue, if K2 can get away
> without needing carveouts and it doesn't even have an MMU then certainly
> OMAP/DRA7x class devices can handle it even better given they *do* have
> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
> which case we are still just hacking around the problem here with
> mandatory specific address memory carveouts.
Optional carveouts on OMAP remoteprocs can be an enhancement in the
future, but at the moment, we won't be able to run use-cases without
this. And I have already given some of the reasons for the same here and
on v4.
regards
Suman
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> Andrew
>>>
>>>
>>>> regards
>>>> Suman
>>>>
>>>>>
>>>>> -Tero
>>>>>
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>>
>>>>>>> -Tero
>>>>>>>
>>>>>>>>
>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>
>>>>>>>>
>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>> allowed by the framework.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>> ---
>>>>>>>>> v5: no changes
>>>>>>>>>
>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>
>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>> #include <linux/module.h>
>>>>>>>>> #include <linux/err.h>
>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>> platform_device *pdev)
>>>>>>>>> if (ret)
>>>>>>>>> goto free_rproc;
>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>> + if (ret) {
>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>> pool\n");
>>>>>>>>> + goto free_rproc;
>>>>>>>>> + }
>>>>>>>>> +
>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>> if (ret)
>>>>>>>>> - goto free_rproc;
>>>>>>>>> + goto release_mem;
>>>>>>>>> return 0;
>>>>>>>>> +release_mem:
>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>> free_rproc:
>>>>>>>>> rproc_free(rproc);
>>>>>>>>> return ret;
>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>> platform_device *pdev)
>>>>>>>>> rproc_del(rproc);
>>>>>>>>> rproc_free(rproc);
>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>> return 0;
>>>>>>>>> }
>>>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>
>>>>> --
>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>
>>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 21:39 ` Suman Anna
0 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 21:39 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:19 PM, Andrew F. Davis wrote:
> On 1/30/20 3:39 PM, Suman Anna wrote:
>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>
>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>> only a few devices.
>>>>>>>>>
>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>
>>>>>>>>
>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>> firmwares
>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>> hard-code them.
>>>>
>>>> As for this statement, this can do with some updating. Post 4.20,
>>>> because of the lazy allocation scheme used for carveouts including the
>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>> have to wait for the vdev synchronization to happen.
>>>>
>>>>>>>
>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>> there was no conclusion on it.
>>>>>>>
>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>
>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>> manner.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> Why not use the default CMA area?
>>>>>
>>>>> I think using default CMA area getting the actual memory block is not
>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>
>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>> could maybe provide feedback why this was designed initially like this
>>>>> and why this would not be a good idea.
>>>>
>>>> I have given some explanation on this on v4 as well, but if it is not
>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>> switched to be assigned from the top of the memory (higher addresses,
>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>
>>>
>>>
>>> If you need 32bit range allocations then
>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>
>>> I'm not saying don't have support for carveouts, just make them
>>> optional, keystone_remoteproc.c does this:
>>>
>>> if (of_reserved_mem_device_init(dev))
>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>
>>> There doesn't even needs to be a warning but that is up to you.
>>
>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>> and most of our firmware images on K2 are actually running out of the
>> DSP internal memory.
>>
>
>
> So again we circle back to it being a firmware issue, if K2 can get away
> without needing carveouts and it doesn't even have an MMU then certainly
> OMAP/DRA7x class devices can handle it even better given they *do* have
> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
> which case we are still just hacking around the problem here with
> mandatory specific address memory carveouts.
Optional carveouts on OMAP remoteprocs can be an enhancement in the
future, but at the moment, we won't be able to run use-cases without
this. And I have already given some of the reasons for the same here and
on v4.
regards
Suman
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> Andrew
>>>
>>>
>>>> regards
>>>> Suman
>>>>
>>>>>
>>>>> -Tero
>>>>>
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>>
>>>>>>> -Tero
>>>>>>>
>>>>>>>>
>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>
>>>>>>>>
>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>> allowed by the framework.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>> ---
>>>>>>>>> v5: no changes
>>>>>>>>>
>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>
>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>> #include <linux/module.h>
>>>>>>>>> #include <linux/err.h>
>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>> platform_device *pdev)
>>>>>>>>> if (ret)
>>>>>>>>> goto free_rproc;
>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>> + if (ret) {
>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>> pool\n");
>>>>>>>>> + goto free_rproc;
>>>>>>>>> + }
>>>>>>>>> +
>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>> if (ret)
>>>>>>>>> - goto free_rproc;
>>>>>>>>> + goto release_mem;
>>>>>>>>> return 0;
>>>>>>>>> +release_mem:
>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>> free_rproc:
>>>>>>>>> rproc_free(rproc);
>>>>>>>>> return ret;
>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>> platform_device *pdev)
>>>>>>>>> rproc_del(rproc);
>>>>>>>>> rproc_free(rproc);
>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>> return 0;
>>>>>>>>> }
>>>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>
>>>>> --
>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>
>>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 21:39 ` Suman Anna
@ 2020-01-30 21:50 ` Andrew F. Davis
-1 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 21:50 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 4:39 PM, Suman Anna wrote:
> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>
>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>> only a few devices.
>>>>>>>>>>
>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>> firmwares
>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>> hard-code them.
>>>>>
>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>> have to wait for the vdev synchronization to happen.
>>>>>
>>>>>>>>
>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>> there was no conclusion on it.
>>>>>>>>
>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>
>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>> manner.
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Why not use the default CMA area?
>>>>>>
>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>
>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>> and why this would not be a good idea.
>>>>>
>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>
>>>>
>>>>
>>>> If you need 32bit range allocations then
>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>
>>>> I'm not saying don't have support for carveouts, just make them
>>>> optional, keystone_remoteproc.c does this:
>>>>
>>>> if (of_reserved_mem_device_init(dev))
>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>
>>>> There doesn't even needs to be a warning but that is up to you.
>>>
>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>> and most of our firmware images on K2 are actually running out of the
>>> DSP internal memory.
>>>
>>
>>
>> So again we circle back to it being a firmware issue, if K2 can get away
>> without needing carveouts and it doesn't even have an MMU then certainly
>> OMAP/DRA7x class devices can handle it even better given they *do* have
>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>> which case we are still just hacking around the problem here with
>> mandatory specific address memory carveouts.
>
> Optional carveouts on OMAP remoteprocs can be an enhancement in the
> future, but at the moment, we won't be able to run use-cases without
> this. And I have already given some of the reasons for the same here and
> on v4.
>
No reason to be dismissive, my questions are valid.
What "use-cases" are we talking about, I have firmware that doesn't need
specific carved-out addresses. If you have misbehaving firmware that
needs statically carved out memory addresses then you can have carveouts
if you want, but it should be optional. If I don't want to pollute my
system's memory space with a bunch of carveout holes then I shouldn't
have to just because your specific firmware needs them.
Andrew
> regards
> Suman
>
>>
>> Andrew
>>
>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> regards
>>>>> Suman
>>>>>
>>>>>>
>>>>>> -Tero
>>>>>>
>>>>>>>
>>>>>>> Andrew
>>>>>>>
>>>>>>>
>>>>>>>> -Tero
>>>>>>>>
>>>>>>>>>
>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>> allowed by the framework.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>> ---
>>>>>>>>>> v5: no changes
>>>>>>>>>>
>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>
>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>> platform_device *pdev)
>>>>>>>>>> if (ret)
>>>>>>>>>> goto free_rproc;
>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>> + if (ret) {
>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>> pool\n");
>>>>>>>>>> + goto free_rproc;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>> if (ret)
>>>>>>>>>> - goto free_rproc;
>>>>>>>>>> + goto release_mem;
>>>>>>>>>> return 0;
>>>>>>>>>> +release_mem:
>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>> free_rproc:
>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>> return ret;
>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>> platform_device *pdev)
>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>> return 0;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>
>>>>>> --
>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>
>>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 21:50 ` Andrew F. Davis
0 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 21:50 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 4:39 PM, Suman Anna wrote:
> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>
>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>> only a few devices.
>>>>>>>>>>
>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>> firmwares
>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>> hard-code them.
>>>>>
>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>> have to wait for the vdev synchronization to happen.
>>>>>
>>>>>>>>
>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>> there was no conclusion on it.
>>>>>>>>
>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>
>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>> manner.
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Why not use the default CMA area?
>>>>>>
>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>
>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>> and why this would not be a good idea.
>>>>>
>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>
>>>>
>>>>
>>>> If you need 32bit range allocations then
>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>
>>>> I'm not saying don't have support for carveouts, just make them
>>>> optional, keystone_remoteproc.c does this:
>>>>
>>>> if (of_reserved_mem_device_init(dev))
>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>
>>>> There doesn't even needs to be a warning but that is up to you.
>>>
>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>> and most of our firmware images on K2 are actually running out of the
>>> DSP internal memory.
>>>
>>
>>
>> So again we circle back to it being a firmware issue, if K2 can get away
>> without needing carveouts and it doesn't even have an MMU then certainly
>> OMAP/DRA7x class devices can handle it even better given they *do* have
>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>> which case we are still just hacking around the problem here with
>> mandatory specific address memory carveouts.
>
> Optional carveouts on OMAP remoteprocs can be an enhancement in the
> future, but at the moment, we won't be able to run use-cases without
> this. And I have already given some of the reasons for the same here and
> on v4.
>
No reason to be dismissive, my questions are valid.
What "use-cases" are we talking about, I have firmware that doesn't need
specific carved-out addresses. If you have misbehaving firmware that
needs statically carved out memory addresses then you can have carveouts
if you want, but it should be optional. If I don't want to pollute my
system's memory space with a bunch of carveout holes then I shouldn't
have to just because your specific firmware needs them.
Andrew
> regards
> Suman
>
>>
>> Andrew
>>
>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> regards
>>>>> Suman
>>>>>
>>>>>>
>>>>>> -Tero
>>>>>>
>>>>>>>
>>>>>>> Andrew
>>>>>>>
>>>>>>>
>>>>>>>> -Tero
>>>>>>>>
>>>>>>>>>
>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>> allowed by the framework.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>> ---
>>>>>>>>>> v5: no changes
>>>>>>>>>>
>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>
>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>> platform_device *pdev)
>>>>>>>>>> if (ret)
>>>>>>>>>> goto free_rproc;
>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>> + if (ret) {
>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>> pool\n");
>>>>>>>>>> + goto free_rproc;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>> if (ret)
>>>>>>>>>> - goto free_rproc;
>>>>>>>>>> + goto release_mem;
>>>>>>>>>> return 0;
>>>>>>>>>> +release_mem:
>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>> free_rproc:
>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>> return ret;
>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>> platform_device *pdev)
>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>> return 0;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>
>>>>>> --
>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>
>>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 21:50 ` Andrew F. Davis
@ 2020-01-30 21:57 ` Suman Anna
-1 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 21:57 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:50 PM, Andrew F. Davis wrote:
> On 1/30/20 4:39 PM, Suman Anna wrote:
>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>
>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>> only a few devices.
>>>>>>>>>>>
>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>> firmwares
>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>> hard-code them.
>>>>>>
>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>
>>>>>>>>>
>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>> there was no conclusion on it.
>>>>>>>>>
>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>
>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>> manner.
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Why not use the default CMA area?
>>>>>>>
>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>
>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>> and why this would not be a good idea.
>>>>>>
>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>
>>>>>
>>>>>
>>>>> If you need 32bit range allocations then
>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>
>>>>> I'm not saying don't have support for carveouts, just make them
>>>>> optional, keystone_remoteproc.c does this:
>>>>>
>>>>> if (of_reserved_mem_device_init(dev))
>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>
>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>
>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>> and most of our firmware images on K2 are actually running out of the
>>>> DSP internal memory.
>>>>
>>>
>>>
>>> So again we circle back to it being a firmware issue, if K2 can get away
>>> without needing carveouts and it doesn't even have an MMU then certainly
>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>> which case we are still just hacking around the problem here with
>>> mandatory specific address memory carveouts.
>>
>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>> future, but at the moment, we won't be able to run use-cases without
>> this. And I have already given some of the reasons for the same here and
>> on v4.
>>
>
>
> No reason to be dismissive, my questions are valid.
>
> What "use-cases" are we talking about, I have firmware that doesn't need
> specific carved-out addresses.
I think you are well aware of all the usecases we provide with the TI
SDKs with IPUs and DSPs. And what is the firmware that you have and what
do you use it for?
If you have misbehaving firmware that
> needs statically carved out memory addresses then you can have carveouts
> if you want, but it should be optional.
If I don't want to pollute my
> system's memory space with a bunch of carveout holes then I shouldn't
> have to just because your specific firmware needs them.
Further follow-up series like early-boot and late-attach will mandate
fixed carveouts actually. You cannot just run out of any random memory.
regards
Suman
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> Andrew
>>>
>>>
>>>> regards
>>>> Suman
>>>>
>>>>>
>>>>> Andrew
>>>>>
>>>>>
>>>>>> regards
>>>>>> Suman
>>>>>>
>>>>>>>
>>>>>>> -Tero
>>>>>>>
>>>>>>>>
>>>>>>>> Andrew
>>>>>>>>
>>>>>>>>
>>>>>>>>> -Tero
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>> ---
>>>>>>>>>>> v5: no changes
>>>>>>>>>>>
>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>> if (ret)
>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>> + if (ret) {
>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>> pool\n");
>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>> + }
>>>>>>>>>>> +
>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>> if (ret)
>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>> return 0;
>>>>>>>>>>> +release_mem:
>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>> free_rproc:
>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>> return ret;
>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>> return 0;
>>>>>>>>>>> }
>>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>
>>>>>>> --
>>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>>
>>>>
>>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 21:57 ` Suman Anna
0 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 21:57 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:50 PM, Andrew F. Davis wrote:
> On 1/30/20 4:39 PM, Suman Anna wrote:
>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>
>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>> only a few devices.
>>>>>>>>>>>
>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>> firmwares
>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>> hard-code them.
>>>>>>
>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>
>>>>>>>>>
>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>> there was no conclusion on it.
>>>>>>>>>
>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>
>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>> manner.
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Why not use the default CMA area?
>>>>>>>
>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>
>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>> and why this would not be a good idea.
>>>>>>
>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>
>>>>>
>>>>>
>>>>> If you need 32bit range allocations then
>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>
>>>>> I'm not saying don't have support for carveouts, just make them
>>>>> optional, keystone_remoteproc.c does this:
>>>>>
>>>>> if (of_reserved_mem_device_init(dev))
>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>
>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>
>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>> and most of our firmware images on K2 are actually running out of the
>>>> DSP internal memory.
>>>>
>>>
>>>
>>> So again we circle back to it being a firmware issue, if K2 can get away
>>> without needing carveouts and it doesn't even have an MMU then certainly
>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>> which case we are still just hacking around the problem here with
>>> mandatory specific address memory carveouts.
>>
>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>> future, but at the moment, we won't be able to run use-cases without
>> this. And I have already given some of the reasons for the same here and
>> on v4.
>>
>
>
> No reason to be dismissive, my questions are valid.
>
> What "use-cases" are we talking about, I have firmware that doesn't need
> specific carved-out addresses.
I think you are well aware of all the usecases we provide with the TI
SDKs with IPUs and DSPs. And what is the firmware that you have and what
do you use it for?
If you have misbehaving firmware that
> needs statically carved out memory addresses then you can have carveouts
> if you want, but it should be optional.
If I don't want to pollute my
> system's memory space with a bunch of carveout holes then I shouldn't
> have to just because your specific firmware needs them.
Further follow-up series like early-boot and late-attach will mandate
fixed carveouts actually. You cannot just run out of any random memory.
regards
Suman
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> Andrew
>>>
>>>
>>>> regards
>>>> Suman
>>>>
>>>>>
>>>>> Andrew
>>>>>
>>>>>
>>>>>> regards
>>>>>> Suman
>>>>>>
>>>>>>>
>>>>>>> -Tero
>>>>>>>
>>>>>>>>
>>>>>>>> Andrew
>>>>>>>>
>>>>>>>>
>>>>>>>>> -Tero
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>> ---
>>>>>>>>>>> v5: no changes
>>>>>>>>>>>
>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>> if (ret)
>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>> + if (ret) {
>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>> pool\n");
>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>> + }
>>>>>>>>>>> +
>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>> if (ret)
>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>> return 0;
>>>>>>>>>>> +release_mem:
>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>> free_rproc:
>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>> return ret;
>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>> return 0;
>>>>>>>>>>> }
>>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>
>>>>>>> --
>>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>>
>>>>
>>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 21:57 ` Suman Anna
@ 2020-01-30 22:06 ` Suman Anna
-1 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 22:06 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:57 PM, Suman Anna wrote:
> On 1/30/20 3:50 PM, Andrew F. Davis wrote:
>> On 1/30/20 4:39 PM, Suman Anna wrote:
>>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>
>>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>>> only a few devices.
>>>>>>>>>>>>
>>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>>> firmwares
>>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>>> hard-code them.
>>>>>>>
>>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>>
>>>>>>>>>>
>>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>>> there was no conclusion on it.
>>>>>>>>>>
>>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>>
>>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>>> manner.
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Why not use the default CMA area?
>>>>>>>>
>>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>>
>>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>>> and why this would not be a good idea.
>>>>>>>
>>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> If you need 32bit range allocations then
>>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>>
>>>>>> I'm not saying don't have support for carveouts, just make them
>>>>>> optional, keystone_remoteproc.c does this:
>>>>>>
>>>>>> if (of_reserved_mem_device_init(dev))
>>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>>
>>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>>
>>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>>> and most of our firmware images on K2 are actually running out of the
>>>>> DSP internal memory.
>>>>>
>>>>
>>>>
>>>> So again we circle back to it being a firmware issue, if K2 can get away
>>>> without needing carveouts and it doesn't even have an MMU then certainly
>>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>>> which case we are still just hacking around the problem here with
>>>> mandatory specific address memory carveouts.
>>>
>>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>>> future, but at the moment, we won't be able to run use-cases without
>>> this. And I have already given some of the reasons for the same here and
>>> on v4.
>>>
>>
>>
>> No reason to be dismissive, my questions are valid.
>>
>> What "use-cases" are we talking about, I have firmware that doesn't need
>> specific carved-out addresses.
>
> I think you are well aware of all the usecases we provide with the TI
> SDKs with IPUs and DSPs. And what is the firmware that you have and what
> do you use it for?
>
> If you have misbehaving firmware that
>> needs statically carved out memory addresses then you can have carveouts
>> if you want, but it should be optional.
> If I don't want to pollute my
>> system's memory space with a bunch of carveout holes then I shouldn't
>> have to just because your specific firmware needs them.
>
> Further follow-up series like early-boot and late-attach will mandate
> fixed carveouts actually. You cannot just run out of any random memory.
Also, these are CMA pools ("reusable"), so they are not actual carveout
holes ("no-map"). This is the preferred method in remoteproc mode so
that the memory is available for kernel when remoteprocs are not in use.
Customers can always choose to make these carveouts so that they do not
run into memory allocation issues when changing firmwares and under
stress conditions. These will have to be carveouts for early-boot usecases.
regards
Suman
>
> regards
> Suman
>
>>
>> Andrew
>>
>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> regards
>>>>> Suman
>>>>>
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>>
>>>>>>> regards
>>>>>>> Suman
>>>>>>>
>>>>>>>>
>>>>>>>> -Tero
>>>>>>>>
>>>>>>>>>
>>>>>>>>> Andrew
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> -Tero
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>>
>>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>>> ---
>>>>>>>>>>>> v5: no changes
>>>>>>>>>>>>
>>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>>
>>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>> if (ret)
>>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>>> + if (ret) {
>>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>>> pool\n");
>>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>>> + }
>>>>>>>>>>>> +
>>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>>> if (ret)
>>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>>> return 0;
>>>>>>>>>>>> +release_mem:
>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>> free_rproc:
>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>> return ret;
>>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>> return 0;
>>>>>>>>>>>> }
>>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>
>>>>>>>> --
>>>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>>>
>>>>>
>>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 22:06 ` Suman Anna
0 siblings, 0 replies; 62+ messages in thread
From: Suman Anna @ 2020-01-30 22:06 UTC (permalink / raw)
To: Andrew F. Davis, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 3:57 PM, Suman Anna wrote:
> On 1/30/20 3:50 PM, Andrew F. Davis wrote:
>> On 1/30/20 4:39 PM, Suman Anna wrote:
>>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>
>>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>>> only a few devices.
>>>>>>>>>>>>
>>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>>> firmwares
>>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>>> hard-code them.
>>>>>>>
>>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>>
>>>>>>>>>>
>>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>>> there was no conclusion on it.
>>>>>>>>>>
>>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>>
>>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>>> manner.
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Why not use the default CMA area?
>>>>>>>>
>>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>>
>>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>>> and why this would not be a good idea.
>>>>>>>
>>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> If you need 32bit range allocations then
>>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>>
>>>>>> I'm not saying don't have support for carveouts, just make them
>>>>>> optional, keystone_remoteproc.c does this:
>>>>>>
>>>>>> if (of_reserved_mem_device_init(dev))
>>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>>
>>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>>
>>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>>> and most of our firmware images on K2 are actually running out of the
>>>>> DSP internal memory.
>>>>>
>>>>
>>>>
>>>> So again we circle back to it being a firmware issue, if K2 can get away
>>>> without needing carveouts and it doesn't even have an MMU then certainly
>>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>>> which case we are still just hacking around the problem here with
>>>> mandatory specific address memory carveouts.
>>>
>>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>>> future, but at the moment, we won't be able to run use-cases without
>>> this. And I have already given some of the reasons for the same here and
>>> on v4.
>>>
>>
>>
>> No reason to be dismissive, my questions are valid.
>>
>> What "use-cases" are we talking about, I have firmware that doesn't need
>> specific carved-out addresses.
>
> I think you are well aware of all the usecases we provide with the TI
> SDKs with IPUs and DSPs. And what is the firmware that you have and what
> do you use it for?
>
> If you have misbehaving firmware that
>> needs statically carved out memory addresses then you can have carveouts
>> if you want, but it should be optional.
> If I don't want to pollute my
>> system's memory space with a bunch of carveout holes then I shouldn't
>> have to just because your specific firmware needs them.
>
> Further follow-up series like early-boot and late-attach will mandate
> fixed carveouts actually. You cannot just run out of any random memory.
Also, these are CMA pools ("reusable"), so they are not actual carveout
holes ("no-map"). This is the preferred method in remoteproc mode so
that the memory is available for kernel when remoteprocs are not in use.
Customers can always choose to make these carveouts so that they do not
run into memory allocation issues when changing firmwares and under
stress conditions. These will have to be carveouts for early-boot usecases.
regards
Suman
>
> regards
> Suman
>
>>
>> Andrew
>>
>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> regards
>>>>> Suman
>>>>>
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>>
>>>>>>> regards
>>>>>>> Suman
>>>>>>>
>>>>>>>>
>>>>>>>> -Tero
>>>>>>>>
>>>>>>>>>
>>>>>>>>> Andrew
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> -Tero
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>>
>>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>>> ---
>>>>>>>>>>>> v5: no changes
>>>>>>>>>>>>
>>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>>
>>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>> if (ret)
>>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>>> + if (ret) {
>>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>>> pool\n");
>>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>>> + }
>>>>>>>>>>>> +
>>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>>> if (ret)
>>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>>> return 0;
>>>>>>>>>>>> +release_mem:
>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>> free_rproc:
>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>> return ret;
>>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>> return 0;
>>>>>>>>>>>> }
>>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>
>>>>>>>> --
>>>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>>>
>>>>>
>>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 22:06 ` Suman Anna
@ 2020-01-30 23:04 ` Andrew F. Davis
-1 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 23:04 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 5:06 PM, Suman Anna wrote:
> On 1/30/20 3:57 PM, Suman Anna wrote:
>> On 1/30/20 3:50 PM, Andrew F. Davis wrote:
>>> On 1/30/20 4:39 PM, Suman Anna wrote:
>>>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>>
>>>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>>>> only a few devices.
>>>>>>>>>>>>>
>>>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>>>> firmwares
>>>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>>>> hard-code them.
>>>>>>>>
>>>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>>>> there was no conclusion on it.
>>>>>>>>>>>
>>>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>>>
>>>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>>>> manner.
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Why not use the default CMA area?
>>>>>>>>>
>>>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>>>
>>>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>>>> and why this would not be a good idea.
>>>>>>>>
>>>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> If you need 32bit range allocations then
>>>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>>>
>>>>>>> I'm not saying don't have support for carveouts, just make them
>>>>>>> optional, keystone_remoteproc.c does this:
>>>>>>>
>>>>>>> if (of_reserved_mem_device_init(dev))
>>>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>>>
>>>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>>>
>>>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>>>> and most of our firmware images on K2 are actually running out of the
>>>>>> DSP internal memory.
>>>>>>
>>>>>
>>>>>
>>>>> So again we circle back to it being a firmware issue, if K2 can get away
>>>>> without needing carveouts and it doesn't even have an MMU then certainly
>>>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>>>> which case we are still just hacking around the problem here with
>>>>> mandatory specific address memory carveouts.
>>>>
>>>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>>>> future, but at the moment, we won't be able to run use-cases without
>>>> this. And I have already given some of the reasons for the same here and
>>>> on v4.
>>>>
>>>
>>>
>>> No reason to be dismissive, my questions are valid.
>>>
>>> What "use-cases" are we talking about, I have firmware that doesn't need
>>> specific carved-out addresses.
>>
>> I think you are well aware of all the usecases we provide with the TI
>> SDKs with IPUs and DSPs. And what is the firmware that you have and what
>> do you use it for?
>>
Yes I know exactly the pieces of TI firmware we are talking about and
why it they still need carveouts. That's not the point, our firmware may
have issues and hard-coding, but we need to allow for correctly built
firmware that doesn't need carveouts also. This driver should not fail
if a carveout is not provided. The remoteproc can run fine without a
carveout, only some firmwares cannot, so it should be optional and not
forced in DT on everyone using our DSP/IPU.
>> If you have misbehaving firmware that
>>> needs statically carved out memory addresses then you can have carveouts
>>> if you want, but it should be optional.
>> If I don't want to pollute my
>>> system's memory space with a bunch of carveout holes then I shouldn't
>>> have to just because your specific firmware needs them.
>>
>> Further follow-up series like early-boot and late-attach will mandate
>> fixed carveouts actually. You cannot just run out of any random memory.
>
Those are different, the location of the loaded firmware in memory will
need to be carved out if it is in use by a remote core before Linux
boots. This carveout is for Linux to allocate from to load the Vrings
and other memory it may need. When late-attach shows up then we can
think about how to handle those.
> Also, these are CMA pools ("reusable"), so they are not actual carveout
> holes ("no-map"). This is the preferred method in remoteproc mode so
> that the memory is available for kernel when remoteprocs are not in use.
> Customers can always choose to make these carveouts so that they do not
> run into memory allocation issues when changing firmwares and under
> stress conditions. These will have to be carveouts for early-boot usecases.
>
Even "reusable" carveouts can only be used by re-locateable memory
(caches and such) so still not a good thing to have your memory space
full of them.
> Customers can always choose to make these carveouts
That is exactly what I am saying, they can choose, but it should be
optional, the current binding and driver make them mandatory or the
driver will not probe.
Andrew
> regards
> Suman
>
>>
>> regards
>> Suman
>>
>>>
>>> Andrew
>>>
>>>
>>>> regards
>>>> Suman
>>>>
>>>>>
>>>>> Andrew
>>>>>
>>>>>
>>>>>> regards
>>>>>> Suman
>>>>>>
>>>>>>>
>>>>>>> Andrew
>>>>>>>
>>>>>>>
>>>>>>>> regards
>>>>>>>> Suman
>>>>>>>>
>>>>>>>>>
>>>>>>>>> -Tero
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Andrew
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> -Tero
>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>>>> ---
>>>>>>>>>>>>> v5: no changes
>>>>>>>>>>>>>
>>>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>>>
>>>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>>>> + if (ret) {
>>>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>>>> pool\n");
>>>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>>>> + }
>>>>>>>>>>>>> +
>>>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>> +release_mem:
>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>> free_rproc:
>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>> return ret;
>>>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>> }
>>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> --
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>>>>
>>>>>>
>>>>
>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-01-30 23:04 ` Andrew F. Davis
0 siblings, 0 replies; 62+ messages in thread
From: Andrew F. Davis @ 2020-01-30 23:04 UTC (permalink / raw)
To: Suman Anna, Tero Kristo, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 1/30/20 5:06 PM, Suman Anna wrote:
> On 1/30/20 3:57 PM, Suman Anna wrote:
>> On 1/30/20 3:50 PM, Andrew F. Davis wrote:
>>> On 1/30/20 4:39 PM, Suman Anna wrote:
>>>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>>
>>>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>>>> only a few devices.
>>>>>>>>>>>>>
>>>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>>>> firmwares
>>>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>>>> hard-code them.
>>>>>>>>
>>>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>>>> there was no conclusion on it.
>>>>>>>>>>>
>>>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>>>
>>>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>>>> manner.
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Why not use the default CMA area?
>>>>>>>>>
>>>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>>>
>>>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>>>> and why this would not be a good idea.
>>>>>>>>
>>>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> If you need 32bit range allocations then
>>>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>>>
>>>>>>> I'm not saying don't have support for carveouts, just make them
>>>>>>> optional, keystone_remoteproc.c does this:
>>>>>>>
>>>>>>> if (of_reserved_mem_device_init(dev))
>>>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>>>
>>>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>>>
>>>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>>>> and most of our firmware images on K2 are actually running out of the
>>>>>> DSP internal memory.
>>>>>>
>>>>>
>>>>>
>>>>> So again we circle back to it being a firmware issue, if K2 can get away
>>>>> without needing carveouts and it doesn't even have an MMU then certainly
>>>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>>>> which case we are still just hacking around the problem here with
>>>>> mandatory specific address memory carveouts.
>>>>
>>>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>>>> future, but at the moment, we won't be able to run use-cases without
>>>> this. And I have already given some of the reasons for the same here and
>>>> on v4.
>>>>
>>>
>>>
>>> No reason to be dismissive, my questions are valid.
>>>
>>> What "use-cases" are we talking about, I have firmware that doesn't need
>>> specific carved-out addresses.
>>
>> I think you are well aware of all the usecases we provide with the TI
>> SDKs with IPUs and DSPs. And what is the firmware that you have and what
>> do you use it for?
>>
Yes I know exactly the pieces of TI firmware we are talking about and
why it they still need carveouts. That's not the point, our firmware may
have issues and hard-coding, but we need to allow for correctly built
firmware that doesn't need carveouts also. This driver should not fail
if a carveout is not provided. The remoteproc can run fine without a
carveout, only some firmwares cannot, so it should be optional and not
forced in DT on everyone using our DSP/IPU.
>> If you have misbehaving firmware that
>>> needs statically carved out memory addresses then you can have carveouts
>>> if you want, but it should be optional.
>> If I don't want to pollute my
>>> system's memory space with a bunch of carveout holes then I shouldn't
>>> have to just because your specific firmware needs them.
>>
>> Further follow-up series like early-boot and late-attach will mandate
>> fixed carveouts actually. You cannot just run out of any random memory.
>
Those are different, the location of the loaded firmware in memory will
need to be carved out if it is in use by a remote core before Linux
boots. This carveout is for Linux to allocate from to load the Vrings
and other memory it may need. When late-attach shows up then we can
think about how to handle those.
> Also, these are CMA pools ("reusable"), so they are not actual carveout
> holes ("no-map"). This is the preferred method in remoteproc mode so
> that the memory is available for kernel when remoteprocs are not in use.
> Customers can always choose to make these carveouts so that they do not
> run into memory allocation issues when changing firmwares and under
> stress conditions. These will have to be carveouts for early-boot usecases.
>
Even "reusable" carveouts can only be used by re-locateable memory
(caches and such) so still not a good thing to have your memory space
full of them.
> Customers can always choose to make these carveouts
That is exactly what I am saying, they can choose, but it should be
optional, the current binding and driver make them mandatory or the
driver will not probe.
Andrew
> regards
> Suman
>
>>
>> regards
>> Suman
>>
>>>
>>> Andrew
>>>
>>>
>>>> regards
>>>> Suman
>>>>
>>>>>
>>>>> Andrew
>>>>>
>>>>>
>>>>>> regards
>>>>>> Suman
>>>>>>
>>>>>>>
>>>>>>> Andrew
>>>>>>>
>>>>>>>
>>>>>>>> regards
>>>>>>>> Suman
>>>>>>>>
>>>>>>>>>
>>>>>>>>> -Tero
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Andrew
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> -Tero
>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>>>> ---
>>>>>>>>>>>>> v5: no changes
>>>>>>>>>>>>>
>>>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>>>
>>>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>>>> + if (ret) {
>>>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>>>> pool\n");
>>>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>>>> + }
>>>>>>>>>>>>> +
>>>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>> +release_mem:
>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>> free_rproc:
>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>> return ret;
>>>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>> }
>>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> --
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>>>>>>>>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>>>>>
>>>>>>
>>>>
>>
>
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
2020-01-30 23:04 ` Andrew F. Davis
@ 2020-02-05 8:51 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-02-05 8:51 UTC (permalink / raw)
To: Andrew F. Davis, Suman Anna, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 31/01/2020 01:04, Andrew F. Davis wrote:
> On 1/30/20 5:06 PM, Suman Anna wrote:
>> On 1/30/20 3:57 PM, Suman Anna wrote:
>>> On 1/30/20 3:50 PM, Andrew F. Davis wrote:
>>>> On 1/30/20 4:39 PM, Suman Anna wrote:
>>>>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>>>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>>>>> only a few devices.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>>>>> firmwares
>>>>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>>>>> hard-code them.
>>>>>>>>>
>>>>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>>>>> there was no conclusion on it.
>>>>>>>>>>>>
>>>>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>>>>
>>>>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>>>>> manner.
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Why not use the default CMA area?
>>>>>>>>>>
>>>>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>>>>
>>>>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>>>>> and why this would not be a good idea.
>>>>>>>>>
>>>>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> If you need 32bit range allocations then
>>>>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>>>>
>>>>>>>> I'm not saying don't have support for carveouts, just make them
>>>>>>>> optional, keystone_remoteproc.c does this:
>>>>>>>>
>>>>>>>> if (of_reserved_mem_device_init(dev))
>>>>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>>>>
>>>>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>>>>
>>>>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>>>>> and most of our firmware images on K2 are actually running out of the
>>>>>>> DSP internal memory.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> So again we circle back to it being a firmware issue, if K2 can get away
>>>>>> without needing carveouts and it doesn't even have an MMU then certainly
>>>>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>>>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>>>>> which case we are still just hacking around the problem here with
>>>>>> mandatory specific address memory carveouts.
>>>>>
>>>>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>>>>> future, but at the moment, we won't be able to run use-cases without
>>>>> this. And I have already given some of the reasons for the same here and
>>>>> on v4.
>>>>>
>>>>
>>>>
>>>> No reason to be dismissive, my questions are valid.
>>>>
>>>> What "use-cases" are we talking about, I have firmware that doesn't need
>>>> specific carved-out addresses.
>>>
>>> I think you are well aware of all the usecases we provide with the TI
>>> SDKs with IPUs and DSPs. And what is the firmware that you have and what
>>> do you use it for?
>>>
>
>
> Yes I know exactly the pieces of TI firmware we are talking about and
> why it they still need carveouts. That's not the point, our firmware may
> have issues and hard-coding, but we need to allow for correctly built
> firmware that doesn't need carveouts also. This driver should not fail
> if a carveout is not provided. The remoteproc can run fine without a
> carveout, only some firmwares cannot, so it should be optional and not
> forced in DT on everyone using our DSP/IPU.
>
>
>>> If you have misbehaving firmware that
>>>> needs statically carved out memory addresses then you can have carveouts
>>>> if you want, but it should be optional.
>>> If I don't want to pollute my
>>>> system's memory space with a bunch of carveout holes then I shouldn't
>>>> have to just because your specific firmware needs them.
>>>
>>> Further follow-up series like early-boot and late-attach will mandate
>>> fixed carveouts actually. You cannot just run out of any random memory.
>>
>
>
> Those are different, the location of the loaded firmware in memory will
> need to be carved out if it is in use by a remote core before Linux
> boots. This carveout is for Linux to allocate from to load the Vrings
> and other memory it may need. When late-attach shows up then we can
> think about how to handle those.
>
>
>> Also, these are CMA pools ("reusable"), so they are not actual carveout
>> holes ("no-map"). This is the preferred method in remoteproc mode so
>> that the memory is available for kernel when remoteprocs are not in use.
>> Customers can always choose to make these carveouts so that they do not
>> run into memory allocation issues when changing firmwares and under
>> stress conditions. These will have to be carveouts for early-boot usecases.
>>
>
>
> Even "reusable" carveouts can only be used by re-locateable memory
> (caches and such) so still not a good thing to have your memory space
> full of them.
>
>> Customers can always choose to make these carveouts
>
> That is exactly what I am saying, they can choose, but it should be
> optional, the current binding and driver make them mandatory or the
> driver will not probe.
Ok, so just to re-cap here, we (me + Suman + Andrew) aligned offline on
this and the plan from our side is to:
- Make the memory-regions part of the binding optional, not mandatory
- Make the omap remoteproc driver portion for the memory-regions also
optional, not fail to probe
- Print a warning from the omap driver if memory regions are not in
place, claiming user knows what he is doing, as right now nothing works
without the memory regions defined.
I'll post v6 against -rc1 with these changes once it is available.
-Tero
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> regards
>>>>> Suman
>>>>>
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>>
>>>>>>> regards
>>>>>>> Suman
>>>>>>>
>>>>>>>>
>>>>>>>> Andrew
>>>>>>>>
>>>>>>>>
>>>>>>>>> regards
>>>>>>>>> Suman
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> -Tero
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Andrew
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>> -Tero
>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>>>>> ---
>>>>>>>>>>>>>> v5: no changes
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>>>>> + if (ret) {
>>>>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>>>>> pool\n");
>>>>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>>>>> + }
>>>>>>>>>>>>>> +
>>>>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>>> +release_mem:
>>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>>> free_rproc:
>>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>>> return ret;
>>>>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>>> }
>>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> --
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>
>>>>>>>
>>>>>
>>>
>>
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread* Re: [PATCHv5 06/14] remoteproc/omap: Initialize and assign reserved memory node
@ 2020-02-05 8:51 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-02-05 8:51 UTC (permalink / raw)
To: Andrew F. Davis, Suman Anna, bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, linux-omap
On 31/01/2020 01:04, Andrew F. Davis wrote:
> On 1/30/20 5:06 PM, Suman Anna wrote:
>> On 1/30/20 3:57 PM, Suman Anna wrote:
>>> On 1/30/20 3:50 PM, Andrew F. Davis wrote:
>>>> On 1/30/20 4:39 PM, Suman Anna wrote:
>>>>> On 1/30/20 3:19 PM, Andrew F. Davis wrote:
>>>>>> On 1/30/20 3:39 PM, Suman Anna wrote:
>>>>>>> On 1/30/20 2:22 PM, Andrew F. Davis wrote:
>>>>>>>> On 1/30/20 2:55 PM, Suman Anna wrote:
>>>>>>>>> On 1/30/20 1:42 PM, Tero Kristo wrote:
>>>>>>>>>> On 30/01/2020 21:20, Andrew F. Davis wrote:
>>>>>>>>>>> On 1/30/20 2:18 PM, Tero Kristo wrote:
>>>>>>>>>>>> On 30/01/2020 20:11, Andrew F. Davis wrote:
>>>>>>>>>>>>> On 1/16/20 8:53 AM, Tero Kristo wrote:
>>>>>>>>>>>>>> From: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> The reserved memory nodes are not assigned to platform devices by
>>>>>>>>>>>>>> default in the driver core to avoid the lookup for every platform
>>>>>>>>>>>>>> device and incur a penalty as the real users are expected to be
>>>>>>>>>>>>>> only a few devices.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> OMAP remoteproc devices fall into the above category and the OMAP
>>>>>>>>>>>>>> remoteproc driver _requires_ specific CMA pools to be assigned
>>>>>>>>>>>>>> for each device at the moment to align on the location of the
>>>>>>>>>>>>>> vrings and vring buffers in the RTOS-side firmware images. So,
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> Same comment as before, this is a firmware issue for only some
>>>>>>>>>>>>> firmwares
>>>>>>>>>>>>> that do not handle being assigned vring locations correctly and instead
>>>>>>>>>>>>> hard-code them.
>>>>>>>>>
>>>>>>>>> As for this statement, this can do with some updating. Post 4.20,
>>>>>>>>> because of the lazy allocation scheme used for carveouts including the
>>>>>>>>> vrings, the resource tables now have to use FW_RSC_ADDR_ANY and will
>>>>>>>>> have to wait for the vdev synchronization to happen.
>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> I believe we discussed this topic in length in previous version but
>>>>>>>>>>>> there was no conclusion on it.
>>>>>>>>>>>>
>>>>>>>>>>>> The commit desc might be a bit misleading, we are not actually forced to
>>>>>>>>>>>> use specific CMA buffers, as we use IOMMU to map these to device
>>>>>>>>>>>> addresses. For example IPU1/IPU2 use internally exact same memory
>>>>>>>>>>>> addresses, iommu is used to map these to specific CMA buffer.
>>>>>>>>>>>>
>>>>>>>>>>>> CMA buffers are mostly used so that we get aligned large chunk of memory
>>>>>>>>>>>> which can be mapped properly with the limited IOMMU OMAP family of chips
>>>>>>>>>>>> have. Not sure if there is any sane way to get this done in any other
>>>>>>>>>>>> manner.
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Why not use the default CMA area?
>>>>>>>>>>
>>>>>>>>>> I think using default CMA area getting the actual memory block is not
>>>>>>>>>> guaranteed and might fail. There are other users for the memory, and it
>>>>>>>>>> might get fragmented at the very late phase we are grabbing the memory
>>>>>>>>>> (omap remoteproc driver probe time.) Some chunks we need are pretty large.
>>>>>>>>>>
>>>>>>>>>> I believe I could experiment with this a bit though and see, or Suman
>>>>>>>>>> could maybe provide feedback why this was designed initially like this
>>>>>>>>>> and why this would not be a good idea.
>>>>>>>>>
>>>>>>>>> I have given some explanation on this on v4 as well, but if it is not
>>>>>>>>> clear, there are restrictions with using default CMA. Default CMA has
>>>>>>>>> switched to be assigned from the top of the memory (higher addresses,
>>>>>>>>> since 3.18 IIRC), and the MMUs on IPUs and DSPs can only address
>>>>>>>>> 32-bits. So, we cannot blindly use the default CMA pool, and this will
>>>>>>>>> definitely not work on boards > 2 GB RAM. And, if you want to add in any
>>>>>>>>> firewall capability, then specific physical addresses becomes mandatory.
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> If you need 32bit range allocations then
>>>>>>>> dma_set_mask(dev, DMA_BIT_MASK(32));
>>>>>>>>
>>>>>>>> I'm not saying don't have support for carveouts, just make them
>>>>>>>> optional, keystone_remoteproc.c does this:
>>>>>>>>
>>>>>>>> if (of_reserved_mem_device_init(dev))
>>>>>>>> dev_warn(dev, "device does not have specific CMA pool\n");
>>>>>>>>
>>>>>>>> There doesn't even needs to be a warning but that is up to you.
>>>>>>>
>>>>>>> It is not exactly an apples to apples comparison. K2s do not have MMUs,
>>>>>>> and most of our firmware images on K2 are actually running out of the
>>>>>>> DSP internal memory.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> So again we circle back to it being a firmware issue, if K2 can get away
>>>>>> without needing carveouts and it doesn't even have an MMU then certainly
>>>>>> OMAP/DRA7x class devices can handle it even better given they *do* have
>>>>>> an IOMMU. Unless someone is hard-coding the IOMMU configuration.. In
>>>>>> which case we are still just hacking around the problem here with
>>>>>> mandatory specific address memory carveouts.
>>>>>
>>>>> Optional carveouts on OMAP remoteprocs can be an enhancement in the
>>>>> future, but at the moment, we won't be able to run use-cases without
>>>>> this. And I have already given some of the reasons for the same here and
>>>>> on v4.
>>>>>
>>>>
>>>>
>>>> No reason to be dismissive, my questions are valid.
>>>>
>>>> What "use-cases" are we talking about, I have firmware that doesn't need
>>>> specific carved-out addresses.
>>>
>>> I think you are well aware of all the usecases we provide with the TI
>>> SDKs with IPUs and DSPs. And what is the firmware that you have and what
>>> do you use it for?
>>>
>
>
> Yes I know exactly the pieces of TI firmware we are talking about and
> why it they still need carveouts. That's not the point, our firmware may
> have issues and hard-coding, but we need to allow for correctly built
> firmware that doesn't need carveouts also. This driver should not fail
> if a carveout is not provided. The remoteproc can run fine without a
> carveout, only some firmwares cannot, so it should be optional and not
> forced in DT on everyone using our DSP/IPU.
>
>
>>> If you have misbehaving firmware that
>>>> needs statically carved out memory addresses then you can have carveouts
>>>> if you want, but it should be optional.
>>> If I don't want to pollute my
>>>> system's memory space with a bunch of carveout holes then I shouldn't
>>>> have to just because your specific firmware needs them.
>>>
>>> Further follow-up series like early-boot and late-attach will mandate
>>> fixed carveouts actually. You cannot just run out of any random memory.
>>
>
>
> Those are different, the location of the loaded firmware in memory will
> need to be carved out if it is in use by a remote core before Linux
> boots. This carveout is for Linux to allocate from to load the Vrings
> and other memory it may need. When late-attach shows up then we can
> think about how to handle those.
>
>
>> Also, these are CMA pools ("reusable"), so they are not actual carveout
>> holes ("no-map"). This is the preferred method in remoteproc mode so
>> that the memory is available for kernel when remoteprocs are not in use.
>> Customers can always choose to make these carveouts so that they do not
>> run into memory allocation issues when changing firmwares and under
>> stress conditions. These will have to be carveouts for early-boot usecases.
>>
>
>
> Even "reusable" carveouts can only be used by re-locateable memory
> (caches and such) so still not a good thing to have your memory space
> full of them.
>
>> Customers can always choose to make these carveouts
>
> That is exactly what I am saying, they can choose, but it should be
> optional, the current binding and driver make them mandatory or the
> driver will not probe.
Ok, so just to re-cap here, we (me + Suman + Andrew) aligned offline on
this and the plan from our side is to:
- Make the memory-regions part of the binding optional, not mandatory
- Make the omap remoteproc driver portion for the memory-regions also
optional, not fail to probe
- Print a warning from the omap driver if memory regions are not in
place, claiming user knows what he is doing, as right now nothing works
without the memory regions defined.
I'll post v6 against -rc1 with these changes once it is available.
-Tero
>
> Andrew
>
>
>> regards
>> Suman
>>
>>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Andrew
>>>>
>>>>
>>>>> regards
>>>>> Suman
>>>>>
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>>
>>>>>>> regards
>>>>>>> Suman
>>>>>>>
>>>>>>>>
>>>>>>>> Andrew
>>>>>>>>
>>>>>>>>
>>>>>>>>> regards
>>>>>>>>> Suman
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> -Tero
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Andrew
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>> -Tero
>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> This is not a requirement of the remote processor itself and so it
>>>>>>>>>>>>> should not fail to probe if a specific memory carveout isn't given.
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>> use the of_reserved_mem_device_init/release() API appropriately
>>>>>>>>>>>>>> to assign the corresponding reserved memory region to the OMAP
>>>>>>>>>>>>>> remoteproc device. Note that only one region per device is
>>>>>>>>>>>>>> allowed by the framework.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Signed-off-by: Suman Anna <s-anna@ti.com>
>>>>>>>>>>>>>> Signed-off-by: Tero Kristo <t-kristo@ti.com>
>>>>>>>>>>>>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>>>>>>>>>>>> ---
>>>>>>>>>>>>>> v5: no changes
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++-
>>>>>>>>>>>>>> 1 file changed, 11 insertions(+), 1 deletion(-)
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> diff --git a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> index 0846839b2c97..194303b860b2 100644
>>>>>>>>>>>>>> --- a/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> +++ b/drivers/remoteproc/omap_remoteproc.c
>>>>>>>>>>>>>> @@ -17,6 +17,7 @@
>>>>>>>>>>>>>> #include <linux/module.h>
>>>>>>>>>>>>>> #include <linux/err.h>
>>>>>>>>>>>>>> #include <linux/of_device.h>
>>>>>>>>>>>>>> +#include <linux/of_reserved_mem.h>
>>>>>>>>>>>>>> #include <linux/platform_device.h>
>>>>>>>>>>>>>> #include <linux/dma-mapping.h>
>>>>>>>>>>>>>> #include <linux/remoteproc.h>
>>>>>>>>>>>>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct
>>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>>> goto free_rproc;
>>>>>>>>>>>>>> + ret = of_reserved_mem_device_init(&pdev->dev);
>>>>>>>>>>>>>> + if (ret) {
>>>>>>>>>>>>>> + dev_err(&pdev->dev, "device does not have specific CMA
>>>>>>>>>>>>>> pool\n");
>>>>>>>>>>>>>> + goto free_rproc;
>>>>>>>>>>>>>> + }
>>>>>>>>>>>>>> +
>>>>>>>>>>>>>> platform_set_drvdata(pdev, rproc);
>>>>>>>>>>>>>> ret = rproc_add(rproc);
>>>>>>>>>>>>>> if (ret)
>>>>>>>>>>>>>> - goto free_rproc;
>>>>>>>>>>>>>> + goto release_mem;
>>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>>> +release_mem:
>>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>>> free_rproc:
>>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>>> return ret;
>>>>>>>>>>>>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct
>>>>>>>>>>>>>> platform_device *pdev)
>>>>>>>>>>>>>> rproc_del(rproc);
>>>>>>>>>>>>>> rproc_free(rproc);
>>>>>>>>>>>>>> + of_reserved_mem_device_release(&pdev->dev);
>>>>>>>>>>>>>> return 0;
>>>>>>>>>>>>>> }
>>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> --
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>
>>>>>>>
>>>>>
>>>
>>
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply [flat|nested] 62+ messages in thread
* [PATCHv5 07/14] remoteproc/omap: Add support for DRA7xx remote processors
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
DRA7xx/AM57xx SoCs have two IPU and up to two DSP processor subsystems
for offloading different computation algorithms. The IPU processor
subsystem contains dual-core ARM Cortex-M4 processors, and is very
similar to those on OMAP5. The DSP processor subsystem is based on
the TI's standard TMS320C66x DSP CorePac core.
Support has been added to the OMAP remoteproc driver through new
DRA7xx specific compatibles for properly probing and booting all
the different processor subsystem instances on DRA7xx/AM57xx
SoCs - IPU1, IPU2, DSP1 & DSP2. A build dependency with SOC_DRA7XX
is added to enable the driver to be built in DRA7xx-only configuration.
The DSP boot address programming needed enhancement for DRA7xx as the
boot register fields are different on DRA7 compared to OMAP4 and OMAP5
SoCs. The register on DRA7xx contains additional fields within the
register and the boot address bit-field is right-shifted by 10 bits.
The internal memory parsing logic has also been updated to compute
the device addresses for the L2 RAM for DSP devices using relative
addressing logic, and to parse two additional RAMs at L1 level - L1P
and L1D. This allows the remoteproc driver to support loading into
these regions for a small subset of firmware images requiring as
such. The most common usage would be to use the L1 programmable
RAMs as L1 Caches.
The firmware lookup logic also has to be adjusted for DRA7xx as
there are (can be) more than one instance of both the IPU and DSP
remote processors for the first time in OMAP4+ SoCs.
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: moved address translation quirks to pdata]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- replaced ti,bootreg-shift property handling with an extra arg
under ti,bootreg
drivers/remoteproc/Kconfig | 2 +-
drivers/remoteproc/omap_remoteproc.c | 38 +++++++++++++++++++++++++++-
2 files changed, 38 insertions(+), 2 deletions(-)
diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
index 94afdde4bc9f..d6450d7fcf92 100644
--- a/drivers/remoteproc/Kconfig
+++ b/drivers/remoteproc/Kconfig
@@ -25,7 +25,7 @@ config IMX_REMOTEPROC
config OMAP_REMOTEPROC
tristate "OMAP remoteproc support"
- depends on ARCH_OMAP4 || SOC_OMAP5
+ depends on ARCH_OMAP4 || SOC_OMAP5 || SOC_DRA7XX
depends on OMAP_IOMMU
select MAILBOX
select OMAP2PLUS_MBOX
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 194303b860b2..27e145181ef4 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -34,10 +34,13 @@
* struct omap_rproc_boot_data - boot data structure for the DSP omap rprocs
* @syscon: regmap handle for the system control configuration module
* @boot_reg: boot register offset within the @syscon regmap
+ * @boot_reg_shift: bit-field shift required for the boot address value in
+ * @boot_reg
*/
struct omap_rproc_boot_data {
struct regmap *syscon;
unsigned int boot_reg;
+ unsigned int boot_reg_shift;
};
/**
@@ -159,6 +162,8 @@ static int omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
struct omap_rproc *oproc = rproc->priv;
struct omap_rproc_boot_data *bdata = oproc->boot_data;
u32 offset = bdata->boot_reg;
+ u32 value;
+ u32 mask;
if (rproc->bootaddr & (SZ_1K - 1)) {
dev_err(dev, "invalid boot address 0x%x, must be aligned on a 1KB boundary\n",
@@ -166,7 +171,10 @@ static int omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
return -EINVAL;
}
- regmap_write(bdata->syscon, offset, rproc->bootaddr);
+ value = rproc->bootaddr >> bdata->boot_reg_shift;
+ mask = ~(SZ_1K - 1) >> bdata->boot_reg_shift;
+
+ regmap_update_bits(bdata->syscon, offset, mask, value);
return 0;
}
@@ -296,6 +304,13 @@ static const struct omap_rproc_mem_data ipu_mems[] = {
{ },
};
+static const struct omap_rproc_mem_data dra7_dsp_mems[] = {
+ { .name = "l2ram", .dev_addr = 0x800000 },
+ { .name = "l1pram", .dev_addr = 0xe00000 },
+ { .name = "l1dram", .dev_addr = 0xf00000 },
+ { },
+};
+
static const struct omap_rproc_dev_data omap4_dsp_dev_data = {
.device_name = "dsp",
};
@@ -314,6 +329,16 @@ static const struct omap_rproc_dev_data omap5_ipu_dev_data = {
.mems = ipu_mems,
};
+static const struct omap_rproc_dev_data dra7_dsp_dev_data = {
+ .device_name = "dsp",
+ .mems = dra7_dsp_mems,
+};
+
+static const struct omap_rproc_dev_data dra7_ipu_dev_data = {
+ .device_name = "ipu",
+ .mems = ipu_mems,
+};
+
static const struct of_device_id omap_rproc_of_match[] = {
{
.compatible = "ti,omap4-dsp",
@@ -331,6 +356,14 @@ static const struct of_device_id omap_rproc_of_match[] = {
.compatible = "ti,omap5-ipu",
.data = &omap5_ipu_dev_data,
},
+ {
+ .compatible = "ti,dra7-dsp",
+ .data = &dra7_dsp_dev_data,
+ },
+ {
+ .compatible = "ti,dra7-ipu",
+ .data = &dra7_ipu_dev_data,
+ },
{
/* end */
},
@@ -383,6 +416,9 @@ static int omap_rproc_get_boot_data(struct platform_device *pdev,
return -EINVAL;
}
+ of_property_read_u32_index(np, "ti,bootreg", 2,
+ &oproc->boot_data->boot_reg_shift);
+
return 0;
}
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 07/14] remoteproc/omap: Add support for DRA7xx remote processors
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
DRA7xx/AM57xx SoCs have two IPU and up to two DSP processor subsystems
for offloading different computation algorithms. The IPU processor
subsystem contains dual-core ARM Cortex-M4 processors, and is very
similar to those on OMAP5. The DSP processor subsystem is based on
the TI's standard TMS320C66x DSP CorePac core.
Support has been added to the OMAP remoteproc driver through new
DRA7xx specific compatibles for properly probing and booting all
the different processor subsystem instances on DRA7xx/AM57xx
SoCs - IPU1, IPU2, DSP1 & DSP2. A build dependency with SOC_DRA7XX
is added to enable the driver to be built in DRA7xx-only configuration.
The DSP boot address programming needed enhancement for DRA7xx as the
boot register fields are different on DRA7 compared to OMAP4 and OMAP5
SoCs. The register on DRA7xx contains additional fields within the
register and the boot address bit-field is right-shifted by 10 bits.
The internal memory parsing logic has also been updated to compute
the device addresses for the L2 RAM for DSP devices using relative
addressing logic, and to parse two additional RAMs at L1 level - L1P
and L1D. This allows the remoteproc driver to support loading into
these regions for a small subset of firmware images requiring as
such. The most common usage would be to use the L1 programmable
RAMs as L1 Caches.
The firmware lookup logic also has to be adjusted for DRA7xx as
there are (can be) more than one instance of both the IPU and DSP
remote processors for the first time in OMAP4+ SoCs.
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: moved address translation quirks to pdata]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- replaced ti,bootreg-shift property handling with an extra arg
under ti,bootreg
drivers/remoteproc/Kconfig | 2 +-
drivers/remoteproc/omap_remoteproc.c | 38 +++++++++++++++++++++++++++-
2 files changed, 38 insertions(+), 2 deletions(-)
diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
index 94afdde4bc9f..d6450d7fcf92 100644
--- a/drivers/remoteproc/Kconfig
+++ b/drivers/remoteproc/Kconfig
@@ -25,7 +25,7 @@ config IMX_REMOTEPROC
config OMAP_REMOTEPROC
tristate "OMAP remoteproc support"
- depends on ARCH_OMAP4 || SOC_OMAP5
+ depends on ARCH_OMAP4 || SOC_OMAP5 || SOC_DRA7XX
depends on OMAP_IOMMU
select MAILBOX
select OMAP2PLUS_MBOX
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 194303b860b2..27e145181ef4 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -34,10 +34,13 @@
* struct omap_rproc_boot_data - boot data structure for the DSP omap rprocs
* @syscon: regmap handle for the system control configuration module
* @boot_reg: boot register offset within the @syscon regmap
+ * @boot_reg_shift: bit-field shift required for the boot address value in
+ * @boot_reg
*/
struct omap_rproc_boot_data {
struct regmap *syscon;
unsigned int boot_reg;
+ unsigned int boot_reg_shift;
};
/**
@@ -159,6 +162,8 @@ static int omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
struct omap_rproc *oproc = rproc->priv;
struct omap_rproc_boot_data *bdata = oproc->boot_data;
u32 offset = bdata->boot_reg;
+ u32 value;
+ u32 mask;
if (rproc->bootaddr & (SZ_1K - 1)) {
dev_err(dev, "invalid boot address 0x%x, must be aligned on a 1KB boundary\n",
@@ -166,7 +171,10 @@ static int omap_rproc_write_dsp_boot_addr(struct rproc *rproc)
return -EINVAL;
}
- regmap_write(bdata->syscon, offset, rproc->bootaddr);
+ value = rproc->bootaddr >> bdata->boot_reg_shift;
+ mask = ~(SZ_1K - 1) >> bdata->boot_reg_shift;
+
+ regmap_update_bits(bdata->syscon, offset, mask, value);
return 0;
}
@@ -296,6 +304,13 @@ static const struct omap_rproc_mem_data ipu_mems[] = {
{ },
};
+static const struct omap_rproc_mem_data dra7_dsp_mems[] = {
+ { .name = "l2ram", .dev_addr = 0x800000 },
+ { .name = "l1pram", .dev_addr = 0xe00000 },
+ { .name = "l1dram", .dev_addr = 0xf00000 },
+ { },
+};
+
static const struct omap_rproc_dev_data omap4_dsp_dev_data = {
.device_name = "dsp",
};
@@ -314,6 +329,16 @@ static const struct omap_rproc_dev_data omap5_ipu_dev_data = {
.mems = ipu_mems,
};
+static const struct omap_rproc_dev_data dra7_dsp_dev_data = {
+ .device_name = "dsp",
+ .mems = dra7_dsp_mems,
+};
+
+static const struct omap_rproc_dev_data dra7_ipu_dev_data = {
+ .device_name = "ipu",
+ .mems = ipu_mems,
+};
+
static const struct of_device_id omap_rproc_of_match[] = {
{
.compatible = "ti,omap4-dsp",
@@ -331,6 +356,14 @@ static const struct of_device_id omap_rproc_of_match[] = {
.compatible = "ti,omap5-ipu",
.data = &omap5_ipu_dev_data,
},
+ {
+ .compatible = "ti,dra7-dsp",
+ .data = &dra7_dsp_dev_data,
+ },
+ {
+ .compatible = "ti,dra7-ipu",
+ .data = &dra7_ipu_dev_data,
+ },
{
/* end */
},
@@ -383,6 +416,9 @@ static int omap_rproc_get_boot_data(struct platform_device *pdev,
return -EINVAL;
}
+ of_property_read_u32_index(np, "ti,bootreg", 2,
+ &oproc->boot_data->boot_reg_shift);
+
return 0;
}
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 08/14] remoteproc/omap: remove the platform_data header
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
The platform data header for OMAP remoteproc is no longer used for
anything post ti-sysc and DT conversion, so just remove it completely.
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Acked-by: Suman Anna <s-anna@ti.com>
---
v5:
- cosmetic changes in patch desc only
include/linux/platform_data/remoteproc-omap.h | 51 -------------------
1 file changed, 51 deletions(-)
delete mode 100644 include/linux/platform_data/remoteproc-omap.h
diff --git a/include/linux/platform_data/remoteproc-omap.h b/include/linux/platform_data/remoteproc-omap.h
deleted file mode 100644
index 7e3a16097672..000000000000
--- a/include/linux/platform_data/remoteproc-omap.h
+++ /dev/null
@@ -1,51 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0-only */
-/*
- * Remote Processor - omap-specific bits
- *
- * Copyright (C) 2011 Texas Instruments, Inc.
- * Copyright (C) 2011 Google, Inc.
- */
-
-#ifndef _PLAT_REMOTEPROC_H
-#define _PLAT_REMOTEPROC_H
-
-struct rproc_ops;
-struct platform_device;
-
-/*
- * struct omap_rproc_pdata - omap remoteproc's platform data
- * @name: the remoteproc's name
- * @oh_name: omap hwmod device
- * @oh_name_opt: optional, secondary omap hwmod device
- * @firmware: name of firmware file to load
- * @mbox_name: name of omap mailbox device to use with this rproc
- * @ops: start/stop rproc handlers
- * @device_enable: omap-specific handler for enabling a device
- * @device_shutdown: omap-specific handler for shutting down a device
- * @set_bootaddr: omap-specific handler for setting the rproc boot address
- */
-struct omap_rproc_pdata {
- const char *name;
- const char *oh_name;
- const char *oh_name_opt;
- const char *firmware;
- const char *mbox_name;
- const struct rproc_ops *ops;
- int (*device_enable)(struct platform_device *pdev);
- int (*device_shutdown)(struct platform_device *pdev);
- void (*set_bootaddr)(u32);
-};
-
-#if defined(CONFIG_OMAP_REMOTEPROC) || defined(CONFIG_OMAP_REMOTEPROC_MODULE)
-
-void __init omap_rproc_reserve_cma(void);
-
-#else
-
-static inline void __init omap_rproc_reserve_cma(void)
-{
-}
-
-#endif
-
-#endif /* _PLAT_REMOTEPROC_H */
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 08/14] remoteproc/omap: remove the platform_data header
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
The platform data header for OMAP remoteproc is no longer used for
anything post ti-sysc and DT conversion, so just remove it completely.
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Acked-by: Suman Anna <s-anna@ti.com>
---
v5:
- cosmetic changes in patch desc only
include/linux/platform_data/remoteproc-omap.h | 51 -------------------
1 file changed, 51 deletions(-)
delete mode 100644 include/linux/platform_data/remoteproc-omap.h
diff --git a/include/linux/platform_data/remoteproc-omap.h b/include/linux/platform_data/remoteproc-omap.h
deleted file mode 100644
index 7e3a16097672..000000000000
--- a/include/linux/platform_data/remoteproc-omap.h
+++ /dev/null
@@ -1,51 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0-only */
-/*
- * Remote Processor - omap-specific bits
- *
- * Copyright (C) 2011 Texas Instruments, Inc.
- * Copyright (C) 2011 Google, Inc.
- */
-
-#ifndef _PLAT_REMOTEPROC_H
-#define _PLAT_REMOTEPROC_H
-
-struct rproc_ops;
-struct platform_device;
-
-/*
- * struct omap_rproc_pdata - omap remoteproc's platform data
- * @name: the remoteproc's name
- * @oh_name: omap hwmod device
- * @oh_name_opt: optional, secondary omap hwmod device
- * @firmware: name of firmware file to load
- * @mbox_name: name of omap mailbox device to use with this rproc
- * @ops: start/stop rproc handlers
- * @device_enable: omap-specific handler for enabling a device
- * @device_shutdown: omap-specific handler for shutting down a device
- * @set_bootaddr: omap-specific handler for setting the rproc boot address
- */
-struct omap_rproc_pdata {
- const char *name;
- const char *oh_name;
- const char *oh_name_opt;
- const char *firmware;
- const char *mbox_name;
- const struct rproc_ops *ops;
- int (*device_enable)(struct platform_device *pdev);
- int (*device_shutdown)(struct platform_device *pdev);
- void (*set_bootaddr)(u32);
-};
-
-#if defined(CONFIG_OMAP_REMOTEPROC) || defined(CONFIG_OMAP_REMOTEPROC_MODULE)
-
-void __init omap_rproc_reserve_cma(void);
-
-#else
-
-static inline void __init omap_rproc_reserve_cma(void)
-{
-}
-
-#endif
-
-#endif /* _PLAT_REMOTEPROC_H */
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 09/14] remoteproc/omap: Check for undefined mailbox messages
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
Add some checks in the mailbox callback function to limit
any processing in the mailbox callback function to only
certain currently valid messages, and drop all the remaining
messages. A debug message is added to print any such invalid
messages when the appropriate trace control is enabled.
Co-developed-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 6 ++++++
drivers/remoteproc/omap_remoteproc.h | 7 +++++++
2 files changed, 13 insertions(+)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 27e145181ef4..777231ce2bae 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -130,6 +130,12 @@ static void omap_rproc_mbox_callback(struct mbox_client *client, void *data)
dev_info(dev, "received echo reply from %s\n", name);
break;
default:
+ if (msg >= RP_MBOX_READY && msg < RP_MBOX_END_MSG)
+ return;
+ if (msg > oproc->rproc->max_notifyid) {
+ dev_dbg(dev, "dropping unknown message 0x%x", msg);
+ return;
+ }
/* msg contains the index of the triggered vring */
if (rproc_vq_interrupt(oproc->rproc, msg) == IRQ_NONE)
dev_dbg(dev, "no message was found in vqid %d\n", msg);
diff --git a/drivers/remoteproc/omap_remoteproc.h b/drivers/remoteproc/omap_remoteproc.h
index f6d2036d383d..72f656c93caa 100644
--- a/drivers/remoteproc/omap_remoteproc.h
+++ b/drivers/remoteproc/omap_remoteproc.h
@@ -56,6 +56,12 @@
*
* @RP_MBOX_ABORT_REQUEST: a "please crash" request, used for testing the
* recovery mechanism (to some extent).
+ *
+ * Introduce new message definitions if any here.
+ *
+ * @RP_MBOX_END_MSG: Indicates end of known/defined messages from remote core
+ * This should be the last definition.
+ *
*/
enum omap_rp_mbox_messages {
RP_MBOX_READY = 0xFFFFFF00,
@@ -64,6 +70,7 @@ enum omap_rp_mbox_messages {
RP_MBOX_ECHO_REQUEST = 0xFFFFFF03,
RP_MBOX_ECHO_REPLY = 0xFFFFFF04,
RP_MBOX_ABORT_REQUEST = 0xFFFFFF05,
+ RP_MBOX_END_MSG = 0xFFFFFF06,
};
#endif /* _OMAP_RPMSG_H */
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 09/14] remoteproc/omap: Check for undefined mailbox messages
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
Add some checks in the mailbox callback function to limit
any processing in the mailbox callback function to only
certain currently valid messages, and drop all the remaining
messages. A debug message is added to print any such invalid
messages when the appropriate trace control is enabled.
Co-developed-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 6 ++++++
drivers/remoteproc/omap_remoteproc.h | 7 +++++++
2 files changed, 13 insertions(+)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 27e145181ef4..777231ce2bae 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -130,6 +130,12 @@ static void omap_rproc_mbox_callback(struct mbox_client *client, void *data)
dev_info(dev, "received echo reply from %s\n", name);
break;
default:
+ if (msg >= RP_MBOX_READY && msg < RP_MBOX_END_MSG)
+ return;
+ if (msg > oproc->rproc->max_notifyid) {
+ dev_dbg(dev, "dropping unknown message 0x%x", msg);
+ return;
+ }
/* msg contains the index of the triggered vring */
if (rproc_vq_interrupt(oproc->rproc, msg) == IRQ_NONE)
dev_dbg(dev, "no message was found in vqid %d\n", msg);
diff --git a/drivers/remoteproc/omap_remoteproc.h b/drivers/remoteproc/omap_remoteproc.h
index f6d2036d383d..72f656c93caa 100644
--- a/drivers/remoteproc/omap_remoteproc.h
+++ b/drivers/remoteproc/omap_remoteproc.h
@@ -56,6 +56,12 @@
*
* @RP_MBOX_ABORT_REQUEST: a "please crash" request, used for testing the
* recovery mechanism (to some extent).
+ *
+ * Introduce new message definitions if any here.
+ *
+ * @RP_MBOX_END_MSG: Indicates end of known/defined messages from remote core
+ * This should be the last definition.
+ *
*/
enum omap_rp_mbox_messages {
RP_MBOX_READY = 0xFFFFFF00,
@@ -64,6 +70,7 @@ enum omap_rp_mbox_messages {
RP_MBOX_ECHO_REQUEST = 0xFFFFFF03,
RP_MBOX_ECHO_REPLY = 0xFFFFFF04,
RP_MBOX_ABORT_REQUEST = 0xFFFFFF05,
+ RP_MBOX_END_MSG = 0xFFFFFF06,
};
#endif /* _OMAP_RPMSG_H */
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 10/14] remoteproc/omap: Request a timer(s) for remoteproc usage
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The remote processors in OMAP4+ SoCs are equipped with internal
timers, like the internal SysTick timer in a Cortex M3/M4 NVIC or
the CTM timer within Unicache in IPU & DSP. However, these timers
are gated when the processor subsystem clock is gated, making
them rather difficult to use as OS tick sources. They will not
be able to wakeup the processor from any processor-sleep induced
clock-gating states.
This can be avoided by using an external timer as the tick source,
which can be controlled independently by the OMAP remoteproc
driver code, but still allowing the processor subsystem clock to
be auto-gated when the remoteproc cores are idle.
This patch adds the support for OMAP remote processors to request
timer(s) to be used by the remoteproc. The timers are enabled and
disabled in line with the enabling/disabling of the remoteproc.
The timer data is not mandatory if the advanced device management
features are not required.
The core timer functionality is provided by the OMAP DMTimer
clocksource driver, which does not export any API. The logic is
implemented through the timer device's platform data ops. The OMAP
remoteproc driver mainly requires ops to request/free a dmtimer,
and to start/stop a timer. The split ops helps in controlling the
timer state without having to request and release a timer everytime
it needs to use the timer.
NOTE: If the gptimer is already in use by the time IPU and/or
DSP are loaded, the processors will fail to boot.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- small fix in error handling logic for enable timers
- build failure fix under enable timers (this was only visible with
compiling the patches sequentially, the failure was masked by patch
#14 in the series)
drivers/remoteproc/omap_remoteproc.c | 276 ++++++++++++++++++++++++++-
1 file changed, 275 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 777231ce2bae..293996af4797 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -26,6 +26,9 @@
#include <linux/regmap.h>
#include <linux/mfd/syscon.h>
#include <linux/reset.h>
+#include <clocksource/timer-ti-dm.h>
+
+#include <linux/platform_data/dmtimer-omap.h>
#include "omap_remoteproc.h"
#include "remoteproc_internal.h"
@@ -57,6 +60,16 @@ struct omap_rproc_mem {
size_t size;
};
+/**
+ * struct omap_rproc_timer - data structure for a timer used by a omap rproc
+ * @odt: timer pointer
+ * @timer_ops: OMAP dmtimer ops for @odt timer
+ */
+struct omap_rproc_timer {
+ struct omap_dm_timer *odt;
+ const struct omap_dm_timer_ops *timer_ops;
+};
+
/**
* struct omap_rproc - omap remote processor state
* @mbox: mailbox channel handle
@@ -64,6 +77,8 @@ struct omap_rproc_mem {
* @boot_data: boot data structure for setting processor boot address
* @mem: internal memory regions data
* @num_mems: number of internal memory regions
+ * @num_timers: number of rproc timer(s)
+ * @timers: timer(s) info used by rproc
* @rproc: rproc handle
* @reset: reset handle
*/
@@ -73,6 +88,8 @@ struct omap_rproc {
struct omap_rproc_boot_data *boot_data;
struct omap_rproc_mem *mem;
int num_mems;
+ int num_timers;
+ struct omap_rproc_timer *timers;
struct rproc *rproc;
struct reset_control *reset;
};
@@ -97,6 +114,227 @@ struct omap_rproc_dev_data {
const struct omap_rproc_mem_data *mems;
};
+/**
+ * omap_rproc_request_timer() - request a timer for a remoteproc
+ * @dev: device requesting the timer
+ * @np: device node pointer to the desired timer
+ * @timer: handle to a struct omap_rproc_timer to return the timer handle
+ *
+ * This helper function is used primarily to request a timer associated with
+ * a remoteproc. The returned handle is stored in the .odt field of the
+ * @timer structure passed in, and is used to invoke other timer specific
+ * ops (like starting a timer either during device initialization or during
+ * a resume operation, or for stopping/freeing a timer).
+ *
+ * Returns 0 on success, otherwise an appropriate failure
+ */
+static int omap_rproc_request_timer(struct device *dev, struct device_node *np,
+ struct omap_rproc_timer *timer)
+{
+ int ret;
+
+ timer->odt = timer->timer_ops->request_by_node(np);
+ if (!timer->odt) {
+ dev_err(dev, "request for timer node %p failed\n", np);
+ return -EBUSY;
+ }
+
+ ret = timer->timer_ops->set_source(timer->odt, OMAP_TIMER_SRC_SYS_CLK);
+ if (ret) {
+ dev_err(dev, "error setting OMAP_TIMER_SRC_SYS_CLK as source for timer node %p\n",
+ np);
+ timer->timer_ops->free(timer->odt);
+ return ret;
+ }
+
+ /* clean counter, remoteproc code will set the value */
+ timer->timer_ops->set_load(timer->odt, 0, 0);
+
+ return 0;
+}
+
+/**
+ * omap_rproc_start_timer - start a timer for a remoteproc
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This helper function is used to start a timer associated with a remoteproc,
+ * obtained using the request_timer ops. The helper function needs to be
+ * invoked by the driver to start the timer (during device initialization)
+ * or to just resume the timer.
+ *
+ * Returns 0 on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_start_timer(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->start(timer->odt);
+}
+
+/**
+ * omap_rproc_stop_timer - stop a timer for a remoteproc
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This helper function is used to disable a timer associated with a
+ * remoteproc, and needs to be called either during a device shutdown
+ * or suspend operation. The separate helper function allows the driver
+ * to just stop a timer without having to release the timer during a
+ * suspend operation.
+ *
+ * Returns 0 on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_stop_timer(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->stop(timer->odt);
+}
+
+/**
+ * omap_rproc_release_timer - release a timer for a remoteproc
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This helper function is used primarily to release a timer associated
+ * with a remoteproc. The dmtimer will be available for other clients to
+ * use once released.
+ *
+ * Returns 0 on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_release_timer(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->free(timer->odt);
+}
+
+/**
+ * omap_rproc_enable_timers - enable the timers for a remoteproc
+ * @rproc: handle of a remote processor
+ * @configure: boolean flag used to acquire and configure the timer handle
+ *
+ * This function is used primarily to enable the timers associated with
+ * a remoteproc. The configure flag is provided to allow the driver to
+ * to either acquire and start a timer (during device initialization) or
+ * to just start a timer (during a resume operation).
+ */
+static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
+{
+ int i;
+ int ret = 0;
+ struct platform_device *tpdev;
+ struct dmtimer_platform_data *tpdata;
+ const struct omap_dm_timer_ops *timer_ops;
+ struct omap_rproc *oproc = rproc->priv;
+ struct omap_rproc_timer *timers = oproc->timers;
+ struct device *dev = rproc->dev.parent;
+ struct device_node *np = NULL;
+
+ if (!oproc->num_timers)
+ return 0;
+
+ if (!configure)
+ goto start_timers;
+
+ for (i = 0; i < oproc->num_timers; i++) {
+ np = of_parse_phandle(dev->of_node, "ti,timers", i);
+ if (!np) {
+ ret = -ENXIO;
+ dev_err(dev, "device node lookup for timer at index %d failed: %d\n",
+ i, ret);
+ goto free_timers;
+ }
+
+ tpdev = of_find_device_by_node(np);
+ if (!tpdev) {
+ ret = -ENODEV;
+ dev_err(dev, "could not get timer platform device\n");
+ goto put_node;
+ }
+
+ tpdata = dev_get_platdata(&tpdev->dev);
+ put_device(&tpdev->dev);
+ if (!tpdata) {
+ ret = -EINVAL;
+ dev_err(dev, "dmtimer pdata structure NULL\n");
+ goto put_node;
+ }
+
+ timer_ops = tpdata->timer_ops;
+ if (!timer_ops || !timer_ops->request_by_node ||
+ !timer_ops->set_source || !timer_ops->set_load ||
+ !timer_ops->free || !timer_ops->start ||
+ !timer_ops->stop) {
+ ret = -EINVAL;
+ dev_err(dev, "device does not have required timer ops\n");
+ goto put_node;
+ }
+
+ timers[i].timer_ops = timer_ops;
+ ret = omap_rproc_request_timer(dev, np, &timers[i]);
+ if (ret) {
+ dev_err(dev, "request for timer %p failed: %d\n", np,
+ ret);
+ goto put_node;
+ }
+ of_node_put(np);
+ }
+
+start_timers:
+ for (i = 0; i < oproc->num_timers; i++) {
+ ret = omap_rproc_start_timer(&timers[i]);
+ if (ret) {
+ dev_err(dev, "start timer %p failed failed: %d\n", np,
+ ret);
+ break;
+ }
+ }
+ if (ret) {
+ while (i >= 0) {
+ omap_rproc_stop_timer(&timers[i]);
+ i--;
+ }
+ goto put_node;
+ }
+ return 0;
+
+put_node:
+ if (configure)
+ of_node_put(np);
+free_timers:
+ while (i--) {
+ omap_rproc_release_timer(&timers[i]);
+ timers[i].odt = NULL;
+ timers[i].timer_ops = NULL;
+ }
+
+ return ret;
+}
+
+/**
+ * omap_rproc_disable_timers - disable the timers for a remoteproc
+ * @rproc: handle of a remote processor
+ * @configure: boolean flag used to release the timer handle
+ *
+ * This function is used primarily to disable the timers associated with
+ * a remoteproc. The configure flag is provided to allow the driver to
+ * to either stop and release a timer (during device shutdown) or to just
+ * stop a timer (during a suspend operation).
+ */
+static int omap_rproc_disable_timers(struct rproc *rproc, bool configure)
+{
+ int i;
+ struct omap_rproc *oproc = rproc->priv;
+ struct omap_rproc_timer *timers = oproc->timers;
+
+ if (!oproc->num_timers)
+ return 0;
+
+ for (i = 0; i < oproc->num_timers; i++) {
+ omap_rproc_stop_timer(&timers[i]);
+ if (configure) {
+ omap_rproc_release_timer(&timers[i]);
+ timers[i].odt = NULL;
+ timers[i].timer_ops = NULL;
+ }
+ }
+
+ return 0;
+}
+
/**
* omap_rproc_mbox_callback() - inbound mailbox message handler
* @client: mailbox client pointer used for requesting the mailbox channel
@@ -232,14 +470,22 @@ static int omap_rproc_start(struct rproc *rproc)
goto put_mbox;
}
+ ret = omap_rproc_enable_timers(rproc, true);
+ if (ret) {
+ dev_err(dev, "omap_rproc_enable_timers failed: %d\n", ret);
+ goto put_mbox;
+ }
+
ret = reset_control_deassert(oproc->reset);
if (ret) {
dev_err(dev, "reset control deassert failed: %d\n", ret);
- goto put_mbox;
+ goto disable_timers;
}
return 0;
+disable_timers:
+ omap_rproc_disable_timers(rproc, true);
put_mbox:
mbox_free_channel(oproc->mbox);
return ret;
@@ -255,6 +501,10 @@ static int omap_rproc_stop(struct rproc *rproc)
if (ret)
return ret;
+ ret = omap_rproc_disable_timers(rproc, true);
+ if (ret)
+ return ret;
+
mbox_free_channel(oproc->mbox);
return 0;
@@ -523,6 +773,30 @@ static int omap_rproc_probe(struct platform_device *pdev)
if (ret)
goto free_rproc;
+ /*
+ * Timer nodes are directly used in client nodes as phandles, so
+ * retrieve the count using appropriate size
+ */
+ oproc->num_timers = of_count_phandle_with_args(np, "ti,timers", NULL);
+ if (oproc->num_timers <= 0) {
+ dev_dbg(&pdev->dev, "device does not have timers, status = %d\n",
+ oproc->num_timers);
+ oproc->num_timers = 0;
+ }
+
+ if (oproc->num_timers) {
+ oproc->timers = devm_kcalloc(&pdev->dev, oproc->num_timers,
+ sizeof(*oproc->timers),
+ GFP_KERNEL);
+ if (!oproc->timers) {
+ ret = -ENOMEM;
+ goto free_rproc;
+ }
+
+ dev_dbg(&pdev->dev, "device has %d tick timers\n",
+ oproc->num_timers);
+ }
+
ret = of_reserved_mem_device_init(&pdev->dev);
if (ret) {
dev_err(&pdev->dev, "device does not have specific CMA pool\n");
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 10/14] remoteproc/omap: Request a timer(s) for remoteproc usage
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The remote processors in OMAP4+ SoCs are equipped with internal
timers, like the internal SysTick timer in a Cortex M3/M4 NVIC or
the CTM timer within Unicache in IPU & DSP. However, these timers
are gated when the processor subsystem clock is gated, making
them rather difficult to use as OS tick sources. They will not
be able to wakeup the processor from any processor-sleep induced
clock-gating states.
This can be avoided by using an external timer as the tick source,
which can be controlled independently by the OMAP remoteproc
driver code, but still allowing the processor subsystem clock to
be auto-gated when the remoteproc cores are idle.
This patch adds the support for OMAP remote processors to request
timer(s) to be used by the remoteproc. The timers are enabled and
disabled in line with the enabling/disabling of the remoteproc.
The timer data is not mandatory if the advanced device management
features are not required.
The core timer functionality is provided by the OMAP DMTimer
clocksource driver, which does not export any API. The logic is
implemented through the timer device's platform data ops. The OMAP
remoteproc driver mainly requires ops to request/free a dmtimer,
and to start/stop a timer. The split ops helps in controlling the
timer state without having to request and release a timer everytime
it needs to use the timer.
NOTE: If the gptimer is already in use by the time IPU and/or
DSP are loaded, the processors will fail to boot.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- small fix in error handling logic for enable timers
- build failure fix under enable timers (this was only visible with
compiling the patches sequentially, the failure was masked by patch
#14 in the series)
drivers/remoteproc/omap_remoteproc.c | 276 ++++++++++++++++++++++++++-
1 file changed, 275 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 777231ce2bae..293996af4797 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -26,6 +26,9 @@
#include <linux/regmap.h>
#include <linux/mfd/syscon.h>
#include <linux/reset.h>
+#include <clocksource/timer-ti-dm.h>
+
+#include <linux/platform_data/dmtimer-omap.h>
#include "omap_remoteproc.h"
#include "remoteproc_internal.h"
@@ -57,6 +60,16 @@ struct omap_rproc_mem {
size_t size;
};
+/**
+ * struct omap_rproc_timer - data structure for a timer used by a omap rproc
+ * @odt: timer pointer
+ * @timer_ops: OMAP dmtimer ops for @odt timer
+ */
+struct omap_rproc_timer {
+ struct omap_dm_timer *odt;
+ const struct omap_dm_timer_ops *timer_ops;
+};
+
/**
* struct omap_rproc - omap remote processor state
* @mbox: mailbox channel handle
@@ -64,6 +77,8 @@ struct omap_rproc_mem {
* @boot_data: boot data structure for setting processor boot address
* @mem: internal memory regions data
* @num_mems: number of internal memory regions
+ * @num_timers: number of rproc timer(s)
+ * @timers: timer(s) info used by rproc
* @rproc: rproc handle
* @reset: reset handle
*/
@@ -73,6 +88,8 @@ struct omap_rproc {
struct omap_rproc_boot_data *boot_data;
struct omap_rproc_mem *mem;
int num_mems;
+ int num_timers;
+ struct omap_rproc_timer *timers;
struct rproc *rproc;
struct reset_control *reset;
};
@@ -97,6 +114,227 @@ struct omap_rproc_dev_data {
const struct omap_rproc_mem_data *mems;
};
+/**
+ * omap_rproc_request_timer() - request a timer for a remoteproc
+ * @dev: device requesting the timer
+ * @np: device node pointer to the desired timer
+ * @timer: handle to a struct omap_rproc_timer to return the timer handle
+ *
+ * This helper function is used primarily to request a timer associated with
+ * a remoteproc. The returned handle is stored in the .odt field of the
+ * @timer structure passed in, and is used to invoke other timer specific
+ * ops (like starting a timer either during device initialization or during
+ * a resume operation, or for stopping/freeing a timer).
+ *
+ * Returns 0 on success, otherwise an appropriate failure
+ */
+static int omap_rproc_request_timer(struct device *dev, struct device_node *np,
+ struct omap_rproc_timer *timer)
+{
+ int ret;
+
+ timer->odt = timer->timer_ops->request_by_node(np);
+ if (!timer->odt) {
+ dev_err(dev, "request for timer node %p failed\n", np);
+ return -EBUSY;
+ }
+
+ ret = timer->timer_ops->set_source(timer->odt, OMAP_TIMER_SRC_SYS_CLK);
+ if (ret) {
+ dev_err(dev, "error setting OMAP_TIMER_SRC_SYS_CLK as source for timer node %p\n",
+ np);
+ timer->timer_ops->free(timer->odt);
+ return ret;
+ }
+
+ /* clean counter, remoteproc code will set the value */
+ timer->timer_ops->set_load(timer->odt, 0, 0);
+
+ return 0;
+}
+
+/**
+ * omap_rproc_start_timer - start a timer for a remoteproc
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This helper function is used to start a timer associated with a remoteproc,
+ * obtained using the request_timer ops. The helper function needs to be
+ * invoked by the driver to start the timer (during device initialization)
+ * or to just resume the timer.
+ *
+ * Returns 0 on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_start_timer(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->start(timer->odt);
+}
+
+/**
+ * omap_rproc_stop_timer - stop a timer for a remoteproc
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This helper function is used to disable a timer associated with a
+ * remoteproc, and needs to be called either during a device shutdown
+ * or suspend operation. The separate helper function allows the driver
+ * to just stop a timer without having to release the timer during a
+ * suspend operation.
+ *
+ * Returns 0 on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_stop_timer(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->stop(timer->odt);
+}
+
+/**
+ * omap_rproc_release_timer - release a timer for a remoteproc
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This helper function is used primarily to release a timer associated
+ * with a remoteproc. The dmtimer will be available for other clients to
+ * use once released.
+ *
+ * Returns 0 on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_release_timer(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->free(timer->odt);
+}
+
+/**
+ * omap_rproc_enable_timers - enable the timers for a remoteproc
+ * @rproc: handle of a remote processor
+ * @configure: boolean flag used to acquire and configure the timer handle
+ *
+ * This function is used primarily to enable the timers associated with
+ * a remoteproc. The configure flag is provided to allow the driver to
+ * to either acquire and start a timer (during device initialization) or
+ * to just start a timer (during a resume operation).
+ */
+static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
+{
+ int i;
+ int ret = 0;
+ struct platform_device *tpdev;
+ struct dmtimer_platform_data *tpdata;
+ const struct omap_dm_timer_ops *timer_ops;
+ struct omap_rproc *oproc = rproc->priv;
+ struct omap_rproc_timer *timers = oproc->timers;
+ struct device *dev = rproc->dev.parent;
+ struct device_node *np = NULL;
+
+ if (!oproc->num_timers)
+ return 0;
+
+ if (!configure)
+ goto start_timers;
+
+ for (i = 0; i < oproc->num_timers; i++) {
+ np = of_parse_phandle(dev->of_node, "ti,timers", i);
+ if (!np) {
+ ret = -ENXIO;
+ dev_err(dev, "device node lookup for timer at index %d failed: %d\n",
+ i, ret);
+ goto free_timers;
+ }
+
+ tpdev = of_find_device_by_node(np);
+ if (!tpdev) {
+ ret = -ENODEV;
+ dev_err(dev, "could not get timer platform device\n");
+ goto put_node;
+ }
+
+ tpdata = dev_get_platdata(&tpdev->dev);
+ put_device(&tpdev->dev);
+ if (!tpdata) {
+ ret = -EINVAL;
+ dev_err(dev, "dmtimer pdata structure NULL\n");
+ goto put_node;
+ }
+
+ timer_ops = tpdata->timer_ops;
+ if (!timer_ops || !timer_ops->request_by_node ||
+ !timer_ops->set_source || !timer_ops->set_load ||
+ !timer_ops->free || !timer_ops->start ||
+ !timer_ops->stop) {
+ ret = -EINVAL;
+ dev_err(dev, "device does not have required timer ops\n");
+ goto put_node;
+ }
+
+ timers[i].timer_ops = timer_ops;
+ ret = omap_rproc_request_timer(dev, np, &timers[i]);
+ if (ret) {
+ dev_err(dev, "request for timer %p failed: %d\n", np,
+ ret);
+ goto put_node;
+ }
+ of_node_put(np);
+ }
+
+start_timers:
+ for (i = 0; i < oproc->num_timers; i++) {
+ ret = omap_rproc_start_timer(&timers[i]);
+ if (ret) {
+ dev_err(dev, "start timer %p failed failed: %d\n", np,
+ ret);
+ break;
+ }
+ }
+ if (ret) {
+ while (i >= 0) {
+ omap_rproc_stop_timer(&timers[i]);
+ i--;
+ }
+ goto put_node;
+ }
+ return 0;
+
+put_node:
+ if (configure)
+ of_node_put(np);
+free_timers:
+ while (i--) {
+ omap_rproc_release_timer(&timers[i]);
+ timers[i].odt = NULL;
+ timers[i].timer_ops = NULL;
+ }
+
+ return ret;
+}
+
+/**
+ * omap_rproc_disable_timers - disable the timers for a remoteproc
+ * @rproc: handle of a remote processor
+ * @configure: boolean flag used to release the timer handle
+ *
+ * This function is used primarily to disable the timers associated with
+ * a remoteproc. The configure flag is provided to allow the driver to
+ * to either stop and release a timer (during device shutdown) or to just
+ * stop a timer (during a suspend operation).
+ */
+static int omap_rproc_disable_timers(struct rproc *rproc, bool configure)
+{
+ int i;
+ struct omap_rproc *oproc = rproc->priv;
+ struct omap_rproc_timer *timers = oproc->timers;
+
+ if (!oproc->num_timers)
+ return 0;
+
+ for (i = 0; i < oproc->num_timers; i++) {
+ omap_rproc_stop_timer(&timers[i]);
+ if (configure) {
+ omap_rproc_release_timer(&timers[i]);
+ timers[i].odt = NULL;
+ timers[i].timer_ops = NULL;
+ }
+ }
+
+ return 0;
+}
+
/**
* omap_rproc_mbox_callback() - inbound mailbox message handler
* @client: mailbox client pointer used for requesting the mailbox channel
@@ -232,14 +470,22 @@ static int omap_rproc_start(struct rproc *rproc)
goto put_mbox;
}
+ ret = omap_rproc_enable_timers(rproc, true);
+ if (ret) {
+ dev_err(dev, "omap_rproc_enable_timers failed: %d\n", ret);
+ goto put_mbox;
+ }
+
ret = reset_control_deassert(oproc->reset);
if (ret) {
dev_err(dev, "reset control deassert failed: %d\n", ret);
- goto put_mbox;
+ goto disable_timers;
}
return 0;
+disable_timers:
+ omap_rproc_disable_timers(rproc, true);
put_mbox:
mbox_free_channel(oproc->mbox);
return ret;
@@ -255,6 +501,10 @@ static int omap_rproc_stop(struct rproc *rproc)
if (ret)
return ret;
+ ret = omap_rproc_disable_timers(rproc, true);
+ if (ret)
+ return ret;
+
mbox_free_channel(oproc->mbox);
return 0;
@@ -523,6 +773,30 @@ static int omap_rproc_probe(struct platform_device *pdev)
if (ret)
goto free_rproc;
+ /*
+ * Timer nodes are directly used in client nodes as phandles, so
+ * retrieve the count using appropriate size
+ */
+ oproc->num_timers = of_count_phandle_with_args(np, "ti,timers", NULL);
+ if (oproc->num_timers <= 0) {
+ dev_dbg(&pdev->dev, "device does not have timers, status = %d\n",
+ oproc->num_timers);
+ oproc->num_timers = 0;
+ }
+
+ if (oproc->num_timers) {
+ oproc->timers = devm_kcalloc(&pdev->dev, oproc->num_timers,
+ sizeof(*oproc->timers),
+ GFP_KERNEL);
+ if (!oproc->timers) {
+ ret = -ENOMEM;
+ goto free_rproc;
+ }
+
+ dev_dbg(&pdev->dev, "device has %d tick timers\n",
+ oproc->num_timers);
+ }
+
ret = of_reserved_mem_device_init(&pdev->dev);
if (ret) {
dev_err(&pdev->dev, "device does not have specific CMA pool\n");
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 11/14] remoteproc/omap: add support for system suspend/resume
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
This patch adds the support for system suspend/resume to the
OMAP remoteproc driver so that the OMAP remoteproc devices can
be suspended/resumed during a system suspend/resume. The support
is added through the driver PM .suspend/.resume callbacks, and
requires appropriate support from the OS running on the remote
processors.
The IPU & DSP remote processors typically have their own private
modules like registers, internal memories, caches etc. The context
of these modules need to be saved and restored properly for a
suspend/resume to work. These are in general not accessible from
the MPU, so the remote processors themselves have to implement
the logic for the context save & restore of these modules.
The OMAP remoteproc driver initiates a suspend by sending a mailbox
message requesting the remote processor to save its context and
enter into an idle/standby state. The remote processor should
usually stop whatever processing it is doing to switch to a context
save mode. The OMAP remoteproc driver detects the completion of
the context save by checking the module standby status for the
remoteproc device. It also stops any resources used by the remote
processors like the timers. The timers need to be running only
when the processor is active and executing, and need to be stopped
otherwise to allow the timer driver to reach low-power states. The
IOMMUs are automatically suspended by the PM core during the late
suspend stage, after the remoteproc suspend process is completed by
putting the remote processor cores into reset. Thereafter, the Linux
kernel can put the domain into further lower power states as possible.
The resume sequence undoes the operations performed in the PM suspend
callback, by starting the timers and finally releasing the processors
from reset. This requires that the remote processor side OS be able to
distinguish a power-resume boot from a power-on/cold boot, restore the
context of its private modules saved during the suspend phase, and
resume executing code from where it was suspended. The IOMMUs would
have been resumed by the PM core during early resume, so they are
already enabled by the time remoteproc resume callback gets invoked.
The remote processors should save their context into System RAM (DDR),
as any internal memories are not guaranteed to retain context as it
depends on the lowest power domain that the remote processor device
is put into. The management of the DDR contents will be managed by
the Linux kernel.
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to use ti-sysc instead of hwmod]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- add error check for reset_control_deassert under rproc_suspend
- removed CONFIG_PM_SLEEP macro usage
drivers/remoteproc/omap_remoteproc.c | 188 +++++++++++++++++++++++++++
drivers/remoteproc/omap_remoteproc.h | 18 ++-
2 files changed, 204 insertions(+), 2 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 293996af4797..452ef6070175 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -15,13 +15,17 @@
#include <linux/kernel.h>
#include <linux/module.h>
+#include <linux/clk.h>
+#include <linux/clk/ti.h>
#include <linux/err.h>
+#include <linux/io.h>
#include <linux/of_device.h>
#include <linux/of_reserved_mem.h>
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
+#include <linux/omap-iommu.h>
#include <linux/omap-mailbox.h>
#include <linux/regmap.h>
#include <linux/mfd/syscon.h>
@@ -81,6 +85,9 @@ struct omap_rproc_timer {
* @timers: timer(s) info used by rproc
* @rproc: rproc handle
* @reset: reset handle
+ * @pm_comp: completion primitive to sync for suspend response
+ * @fck: functional clock for the remoteproc
+ * @suspend_acked: state machine flag to store the suspend request ack
*/
struct omap_rproc {
struct mbox_chan *mbox;
@@ -92,6 +99,9 @@ struct omap_rproc {
struct omap_rproc_timer *timers;
struct rproc *rproc;
struct reset_control *reset;
+ struct completion pm_comp;
+ struct clk *fck;
+ bool suspend_acked;
};
/**
@@ -367,6 +377,12 @@ static void omap_rproc_mbox_callback(struct mbox_client *client, void *data)
case RP_MBOX_ECHO_REPLY:
dev_info(dev, "received echo reply from %s\n", name);
break;
+ case RP_MBOX_SUSPEND_ACK:
+ /* Fall through */
+ case RP_MBOX_SUSPEND_CANCEL:
+ oproc->suspend_acked = msg == RP_MBOX_SUSPEND_ACK;
+ complete(&oproc->pm_comp);
+ break;
default:
if (msg >= RP_MBOX_READY && msg < RP_MBOX_END_MSG)
return;
@@ -555,6 +571,165 @@ static const struct rproc_ops omap_rproc_ops = {
.da_to_va = omap_rproc_da_to_va,
};
+#ifdef CONFIG_PM
+static bool _is_rproc_in_standby(struct omap_rproc *oproc)
+{
+ return ti_clk_is_in_standby(oproc->fck);
+}
+
+/* 1 sec is long enough time to let the remoteproc side suspend the device */
+#define DEF_SUSPEND_TIMEOUT 1000
+static int _omap_rproc_suspend(struct rproc *rproc)
+{
+ struct device *dev = rproc->dev.parent;
+ struct omap_rproc *oproc = rproc->priv;
+ unsigned long to = msecs_to_jiffies(DEF_SUSPEND_TIMEOUT);
+ unsigned long ta = jiffies + to;
+ int ret;
+
+ reinit_completion(&oproc->pm_comp);
+ oproc->suspend_acked = false;
+ ret = mbox_send_message(oproc->mbox, (void *)RP_MBOX_SUSPEND_SYSTEM);
+ if (ret < 0) {
+ dev_err(dev, "PM mbox_send_message failed: %d\n", ret);
+ return ret;
+ }
+
+ ret = wait_for_completion_timeout(&oproc->pm_comp, to);
+ if (!oproc->suspend_acked)
+ return -EBUSY;
+
+ /*
+ * The remoteproc side is returning the ACK message before saving the
+ * context, because the context saving is performed within a SYS/BIOS
+ * function, and it cannot have any inter-dependencies against the IPC
+ * layer. Also, as the SYS/BIOS needs to preserve properly the processor
+ * register set, sending this ACK or signalling the completion of the
+ * context save through a shared memory variable can never be the
+ * absolute last thing to be executed on the remoteproc side, and the
+ * MPU cannot use the ACK message as a sync point to put the remoteproc
+ * into reset. The only way to ensure that the remote processor has
+ * completed saving the context is to check that the module has reached
+ * STANDBY state (after saving the context, the SYS/BIOS executes the
+ * appropriate target-specific WFI instruction causing the module to
+ * enter STANDBY).
+ */
+ while (!_is_rproc_in_standby(oproc)) {
+ if (time_after(jiffies, ta))
+ return -ETIME;
+ schedule();
+ }
+
+ reset_control_assert(oproc->reset);
+
+ ret = omap_rproc_disable_timers(rproc, false);
+ if (ret) {
+ dev_err(dev, "disabling timers during suspend failed %d\n",
+ ret);
+ goto enable_device;
+ }
+
+ return 0;
+
+enable_device:
+ reset_control_deassert(oproc->reset);
+ return ret;
+}
+
+static int _omap_rproc_resume(struct rproc *rproc)
+{
+ struct device *dev = rproc->dev.parent;
+ struct omap_rproc *oproc = rproc->priv;
+ int ret;
+
+ /* boot address could be lost after suspend, so restore it */
+ if (oproc->boot_data) {
+ ret = omap_rproc_write_dsp_boot_addr(rproc);
+ if (ret) {
+ dev_err(dev, "boot address restore failed %d\n", ret);
+ goto out;
+ }
+ }
+
+ ret = omap_rproc_enable_timers(rproc, false);
+ if (ret) {
+ dev_err(dev, "enabling timers during resume failed %d\n", ret);
+ goto out;
+ }
+
+ ret = reset_control_deassert(oproc->reset);
+ if (ret) {
+ dev_err(dev, "reset deassert failed %d\n", ret);
+ goto disable_timers;
+ }
+
+ return 0;
+
+disable_timers:
+ omap_rproc_disable_timers(rproc, false);
+
+out:
+ return ret;
+}
+
+static int __maybe_unused omap_rproc_suspend(struct device *dev)
+{
+ struct platform_device *pdev = to_platform_device(dev);
+ struct rproc *rproc = platform_get_drvdata(pdev);
+ int ret = 0;
+
+ mutex_lock(&rproc->lock);
+ if (rproc->state == RPROC_OFFLINE)
+ goto out;
+
+ if (rproc->state == RPROC_SUSPENDED)
+ goto out;
+
+ if (rproc->state != RPROC_RUNNING) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ ret = _omap_rproc_suspend(rproc);
+ if (ret) {
+ dev_err(dev, "suspend failed %d\n", ret);
+ goto out;
+ }
+
+ rproc->state = RPROC_SUSPENDED;
+out:
+ mutex_unlock(&rproc->lock);
+ return ret;
+}
+
+static int __maybe_unused omap_rproc_resume(struct device *dev)
+{
+ struct platform_device *pdev = to_platform_device(dev);
+ struct rproc *rproc = platform_get_drvdata(pdev);
+ int ret = 0;
+
+ mutex_lock(&rproc->lock);
+ if (rproc->state == RPROC_OFFLINE)
+ goto out;
+
+ if (rproc->state != RPROC_SUSPENDED) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ ret = _omap_rproc_resume(rproc);
+ if (ret) {
+ dev_err(dev, "resume failed %d\n", ret);
+ goto out;
+ }
+
+ rproc->state = RPROC_RUNNING;
+out:
+ mutex_unlock(&rproc->lock);
+ return ret;
+}
+#endif /* CONFIG_PM */
+
static const struct omap_rproc_mem_data ipu_mems[] = {
{ .name = "l2ram", .dev_addr = 0x20000000 },
{ },
@@ -797,6 +972,14 @@ static int omap_rproc_probe(struct platform_device *pdev)
oproc->num_timers);
}
+ init_completion(&oproc->pm_comp);
+
+ oproc->fck = devm_clk_get(&pdev->dev, 0);
+ if (IS_ERR(oproc->fck)) {
+ ret = PTR_ERR(oproc->fck);
+ goto free_rproc;
+ }
+
ret = of_reserved_mem_device_init(&pdev->dev);
if (ret) {
dev_err(&pdev->dev, "device does not have specific CMA pool\n");
@@ -829,11 +1012,16 @@ static int omap_rproc_remove(struct platform_device *pdev)
return 0;
}
+static const struct dev_pm_ops omap_rproc_pm_ops = {
+ SET_SYSTEM_SLEEP_PM_OPS(omap_rproc_suspend, omap_rproc_resume)
+};
+
static struct platform_driver omap_rproc_driver = {
.probe = omap_rproc_probe,
.remove = omap_rproc_remove,
.driver = {
.name = "omap-rproc",
+ .pm = &omap_rproc_pm_ops,
.of_match_table = omap_rproc_of_match,
},
};
diff --git a/drivers/remoteproc/omap_remoteproc.h b/drivers/remoteproc/omap_remoteproc.h
index 72f656c93caa..13f17d9135c0 100644
--- a/drivers/remoteproc/omap_remoteproc.h
+++ b/drivers/remoteproc/omap_remoteproc.h
@@ -1,7 +1,7 @@
/*
* Remote processor messaging
*
- * Copyright (C) 2011 Texas Instruments, Inc.
+ * Copyright (C) 2011-2020 Texas Instruments, Inc.
* Copyright (C) 2011 Google, Inc.
* All rights reserved.
*
@@ -57,6 +57,16 @@
* @RP_MBOX_ABORT_REQUEST: a "please crash" request, used for testing the
* recovery mechanism (to some extent).
*
+ * @RP_MBOX_SUSPEND_AUTO: auto suspend request for the remote processor
+ *
+ * @RP_MBOX_SUSPEND_SYSTEM: system suspend request for the remote processor
+ *
+ * @RP_MBOX_SUSPEND_ACK: successful response from remote processor for a
+ * suspend request
+ *
+ * @RP_MBOX_SUSPEND_CANCEL: a cancel suspend response from a remote processor
+ * on a suspend request
+ *
* Introduce new message definitions if any here.
*
* @RP_MBOX_END_MSG: Indicates end of known/defined messages from remote core
@@ -70,7 +80,11 @@ enum omap_rp_mbox_messages {
RP_MBOX_ECHO_REQUEST = 0xFFFFFF03,
RP_MBOX_ECHO_REPLY = 0xFFFFFF04,
RP_MBOX_ABORT_REQUEST = 0xFFFFFF05,
- RP_MBOX_END_MSG = 0xFFFFFF06,
+ RP_MBOX_SUSPEND_AUTO = 0xFFFFFF10,
+ RP_MBOX_SUSPEND_SYSTEM = 0xFFFFFF11,
+ RP_MBOX_SUSPEND_ACK = 0xFFFFFF12,
+ RP_MBOX_SUSPEND_CANCEL = 0xFFFFFF13,
+ RP_MBOX_END_MSG = 0xFFFFFF14,
};
#endif /* _OMAP_RPMSG_H */
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 11/14] remoteproc/omap: add support for system suspend/resume
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
This patch adds the support for system suspend/resume to the
OMAP remoteproc driver so that the OMAP remoteproc devices can
be suspended/resumed during a system suspend/resume. The support
is added through the driver PM .suspend/.resume callbacks, and
requires appropriate support from the OS running on the remote
processors.
The IPU & DSP remote processors typically have their own private
modules like registers, internal memories, caches etc. The context
of these modules need to be saved and restored properly for a
suspend/resume to work. These are in general not accessible from
the MPU, so the remote processors themselves have to implement
the logic for the context save & restore of these modules.
The OMAP remoteproc driver initiates a suspend by sending a mailbox
message requesting the remote processor to save its context and
enter into an idle/standby state. The remote processor should
usually stop whatever processing it is doing to switch to a context
save mode. The OMAP remoteproc driver detects the completion of
the context save by checking the module standby status for the
remoteproc device. It also stops any resources used by the remote
processors like the timers. The timers need to be running only
when the processor is active and executing, and need to be stopped
otherwise to allow the timer driver to reach low-power states. The
IOMMUs are automatically suspended by the PM core during the late
suspend stage, after the remoteproc suspend process is completed by
putting the remote processor cores into reset. Thereafter, the Linux
kernel can put the domain into further lower power states as possible.
The resume sequence undoes the operations performed in the PM suspend
callback, by starting the timers and finally releasing the processors
from reset. This requires that the remote processor side OS be able to
distinguish a power-resume boot from a power-on/cold boot, restore the
context of its private modules saved during the suspend phase, and
resume executing code from where it was suspended. The IOMMUs would
have been resumed by the PM core during early resume, so they are
already enabled by the time remoteproc resume callback gets invoked.
The remote processors should save their context into System RAM (DDR),
as any internal memories are not guaranteed to retain context as it
depends on the lowest power domain that the remote processor device
is put into. The management of the DDR contents will be managed by
the Linux kernel.
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to use ti-sysc instead of hwmod]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- add error check for reset_control_deassert under rproc_suspend
- removed CONFIG_PM_SLEEP macro usage
drivers/remoteproc/omap_remoteproc.c | 188 +++++++++++++++++++++++++++
drivers/remoteproc/omap_remoteproc.h | 18 ++-
2 files changed, 204 insertions(+), 2 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 293996af4797..452ef6070175 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -15,13 +15,17 @@
#include <linux/kernel.h>
#include <linux/module.h>
+#include <linux/clk.h>
+#include <linux/clk/ti.h>
#include <linux/err.h>
+#include <linux/io.h>
#include <linux/of_device.h>
#include <linux/of_reserved_mem.h>
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
+#include <linux/omap-iommu.h>
#include <linux/omap-mailbox.h>
#include <linux/regmap.h>
#include <linux/mfd/syscon.h>
@@ -81,6 +85,9 @@ struct omap_rproc_timer {
* @timers: timer(s) info used by rproc
* @rproc: rproc handle
* @reset: reset handle
+ * @pm_comp: completion primitive to sync for suspend response
+ * @fck: functional clock for the remoteproc
+ * @suspend_acked: state machine flag to store the suspend request ack
*/
struct omap_rproc {
struct mbox_chan *mbox;
@@ -92,6 +99,9 @@ struct omap_rproc {
struct omap_rproc_timer *timers;
struct rproc *rproc;
struct reset_control *reset;
+ struct completion pm_comp;
+ struct clk *fck;
+ bool suspend_acked;
};
/**
@@ -367,6 +377,12 @@ static void omap_rproc_mbox_callback(struct mbox_client *client, void *data)
case RP_MBOX_ECHO_REPLY:
dev_info(dev, "received echo reply from %s\n", name);
break;
+ case RP_MBOX_SUSPEND_ACK:
+ /* Fall through */
+ case RP_MBOX_SUSPEND_CANCEL:
+ oproc->suspend_acked = msg == RP_MBOX_SUSPEND_ACK;
+ complete(&oproc->pm_comp);
+ break;
default:
if (msg >= RP_MBOX_READY && msg < RP_MBOX_END_MSG)
return;
@@ -555,6 +571,165 @@ static const struct rproc_ops omap_rproc_ops = {
.da_to_va = omap_rproc_da_to_va,
};
+#ifdef CONFIG_PM
+static bool _is_rproc_in_standby(struct omap_rproc *oproc)
+{
+ return ti_clk_is_in_standby(oproc->fck);
+}
+
+/* 1 sec is long enough time to let the remoteproc side suspend the device */
+#define DEF_SUSPEND_TIMEOUT 1000
+static int _omap_rproc_suspend(struct rproc *rproc)
+{
+ struct device *dev = rproc->dev.parent;
+ struct omap_rproc *oproc = rproc->priv;
+ unsigned long to = msecs_to_jiffies(DEF_SUSPEND_TIMEOUT);
+ unsigned long ta = jiffies + to;
+ int ret;
+
+ reinit_completion(&oproc->pm_comp);
+ oproc->suspend_acked = false;
+ ret = mbox_send_message(oproc->mbox, (void *)RP_MBOX_SUSPEND_SYSTEM);
+ if (ret < 0) {
+ dev_err(dev, "PM mbox_send_message failed: %d\n", ret);
+ return ret;
+ }
+
+ ret = wait_for_completion_timeout(&oproc->pm_comp, to);
+ if (!oproc->suspend_acked)
+ return -EBUSY;
+
+ /*
+ * The remoteproc side is returning the ACK message before saving the
+ * context, because the context saving is performed within a SYS/BIOS
+ * function, and it cannot have any inter-dependencies against the IPC
+ * layer. Also, as the SYS/BIOS needs to preserve properly the processor
+ * register set, sending this ACK or signalling the completion of the
+ * context save through a shared memory variable can never be the
+ * absolute last thing to be executed on the remoteproc side, and the
+ * MPU cannot use the ACK message as a sync point to put the remoteproc
+ * into reset. The only way to ensure that the remote processor has
+ * completed saving the context is to check that the module has reached
+ * STANDBY state (after saving the context, the SYS/BIOS executes the
+ * appropriate target-specific WFI instruction causing the module to
+ * enter STANDBY).
+ */
+ while (!_is_rproc_in_standby(oproc)) {
+ if (time_after(jiffies, ta))
+ return -ETIME;
+ schedule();
+ }
+
+ reset_control_assert(oproc->reset);
+
+ ret = omap_rproc_disable_timers(rproc, false);
+ if (ret) {
+ dev_err(dev, "disabling timers during suspend failed %d\n",
+ ret);
+ goto enable_device;
+ }
+
+ return 0;
+
+enable_device:
+ reset_control_deassert(oproc->reset);
+ return ret;
+}
+
+static int _omap_rproc_resume(struct rproc *rproc)
+{
+ struct device *dev = rproc->dev.parent;
+ struct omap_rproc *oproc = rproc->priv;
+ int ret;
+
+ /* boot address could be lost after suspend, so restore it */
+ if (oproc->boot_data) {
+ ret = omap_rproc_write_dsp_boot_addr(rproc);
+ if (ret) {
+ dev_err(dev, "boot address restore failed %d\n", ret);
+ goto out;
+ }
+ }
+
+ ret = omap_rproc_enable_timers(rproc, false);
+ if (ret) {
+ dev_err(dev, "enabling timers during resume failed %d\n", ret);
+ goto out;
+ }
+
+ ret = reset_control_deassert(oproc->reset);
+ if (ret) {
+ dev_err(dev, "reset deassert failed %d\n", ret);
+ goto disable_timers;
+ }
+
+ return 0;
+
+disable_timers:
+ omap_rproc_disable_timers(rproc, false);
+
+out:
+ return ret;
+}
+
+static int __maybe_unused omap_rproc_suspend(struct device *dev)
+{
+ struct platform_device *pdev = to_platform_device(dev);
+ struct rproc *rproc = platform_get_drvdata(pdev);
+ int ret = 0;
+
+ mutex_lock(&rproc->lock);
+ if (rproc->state == RPROC_OFFLINE)
+ goto out;
+
+ if (rproc->state == RPROC_SUSPENDED)
+ goto out;
+
+ if (rproc->state != RPROC_RUNNING) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ ret = _omap_rproc_suspend(rproc);
+ if (ret) {
+ dev_err(dev, "suspend failed %d\n", ret);
+ goto out;
+ }
+
+ rproc->state = RPROC_SUSPENDED;
+out:
+ mutex_unlock(&rproc->lock);
+ return ret;
+}
+
+static int __maybe_unused omap_rproc_resume(struct device *dev)
+{
+ struct platform_device *pdev = to_platform_device(dev);
+ struct rproc *rproc = platform_get_drvdata(pdev);
+ int ret = 0;
+
+ mutex_lock(&rproc->lock);
+ if (rproc->state == RPROC_OFFLINE)
+ goto out;
+
+ if (rproc->state != RPROC_SUSPENDED) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ ret = _omap_rproc_resume(rproc);
+ if (ret) {
+ dev_err(dev, "resume failed %d\n", ret);
+ goto out;
+ }
+
+ rproc->state = RPROC_RUNNING;
+out:
+ mutex_unlock(&rproc->lock);
+ return ret;
+}
+#endif /* CONFIG_PM */
+
static const struct omap_rproc_mem_data ipu_mems[] = {
{ .name = "l2ram", .dev_addr = 0x20000000 },
{ },
@@ -797,6 +972,14 @@ static int omap_rproc_probe(struct platform_device *pdev)
oproc->num_timers);
}
+ init_completion(&oproc->pm_comp);
+
+ oproc->fck = devm_clk_get(&pdev->dev, 0);
+ if (IS_ERR(oproc->fck)) {
+ ret = PTR_ERR(oproc->fck);
+ goto free_rproc;
+ }
+
ret = of_reserved_mem_device_init(&pdev->dev);
if (ret) {
dev_err(&pdev->dev, "device does not have specific CMA pool\n");
@@ -829,11 +1012,16 @@ static int omap_rproc_remove(struct platform_device *pdev)
return 0;
}
+static const struct dev_pm_ops omap_rproc_pm_ops = {
+ SET_SYSTEM_SLEEP_PM_OPS(omap_rproc_suspend, omap_rproc_resume)
+};
+
static struct platform_driver omap_rproc_driver = {
.probe = omap_rproc_probe,
.remove = omap_rproc_remove,
.driver = {
.name = "omap-rproc",
+ .pm = &omap_rproc_pm_ops,
.of_match_table = omap_rproc_of_match,
},
};
diff --git a/drivers/remoteproc/omap_remoteproc.h b/drivers/remoteproc/omap_remoteproc.h
index 72f656c93caa..13f17d9135c0 100644
--- a/drivers/remoteproc/omap_remoteproc.h
+++ b/drivers/remoteproc/omap_remoteproc.h
@@ -1,7 +1,7 @@
/*
* Remote processor messaging
*
- * Copyright (C) 2011 Texas Instruments, Inc.
+ * Copyright (C) 2011-2020 Texas Instruments, Inc.
* Copyright (C) 2011 Google, Inc.
* All rights reserved.
*
@@ -57,6 +57,16 @@
* @RP_MBOX_ABORT_REQUEST: a "please crash" request, used for testing the
* recovery mechanism (to some extent).
*
+ * @RP_MBOX_SUSPEND_AUTO: auto suspend request for the remote processor
+ *
+ * @RP_MBOX_SUSPEND_SYSTEM: system suspend request for the remote processor
+ *
+ * @RP_MBOX_SUSPEND_ACK: successful response from remote processor for a
+ * suspend request
+ *
+ * @RP_MBOX_SUSPEND_CANCEL: a cancel suspend response from a remote processor
+ * on a suspend request
+ *
* Introduce new message definitions if any here.
*
* @RP_MBOX_END_MSG: Indicates end of known/defined messages from remote core
@@ -70,7 +80,11 @@ enum omap_rp_mbox_messages {
RP_MBOX_ECHO_REQUEST = 0xFFFFFF03,
RP_MBOX_ECHO_REPLY = 0xFFFFFF04,
RP_MBOX_ABORT_REQUEST = 0xFFFFFF05,
- RP_MBOX_END_MSG = 0xFFFFFF06,
+ RP_MBOX_SUSPEND_AUTO = 0xFFFFFF10,
+ RP_MBOX_SUSPEND_SYSTEM = 0xFFFFFF11,
+ RP_MBOX_SUSPEND_ACK = 0xFFFFFF12,
+ RP_MBOX_SUSPEND_CANCEL = 0xFFFFFF13,
+ RP_MBOX_END_MSG = 0xFFFFFF14,
};
#endif /* _OMAP_RPMSG_H */
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 12/14] remoteproc/omap: add support for runtime auto-suspend/resume
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
This patch enhances the PM support in the OMAP remoteproc driver to
support the runtime auto-suspend. A remoteproc may not be required to
be running all the time, and typically will need to be active only
during certain usecases. As such, to save power, it should be turned
off during potential long periods of inactivity between usecases.
This suspend and resume of the device is a relatively heavy process
in terms of latencies, so a remoteproc should be suspended only after
a certain period of prolonged inactivity. The OMAP remoteproc driver
leverages the runtime pm framework's auto_suspend feature to accomplish
this functionality. This feature is automatically enabled when a remote
processor has successfully booted. The 'autosuspend_delay_ms' for each
device dictates the inactivity period/time to wait for before
suspending the device.
The runtime auto-suspend design relies on marking the last busy time
on every communication (virtqueue kick) to and from the remote processor.
When there has been no activity for 'autosuspend_delay_ms' time, the
runtime PM framework invokes the driver's runtime pm suspend callback
to suspend the device. The remote processor will be woken up on the
initiation of the next communication message through the runtime pm
resume callback. The current auto-suspend design also allows a remote
processor to deny a auto-suspend attempt, if it wishes to, by sending a
NACK response to the initial suspend request message sent to the remote
processor as part of the suspend process. The auto-suspend request is
also only attempted if the remote processor is idled and in standby at
the time of inactivity timer expiry. This choice is made to avoid
unnecessary messaging, and the auto-suspend is simply rescheduled to
be attempted again after a further lapse of autosuspend_delay_ms.
The runtime pm callbacks functionality in this patch reuses most of the
core logic from the suspend/resume support code, and make use of an
additional auto_suspend flag to differentiate the logic in common code
from system suspend. The system suspend/resume sequences are also updated
to reflect the proper pm_runtime statuses, and also to really perform a
suspend/resume only if the remoteproc has not been auto-suspended at the
time of request. The remote processor is left in suspended state on a
system resume if it has been auto-suspended before, and will be woken up
only when a usecase needs to run.
The OMAP remoteproc driver currently uses a default value of 10 seconds
for all OMAP remoteprocs, and a different value can be chosen either by
choosing a positive value for the 'ti,autosuspend-delay' under DT or by
updating the 'autosuspend_delay_ms' field at runtime through the sysfs
interface.
Eg: To use 25 seconds for IPU2 on DRA7xx,
echo 25000 > /sys/bus/platform/devices/55020000.ipu/power/autosuspend_delay_ms
The runtime suspend feature can also be similarly enabled or disabled by
writing 'auto' or 'on' to the device's 'control' power field. The default
is enabled.
Eg: To disable auto-suspend for IPU2 on DRA7xx SoC,
echo on > /sys/bus/platform/devices/55020000.ipu/power/control
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to use ti-sysc instead of hwmod]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- add new pdata status variable for reset state, this avoids
asserting/deasserting the rproc reset twice in a row resulting in timeout
drivers/remoteproc/omap_remoteproc.c | 189 +++++++++++++++++++++++++--
1 file changed, 178 insertions(+), 11 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 452ef6070175..0700d2b332f7 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -22,6 +22,7 @@
#include <linux/of_device.h>
#include <linux/of_reserved_mem.h>
#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
@@ -37,6 +38,9 @@
#include "omap_remoteproc.h"
#include "remoteproc_internal.h"
+/* default auto-suspend delay (ms) */
+#define DEFAULT_AUTOSUSPEND_DELAY 10000
+
/**
* struct omap_rproc_boot_data - boot data structure for the DSP omap rprocs
* @syscon: regmap handle for the system control configuration module
@@ -83,11 +87,14 @@ struct omap_rproc_timer {
* @num_mems: number of internal memory regions
* @num_timers: number of rproc timer(s)
* @timers: timer(s) info used by rproc
+ * @autosuspend_delay: auto-suspend delay value to be used for runtime pm
+ * @need_resume: if true a resume is needed in the system resume callback
* @rproc: rproc handle
* @reset: reset handle
* @pm_comp: completion primitive to sync for suspend response
* @fck: functional clock for the remoteproc
* @suspend_acked: state machine flag to store the suspend request ack
+ * @in_reset: if remoteproc is in reset or not
*/
struct omap_rproc {
struct mbox_chan *mbox;
@@ -97,11 +104,14 @@ struct omap_rproc {
int num_mems;
int num_timers;
struct omap_rproc_timer *timers;
+ int autosuspend_delay;
+ bool need_resume;
struct rproc *rproc;
struct reset_control *reset;
struct completion pm_comp;
struct clk *fck;
bool suspend_acked;
+ bool in_reset;
};
/**
@@ -403,11 +413,23 @@ static void omap_rproc_kick(struct rproc *rproc, int vqid)
struct device *dev = rproc->dev.parent;
int ret;
+ /* wake up the rproc before kicking it */
+ ret = pm_runtime_get_sync(dev);
+ if (WARN_ON(ret < 0)) {
+ dev_err(dev, "pm_runtime_get_sync() failed during kick, ret = %d\n",
+ ret);
+ pm_runtime_put_noidle(dev);
+ return;
+ }
+
/* send the index of the triggered virtqueue in the mailbox payload */
ret = mbox_send_message(oproc->mbox, (void *)vqid);
if (ret < 0)
dev_err(dev, "failed to send mailbox message, status = %d\n",
ret);
+
+ pm_runtime_mark_last_busy(dev);
+ pm_runtime_put_autosuspend(dev);
}
/**
@@ -498,6 +520,12 @@ static int omap_rproc_start(struct rproc *rproc)
goto disable_timers;
}
+ oproc->in_reset = false;
+
+ pm_runtime_get_sync(dev);
+ pm_runtime_mark_last_busy(dev);
+ pm_runtime_put_autosuspend(dev);
+
return 0;
disable_timers:
@@ -510,6 +538,7 @@ static int omap_rproc_start(struct rproc *rproc)
/* power off the remote processor */
static int omap_rproc_stop(struct rproc *rproc)
{
+ struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
int ret;
@@ -517,12 +546,20 @@ static int omap_rproc_stop(struct rproc *rproc)
if (ret)
return ret;
+ oproc->in_reset = true;
+
ret = omap_rproc_disable_timers(rproc, true);
if (ret)
return ret;
mbox_free_channel(oproc->mbox);
+ /*
+ * update the runtime pm states and status now that the remoteproc
+ * has stopped
+ */
+ pm_runtime_put_noidle(dev);
+
return 0;
}
@@ -579,17 +616,19 @@ static bool _is_rproc_in_standby(struct omap_rproc *oproc)
/* 1 sec is long enough time to let the remoteproc side suspend the device */
#define DEF_SUSPEND_TIMEOUT 1000
-static int _omap_rproc_suspend(struct rproc *rproc)
+static int _omap_rproc_suspend(struct rproc *rproc, bool auto_suspend)
{
struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
unsigned long to = msecs_to_jiffies(DEF_SUSPEND_TIMEOUT);
unsigned long ta = jiffies + to;
+ u32 suspend_msg = auto_suspend ?
+ RP_MBOX_SUSPEND_AUTO : RP_MBOX_SUSPEND_SYSTEM;
int ret;
reinit_completion(&oproc->pm_comp);
oproc->suspend_acked = false;
- ret = mbox_send_message(oproc->mbox, (void *)RP_MBOX_SUSPEND_SYSTEM);
+ ret = mbox_send_message(oproc->mbox, (void *)suspend_msg);
if (ret < 0) {
dev_err(dev, "PM mbox_send_message failed: %d\n", ret);
return ret;
@@ -622,6 +661,8 @@ static int _omap_rproc_suspend(struct rproc *rproc)
reset_control_assert(oproc->reset);
+ oproc->in_reset = true;
+
ret = omap_rproc_disable_timers(rproc, false);
if (ret) {
dev_err(dev, "disabling timers during suspend failed %d\n",
@@ -629,42 +670,79 @@ static int _omap_rproc_suspend(struct rproc *rproc)
goto enable_device;
}
+ /*
+ * IOMMUs would have to be disabled specifically for runtime suspend.
+ * They are handled automatically through System PM callbacks for
+ * regular system suspend
+ */
+ if (auto_suspend) {
+ ret = omap_iommu_domain_deactivate(rproc->domain);
+ if (ret) {
+ dev_err(dev, "iommu domain deactivate failed %d\n",
+ ret);
+ goto enable_timers;
+ }
+ }
+
return 0;
+enable_timers:
+ /* ignore errors on re-enabling code */
+ omap_rproc_enable_timers(rproc, false);
enable_device:
reset_control_deassert(oproc->reset);
return ret;
}
-static int _omap_rproc_resume(struct rproc *rproc)
+static int _omap_rproc_resume(struct rproc *rproc, bool auto_suspend)
{
struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
int ret;
+ /*
+ * IOMMUs would have to be enabled specifically for runtime resume.
+ * They would have been already enabled automatically through System
+ * PM callbacks for regular system resume
+ */
+ if (auto_suspend) {
+ ret = omap_iommu_domain_activate(rproc->domain);
+ if (ret) {
+ dev_err(dev, "omap_iommu activate failed %d\n", ret);
+ goto out;
+ }
+ }
+
/* boot address could be lost after suspend, so restore it */
if (oproc->boot_data) {
ret = omap_rproc_write_dsp_boot_addr(rproc);
if (ret) {
dev_err(dev, "boot address restore failed %d\n", ret);
- goto out;
+ goto suspend_iommu;
}
}
ret = omap_rproc_enable_timers(rproc, false);
if (ret) {
dev_err(dev, "enabling timers during resume failed %d\n", ret);
- goto out;
+ goto suspend_iommu;
}
- ret = reset_control_deassert(oproc->reset);
- if (ret) {
- dev_err(dev, "reset deassert failed %d\n", ret);
- goto disable_timers;
+ if (oproc->in_reset) {
+ ret = reset_control_deassert(oproc->reset);
+ if (ret) {
+ dev_err(dev, "reset deassert failed %d\n", ret);
+ goto disable_timers;
+ }
+
+ oproc->in_reset = false;
}
return 0;
+suspend_iommu:
+ if (auto_suspend)
+ omap_iommu_domain_deactivate(rproc->domain);
disable_timers:
omap_rproc_disable_timers(rproc, false);
@@ -676,6 +754,7 @@ static int __maybe_unused omap_rproc_suspend(struct device *dev)
{
struct platform_device *pdev = to_platform_device(dev);
struct rproc *rproc = platform_get_drvdata(pdev);
+ struct omap_rproc *oproc = rproc->priv;
int ret = 0;
mutex_lock(&rproc->lock);
@@ -690,13 +769,19 @@ static int __maybe_unused omap_rproc_suspend(struct device *dev)
goto out;
}
- ret = _omap_rproc_suspend(rproc);
+ ret = _omap_rproc_suspend(rproc, false);
if (ret) {
dev_err(dev, "suspend failed %d\n", ret);
goto out;
}
+ /*
+ * remoteproc is running at the time of system suspend, so remember
+ * it so as to wake it up during system resume
+ */
+ oproc->need_resume = true;
rproc->state = RPROC_SUSPENDED;
+
out:
mutex_unlock(&rproc->lock);
return ret;
@@ -706,6 +791,7 @@ static int __maybe_unused omap_rproc_resume(struct device *dev)
{
struct platform_device *pdev = to_platform_device(dev);
struct rproc *rproc = platform_get_drvdata(pdev);
+ struct omap_rproc *oproc = rproc->priv;
int ret = 0;
mutex_lock(&rproc->lock);
@@ -717,17 +803,87 @@ static int __maybe_unused omap_rproc_resume(struct device *dev)
goto out;
}
- ret = _omap_rproc_resume(rproc);
+ /*
+ * remoteproc was auto-suspended at the time of system suspend,
+ * so no need to wake-up the processor (leave it in suspended
+ * state, will be woken up during a subsequent runtime_resume)
+ */
+ if (!oproc->need_resume)
+ goto out;
+
+ ret = _omap_rproc_resume(rproc, false);
if (ret) {
dev_err(dev, "resume failed %d\n", ret);
goto out;
}
+ oproc->need_resume = false;
rproc->state = RPROC_RUNNING;
+
+ pm_runtime_mark_last_busy(dev);
out:
mutex_unlock(&rproc->lock);
return ret;
}
+
+static int omap_rproc_runtime_suspend(struct device *dev)
+{
+ struct rproc *rproc = dev_get_drvdata(dev);
+ struct omap_rproc *oproc = rproc->priv;
+ int ret;
+
+ if (rproc->state == RPROC_CRASHED) {
+ dev_dbg(dev, "rproc cannot be runtime suspended when crashed!\n");
+ return -EBUSY;
+ }
+
+ if (WARN_ON(rproc->state != RPROC_RUNNING)) {
+ dev_err(dev, "rproc cannot be runtime suspended when not running!\n");
+ return -EBUSY;
+ }
+
+ /*
+ * do not even attempt suspend if the remote processor is not
+ * idled for runtime auto-suspend
+ */
+ if (!_is_rproc_in_standby(oproc)) {
+ ret = -EBUSY;
+ goto abort;
+ }
+
+ ret = _omap_rproc_suspend(rproc, true);
+ if (ret)
+ goto abort;
+
+ rproc->state = RPROC_SUSPENDED;
+ return 0;
+
+abort:
+ pm_runtime_mark_last_busy(dev);
+ return ret;
+}
+
+static int omap_rproc_runtime_resume(struct device *dev)
+{
+ struct rproc *rproc = dev_get_drvdata(dev);
+ int ret;
+
+ if (WARN_ON(rproc->state != RPROC_SUSPENDED &&
+ rproc->state != RPROC_OFFLINE)) {
+ dev_err(dev, "rproc cannot be runtime resumed if not suspended! state=%d\n", rproc->state);
+ return -EBUSY;
+ }
+
+ ret = _omap_rproc_resume(rproc, rproc->state == RPROC_SUSPENDED);
+ if (ret) {
+ dev_err(dev, "runtime resume failed %d\n", ret);
+ return ret;
+ }
+
+ rproc->state = RPROC_RUNNING;
+
+ return 0;
+}
#endif /* CONFIG_PM */
static const struct omap_rproc_mem_data ipu_mems[] = {
@@ -973,6 +1129,14 @@ static int omap_rproc_probe(struct platform_device *pdev)
}
init_completion(&oproc->pm_comp);
+ oproc->autosuspend_delay = DEFAULT_AUTOSUSPEND_DELAY;
+
+ of_property_read_u32(pdev->dev.of_node, "ti,autosuspend-delay-ms",
+ &oproc->autosuspend_delay);
+
+ pm_runtime_set_autosuspend_delay(&pdev->dev, oproc->autosuspend_delay);
+ pm_runtime_use_autosuspend(&pdev->dev);
+ pm_runtime_enable(&pdev->dev);
oproc->fck = devm_clk_get(&pdev->dev, 0);
if (IS_ERR(oproc->fck)) {
@@ -1008,12 +1172,15 @@ static int omap_rproc_remove(struct platform_device *pdev)
rproc_del(rproc);
rproc_free(rproc);
of_reserved_mem_device_release(&pdev->dev);
+ pm_runtime_disable(&pdev->dev);
return 0;
}
static const struct dev_pm_ops omap_rproc_pm_ops = {
SET_SYSTEM_SLEEP_PM_OPS(omap_rproc_suspend, omap_rproc_resume)
+ SET_RUNTIME_PM_OPS(omap_rproc_runtime_suspend,
+ omap_rproc_runtime_resume, NULL)
};
static struct platform_driver omap_rproc_driver = {
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 12/14] remoteproc/omap: add support for runtime auto-suspend/resume
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
This patch enhances the PM support in the OMAP remoteproc driver to
support the runtime auto-suspend. A remoteproc may not be required to
be running all the time, and typically will need to be active only
during certain usecases. As such, to save power, it should be turned
off during potential long periods of inactivity between usecases.
This suspend and resume of the device is a relatively heavy process
in terms of latencies, so a remoteproc should be suspended only after
a certain period of prolonged inactivity. The OMAP remoteproc driver
leverages the runtime pm framework's auto_suspend feature to accomplish
this functionality. This feature is automatically enabled when a remote
processor has successfully booted. The 'autosuspend_delay_ms' for each
device dictates the inactivity period/time to wait for before
suspending the device.
The runtime auto-suspend design relies on marking the last busy time
on every communication (virtqueue kick) to and from the remote processor.
When there has been no activity for 'autosuspend_delay_ms' time, the
runtime PM framework invokes the driver's runtime pm suspend callback
to suspend the device. The remote processor will be woken up on the
initiation of the next communication message through the runtime pm
resume callback. The current auto-suspend design also allows a remote
processor to deny a auto-suspend attempt, if it wishes to, by sending a
NACK response to the initial suspend request message sent to the remote
processor as part of the suspend process. The auto-suspend request is
also only attempted if the remote processor is idled and in standby at
the time of inactivity timer expiry. This choice is made to avoid
unnecessary messaging, and the auto-suspend is simply rescheduled to
be attempted again after a further lapse of autosuspend_delay_ms.
The runtime pm callbacks functionality in this patch reuses most of the
core logic from the suspend/resume support code, and make use of an
additional auto_suspend flag to differentiate the logic in common code
from system suspend. The system suspend/resume sequences are also updated
to reflect the proper pm_runtime statuses, and also to really perform a
suspend/resume only if the remoteproc has not been auto-suspended at the
time of request. The remote processor is left in suspended state on a
system resume if it has been auto-suspended before, and will be woken up
only when a usecase needs to run.
The OMAP remoteproc driver currently uses a default value of 10 seconds
for all OMAP remoteprocs, and a different value can be chosen either by
choosing a positive value for the 'ti,autosuspend-delay' under DT or by
updating the 'autosuspend_delay_ms' field at runtime through the sysfs
interface.
Eg: To use 25 seconds for IPU2 on DRA7xx,
echo 25000 > /sys/bus/platform/devices/55020000.ipu/power/autosuspend_delay_ms
The runtime suspend feature can also be similarly enabled or disabled by
writing 'auto' or 'on' to the device's 'control' power field. The default
is enabled.
Eg: To disable auto-suspend for IPU2 on DRA7xx SoC,
echo on > /sys/bus/platform/devices/55020000.ipu/power/control
Signed-off-by: Suman Anna <s-anna@ti.com>
[t-kristo@ti.com: converted to use ti-sysc instead of hwmod]
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- add new pdata status variable for reset state, this avoids
asserting/deasserting the rproc reset twice in a row resulting in timeout
drivers/remoteproc/omap_remoteproc.c | 189 +++++++++++++++++++++++++--
1 file changed, 178 insertions(+), 11 deletions(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 452ef6070175..0700d2b332f7 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -22,6 +22,7 @@
#include <linux/of_device.h>
#include <linux/of_reserved_mem.h>
#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
#include <linux/dma-mapping.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
@@ -37,6 +38,9 @@
#include "omap_remoteproc.h"
#include "remoteproc_internal.h"
+/* default auto-suspend delay (ms) */
+#define DEFAULT_AUTOSUSPEND_DELAY 10000
+
/**
* struct omap_rproc_boot_data - boot data structure for the DSP omap rprocs
* @syscon: regmap handle for the system control configuration module
@@ -83,11 +87,14 @@ struct omap_rproc_timer {
* @num_mems: number of internal memory regions
* @num_timers: number of rproc timer(s)
* @timers: timer(s) info used by rproc
+ * @autosuspend_delay: auto-suspend delay value to be used for runtime pm
+ * @need_resume: if true a resume is needed in the system resume callback
* @rproc: rproc handle
* @reset: reset handle
* @pm_comp: completion primitive to sync for suspend response
* @fck: functional clock for the remoteproc
* @suspend_acked: state machine flag to store the suspend request ack
+ * @in_reset: if remoteproc is in reset or not
*/
struct omap_rproc {
struct mbox_chan *mbox;
@@ -97,11 +104,14 @@ struct omap_rproc {
int num_mems;
int num_timers;
struct omap_rproc_timer *timers;
+ int autosuspend_delay;
+ bool need_resume;
struct rproc *rproc;
struct reset_control *reset;
struct completion pm_comp;
struct clk *fck;
bool suspend_acked;
+ bool in_reset;
};
/**
@@ -403,11 +413,23 @@ static void omap_rproc_kick(struct rproc *rproc, int vqid)
struct device *dev = rproc->dev.parent;
int ret;
+ /* wake up the rproc before kicking it */
+ ret = pm_runtime_get_sync(dev);
+ if (WARN_ON(ret < 0)) {
+ dev_err(dev, "pm_runtime_get_sync() failed during kick, ret = %d\n",
+ ret);
+ pm_runtime_put_noidle(dev);
+ return;
+ }
+
/* send the index of the triggered virtqueue in the mailbox payload */
ret = mbox_send_message(oproc->mbox, (void *)vqid);
if (ret < 0)
dev_err(dev, "failed to send mailbox message, status = %d\n",
ret);
+
+ pm_runtime_mark_last_busy(dev);
+ pm_runtime_put_autosuspend(dev);
}
/**
@@ -498,6 +520,12 @@ static int omap_rproc_start(struct rproc *rproc)
goto disable_timers;
}
+ oproc->in_reset = false;
+
+ pm_runtime_get_sync(dev);
+ pm_runtime_mark_last_busy(dev);
+ pm_runtime_put_autosuspend(dev);
+
return 0;
disable_timers:
@@ -510,6 +538,7 @@ static int omap_rproc_start(struct rproc *rproc)
/* power off the remote processor */
static int omap_rproc_stop(struct rproc *rproc)
{
+ struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
int ret;
@@ -517,12 +546,20 @@ static int omap_rproc_stop(struct rproc *rproc)
if (ret)
return ret;
+ oproc->in_reset = true;
+
ret = omap_rproc_disable_timers(rproc, true);
if (ret)
return ret;
mbox_free_channel(oproc->mbox);
+ /*
+ * update the runtime pm states and status now that the remoteproc
+ * has stopped
+ */
+ pm_runtime_put_noidle(dev);
+
return 0;
}
@@ -579,17 +616,19 @@ static bool _is_rproc_in_standby(struct omap_rproc *oproc)
/* 1 sec is long enough time to let the remoteproc side suspend the device */
#define DEF_SUSPEND_TIMEOUT 1000
-static int _omap_rproc_suspend(struct rproc *rproc)
+static int _omap_rproc_suspend(struct rproc *rproc, bool auto_suspend)
{
struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
unsigned long to = msecs_to_jiffies(DEF_SUSPEND_TIMEOUT);
unsigned long ta = jiffies + to;
+ u32 suspend_msg = auto_suspend ?
+ RP_MBOX_SUSPEND_AUTO : RP_MBOX_SUSPEND_SYSTEM;
int ret;
reinit_completion(&oproc->pm_comp);
oproc->suspend_acked = false;
- ret = mbox_send_message(oproc->mbox, (void *)RP_MBOX_SUSPEND_SYSTEM);
+ ret = mbox_send_message(oproc->mbox, (void *)suspend_msg);
if (ret < 0) {
dev_err(dev, "PM mbox_send_message failed: %d\n", ret);
return ret;
@@ -622,6 +661,8 @@ static int _omap_rproc_suspend(struct rproc *rproc)
reset_control_assert(oproc->reset);
+ oproc->in_reset = true;
+
ret = omap_rproc_disable_timers(rproc, false);
if (ret) {
dev_err(dev, "disabling timers during suspend failed %d\n",
@@ -629,42 +670,79 @@ static int _omap_rproc_suspend(struct rproc *rproc)
goto enable_device;
}
+ /*
+ * IOMMUs would have to be disabled specifically for runtime suspend.
+ * They are handled automatically through System PM callbacks for
+ * regular system suspend
+ */
+ if (auto_suspend) {
+ ret = omap_iommu_domain_deactivate(rproc->domain);
+ if (ret) {
+ dev_err(dev, "iommu domain deactivate failed %d\n",
+ ret);
+ goto enable_timers;
+ }
+ }
+
return 0;
+enable_timers:
+ /* ignore errors on re-enabling code */
+ omap_rproc_enable_timers(rproc, false);
enable_device:
reset_control_deassert(oproc->reset);
return ret;
}
-static int _omap_rproc_resume(struct rproc *rproc)
+static int _omap_rproc_resume(struct rproc *rproc, bool auto_suspend)
{
struct device *dev = rproc->dev.parent;
struct omap_rproc *oproc = rproc->priv;
int ret;
+ /*
+ * IOMMUs would have to be enabled specifically for runtime resume.
+ * They would have been already enabled automatically through System
+ * PM callbacks for regular system resume
+ */
+ if (auto_suspend) {
+ ret = omap_iommu_domain_activate(rproc->domain);
+ if (ret) {
+ dev_err(dev, "omap_iommu activate failed %d\n", ret);
+ goto out;
+ }
+ }
+
/* boot address could be lost after suspend, so restore it */
if (oproc->boot_data) {
ret = omap_rproc_write_dsp_boot_addr(rproc);
if (ret) {
dev_err(dev, "boot address restore failed %d\n", ret);
- goto out;
+ goto suspend_iommu;
}
}
ret = omap_rproc_enable_timers(rproc, false);
if (ret) {
dev_err(dev, "enabling timers during resume failed %d\n", ret);
- goto out;
+ goto suspend_iommu;
}
- ret = reset_control_deassert(oproc->reset);
- if (ret) {
- dev_err(dev, "reset deassert failed %d\n", ret);
- goto disable_timers;
+ if (oproc->in_reset) {
+ ret = reset_control_deassert(oproc->reset);
+ if (ret) {
+ dev_err(dev, "reset deassert failed %d\n", ret);
+ goto disable_timers;
+ }
+
+ oproc->in_reset = false;
}
return 0;
+suspend_iommu:
+ if (auto_suspend)
+ omap_iommu_domain_deactivate(rproc->domain);
disable_timers:
omap_rproc_disable_timers(rproc, false);
@@ -676,6 +754,7 @@ static int __maybe_unused omap_rproc_suspend(struct device *dev)
{
struct platform_device *pdev = to_platform_device(dev);
struct rproc *rproc = platform_get_drvdata(pdev);
+ struct omap_rproc *oproc = rproc->priv;
int ret = 0;
mutex_lock(&rproc->lock);
@@ -690,13 +769,19 @@ static int __maybe_unused omap_rproc_suspend(struct device *dev)
goto out;
}
- ret = _omap_rproc_suspend(rproc);
+ ret = _omap_rproc_suspend(rproc, false);
if (ret) {
dev_err(dev, "suspend failed %d\n", ret);
goto out;
}
+ /*
+ * remoteproc is running at the time of system suspend, so remember
+ * it so as to wake it up during system resume
+ */
+ oproc->need_resume = true;
rproc->state = RPROC_SUSPENDED;
+
out:
mutex_unlock(&rproc->lock);
return ret;
@@ -706,6 +791,7 @@ static int __maybe_unused omap_rproc_resume(struct device *dev)
{
struct platform_device *pdev = to_platform_device(dev);
struct rproc *rproc = platform_get_drvdata(pdev);
+ struct omap_rproc *oproc = rproc->priv;
int ret = 0;
mutex_lock(&rproc->lock);
@@ -717,17 +803,87 @@ static int __maybe_unused omap_rproc_resume(struct device *dev)
goto out;
}
- ret = _omap_rproc_resume(rproc);
+ /*
+ * remoteproc was auto-suspended at the time of system suspend,
+ * so no need to wake-up the processor (leave it in suspended
+ * state, will be woken up during a subsequent runtime_resume)
+ */
+ if (!oproc->need_resume)
+ goto out;
+
+ ret = _omap_rproc_resume(rproc, false);
if (ret) {
dev_err(dev, "resume failed %d\n", ret);
goto out;
}
+ oproc->need_resume = false;
rproc->state = RPROC_RUNNING;
+
+ pm_runtime_mark_last_busy(dev);
out:
mutex_unlock(&rproc->lock);
return ret;
}
+
+static int omap_rproc_runtime_suspend(struct device *dev)
+{
+ struct rproc *rproc = dev_get_drvdata(dev);
+ struct omap_rproc *oproc = rproc->priv;
+ int ret;
+
+ if (rproc->state == RPROC_CRASHED) {
+ dev_dbg(dev, "rproc cannot be runtime suspended when crashed!\n");
+ return -EBUSY;
+ }
+
+ if (WARN_ON(rproc->state != RPROC_RUNNING)) {
+ dev_err(dev, "rproc cannot be runtime suspended when not running!\n");
+ return -EBUSY;
+ }
+
+ /*
+ * do not even attempt suspend if the remote processor is not
+ * idled for runtime auto-suspend
+ */
+ if (!_is_rproc_in_standby(oproc)) {
+ ret = -EBUSY;
+ goto abort;
+ }
+
+ ret = _omap_rproc_suspend(rproc, true);
+ if (ret)
+ goto abort;
+
+ rproc->state = RPROC_SUSPENDED;
+ return 0;
+
+abort:
+ pm_runtime_mark_last_busy(dev);
+ return ret;
+}
+
+static int omap_rproc_runtime_resume(struct device *dev)
+{
+ struct rproc *rproc = dev_get_drvdata(dev);
+ int ret;
+
+ if (WARN_ON(rproc->state != RPROC_SUSPENDED &&
+ rproc->state != RPROC_OFFLINE)) {
+ dev_err(dev, "rproc cannot be runtime resumed if not suspended! state=%d\n", rproc->state);
+ return -EBUSY;
+ }
+
+ ret = _omap_rproc_resume(rproc, rproc->state == RPROC_SUSPENDED);
+ if (ret) {
+ dev_err(dev, "runtime resume failed %d\n", ret);
+ return ret;
+ }
+
+ rproc->state = RPROC_RUNNING;
+
+ return 0;
+}
#endif /* CONFIG_PM */
static const struct omap_rproc_mem_data ipu_mems[] = {
@@ -973,6 +1129,14 @@ static int omap_rproc_probe(struct platform_device *pdev)
}
init_completion(&oproc->pm_comp);
+ oproc->autosuspend_delay = DEFAULT_AUTOSUSPEND_DELAY;
+
+ of_property_read_u32(pdev->dev.of_node, "ti,autosuspend-delay-ms",
+ &oproc->autosuspend_delay);
+
+ pm_runtime_set_autosuspend_delay(&pdev->dev, oproc->autosuspend_delay);
+ pm_runtime_use_autosuspend(&pdev->dev);
+ pm_runtime_enable(&pdev->dev);
oproc->fck = devm_clk_get(&pdev->dev, 0);
if (IS_ERR(oproc->fck)) {
@@ -1008,12 +1172,15 @@ static int omap_rproc_remove(struct platform_device *pdev)
rproc_del(rproc);
rproc_free(rproc);
of_reserved_mem_device_release(&pdev->dev);
+ pm_runtime_disable(&pdev->dev);
return 0;
}
static const struct dev_pm_ops omap_rproc_pm_ops = {
SET_SYSTEM_SLEEP_PM_OPS(omap_rproc_suspend, omap_rproc_resume)
+ SET_RUNTIME_PM_OPS(omap_rproc_runtime_suspend,
+ omap_rproc_runtime_resume, NULL)
};
static struct platform_driver omap_rproc_driver = {
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 13/14] remoteproc/omap: report device exceptions and trigger recovery
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The OMAP remote processors send a special mailbox message
(RP_MBOX_CRASH) when they crash and detect an internal device
exception.
Add support to the mailbox handling function upon detection of
this special message to report this crash to the remoteproc core.
The remoteproc core can trigger a recovery using the prevailing
recovery mechanism, already in use for MMU Fault recovery.
Co-developed-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 0700d2b332f7..ebc9ebba506c 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -381,8 +381,12 @@ static void omap_rproc_mbox_callback(struct mbox_client *client, void *data)
switch (msg) {
case RP_MBOX_CRASH:
- /* just log this for now. later, we'll also do recovery */
+ /*
+ * remoteproc detected an exception, notify the rproc core.
+ * The remoteproc core will handle the recovery.
+ */
dev_err(dev, "omap rproc %s crashed\n", name);
+ rproc_report_crash(oproc->rproc, RPROC_FATAL_ERROR);
break;
case RP_MBOX_ECHO_REPLY:
dev_info(dev, "received echo reply from %s\n", name);
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 13/14] remoteproc/omap: report device exceptions and trigger recovery
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
The OMAP remote processors send a special mailbox message
(RP_MBOX_CRASH) when they crash and detect an internal device
exception.
Add support to the mailbox handling function upon detection of
this special message to report this crash to the remoteproc core.
The remoteproc core can trigger a recovery using the prevailing
recovery mechanism, already in use for MMU Fault recovery.
Co-developed-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Subramaniam Chanderashekarapuram <subramaniam.ca@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
v5: no changes
drivers/remoteproc/omap_remoteproc.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index 0700d2b332f7..ebc9ebba506c 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -381,8 +381,12 @@ static void omap_rproc_mbox_callback(struct mbox_client *client, void *data)
switch (msg) {
case RP_MBOX_CRASH:
- /* just log this for now. later, we'll also do recovery */
+ /*
+ * remoteproc detected an exception, notify the rproc core.
+ * The remoteproc core will handle the recovery.
+ */
dev_err(dev, "omap rproc %s crashed\n", name);
+ rproc_report_crash(oproc->rproc, RPROC_FATAL_ERROR);
break;
case RP_MBOX_ECHO_REPLY:
dev_info(dev, "received echo reply from %s\n", name);
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread
* [PATCHv5 14/14] remoteproc/omap: add watchdog functionality for remote processors
2020-01-16 13:53 ` Tero Kristo
@ 2020-01-16 13:53 ` Tero Kristo
-1 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
Remote processors can be stuck in a loop, and may not be recoverable
if they do not have a built-in watchdog. The watchdog implementation
for OMAP remote processors uses external gptimers that can be used
to interrupt both the Linux host as well as the remote processor.
Each remote processor is responsible for refreshing the timer during
normal behavior - during OS task scheduling or entering the idle loop
properly. During a watchdog condition (executing a tight loop causing
no scheduling), the host processor gets interrupts and schedules a
recovery for the corresponding remote processor. The remote processor
may also get interrupted to be able to print a back trace.
A menuconfig option has also been added to enable/disable the Watchdog
functionality, with the default as disabled.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- some error hardening changes
- modified some checks to add also number of watchdog timers
drivers/remoteproc/Kconfig | 12 +++
drivers/remoteproc/omap_remoteproc.c | 155 ++++++++++++++++++++++++---
2 files changed, 155 insertions(+), 12 deletions(-)
diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
index d6450d7fcf92..b2eaa18ad503 100644
--- a/drivers/remoteproc/Kconfig
+++ b/drivers/remoteproc/Kconfig
@@ -42,6 +42,18 @@ config OMAP_REMOTEPROC
It's safe to say N here if you're not interested in multimedia
offloading or just want a bare minimum kernel.
+config OMAP_REMOTEPROC_WATCHDOG
+ bool "OMAP remoteproc watchdog timer"
+ depends on OMAP_REMOTEPROC
+ default n
+ help
+ Say Y here to enable watchdog timer for remote processors.
+
+ This option controls the watchdog functionality for the remote
+ processors in OMAP. Dedicated OMAP DMTimers are used by the remote
+ processors and triggers the timer interrupt upon a watchdog
+ detection.
+
config WKUP_M3_RPROC
tristate "AMx3xx Wakeup M3 remoteproc support"
depends on SOC_AM33XX || SOC_AM43XX
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index ebc9ebba506c..e8f63c9add63 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -24,6 +24,7 @@
#include <linux/platform_device.h>
#include <linux/pm_runtime.h>
#include <linux/dma-mapping.h>
+#include <linux/interrupt.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
#include <linux/omap-iommu.h>
@@ -72,10 +73,12 @@ struct omap_rproc_mem {
* struct omap_rproc_timer - data structure for a timer used by a omap rproc
* @odt: timer pointer
* @timer_ops: OMAP dmtimer ops for @odt timer
+ * @irq: timer irq
*/
struct omap_rproc_timer {
struct omap_dm_timer *odt;
const struct omap_dm_timer_ops *timer_ops;
+ int irq;
};
/**
@@ -86,6 +89,7 @@ struct omap_rproc_timer {
* @mem: internal memory regions data
* @num_mems: number of internal memory regions
* @num_timers: number of rproc timer(s)
+ * @num_wd_timers: number of rproc watchdog timers
* @timers: timer(s) info used by rproc
* @autosuspend_delay: auto-suspend delay value to be used for runtime pm
* @need_resume: if true a resume is needed in the system resume callback
@@ -103,6 +107,7 @@ struct omap_rproc {
struct omap_rproc_mem *mem;
int num_mems;
int num_timers;
+ int num_wd_timers;
struct omap_rproc_timer *timers;
int autosuspend_delay;
bool need_resume;
@@ -221,6 +226,81 @@ static inline int omap_rproc_release_timer(struct omap_rproc_timer *timer)
return timer->timer_ops->free(timer->odt);
}
+/**
+ * omap_rproc_get_timer_irq - get the irq for a timer
+ * @timer - handle to a OMAP rproc timer
+ *
+ * This function is used to get the irq associated with a watchdog timer. The
+ * function is called by the OMAP remoteproc driver to register a interrupt
+ * handler to handle watchdog events on the remote processor.
+ *
+ * Returns the irq id on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_get_timer_irq(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->get_irq(timer->odt);
+}
+
+/**
+ * omap_rproc_ack_timer_irq - acknowledge a timer irq
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This function is used to clear the irq associated with a watchdog timer. The
+ * The function is called by the OMAP remoteproc upon a watchdog event on the
+ * remote processor to clear the interrupt status of the watchdog timer.
+ *
+ * Returns the irq id on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline void omap_rproc_ack_timer_irq(struct omap_rproc_timer *timer)
+{
+ timer->timer_ops->write_status(timer->odt, OMAP_TIMER_INT_OVERFLOW);
+}
+
+/**
+ * omap_rproc_watchdog_isr - Watchdog ISR handler for remoteproc device
+ * @irq: IRQ number associated with a watchdog timer
+ * @data: IRQ handler data
+ *
+ * This ISR routine executes the required necessary low-level code to
+ * acknowledge a watchdog timer interrupt. There can be multiple watchdog
+ * timers associated with a rproc (like IPUs which have 2 watchdog timers,
+ * one per Cortex M3/M4 core), so a lookup has to be performed to identify
+ * the timer to acknowledge its interrupt.
+ *
+ * The function also invokes rproc_report_crash to report the watchdog event
+ * to the remoteproc driver core, to trigger a recovery.
+ *
+ * Return: IRQ_HANDLED or IRQ_NONE
+ */
+static irqreturn_t omap_rproc_watchdog_isr(int irq, void *data)
+{
+ struct rproc *rproc = data;
+ struct omap_rproc *oproc = rproc->priv;
+ struct device *dev = rproc->dev.parent;
+ struct omap_rproc_timer *timers = oproc->timers;
+ struct omap_rproc_timer *wd_timer = NULL;
+ int num_timers = oproc->num_timers + oproc->num_wd_timers;
+ int i;
+
+ for (i = oproc->num_timers; i < num_timers; i++) {
+ if (timers[i].irq > 0 && irq == timers[i].irq) {
+ wd_timer = &timers[i];
+ break;
+ }
+ }
+
+ if (!wd_timer) {
+ dev_err(dev, "invalid timer\n");
+ return IRQ_NONE;
+ }
+
+ omap_rproc_ack_timer_irq(wd_timer);
+
+ rproc_report_crash(rproc, RPROC_WATCHDOG);
+
+ return IRQ_HANDLED;
+}
+
/**
* omap_rproc_enable_timers - enable the timers for a remoteproc
* @rproc: handle of a remote processor
@@ -242,19 +322,26 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
struct omap_rproc_timer *timers = oproc->timers;
struct device *dev = rproc->dev.parent;
struct device_node *np = NULL;
+ int num_timers = oproc->num_timers + oproc->num_wd_timers;
- if (!oproc->num_timers)
+ if (!num_timers)
return 0;
if (!configure)
goto start_timers;
- for (i = 0; i < oproc->num_timers; i++) {
- np = of_parse_phandle(dev->of_node, "ti,timers", i);
+ for (i = 0; i < num_timers; i++) {
+ if (i < oproc->num_timers)
+ np = of_parse_phandle(dev->of_node, "ti,timers", i);
+ else
+ np = of_parse_phandle(dev->of_node,
+ "ti,watchdog-timers",
+ (i - oproc->num_timers));
if (!np) {
ret = -ENXIO;
dev_err(dev, "device node lookup for timer at index %d failed: %d\n",
- i, ret);
+ i < oproc->num_timers ? i :
+ i - oproc->num_timers, ret);
goto free_timers;
}
@@ -277,12 +364,14 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
if (!timer_ops || !timer_ops->request_by_node ||
!timer_ops->set_source || !timer_ops->set_load ||
!timer_ops->free || !timer_ops->start ||
- !timer_ops->stop) {
+ !timer_ops->stop || !timer_ops->get_irq ||
+ !timer_ops->write_status) {
ret = -EINVAL;
dev_err(dev, "device does not have required timer ops\n");
goto put_node;
}
+ timers[i].irq = -1;
timers[i].timer_ops = timer_ops;
ret = omap_rproc_request_timer(dev, np, &timers[i]);
if (ret) {
@@ -291,10 +380,33 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
goto put_node;
}
of_node_put(np);
+
+ if (i >= oproc->num_timers) {
+ timers[i].irq = omap_rproc_get_timer_irq(&timers[i]);
+ if (timers[i].irq < 0) {
+ dev_err(dev, "get_irq for timer %p failed: %d\n",
+ np, timers[i].irq);
+ ret = -EBUSY;
+ goto free_timers;
+ }
+
+ ret = request_irq(timers[i].irq,
+ omap_rproc_watchdog_isr, IRQF_SHARED,
+ "rproc-wdt", rproc);
+ if (ret) {
+ dev_err(dev, "error requesting irq for timer %p\n",
+ np);
+ omap_rproc_release_timer(&timers[i]);
+ timers[i].odt = NULL;
+ timers[i].timer_ops = NULL;
+ timers[i].irq = -1;
+ goto free_timers;
+ }
+ }
}
start_timers:
- for (i = 0; i < oproc->num_timers; i++) {
+ for (i = 0; i < num_timers; i++) {
ret = omap_rproc_start_timer(&timers[i]);
if (ret) {
dev_err(dev, "start timer %p failed failed: %d\n", np,
@@ -316,9 +428,12 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
of_node_put(np);
free_timers:
while (i--) {
+ if (i >= oproc->num_timers)
+ free_irq(timers[i].irq, rproc);
omap_rproc_release_timer(&timers[i]);
timers[i].odt = NULL;
timers[i].timer_ops = NULL;
+ timers[i].irq = -1;
}
return ret;
@@ -339,16 +454,20 @@ static int omap_rproc_disable_timers(struct rproc *rproc, bool configure)
int i;
struct omap_rproc *oproc = rproc->priv;
struct omap_rproc_timer *timers = oproc->timers;
+ int num_timers = oproc->num_timers + oproc->num_wd_timers;
- if (!oproc->num_timers)
+ if (!num_timers)
return 0;
- for (i = 0; i < oproc->num_timers; i++) {
+ for (i = 0; i < num_timers; i++) {
omap_rproc_stop_timer(&timers[i]);
if (configure) {
+ if (i >= oproc->num_timers)
+ free_irq(timers[i].irq, rproc);
omap_rproc_release_timer(&timers[i]);
timers[i].odt = NULL;
timers[i].timer_ops = NULL;
+ timers[i].irq = -1;
}
}
@@ -1067,6 +1186,7 @@ static int omap_rproc_probe(struct platform_device *pdev)
struct omap_rproc *oproc;
struct rproc *rproc;
const char *firmware;
+ int num_timers;
int ret;
struct reset_control *reset;
@@ -1119,8 +1239,19 @@ static int omap_rproc_probe(struct platform_device *pdev)
oproc->num_timers = 0;
}
- if (oproc->num_timers) {
- oproc->timers = devm_kcalloc(&pdev->dev, oproc->num_timers,
+#ifdef CONFIG_OMAP_REMOTEPROC_WATCHDOG
+ oproc->num_wd_timers =
+ of_count_phandle_with_args(np, "ti,watchdog-timers", NULL);
+ if (oproc->num_wd_timers <= 0) {
+ dev_dbg(&pdev->dev, "device does not have watchdog timers, status = %d\n",
+ oproc->num_wd_timers);
+ oproc->num_wd_timers = 0;
+ }
+#endif
+
+ if (oproc->num_timers || oproc->num_wd_timers) {
+ num_timers = oproc->num_timers + oproc->num_wd_timers;
+ oproc->timers = devm_kcalloc(&pdev->dev, num_timers,
sizeof(*oproc->timers),
GFP_KERNEL);
if (!oproc->timers) {
@@ -1128,8 +1259,8 @@ static int omap_rproc_probe(struct platform_device *pdev)
goto free_rproc;
}
- dev_dbg(&pdev->dev, "device has %d tick timers\n",
- oproc->num_timers);
+ dev_dbg(&pdev->dev, "device has %d tick timers and %d watchdog timers\n",
+ oproc->num_timers, oproc->num_wd_timers);
}
init_completion(&oproc->pm_comp);
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread* [PATCHv5 14/14] remoteproc/omap: add watchdog functionality for remote processors
@ 2020-01-16 13:53 ` Tero Kristo
0 siblings, 0 replies; 62+ messages in thread
From: Tero Kristo @ 2020-01-16 13:53 UTC (permalink / raw)
To: bjorn.andersson, ohad, linux-remoteproc
Cc: linux-kernel, mathieu.poirier, s-anna, linux-omap, Tero Kristo
From: Suman Anna <s-anna@ti.com>
Remote processors can be stuck in a loop, and may not be recoverable
if they do not have a built-in watchdog. The watchdog implementation
for OMAP remote processors uses external gptimers that can be used
to interrupt both the Linux host as well as the remote processor.
Each remote processor is responsible for refreshing the timer during
normal behavior - during OS task scheduling or entering the idle loop
properly. During a watchdog condition (executing a tight loop causing
no scheduling), the host processor gets interrupts and schedules a
recovery for the corresponding remote processor. The remote processor
may also get interrupted to be able to print a back trace.
A menuconfig option has also been added to enable/disable the Watchdog
functionality, with the default as disabled.
Signed-off-by: Suman Anna <s-anna@ti.com>
Signed-off-by: Tero Kristo <t-kristo@ti.com>
---
v5:
- some error hardening changes
- modified some checks to add also number of watchdog timers
drivers/remoteproc/Kconfig | 12 +++
drivers/remoteproc/omap_remoteproc.c | 155 ++++++++++++++++++++++++---
2 files changed, 155 insertions(+), 12 deletions(-)
diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
index d6450d7fcf92..b2eaa18ad503 100644
--- a/drivers/remoteproc/Kconfig
+++ b/drivers/remoteproc/Kconfig
@@ -42,6 +42,18 @@ config OMAP_REMOTEPROC
It's safe to say N here if you're not interested in multimedia
offloading or just want a bare minimum kernel.
+config OMAP_REMOTEPROC_WATCHDOG
+ bool "OMAP remoteproc watchdog timer"
+ depends on OMAP_REMOTEPROC
+ default n
+ help
+ Say Y here to enable watchdog timer for remote processors.
+
+ This option controls the watchdog functionality for the remote
+ processors in OMAP. Dedicated OMAP DMTimers are used by the remote
+ processors and triggers the timer interrupt upon a watchdog
+ detection.
+
config WKUP_M3_RPROC
tristate "AMx3xx Wakeup M3 remoteproc support"
depends on SOC_AM33XX || SOC_AM43XX
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c
index ebc9ebba506c..e8f63c9add63 100644
--- a/drivers/remoteproc/omap_remoteproc.c
+++ b/drivers/remoteproc/omap_remoteproc.c
@@ -24,6 +24,7 @@
#include <linux/platform_device.h>
#include <linux/pm_runtime.h>
#include <linux/dma-mapping.h>
+#include <linux/interrupt.h>
#include <linux/remoteproc.h>
#include <linux/mailbox_client.h>
#include <linux/omap-iommu.h>
@@ -72,10 +73,12 @@ struct omap_rproc_mem {
* struct omap_rproc_timer - data structure for a timer used by a omap rproc
* @odt: timer pointer
* @timer_ops: OMAP dmtimer ops for @odt timer
+ * @irq: timer irq
*/
struct omap_rproc_timer {
struct omap_dm_timer *odt;
const struct omap_dm_timer_ops *timer_ops;
+ int irq;
};
/**
@@ -86,6 +89,7 @@ struct omap_rproc_timer {
* @mem: internal memory regions data
* @num_mems: number of internal memory regions
* @num_timers: number of rproc timer(s)
+ * @num_wd_timers: number of rproc watchdog timers
* @timers: timer(s) info used by rproc
* @autosuspend_delay: auto-suspend delay value to be used for runtime pm
* @need_resume: if true a resume is needed in the system resume callback
@@ -103,6 +107,7 @@ struct omap_rproc {
struct omap_rproc_mem *mem;
int num_mems;
int num_timers;
+ int num_wd_timers;
struct omap_rproc_timer *timers;
int autosuspend_delay;
bool need_resume;
@@ -221,6 +226,81 @@ static inline int omap_rproc_release_timer(struct omap_rproc_timer *timer)
return timer->timer_ops->free(timer->odt);
}
+/**
+ * omap_rproc_get_timer_irq - get the irq for a timer
+ * @timer - handle to a OMAP rproc timer
+ *
+ * This function is used to get the irq associated with a watchdog timer. The
+ * function is called by the OMAP remoteproc driver to register a interrupt
+ * handler to handle watchdog events on the remote processor.
+ *
+ * Returns the irq id on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline int omap_rproc_get_timer_irq(struct omap_rproc_timer *timer)
+{
+ return timer->timer_ops->get_irq(timer->odt);
+}
+
+/**
+ * omap_rproc_ack_timer_irq - acknowledge a timer irq
+ * @timer: handle to a OMAP rproc timer
+ *
+ * This function is used to clear the irq associated with a watchdog timer. The
+ * The function is called by the OMAP remoteproc upon a watchdog event on the
+ * remote processor to clear the interrupt status of the watchdog timer.
+ *
+ * Returns the irq id on success, otherwise a failure as returned by DMTimer ops
+ */
+static inline void omap_rproc_ack_timer_irq(struct omap_rproc_timer *timer)
+{
+ timer->timer_ops->write_status(timer->odt, OMAP_TIMER_INT_OVERFLOW);
+}
+
+/**
+ * omap_rproc_watchdog_isr - Watchdog ISR handler for remoteproc device
+ * @irq: IRQ number associated with a watchdog timer
+ * @data: IRQ handler data
+ *
+ * This ISR routine executes the required necessary low-level code to
+ * acknowledge a watchdog timer interrupt. There can be multiple watchdog
+ * timers associated with a rproc (like IPUs which have 2 watchdog timers,
+ * one per Cortex M3/M4 core), so a lookup has to be performed to identify
+ * the timer to acknowledge its interrupt.
+ *
+ * The function also invokes rproc_report_crash to report the watchdog event
+ * to the remoteproc driver core, to trigger a recovery.
+ *
+ * Return: IRQ_HANDLED or IRQ_NONE
+ */
+static irqreturn_t omap_rproc_watchdog_isr(int irq, void *data)
+{
+ struct rproc *rproc = data;
+ struct omap_rproc *oproc = rproc->priv;
+ struct device *dev = rproc->dev.parent;
+ struct omap_rproc_timer *timers = oproc->timers;
+ struct omap_rproc_timer *wd_timer = NULL;
+ int num_timers = oproc->num_timers + oproc->num_wd_timers;
+ int i;
+
+ for (i = oproc->num_timers; i < num_timers; i++) {
+ if (timers[i].irq > 0 && irq == timers[i].irq) {
+ wd_timer = &timers[i];
+ break;
+ }
+ }
+
+ if (!wd_timer) {
+ dev_err(dev, "invalid timer\n");
+ return IRQ_NONE;
+ }
+
+ omap_rproc_ack_timer_irq(wd_timer);
+
+ rproc_report_crash(rproc, RPROC_WATCHDOG);
+
+ return IRQ_HANDLED;
+}
+
/**
* omap_rproc_enable_timers - enable the timers for a remoteproc
* @rproc: handle of a remote processor
@@ -242,19 +322,26 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
struct omap_rproc_timer *timers = oproc->timers;
struct device *dev = rproc->dev.parent;
struct device_node *np = NULL;
+ int num_timers = oproc->num_timers + oproc->num_wd_timers;
- if (!oproc->num_timers)
+ if (!num_timers)
return 0;
if (!configure)
goto start_timers;
- for (i = 0; i < oproc->num_timers; i++) {
- np = of_parse_phandle(dev->of_node, "ti,timers", i);
+ for (i = 0; i < num_timers; i++) {
+ if (i < oproc->num_timers)
+ np = of_parse_phandle(dev->of_node, "ti,timers", i);
+ else
+ np = of_parse_phandle(dev->of_node,
+ "ti,watchdog-timers",
+ (i - oproc->num_timers));
if (!np) {
ret = -ENXIO;
dev_err(dev, "device node lookup for timer at index %d failed: %d\n",
- i, ret);
+ i < oproc->num_timers ? i :
+ i - oproc->num_timers, ret);
goto free_timers;
}
@@ -277,12 +364,14 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
if (!timer_ops || !timer_ops->request_by_node ||
!timer_ops->set_source || !timer_ops->set_load ||
!timer_ops->free || !timer_ops->start ||
- !timer_ops->stop) {
+ !timer_ops->stop || !timer_ops->get_irq ||
+ !timer_ops->write_status) {
ret = -EINVAL;
dev_err(dev, "device does not have required timer ops\n");
goto put_node;
}
+ timers[i].irq = -1;
timers[i].timer_ops = timer_ops;
ret = omap_rproc_request_timer(dev, np, &timers[i]);
if (ret) {
@@ -291,10 +380,33 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
goto put_node;
}
of_node_put(np);
+
+ if (i >= oproc->num_timers) {
+ timers[i].irq = omap_rproc_get_timer_irq(&timers[i]);
+ if (timers[i].irq < 0) {
+ dev_err(dev, "get_irq for timer %p failed: %d\n",
+ np, timers[i].irq);
+ ret = -EBUSY;
+ goto free_timers;
+ }
+
+ ret = request_irq(timers[i].irq,
+ omap_rproc_watchdog_isr, IRQF_SHARED,
+ "rproc-wdt", rproc);
+ if (ret) {
+ dev_err(dev, "error requesting irq for timer %p\n",
+ np);
+ omap_rproc_release_timer(&timers[i]);
+ timers[i].odt = NULL;
+ timers[i].timer_ops = NULL;
+ timers[i].irq = -1;
+ goto free_timers;
+ }
+ }
}
start_timers:
- for (i = 0; i < oproc->num_timers; i++) {
+ for (i = 0; i < num_timers; i++) {
ret = omap_rproc_start_timer(&timers[i]);
if (ret) {
dev_err(dev, "start timer %p failed failed: %d\n", np,
@@ -316,9 +428,12 @@ static int omap_rproc_enable_timers(struct rproc *rproc, bool configure)
of_node_put(np);
free_timers:
while (i--) {
+ if (i >= oproc->num_timers)
+ free_irq(timers[i].irq, rproc);
omap_rproc_release_timer(&timers[i]);
timers[i].odt = NULL;
timers[i].timer_ops = NULL;
+ timers[i].irq = -1;
}
return ret;
@@ -339,16 +454,20 @@ static int omap_rproc_disable_timers(struct rproc *rproc, bool configure)
int i;
struct omap_rproc *oproc = rproc->priv;
struct omap_rproc_timer *timers = oproc->timers;
+ int num_timers = oproc->num_timers + oproc->num_wd_timers;
- if (!oproc->num_timers)
+ if (!num_timers)
return 0;
- for (i = 0; i < oproc->num_timers; i++) {
+ for (i = 0; i < num_timers; i++) {
omap_rproc_stop_timer(&timers[i]);
if (configure) {
+ if (i >= oproc->num_timers)
+ free_irq(timers[i].irq, rproc);
omap_rproc_release_timer(&timers[i]);
timers[i].odt = NULL;
timers[i].timer_ops = NULL;
+ timers[i].irq = -1;
}
}
@@ -1067,6 +1186,7 @@ static int omap_rproc_probe(struct platform_device *pdev)
struct omap_rproc *oproc;
struct rproc *rproc;
const char *firmware;
+ int num_timers;
int ret;
struct reset_control *reset;
@@ -1119,8 +1239,19 @@ static int omap_rproc_probe(struct platform_device *pdev)
oproc->num_timers = 0;
}
- if (oproc->num_timers) {
- oproc->timers = devm_kcalloc(&pdev->dev, oproc->num_timers,
+#ifdef CONFIG_OMAP_REMOTEPROC_WATCHDOG
+ oproc->num_wd_timers =
+ of_count_phandle_with_args(np, "ti,watchdog-timers", NULL);
+ if (oproc->num_wd_timers <= 0) {
+ dev_dbg(&pdev->dev, "device does not have watchdog timers, status = %d\n",
+ oproc->num_wd_timers);
+ oproc->num_wd_timers = 0;
+ }
+#endif
+
+ if (oproc->num_timers || oproc->num_wd_timers) {
+ num_timers = oproc->num_timers + oproc->num_wd_timers;
+ oproc->timers = devm_kcalloc(&pdev->dev, num_timers,
sizeof(*oproc->timers),
GFP_KERNEL);
if (!oproc->timers) {
@@ -1128,8 +1259,8 @@ static int omap_rproc_probe(struct platform_device *pdev)
goto free_rproc;
}
- dev_dbg(&pdev->dev, "device has %d tick timers\n",
- oproc->num_timers);
+ dev_dbg(&pdev->dev, "device has %d tick timers and %d watchdog timers\n",
+ oproc->num_timers, oproc->num_wd_timers);
}
init_completion(&oproc->pm_comp);
--
2.17.1
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
^ permalink raw reply related [flat|nested] 62+ messages in thread