[PATCH v6 0/7] qcom: add OCMEM support

[PATCH v6 0/7] qcom: add OCMEM support

[Brian Masney]

This patch series adds support for Qualcomm's On Chip MEMory (OCMEM)
that is needed in order to support some a3xx and a4xx-based GPUs
upstream. This is based on Rob Clark's patch series that he submitted
in October 2015 and I am resubmitting updated patches with his
permission. See the individual patches for the changelog.

This was tested with the GPU on a LG Nexus 5 (hammerhead) phone and
this will work on other msm8974-based systems. For a summary of what
currently works upstream on the Nexus 5, see my status page at
https://masneyb.github.io/nexus-5-upstream/.

Changes since v5:
- Rename ocmem device tree property to sram

See individual patches for changelogs for previous versions.

Brian Masney (5):
  dt-bindings: soc: qcom: add On Chip MEMory (OCMEM) bindings
  dt-bindings: display: msm: gmu: add optional ocmem property
  soc: qcom: add OCMEM driver
  drm/msm/gpu: add ocmem init/cleanup functions
  ARM: qcom_defconfig: add ocmem support

Rob Clark (2):
  firmware: qcom: scm: add OCMEM lock/unlock interface
  firmware: qcom: scm: add support to restore secure config to
    qcm_scm-32

 .../devicetree/bindings/display/msm/gmu.txt   |  50 ++
 .../devicetree/bindings/sram/qcom,ocmem.yaml  |  96 ++++
 arch/arm/configs/qcom_defconfig               |   1 +
 drivers/firmware/qcom_scm-32.c                |  52 ++-
 drivers/firmware/qcom_scm-64.c                |  12 +
 drivers/firmware/qcom_scm.c                   |  53 +++
 drivers/firmware/qcom_scm.h                   |   9 +
 drivers/gpu/drm/msm/Kconfig                   |   1 +
 drivers/gpu/drm/msm/adreno/a3xx_gpu.c         |  28 +-
 drivers/gpu/drm/msm/adreno/a3xx_gpu.h         |   3 +-
 drivers/gpu/drm/msm/adreno/a4xx_gpu.c         |  25 +-
 drivers/gpu/drm/msm/adreno/a4xx_gpu.h         |   3 +-
 drivers/gpu/drm/msm/adreno/adreno_gpu.c       |  40 ++
 drivers/gpu/drm/msm/adreno/adreno_gpu.h       |  10 +
 drivers/soc/qcom/Kconfig                      |  10 +
 drivers/soc/qcom/Makefile                     |   1 +
 drivers/soc/qcom/ocmem.c                      | 433 ++++++++++++++++++
 include/linux/qcom_scm.h                      |  26 ++
 include/soc/qcom/ocmem.h                      |  62 +++
 19 files changed, 870 insertions(+), 45 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/sram/qcom,ocmem.yaml
 create mode 100644 drivers/soc/qcom/ocmem.c
 create mode 100644 include/soc/qcom/ocmem.h

-- 
2.21.0

[PATCH v6 1/7] dt-bindings: soc: qcom: add On Chip MEMory (OCMEM) bindings

[Brian Masney]

Add device tree bindings for the On Chip Memory (OCMEM) that is present
on some Qualcomm Snapdragon SoCs.

Signed-off-by: Brian Masney <masneyb@onstation.org>
Reviewed-by: Rob Herring <robh@kernel.org>
---
Changes since v5:
- None

Changes since v4:
- remove qcom from path in $id

Changes since v3:
- add ranges property
- remove unnecessary literal block |
- add #address-cells and #size-cells to binding
- rename path devicetree/bindings/sram/qcom/ to devicetree/bindings/sram/ since
  this is the only qcom binding in the sram namespace. That was a holdover from
  when I originally put this in the soc namespace.

Changes since v2:
- Add *-sram node and gmu-sram to example.

Changes since v1:
- Rename qcom,ocmem-msm8974 to qcom,msm8974-ocmem
- Renamed reg-names to ctrl and mem
- update hardware description
- moved from soc to sram namespace in the device tree bindings

 .../devicetree/bindings/sram/qcom,ocmem.yaml  | 96 +++++++++++++++++++
 1 file changed, 96 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/sram/qcom,ocmem.yaml

diff --git a/Documentation/devicetree/bindings/sram/qcom,ocmem.yaml b/Documentation/devicetree/bindings/sram/qcom,ocmem.yaml
new file mode 100644
index 000000000000..222990f9923c
--- /dev/null
+++ b/Documentation/devicetree/bindings/sram/qcom,ocmem.yaml
@@ -0,0 +1,96 @@
+# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/sram/qcom,ocmem.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: On Chip Memory (OCMEM) that is present on some Qualcomm Snapdragon SoCs.
+
+maintainers:
+  - Brian Masney <masneyb@onstation.org>
+
+description: |
+  The On Chip Memory (OCMEM) is typically used by the GPU, camera/video, and
+  audio components on some Snapdragon SoCs.
+
+properties:
+  compatible:
+    const: qcom,msm8974-ocmem
+
+  reg:
+    items:
+      - description: Control registers
+      - description: OCMEM address range
+
+  reg-names:
+    items:
+      - const: ctrl
+      - const: mem
+
+  clocks:
+    items:
+      - description: Core clock
+      - description: Interface clock
+
+  clock-names:
+    items:
+      - const: core
+      - const: iface
+
+  '#address-cells':
+    const: 1
+
+  '#size-cells':
+    const: 1
+
+required:
+  - compatible
+  - reg
+  - reg-names
+  - clocks
+  - clock-names
+  - '#address-cells'
+  - '#size-cells'
+
+patternProperties:
+  "^.+-sram$":
+    type: object
+    description: A region of reserved memory.
+
+    properties:
+      reg:
+        maxItems: 1
+
+      ranges:
+        maxItems: 1
+
+    required:
+      - reg
+      - ranges
+
+examples:
+  - |
+      #include <dt-bindings/clock/qcom,rpmcc.h>
+      #include <dt-bindings/clock/qcom,mmcc-msm8974.h>
+
+      ocmem: ocmem@fdd00000 {
+        compatible = "qcom,msm8974-ocmem";
+
+        reg = <0xfdd00000 0x2000>,
+              <0xfec00000 0x180000>;
+        reg-names = "ctrl",
+                    "mem";
+
+        clocks = <&rpmcc RPM_SMD_OCMEMGX_CLK>,
+                 <&mmcc OCMEMCX_OCMEMNOC_CLK>;
+        clock-names = "core",
+                      "iface";
+
+        #address-cells = <1>;
+        #size-cells = <1>;
+
+        gmu-sram@0 {
+                reg = <0x0 0x100000>;
+                ranges = <0 0 0xfec00000 0x100000>;
+        };
+      };
-- 
2.21.0

[PATCH v6 2/7] dt-bindings: display: msm: gmu: add optional ocmem property

[Brian Masney]

Some A3xx and A4xx Adreno GPUs do not have GMEM inside the GPU core and
must use the On Chip MEMory (OCMEM) in order to be functional. Add the
optional ocmem property to the Adreno Graphics Management Unit bindings.

Signed-off-by: Brian Masney <masneyb@onstation.org>
---
Changes since v5:
- rename ocmem property to sram to match what TI currently has.

Changes since v4:
- None

Changes since v3:
- correct link to qcom,ocmem.yaml

Changes since v2:
- Add a3xx example with OCMEM

Changes since v1:
- None

 .../devicetree/bindings/display/msm/gmu.txt   | 50 +++++++++++++++++++
 1 file changed, 50 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/msm/gmu.txt b/Documentation/devicetree/bindings/display/msm/gmu.txt
index 90af5b0a56a9..2305a2aede5a 100644
--- a/Documentation/devicetree/bindings/display/msm/gmu.txt
+++ b/Documentation/devicetree/bindings/display/msm/gmu.txt
@@ -31,6 +31,10 @@ Required properties:
 - iommus: phandle to the adreno iommu
 - operating-points-v2: phandle to the OPP operating points
 
+Optional properties:
+- sram: phandle to the On Chip Memory (OCMEM) that's present on some Snapdragon
+        SoCs. See Documentation/devicetree/bindings/sram/qcom,ocmem.yaml.
+
 Example:
 
 / {
@@ -63,3 +67,49 @@ Example:
 		operating-points-v2 = <&gmu_opp_table>;
 	};
 };
+
+a3xx example with OCMEM support:
+
+/ {
+	...
+
+	gpu: adreno@fdb00000 {
+		compatible = "qcom,adreno-330.2",
+		             "qcom,adreno";
+		reg = <0xfdb00000 0x10000>;
+		reg-names = "kgsl_3d0_reg_memory";
+		interrupts = <GIC_SPI 33 IRQ_TYPE_LEVEL_HIGH>;
+		interrupt-names = "kgsl_3d0_irq";
+		clock-names = "core",
+		              "iface",
+		              "mem_iface";
+		clocks = <&mmcc OXILI_GFX3D_CLK>,
+		         <&mmcc OXILICX_AHB_CLK>,
+		         <&mmcc OXILICX_AXI_CLK>;
+		sram = <&ocmem>;
+		power-domains = <&mmcc OXILICX_GDSC>;
+		operating-points-v2 = <&gpu_opp_table>;
+		iommus = <&gpu_iommu 0>;
+	};
+
+	ocmem: ocmem@fdd00000 {
+		compatible = "qcom,msm8974-ocmem";
+
+		reg = <0xfdd00000 0x2000>,
+		      <0xfec00000 0x180000>;
+		reg-names = "ctrl",
+		             "mem";
+
+		clocks = <&rpmcc RPM_SMD_OCMEMGX_CLK>,
+		         <&mmcc OCMEMCX_OCMEMNOC_CLK>;
+		clock-names = "core",
+		              "iface";
+
+		#address-cells = <1>;
+		#size-cells = <1>;
+
+		gmu-sram@0 {
+			reg = <0x0 0x100000>;
+		};
+	};
+};
-- 
2.21.0

[PATCH v6 3/7] firmware: qcom: scm: add OCMEM lock/unlock interface

[Brian Masney]

From: Rob Clark <robdclark@gmail.com>

Add support for the OCMEM lock/unlock interface that is needed by the
On Chip MEMory (OCMEM) that is present on some Snapdragon devices.

Signed-off-by: Rob Clark <robdclark@gmail.com>
[masneyb@onstation.org: ported to latest kernel; minor reformatting.]
Signed-off-by: Brian Masney <masneyb@onstation.org>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
Changes since v5:
- None

Changes since v4:
- None

Changes since v3:
- None

Changes since v2:
- None

Changes since v1:
- None

Rob's last version of this patch:
https://patchwork.kernel.org/patch/7340711/

 drivers/firmware/qcom_scm-32.c | 35 +++++++++++++++++++++++++++++
 drivers/firmware/qcom_scm-64.c | 12 ++++++++++
 drivers/firmware/qcom_scm.c    | 40 ++++++++++++++++++++++++++++++++++
 drivers/firmware/qcom_scm.h    |  9 ++++++++
 include/linux/qcom_scm.h       | 15 +++++++++++++
 5 files changed, 111 insertions(+)

diff --git a/drivers/firmware/qcom_scm-32.c b/drivers/firmware/qcom_scm-32.c
index 215061c581e1..4c2514e5e249 100644
--- a/drivers/firmware/qcom_scm-32.c
+++ b/drivers/firmware/qcom_scm-32.c
@@ -442,6 +442,41 @@ int __qcom_scm_hdcp_req(struct device *dev, struct qcom_scm_hdcp_req *req,
 		req, req_cnt * sizeof(*req), resp, sizeof(*resp));
 }
 
+int __qcom_scm_ocmem_lock(struct device *dev, u32 id, u32 offset, u32 size,
+			  u32 mode)
+{
+	struct ocmem_tz_lock {
+		__le32 id;
+		__le32 offset;
+		__le32 size;
+		__le32 mode;
+	} request;
+
+	request.id = cpu_to_le32(id);
+	request.offset = cpu_to_le32(offset);
+	request.size = cpu_to_le32(size);
+	request.mode = cpu_to_le32(mode);
+
+	return qcom_scm_call(dev, QCOM_SCM_OCMEM_SVC, QCOM_SCM_OCMEM_LOCK_CMD,
+			     &request, sizeof(request), NULL, 0);
+}
+
+int __qcom_scm_ocmem_unlock(struct device *dev, u32 id, u32 offset, u32 size)
+{
+	struct ocmem_tz_unlock {
+		__le32 id;
+		__le32 offset;
+		__le32 size;
+	} request;
+
+	request.id = cpu_to_le32(id);
+	request.offset = cpu_to_le32(offset);
+	request.size = cpu_to_le32(size);
+
+	return qcom_scm_call(dev, QCOM_SCM_OCMEM_SVC, QCOM_SCM_OCMEM_UNLOCK_CMD,
+			     &request, sizeof(request), NULL, 0);
+}
+
 void __qcom_scm_init(void)
 {
 }
diff --git a/drivers/firmware/qcom_scm-64.c b/drivers/firmware/qcom_scm-64.c
index 91d5ad7cf58b..c3a3d9874def 100644
--- a/drivers/firmware/qcom_scm-64.c
+++ b/drivers/firmware/qcom_scm-64.c
@@ -241,6 +241,18 @@ int __qcom_scm_hdcp_req(struct device *dev, struct qcom_scm_hdcp_req *req,
 	return ret;
 }
 
+int __qcom_scm_ocmem_lock(struct device *dev, uint32_t id, uint32_t offset,
+			  uint32_t size, uint32_t mode)
+{
+	return -ENOTSUPP;
+}
+
+int __qcom_scm_ocmem_unlock(struct device *dev, uint32_t id, uint32_t offset,
+			    uint32_t size)
+{
+	return -ENOTSUPP;
+}
+
 void __qcom_scm_init(void)
 {
 	u64 cmd;
diff --git a/drivers/firmware/qcom_scm.c b/drivers/firmware/qcom_scm.c
index 4802ab170fe5..7e285ff3961d 100644
--- a/drivers/firmware/qcom_scm.c
+++ b/drivers/firmware/qcom_scm.c
@@ -191,6 +191,46 @@ bool qcom_scm_pas_supported(u32 peripheral)
 }
 EXPORT_SYMBOL(qcom_scm_pas_supported);
 
+/**
+ * qcom_scm_ocmem_lock_available() - is OCMEM lock/unlock interface available
+ */
+bool qcom_scm_ocmem_lock_available(void)
+{
+	return __qcom_scm_is_call_available(__scm->dev, QCOM_SCM_OCMEM_SVC,
+					    QCOM_SCM_OCMEM_LOCK_CMD);
+}
+EXPORT_SYMBOL(qcom_scm_ocmem_lock_available);
+
+/**
+ * qcom_scm_ocmem_lock() - call OCMEM lock interface to assign an OCMEM
+ * region to the specified initiator
+ *
+ * @id:     tz initiator id
+ * @offset: OCMEM offset
+ * @size:   OCMEM size
+ * @mode:   access mode (WIDE/NARROW)
+ */
+int qcom_scm_ocmem_lock(enum qcom_scm_ocmem_client id, u32 offset, u32 size,
+			u32 mode)
+{
+	return __qcom_scm_ocmem_lock(__scm->dev, id, offset, size, mode);
+}
+EXPORT_SYMBOL(qcom_scm_ocmem_lock);
+
+/**
+ * qcom_scm_ocmem_unlock() - call OCMEM unlock interface to release an OCMEM
+ * region from the specified initiator
+ *
+ * @id:     tz initiator id
+ * @offset: OCMEM offset
+ * @size:   OCMEM size
+ */
+int qcom_scm_ocmem_unlock(enum qcom_scm_ocmem_client id, u32 offset, u32 size)
+{
+	return __qcom_scm_ocmem_unlock(__scm->dev, id, offset, size);
+}
+EXPORT_SYMBOL(qcom_scm_ocmem_unlock);
+
 /**
  * qcom_scm_pas_init_image() - Initialize peripheral authentication service
  *			       state machine for a given peripheral, using the
diff --git a/drivers/firmware/qcom_scm.h b/drivers/firmware/qcom_scm.h
index 99506bd873c0..ef293ee67ec1 100644
--- a/drivers/firmware/qcom_scm.h
+++ b/drivers/firmware/qcom_scm.h
@@ -42,6 +42,15 @@ extern int __qcom_scm_hdcp_req(struct device *dev,
 
 extern void __qcom_scm_init(void);
 
+#define QCOM_SCM_OCMEM_SVC			0xf
+#define QCOM_SCM_OCMEM_LOCK_CMD		0x1
+#define QCOM_SCM_OCMEM_UNLOCK_CMD		0x2
+
+extern int __qcom_scm_ocmem_lock(struct device *dev, u32 id, u32 offset,
+				 u32 size, u32 mode);
+extern int __qcom_scm_ocmem_unlock(struct device *dev, u32 id, u32 offset,
+				   u32 size);
+
 #define QCOM_SCM_SVC_PIL		0x2
 #define QCOM_SCM_PAS_INIT_IMAGE_CMD	0x1
 #define QCOM_SCM_PAS_MEM_SETUP_CMD	0x2
diff --git a/include/linux/qcom_scm.h b/include/linux/qcom_scm.h
index 2d5eff506e13..b49b734d662c 100644
--- a/include/linux/qcom_scm.h
+++ b/include/linux/qcom_scm.h
@@ -24,6 +24,16 @@ struct qcom_scm_vmperm {
 	int perm;
 };
 
+enum qcom_scm_ocmem_client {
+	QCOM_SCM_OCMEM_UNUSED_ID = 0x0,
+	QCOM_SCM_OCMEM_GRAPHICS_ID,
+	QCOM_SCM_OCMEM_VIDEO_ID,
+	QCOM_SCM_OCMEM_LP_AUDIO_ID,
+	QCOM_SCM_OCMEM_SENSORS_ID,
+	QCOM_SCM_OCMEM_OTHER_OS_ID,
+	QCOM_SCM_OCMEM_DEBUG_ID,
+};
+
 #define QCOM_SCM_VMID_HLOS       0x3
 #define QCOM_SCM_VMID_MSS_MSA    0xF
 #define QCOM_SCM_VMID_WLAN       0x18
@@ -41,6 +51,11 @@ extern bool qcom_scm_is_available(void);
 extern bool qcom_scm_hdcp_available(void);
 extern int qcom_scm_hdcp_req(struct qcom_scm_hdcp_req *req, u32 req_cnt,
 			     u32 *resp);
+extern bool qcom_scm_ocmem_lock_available(void);
+extern int qcom_scm_ocmem_lock(enum qcom_scm_ocmem_client id, u32 offset,
+			       u32 size, u32 mode);
+extern int qcom_scm_ocmem_unlock(enum qcom_scm_ocmem_client id, u32 offset,
+				 u32 size);
 extern bool qcom_scm_pas_supported(u32 peripheral);
 extern int qcom_scm_pas_init_image(u32 peripheral, const void *metadata,
 				   size_t size);
-- 
2.21.0

[PATCH v6 4/7] firmware: qcom: scm: add support to restore secure config to qcm_scm-32

[Brian Masney]

From: Rob Clark <robdclark@gmail.com>

Add support to restore the secure configuration for qcm_scm-32.c. This
is needed by the On Chip MEMory (OCMEM) that is present on some
Snapdragon devices.

Signed-off-by: Rob Clark <robdclark@gmail.com>
[masneyb@onstation.org: ported to latest kernel; set ctx_bank_num to
 spare parameter.]
Signed-off-by: Brian Masney <masneyb@onstation.org>
---
Changes since v5:
- None

Changes since v4:
- None

Changes since v3:
- None

Changes since v2:
- None

Changes since v1:
- Use existing __qcom_scm_restore_sec_cfg() function stub in
  qcom_scm-32.c that was unimplemented
- Set the cfg.ctx_bank_num to the spare function parameter. It was
  previously set to the device_id.

 drivers/firmware/qcom_scm-32.c | 17 ++++++++++++++++-
 drivers/firmware/qcom_scm.c    | 13 +++++++++++++
 include/linux/qcom_scm.h       | 11 +++++++++++
 3 files changed, 40 insertions(+), 1 deletion(-)

diff --git a/drivers/firmware/qcom_scm-32.c b/drivers/firmware/qcom_scm-32.c
index 4c2514e5e249..5d90b7f5ab5a 100644
--- a/drivers/firmware/qcom_scm-32.c
+++ b/drivers/firmware/qcom_scm-32.c
@@ -617,7 +617,22 @@ int __qcom_scm_assign_mem(struct device *dev, phys_addr_t mem_region,
 int __qcom_scm_restore_sec_cfg(struct device *dev, u32 device_id,
 			       u32 spare)
 {
-	return -ENODEV;
+	struct msm_scm_sec_cfg {
+		__le32 id;
+		__le32 ctx_bank_num;
+	} cfg;
+	int ret, scm_ret = 0;
+
+	cfg.id = cpu_to_le32(device_id);
+	cfg.ctx_bank_num = cpu_to_le32(spare);
+
+	ret = qcom_scm_call(dev, QCOM_SCM_SVC_MP, QCOM_SCM_RESTORE_SEC_CFG,
+			    &cfg, sizeof(cfg), &scm_ret, sizeof(scm_ret));
+
+	if (ret || scm_ret)
+		return ret ? ret : -EINVAL;
+
+	return 0;
 }
 
 int __qcom_scm_iommu_secure_ptbl_size(struct device *dev, u32 spare,
diff --git a/drivers/firmware/qcom_scm.c b/drivers/firmware/qcom_scm.c
index 7e285ff3961d..27c1d98a34e6 100644
--- a/drivers/firmware/qcom_scm.c
+++ b/drivers/firmware/qcom_scm.c
@@ -367,6 +367,19 @@ static const struct reset_control_ops qcom_scm_pas_reset_ops = {
 	.deassert = qcom_scm_pas_reset_deassert,
 };
 
+/**
+ * qcom_scm_restore_sec_cfg_available() - Check if secure environment
+ * supports restore security config interface.
+ *
+ * Return true if restore-cfg interface is supported, false if not.
+ */
+bool qcom_scm_restore_sec_cfg_available(void)
+{
+	return __qcom_scm_is_call_available(__scm->dev, QCOM_SCM_SVC_MP,
+					    QCOM_SCM_RESTORE_SEC_CFG);
+}
+EXPORT_SYMBOL(qcom_scm_restore_sec_cfg_available);
+
 int qcom_scm_restore_sec_cfg(u32 device_id, u32 spare)
 {
 	return __qcom_scm_restore_sec_cfg(__scm->dev, device_id, spare);
diff --git a/include/linux/qcom_scm.h b/include/linux/qcom_scm.h
index b49b734d662c..04382e1798e4 100644
--- a/include/linux/qcom_scm.h
+++ b/include/linux/qcom_scm.h
@@ -34,6 +34,16 @@ enum qcom_scm_ocmem_client {
 	QCOM_SCM_OCMEM_DEBUG_ID,
 };
 
+enum qcom_scm_sec_dev_id {
+	QCOM_SCM_MDSS_DEV_ID    = 1,
+	QCOM_SCM_OCMEM_DEV_ID   = 5,
+	QCOM_SCM_PCIE0_DEV_ID   = 11,
+	QCOM_SCM_PCIE1_DEV_ID   = 12,
+	QCOM_SCM_GFX_DEV_ID     = 18,
+	QCOM_SCM_UFS_DEV_ID     = 19,
+	QCOM_SCM_ICE_DEV_ID     = 20,
+};
+
 #define QCOM_SCM_VMID_HLOS       0x3
 #define QCOM_SCM_VMID_MSS_MSA    0xF
 #define QCOM_SCM_VMID_WLAN       0x18
@@ -70,6 +80,7 @@ extern int qcom_scm_assign_mem(phys_addr_t mem_addr, size_t mem_sz,
 extern void qcom_scm_cpu_power_down(u32 flags);
 extern u32 qcom_scm_get_version(void);
 extern int qcom_scm_set_remote_state(u32 state, u32 id);
+extern bool qcom_scm_restore_sec_cfg_available(void);
 extern int qcom_scm_restore_sec_cfg(u32 device_id, u32 spare);
 extern int qcom_scm_iommu_secure_ptbl_size(u32 spare, size_t *size);
 extern int qcom_scm_iommu_secure_ptbl_init(u64 addr, u32 size, u32 spare);
-- 
2.21.0

[PATCH v6 5/7] soc: qcom: add OCMEM driver

[Brian Masney]

The OCMEM driver handles allocation and configuration of the On Chip
MEMory that is present on some Snapdragon SoCs. Devices which have
OCMEM do not have GMEM inside the GPU core, so the GPU must instead
use OCMEM to be functional. Since the GPU is currently the only OCMEM
user with an upstream driver, this is just a minimal implementation
sufficient for statically allocating to the GPU it's chunk of OCMEM.

This driver currently does not read the gmu-sram node that is described
in the device tree bindings. The starting memory address of the GPU's
reserved memory region is hardcoded to zero to match what the hardware
expects. The driver can be updated to read the reserved memory regions
from device tree once other users of OCMEM are added upstream.

Signed-off-by: Brian Masney <masneyb@onstation.org>
Co-developed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
Changes since v5:
- rename ocmem property to sram

Changes since v4:
- Change 'GPL' to 'GPL v2' in MODULE_LICENSE

Changes since v3:
- None

Changes since v2
- Changed static inline stubs return -ENODEV when OCMEM is not
  configured into the kernel.

Changes since v1:
- ocmem_allocate(): check for alignment and minimum allocation size.
  The 64K values came from the downstream MSM kernel sources.
- add locking to memory allocations based on the client
- use clk_bulk_*() functions
- rename qcom,ocmem-msm8974 to qcom,msm8974-ocmem
- rename reg-names to ctrl and mem
- remove ocmem.xml.h file; use FIELD_PREP() instead for some nice cleanups
- add static inline noop versions of public-facing functions when ocmem
  is disabled to remove #ifdefs in adrenu_gpu.c
- use unsigned long for memory addresses
- move ocmem_dev_remove() below _probe() function
- remove error check from platform_get_resource_byname for ctrl resource
- add MODULE_DESCRIPTION() and MODULE_LICENSE()
- add description to top of ocmem.[ch]
- correct thin mode bit in update_ocmem()
- add 'WARN_ON(client != OCMEM_GRAPHICS)' to device_address()
- make of_get_ocmem return error codes via ERR_PTR instead of NULL
- ocmem_{allocate,free} - WARN_ON() if client != OCMEM_GRAPHICS.
  Simplify if statements.
- allow NULL to be passed into ocmem_free
- remove unnecessary initialization of i in update_ocmem()
- add dev_dbg to ocmem_allocate

Changes since Rob's last version of this patch from 2015:
https://patchwork.kernel.org/patch/7379801/
- reformatted driver to allow multiple instances
- updated logging of error paths during device probing
- remove unused psgsc_ctrl
- remove _clk from clock names
- propagate error code from devm_ioremap_resource()
- use device_get_match_data()
- SPDX license tags
- remove QCOM_SMD in Kconfig
- select ARCH_QCOM in Kconfig
- select ARCH_QCOM in Kconfig
- select QCOM_SCM in Kconfig
- longer description in Kconfig

 drivers/soc/qcom/Kconfig  |  10 +
 drivers/soc/qcom/Makefile |   1 +
 drivers/soc/qcom/ocmem.c  | 433 ++++++++++++++++++++++++++++++++++++++
 include/soc/qcom/ocmem.h  |  62 ++++++
 4 files changed, 506 insertions(+)
 create mode 100644 drivers/soc/qcom/ocmem.c
 create mode 100644 include/soc/qcom/ocmem.h

diff --git a/drivers/soc/qcom/Kconfig b/drivers/soc/qcom/Kconfig
index 661e47acc354..adbf3bfae805 100644
--- a/drivers/soc/qcom/Kconfig
+++ b/drivers/soc/qcom/Kconfig
@@ -74,6 +74,16 @@ config QCOM_MDT_LOADER
 	tristate
 	select QCOM_SCM
 
+config QCOM_OCMEM
+	tristate "Qualcomm On Chip Memory (OCMEM) driver"
+	depends on ARCH_QCOM
+	select QCOM_SCM
+	help
+          The On Chip Memory (OCMEM) allocator allows various clients to
+          allocate memory from OCMEM based on performance, latency and power
+          requirements. This is typically used by the GPU, camera/video, and
+          audio components on some Snapdragon SoCs.
+
 config QCOM_PM
 	bool "Qualcomm Power Management"
 	depends on ARCH_QCOM && !ARM64
diff --git a/drivers/soc/qcom/Makefile b/drivers/soc/qcom/Makefile
index 162788701a77..aab333720bfd 100644
--- a/drivers/soc/qcom/Makefile
+++ b/drivers/soc/qcom/Makefile
@@ -6,6 +6,7 @@ obj-$(CONFIG_QCOM_COMMAND_DB) += cmd-db.o
 obj-$(CONFIG_QCOM_GLINK_SSR) +=	glink_ssr.o
 obj-$(CONFIG_QCOM_GSBI)	+=	qcom_gsbi.o
 obj-$(CONFIG_QCOM_MDT_LOADER)	+= mdt_loader.o
+obj-$(CONFIG_QCOM_OCMEM)	+= ocmem.o
 obj-$(CONFIG_QCOM_PM)	+=	spm.o
 obj-$(CONFIG_QCOM_QMI_HELPERS)	+= qmi_helpers.o
 qmi_helpers-y	+= qmi_encdec.o qmi_interface.o
diff --git a/drivers/soc/qcom/ocmem.c b/drivers/soc/qcom/ocmem.c
new file mode 100644
index 000000000000..f9e31be5d844
--- /dev/null
+++ b/drivers/soc/qcom/ocmem.c
@@ -0,0 +1,433 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * The On Chip Memory (OCMEM) allocator allows various clients to allocate
+ * memory from OCMEM based on performance, latency and power requirements.
+ * This is typically used by the GPU, camera/video, and audio components on
+ * some Snapdragon SoCs.
+ *
+ * Copyright (C) 2019 Brian Masney <masneyb@onstation.org>
+ * Copyright (C) 2015 Red Hat. Author: Rob Clark <robdclark@gmail.com>
+ */
+
+#include <linux/bitfield.h>
+#include <linux/clk.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/qcom_scm.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <soc/qcom/ocmem.h>
+
+enum region_mode {
+	WIDE_MODE = 0x0,
+	THIN_MODE,
+	MODE_DEFAULT = WIDE_MODE,
+};
+
+enum ocmem_macro_state {
+	PASSTHROUGH = 0,
+	PERI_ON = 1,
+	CORE_ON = 2,
+	CLK_OFF = 4,
+};
+
+struct ocmem_region {
+	bool interleaved;
+	enum region_mode mode;
+	unsigned int num_macros;
+	enum ocmem_macro_state macro_state[4];
+	unsigned long macro_size;
+	unsigned long region_size;
+};
+
+struct ocmem_config {
+	uint8_t num_regions;
+	unsigned long macro_size;
+};
+
+struct ocmem {
+	struct device *dev;
+	const struct ocmem_config *config;
+	struct resource *memory;
+	void __iomem *mmio;
+	unsigned int num_ports;
+	unsigned int num_macros;
+	bool interleaved;
+	struct ocmem_region *regions;
+	unsigned long active_allocations;
+};
+
+#define OCMEM_MIN_ALIGN				SZ_64K
+#define OCMEM_MIN_ALLOC				SZ_64K
+
+#define OCMEM_REG_HW_VERSION			0x00000000
+#define OCMEM_REG_HW_PROFILE			0x00000004
+
+#define OCMEM_REG_REGION_MODE_CTL		0x00001000
+#define OCMEM_REGION_MODE_CTL_REG0_THIN		0x00000001
+#define OCMEM_REGION_MODE_CTL_REG1_THIN		0x00000002
+#define OCMEM_REGION_MODE_CTL_REG2_THIN		0x00000004
+#define OCMEM_REGION_MODE_CTL_REG3_THIN		0x00000008
+
+#define OCMEM_REG_GFX_MPU_START			0x00001004
+#define OCMEM_REG_GFX_MPU_END			0x00001008
+
+#define OCMEM_HW_PROFILE_NUM_PORTS(val)		FIELD_PREP(0x0000000f, (val))
+#define OCMEM_HW_PROFILE_NUM_MACROS(val)	FIELD_PREP(0x00003f00, (val))
+
+#define OCMEM_HW_PROFILE_LAST_REGN_HALFSIZE	0x00010000
+#define OCMEM_HW_PROFILE_INTERLEAVING		0x00020000
+#define OCMEM_REG_GEN_STATUS			0x0000000c
+
+#define OCMEM_REG_PSGSC_STATUS			0x00000038
+#define OCMEM_REG_PSGSC_CTL(i0)			(0x0000003c + 0x1*(i0))
+
+#define OCMEM_PSGSC_CTL_MACRO0_MODE(val)	FIELD_PREP(0x00000007, (val))
+#define OCMEM_PSGSC_CTL_MACRO1_MODE(val)	FIELD_PREP(0x00000070, (val))
+#define OCMEM_PSGSC_CTL_MACRO2_MODE(val)	FIELD_PREP(0x00000700, (val))
+#define OCMEM_PSGSC_CTL_MACRO3_MODE(val)	FIELD_PREP(0x00007000, (val))
+
+#define OCMEM_CLK_CORE_IDX			0
+static struct clk_bulk_data ocmem_clks[] = {
+	{
+		.id = "core",
+	},
+	{
+		.id = "iface",
+	},
+};
+
+static inline void ocmem_write(struct ocmem *ocmem, u32 reg, u32 data)
+{
+	writel(data, ocmem->mmio + reg);
+}
+
+static inline u32 ocmem_read(struct ocmem *ocmem, u32 reg)
+{
+	return readl(ocmem->mmio + reg);
+}
+
+static void update_ocmem(struct ocmem *ocmem)
+{
+	uint32_t region_mode_ctrl = 0x0;
+	int i;
+
+	if (!qcom_scm_ocmem_lock_available()) {
+		for (i = 0; i < ocmem->config->num_regions; i++) {
+			struct ocmem_region *region = &ocmem->regions[i];
+
+			if (region->mode == THIN_MODE)
+				region_mode_ctrl |= BIT(i);
+		}
+
+		dev_dbg(ocmem->dev, "ocmem_region_mode_control %x\n",
+			region_mode_ctrl);
+		ocmem_write(ocmem, OCMEM_REG_REGION_MODE_CTL, region_mode_ctrl);
+	}
+
+	for (i = 0; i < ocmem->config->num_regions; i++) {
+		struct ocmem_region *region = &ocmem->regions[i];
+		u32 data;
+
+		data = OCMEM_PSGSC_CTL_MACRO0_MODE(region->macro_state[0]) |
+			OCMEM_PSGSC_CTL_MACRO1_MODE(region->macro_state[1]) |
+			OCMEM_PSGSC_CTL_MACRO2_MODE(region->macro_state[2]) |
+			OCMEM_PSGSC_CTL_MACRO3_MODE(region->macro_state[3]);
+
+		ocmem_write(ocmem, OCMEM_REG_PSGSC_CTL(i), data);
+	}
+}
+
+static unsigned long phys_to_offset(struct ocmem *ocmem,
+				    unsigned long addr)
+{
+	if (addr < ocmem->memory->start || addr >= ocmem->memory->end)
+		return 0;
+
+	return addr - ocmem->memory->start;
+}
+
+static unsigned long device_address(struct ocmem *ocmem,
+				    enum ocmem_client client,
+				    unsigned long addr)
+{
+	WARN_ON(client != OCMEM_GRAPHICS);
+
+	/* TODO: gpu uses phys_to_offset, but others do not.. */
+	return phys_to_offset(ocmem, addr);
+}
+
+static void update_range(struct ocmem *ocmem, struct ocmem_buf *buf,
+			 enum ocmem_macro_state mstate, enum region_mode rmode)
+{
+	unsigned long offset = 0;
+	int i, j;
+
+	for (i = 0; i < ocmem->config->num_regions; i++) {
+		struct ocmem_region *region = &ocmem->regions[i];
+
+		if (buf->offset <= offset && offset < buf->offset + buf->len)
+			region->mode = rmode;
+
+		for (j = 0; j < region->num_macros; j++) {
+			if (buf->offset <= offset &&
+			    offset < buf->offset + buf->len)
+				region->macro_state[j] = mstate;
+
+			offset += region->macro_size;
+		}
+	}
+
+	update_ocmem(ocmem);
+}
+
+struct ocmem *of_get_ocmem(struct device *dev)
+{
+	struct platform_device *pdev;
+	struct device_node *devnode;
+
+	devnode = of_parse_phandle(dev->of_node, "sram", 0);
+	if (!devnode) {
+		dev_err(dev, "Cannot look up sram phandle\n");
+		return ERR_PTR(-ENODEV);
+	}
+
+	pdev = of_find_device_by_node(devnode);
+	if (!pdev) {
+		dev_err(dev, "Cannot find device node %s\n", devnode->name);
+		return ERR_PTR(-EPROBE_DEFER);
+	}
+
+	return platform_get_drvdata(pdev);
+}
+EXPORT_SYMBOL(of_get_ocmem);
+
+struct ocmem_buf *ocmem_allocate(struct ocmem *ocmem, enum ocmem_client client,
+				 unsigned long size)
+{
+	struct ocmem_buf *buf;
+	int ret;
+
+	/* TODO: add support for other clients... */
+	if (WARN_ON(client != OCMEM_GRAPHICS))
+		return ERR_PTR(-ENODEV);
+
+	if (size < OCMEM_MIN_ALLOC || !IS_ALIGNED(size, OCMEM_MIN_ALIGN))
+		return ERR_PTR(-EINVAL);
+
+	if (test_and_set_bit_lock(BIT(client), &ocmem->active_allocations))
+		return ERR_PTR(-EBUSY);
+
+	buf = kzalloc(sizeof(*buf), GFP_KERNEL);
+	if (!buf) {
+		ret = -ENOMEM;
+		goto err_unlock;
+	}
+
+	buf->offset = 0;
+	buf->addr = device_address(ocmem, client, buf->offset);
+	buf->len = size;
+
+	update_range(ocmem, buf, CORE_ON, WIDE_MODE);
+
+	if (qcom_scm_ocmem_lock_available()) {
+		ret = qcom_scm_ocmem_lock(QCOM_SCM_OCMEM_GRAPHICS_ID,
+					  buf->offset, buf->len, WIDE_MODE);
+		if (ret) {
+			dev_err(ocmem->dev, "could not lock: %d\n", ret);
+			ret = -EINVAL;
+			goto err_kfree;
+		}
+	} else {
+		ocmem_write(ocmem, OCMEM_REG_GFX_MPU_START, buf->offset);
+		ocmem_write(ocmem, OCMEM_REG_GFX_MPU_END,
+			    buf->offset + buf->len);
+	}
+
+	dev_dbg(ocmem->dev, "using %ldK of OCMEM at 0x%08lx for client %d\n",
+		size / 1024, buf->addr, client);
+
+	return buf;
+
+err_kfree:
+	kfree(buf);
+err_unlock:
+	clear_bit_unlock(BIT(client), &ocmem->active_allocations);
+
+	return ERR_PTR(ret);
+}
+EXPORT_SYMBOL(ocmem_allocate);
+
+void ocmem_free(struct ocmem *ocmem, enum ocmem_client client,
+		struct ocmem_buf *buf)
+{
+	/* TODO: add support for other clients... */
+	if (WARN_ON(client != OCMEM_GRAPHICS))
+		return;
+
+	update_range(ocmem, buf, CLK_OFF, MODE_DEFAULT);
+
+	if (qcom_scm_ocmem_lock_available()) {
+		int ret;
+
+		ret = qcom_scm_ocmem_unlock(QCOM_SCM_OCMEM_GRAPHICS_ID,
+					    buf->offset, buf->len);
+		if (ret)
+			dev_err(ocmem->dev, "could not unlock: %d\n", ret);
+	} else {
+		ocmem_write(ocmem, OCMEM_REG_GFX_MPU_START, 0x0);
+		ocmem_write(ocmem, OCMEM_REG_GFX_MPU_END, 0x0);
+	}
+
+	kfree(buf);
+
+	clear_bit_unlock(BIT(client), &ocmem->active_allocations);
+}
+EXPORT_SYMBOL(ocmem_free);
+
+static int ocmem_dev_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	unsigned long reg, region_size;
+	int i, j, ret, num_banks;
+	struct resource *res;
+	struct ocmem *ocmem;
+
+	if (!qcom_scm_is_available())
+		return -EPROBE_DEFER;
+
+	ocmem = devm_kzalloc(dev, sizeof(*ocmem), GFP_KERNEL);
+	if (!ocmem)
+		return -ENOMEM;
+
+	ocmem->dev = dev;
+	ocmem->config = device_get_match_data(dev);
+
+	ret = devm_clk_bulk_get(dev, ARRAY_SIZE(ocmem_clks), ocmem_clks);
+	if (ret) {
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Unable to get clocks\n");
+
+		return ret;
+	}
+
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "ctrl");
+	ocmem->mmio = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(ocmem->mmio)) {
+		dev_err(&pdev->dev, "Failed to ioremap ocmem_ctrl resource\n");
+		return PTR_ERR(ocmem->mmio);
+	}
+
+	ocmem->memory = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+						     "mem");
+	if (!ocmem->memory) {
+		dev_err(dev, "Could not get mem region\n");
+		return -ENXIO;
+	}
+
+	/* The core clock is synchronous with graphics */
+	WARN_ON(clk_set_rate(ocmem_clks[OCMEM_CLK_CORE_IDX].clk, 1000) < 0);
+
+	ret = clk_bulk_prepare_enable(ARRAY_SIZE(ocmem_clks), ocmem_clks);
+	if (ret) {
+		dev_info(ocmem->dev, "Failed to enable clocks\n");
+		return ret;
+	}
+
+	if (qcom_scm_restore_sec_cfg_available()) {
+		dev_dbg(dev, "configuring scm\n");
+		ret = qcom_scm_restore_sec_cfg(QCOM_SCM_OCMEM_DEV_ID, 0);
+		if (ret) {
+			dev_err(dev, "Could not enable secure configuration\n");
+			goto err_clk_disable;
+		}
+	}
+
+	reg = ocmem_read(ocmem, OCMEM_REG_HW_PROFILE);
+	ocmem->num_ports = OCMEM_HW_PROFILE_NUM_PORTS(reg);
+	ocmem->num_macros = OCMEM_HW_PROFILE_NUM_MACROS(reg);
+	ocmem->interleaved = !!(reg & OCMEM_HW_PROFILE_INTERLEAVING);
+
+	num_banks = ocmem->num_ports / 2;
+	region_size = ocmem->config->macro_size * num_banks;
+
+	dev_info(dev, "%u ports, %u regions, %u macros, %sinterleaved\n",
+		 ocmem->num_ports, ocmem->config->num_regions,
+		 ocmem->num_macros, ocmem->interleaved ? "" : "not ");
+
+	ocmem->regions = devm_kcalloc(dev, ocmem->config->num_regions,
+				      sizeof(struct ocmem_region), GFP_KERNEL);
+	if (!ocmem->regions) {
+		ret = -ENOMEM;
+		goto err_clk_disable;
+	}
+
+	for (i = 0; i < ocmem->config->num_regions; i++) {
+		struct ocmem_region *region = &ocmem->regions[i];
+
+		if (WARN_ON(num_banks > ARRAY_SIZE(region->macro_state))) {
+			ret = -EINVAL;
+			goto err_clk_disable;
+		}
+
+		region->mode = MODE_DEFAULT;
+		region->num_macros = num_banks;
+
+		if (i == (ocmem->config->num_regions - 1) &&
+		    reg & OCMEM_HW_PROFILE_LAST_REGN_HALFSIZE) {
+			region->macro_size = ocmem->config->macro_size / 2;
+			region->region_size = region_size / 2;
+		} else {
+			region->macro_size = ocmem->config->macro_size;
+			region->region_size = region_size;
+		}
+
+		for (j = 0; j < ARRAY_SIZE(region->macro_state); j++)
+			region->macro_state[j] = CLK_OFF;
+	}
+
+	platform_set_drvdata(pdev, ocmem);
+
+	return 0;
+
+err_clk_disable:
+	clk_bulk_disable_unprepare(ARRAY_SIZE(ocmem_clks), ocmem_clks);
+	return ret;
+}
+
+static int ocmem_dev_remove(struct platform_device *pdev)
+{
+	clk_bulk_disable_unprepare(ARRAY_SIZE(ocmem_clks), ocmem_clks);
+
+	return 0;
+}
+
+static const struct ocmem_config ocmem_8974_config = {
+	.num_regions = 3,
+	.macro_size = SZ_128K,
+};
+
+static const struct of_device_id ocmem_of_match[] = {
+	{ .compatible = "qcom,msm8974-ocmem", .data = &ocmem_8974_config },
+	{ }
+};
+
+MODULE_DEVICE_TABLE(of, ocmem_of_match);
+
+static struct platform_driver ocmem_driver = {
+	.probe = ocmem_dev_probe,
+	.remove = ocmem_dev_remove,
+	.driver = {
+		.name = "ocmem",
+		.of_match_table = ocmem_of_match,
+	},
+};
+
+module_platform_driver(ocmem_driver);
+
+MODULE_DESCRIPTION("On Chip Memory (OCMEM) allocator for some Snapdragon SoCs");
+MODULE_LICENSE("GPL v2");
diff --git a/include/soc/qcom/ocmem.h b/include/soc/qcom/ocmem.h
new file mode 100644
index 000000000000..a0ae336ba78b
--- /dev/null
+++ b/include/soc/qcom/ocmem.h
@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * The On Chip Memory (OCMEM) allocator allows various clients to allocate
+ * memory from OCMEM based on performance, latency and power requirements.
+ * This is typically used by the GPU, camera/video, and audio components on
+ * some Snapdragon SoCs.
+ *
+ * Copyright (C) 2019 Brian Masney <masneyb@onstation.org>
+ * Copyright (C) 2015 Red Hat. Author: Rob Clark <robdclark@gmail.com>
+ */
+
+#ifndef __OCMEM_H__
+#define __OCMEM_H__
+
+enum ocmem_client {
+	/* GMEM clients */
+	OCMEM_GRAPHICS = 0x0,
+	/*
+	 * TODO add more once ocmem_allocate() is clever enough to
+	 * deal with multiple clients.
+	 */
+	OCMEM_CLIENT_MAX,
+};
+
+struct ocmem;
+
+struct ocmem_buf {
+	unsigned long offset;
+	unsigned long addr;
+	unsigned long len;
+};
+
+#if IS_ENABLED(CONFIG_QCOM_OCMEM)
+
+struct ocmem *of_get_ocmem(struct device *dev);
+struct ocmem_buf *ocmem_allocate(struct ocmem *ocmem, enum ocmem_client client,
+				 unsigned long size);
+void ocmem_free(struct ocmem *ocmem, enum ocmem_client client,
+		struct ocmem_buf *buf);
+
+#else /* IS_ENABLED(CONFIG_QCOM_OCMEM) */
+
+static inline struct ocmem *of_get_ocmem(struct device *dev)
+{
+	return ERR_PTR(-ENODEV);
+}
+
+static inline struct ocmem_buf *ocmem_allocate(struct ocmem *ocmem,
+					       enum ocmem_client client,
+					       unsigned long size)
+{
+	return ERR_PTR(-ENODEV);
+}
+
+static inline void ocmem_free(struct ocmem *ocmem, enum ocmem_client client,
+			      struct ocmem_buf *buf)
+{
+}
+
+#endif /* IS_ENABLED(CONFIG_QCOM_OCMEM) */
+
+#endif /* __OCMEM_H__ */
-- 
2.21.0

[PATCH v6 6/7] drm/msm/gpu: add ocmem init/cleanup functions

[Brian Masney]

The files a3xx_gpu.c and a4xx_gpu.c have ifdefs for the OCMEM support
that was missing upstream. Add two new functions (adreno_gpu_ocmem_init
and adreno_gpu_ocmem_cleanup) that removes some duplicated code.

Signed-off-by: Brian Masney <masneyb@onstation.org>
---
Changes since v5:
- None

Changes since v4:
- None

Changes since v3:
- None

Changes since v2:
- Check for -ENODEV error of_get_ocmem()
- remove fail_cleanup_ocmem label in a[34]xx_gpu_init

Changes since v1:
- remove CONFIG_QCOM_OCMEM #ifdefs
- use unsigned long for memory addresses instead of uint32_t
- add 'depends on QCOM_OCMEM || QCOM_OCMEM=n' to Kconfig

 drivers/gpu/drm/msm/Kconfig             |  1 +
 drivers/gpu/drm/msm/adreno/a3xx_gpu.c   | 28 +++++------------
 drivers/gpu/drm/msm/adreno/a3xx_gpu.h   |  3 +-
 drivers/gpu/drm/msm/adreno/a4xx_gpu.c   | 25 ++++------------
 drivers/gpu/drm/msm/adreno/a4xx_gpu.h   |  3 +-
 drivers/gpu/drm/msm/adreno/adreno_gpu.c | 40 +++++++++++++++++++++++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.h | 10 +++++++
 7 files changed, 66 insertions(+), 44 deletions(-)

diff --git a/drivers/gpu/drm/msm/Kconfig b/drivers/gpu/drm/msm/Kconfig
index 9c37e4de5896..b3d3b2172659 100644
--- a/drivers/gpu/drm/msm/Kconfig
+++ b/drivers/gpu/drm/msm/Kconfig
@@ -7,6 +7,7 @@ config DRM_MSM
 	depends on OF && COMMON_CLK
 	depends on MMU
 	depends on INTERCONNECT || !INTERCONNECT
+	depends on QCOM_OCMEM || QCOM_OCMEM=n
 	select QCOM_MDT_LOADER if ARCH_QCOM
 	select REGULATOR
 	select DRM_KMS_HELPER
diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
index 5f7e98028eaf..7ad14937fcdf 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
@@ -6,10 +6,6 @@
  * Copyright (c) 2014 The Linux Foundation. All rights reserved.
  */
 
-#ifdef CONFIG_MSM_OCMEM
-#  include <mach/ocmem.h>
-#endif
-
 #include "a3xx_gpu.h"
 
 #define A3XX_INT0_MASK \
@@ -195,9 +191,9 @@ static int a3xx_hw_init(struct msm_gpu *gpu)
 		gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x00000000);
 
 	/* Set the OCMEM base address for A330, etc */
-	if (a3xx_gpu->ocmem_hdl) {
+	if (a3xx_gpu->ocmem.hdl) {
 		gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
-			(unsigned int)(a3xx_gpu->ocmem_base >> 14));
+			(unsigned int)(a3xx_gpu->ocmem.base >> 14));
 	}
 
 	/* Turn on performance counters: */
@@ -318,10 +314,7 @@ static void a3xx_destroy(struct msm_gpu *gpu)
 
 	adreno_gpu_cleanup(adreno_gpu);
 
-#ifdef CONFIG_MSM_OCMEM
-	if (a3xx_gpu->ocmem_base)
-		ocmem_free(OCMEM_GRAPHICS, a3xx_gpu->ocmem_hdl);
-#endif
+	adreno_gpu_ocmem_cleanup(&a3xx_gpu->ocmem);
 
 	kfree(a3xx_gpu);
 }
@@ -494,17 +487,10 @@ struct msm_gpu *a3xx_gpu_init(struct drm_device *dev)
 
 	/* if needed, allocate gmem: */
 	if (adreno_is_a330(adreno_gpu)) {
-#ifdef CONFIG_MSM_OCMEM
-		/* TODO this is different/missing upstream: */
-		struct ocmem_buf *ocmem_hdl =
-				ocmem_allocate(OCMEM_GRAPHICS, adreno_gpu->gmem);
-
-		a3xx_gpu->ocmem_hdl = ocmem_hdl;
-		a3xx_gpu->ocmem_base = ocmem_hdl->addr;
-		adreno_gpu->gmem = ocmem_hdl->len;
-		DBG("using %dK of OCMEM at 0x%08x", adreno_gpu->gmem / 1024,
-				a3xx_gpu->ocmem_base);
-#endif
+		ret = adreno_gpu_ocmem_init(&adreno_gpu->base.pdev->dev,
+					    adreno_gpu, &a3xx_gpu->ocmem);
+		if (ret)
+			goto fail;
 	}
 
 	if (!gpu->aspace) {
diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.h b/drivers/gpu/drm/msm/adreno/a3xx_gpu.h
index 5dc33e5ea53b..c555fb13e0d7 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.h
@@ -19,8 +19,7 @@ struct a3xx_gpu {
 	struct adreno_gpu base;
 
 	/* if OCMEM is used for GMEM: */
-	uint32_t ocmem_base;
-	void *ocmem_hdl;
+	struct adreno_ocmem ocmem;
 };
 #define to_a3xx_gpu(x) container_of(x, struct a3xx_gpu, base)
 
diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
index ab2b752566d8..b01388a9e89e 100644
--- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
@@ -2,9 +2,6 @@
 /* Copyright (c) 2014 The Linux Foundation. All rights reserved.
  */
 #include "a4xx_gpu.h"
-#ifdef CONFIG_MSM_OCMEM
-#  include <soc/qcom/ocmem.h>
-#endif
 
 #define A4XX_INT0_MASK \
 	(A4XX_INT0_RBBM_AHB_ERROR |        \
@@ -188,7 +185,7 @@ static int a4xx_hw_init(struct msm_gpu *gpu)
 			(1 << 30) | 0xFFFF);
 
 	gpu_write(gpu, REG_A4XX_RB_GMEM_BASE_ADDR,
-			(unsigned int)(a4xx_gpu->ocmem_base >> 14));
+			(unsigned int)(a4xx_gpu->ocmem.base >> 14));
 
 	/* Turn on performance counters: */
 	gpu_write(gpu, REG_A4XX_RBBM_PERFCTR_CTL, 0x01);
@@ -318,10 +315,7 @@ static void a4xx_destroy(struct msm_gpu *gpu)
 
 	adreno_gpu_cleanup(adreno_gpu);
 
-#ifdef CONFIG_MSM_OCMEM
-	if (a4xx_gpu->ocmem_base)
-		ocmem_free(OCMEM_GRAPHICS, a4xx_gpu->ocmem_hdl);
-#endif
+	adreno_gpu_ocmem_cleanup(&a4xx_gpu->ocmem);
 
 	kfree(a4xx_gpu);
 }
@@ -578,17 +572,10 @@ struct msm_gpu *a4xx_gpu_init(struct drm_device *dev)
 
 	/* if needed, allocate gmem: */
 	if (adreno_is_a4xx(adreno_gpu)) {
-#ifdef CONFIG_MSM_OCMEM
-		/* TODO this is different/missing upstream: */
-		struct ocmem_buf *ocmem_hdl =
-				ocmem_allocate(OCMEM_GRAPHICS, adreno_gpu->gmem);
-
-		a4xx_gpu->ocmem_hdl = ocmem_hdl;
-		a4xx_gpu->ocmem_base = ocmem_hdl->addr;
-		adreno_gpu->gmem = ocmem_hdl->len;
-		DBG("using %dK of OCMEM at 0x%08x", adreno_gpu->gmem / 1024,
-				a4xx_gpu->ocmem_base);
-#endif
+		ret = adreno_gpu_ocmem_init(dev->dev, adreno_gpu,
+					    &a4xx_gpu->ocmem);
+		if (ret)
+			goto fail;
 	}
 
 	if (!gpu->aspace) {
diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.h b/drivers/gpu/drm/msm/adreno/a4xx_gpu.h
index d506311ee240..a01448cba2ea 100644
--- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.h
@@ -16,8 +16,7 @@ struct a4xx_gpu {
 	struct adreno_gpu base;
 
 	/* if OCMEM is used for GMEM: */
-	uint32_t ocmem_base;
-	void *ocmem_hdl;
+	struct adreno_ocmem ocmem;
 };
 #define to_a4xx_gpu(x) container_of(x, struct a4xx_gpu, base)
 
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 048c8be426f3..0783e4b5486a 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -14,6 +14,7 @@
 #include <linux/pm_opp.h>
 #include <linux/slab.h>
 #include <linux/soc/qcom/mdt_loader.h>
+#include <soc/qcom/ocmem.h>
 #include "adreno_gpu.h"
 #include "msm_gem.h"
 #include "msm_mmu.h"
@@ -893,6 +894,45 @@ static int adreno_get_pwrlevels(struct device *dev,
 	return 0;
 }
 
+int adreno_gpu_ocmem_init(struct device *dev, struct adreno_gpu *adreno_gpu,
+			  struct adreno_ocmem *adreno_ocmem)
+{
+	struct ocmem_buf *ocmem_hdl;
+	struct ocmem *ocmem;
+
+	ocmem = of_get_ocmem(dev);
+	if (IS_ERR(ocmem)) {
+		if (PTR_ERR(ocmem) == -ENODEV) {
+			/*
+			 * Return success since either the ocmem property was
+			 * not specified in device tree, or ocmem support is
+			 * not compiled into the kernel.
+			 */
+			return 0;
+		}
+
+		return PTR_ERR(ocmem);
+	}
+
+	ocmem_hdl = ocmem_allocate(ocmem, OCMEM_GRAPHICS, adreno_gpu->gmem);
+	if (IS_ERR(ocmem_hdl))
+		return PTR_ERR(ocmem_hdl);
+
+	adreno_ocmem->ocmem = ocmem;
+	adreno_ocmem->base = ocmem_hdl->addr;
+	adreno_ocmem->hdl = ocmem_hdl;
+	adreno_gpu->gmem = ocmem_hdl->len;
+
+	return 0;
+}
+
+void adreno_gpu_ocmem_cleanup(struct adreno_ocmem *adreno_ocmem)
+{
+	if (adreno_ocmem && adreno_ocmem->base)
+		ocmem_free(adreno_ocmem->ocmem, OCMEM_GRAPHICS,
+			   adreno_ocmem->hdl);
+}
+
 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 		struct adreno_gpu *adreno_gpu,
 		const struct adreno_gpu_funcs *funcs, int nr_rings)
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index c7441fb8313e..d225a8a92e58 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -126,6 +126,12 @@ struct adreno_gpu {
 };
 #define to_adreno_gpu(x) container_of(x, struct adreno_gpu, base)
 
+struct adreno_ocmem {
+	struct ocmem *ocmem;
+	unsigned long base;
+	void *hdl;
+};
+
 /* platform config data (ie. from DT, or pdata) */
 struct adreno_platform_config {
 	struct adreno_rev rev;
@@ -236,6 +242,10 @@ void adreno_dump(struct msm_gpu *gpu);
 void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords);
 struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu);
 
+int adreno_gpu_ocmem_init(struct device *dev, struct adreno_gpu *adreno_gpu,
+			  struct adreno_ocmem *ocmem);
+void adreno_gpu_ocmem_cleanup(struct adreno_ocmem *ocmem);
+
 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 		struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs,
 		int nr_rings);
-- 
2.21.0

[PATCH v6 7/7] ARM: qcom_defconfig: add ocmem support

[Brian Masney]

Add ocmem driver that's needed for some a3xx and a4xx based systems.

Signed-off-by: Brian Masney <masneyb@onstation.org>
---
Changes since v5:
- None

This patch was introduced in v5.

 arch/arm/configs/qcom_defconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm/configs/qcom_defconfig b/arch/arm/configs/qcom_defconfig
index 34433bf5885d..595e1910ba78 100644
--- a/arch/arm/configs/qcom_defconfig
+++ b/arch/arm/configs/qcom_defconfig
@@ -225,6 +225,7 @@ CONFIG_QCOM_WCNSS_PIL=y
 CONFIG_RPMSG_CHAR=y
 CONFIG_RPMSG_QCOM_SMD=y
 CONFIG_QCOM_GSBI=y
+CONFIG_QCOM_OCMEM=y
 CONFIG_QCOM_PM=y
 CONFIG_QCOM_SMEM=y
 CONFIG_QCOM_SMD_RPM=y
-- 
2.21.0

Re: [PATCH v6 4/7] firmware: qcom: scm: add support to restore secure config to qcm_scm-32

[Bjorn Andersson]

On Thu 22 Aug 07:37 PDT 2019, Brian Masney wrote:

> From: Rob Clark <robdclark@gmail.com>
> 
> Add support to restore the secure configuration for qcm_scm-32.c. This
> is needed by the On Chip MEMory (OCMEM) that is present on some
> Snapdragon devices.
> 
> Signed-off-by: Rob Clark <robdclark@gmail.com>
> [masneyb@onstation.org: ported to latest kernel; set ctx_bank_num to
>  spare parameter.]
> Signed-off-by: Brian Masney <masneyb@onstation.org>

Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>

> ---
> Changes since v5:
> - None
> 
> Changes since v4:
> - None
> 
> Changes since v3:
> - None
> 
> Changes since v2:
> - None
> 
> Changes since v1:
> - Use existing __qcom_scm_restore_sec_cfg() function stub in
>   qcom_scm-32.c that was unimplemented
> - Set the cfg.ctx_bank_num to the spare function parameter. It was
>   previously set to the device_id.
> 
>  drivers/firmware/qcom_scm-32.c | 17 ++++++++++++++++-
>  drivers/firmware/qcom_scm.c    | 13 +++++++++++++
>  include/linux/qcom_scm.h       | 11 +++++++++++
>  3 files changed, 40 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/firmware/qcom_scm-32.c b/drivers/firmware/qcom_scm-32.c
> index 4c2514e5e249..5d90b7f5ab5a 100644
> --- a/drivers/firmware/qcom_scm-32.c
> +++ b/drivers/firmware/qcom_scm-32.c
> @@ -617,7 +617,22 @@ int __qcom_scm_assign_mem(struct device *dev, phys_addr_t mem_region,
>  int __qcom_scm_restore_sec_cfg(struct device *dev, u32 device_id,
>  			       u32 spare)
>  {
> -	return -ENODEV;
> +	struct msm_scm_sec_cfg {
> +		__le32 id;
> +		__le32 ctx_bank_num;
> +	} cfg;
> +	int ret, scm_ret = 0;
> +
> +	cfg.id = cpu_to_le32(device_id);
> +	cfg.ctx_bank_num = cpu_to_le32(spare);
> +
> +	ret = qcom_scm_call(dev, QCOM_SCM_SVC_MP, QCOM_SCM_RESTORE_SEC_CFG,
> +			    &cfg, sizeof(cfg), &scm_ret, sizeof(scm_ret));
> +
> +	if (ret || scm_ret)
> +		return ret ? ret : -EINVAL;
> +
> +	return 0;
>  }
>  
>  int __qcom_scm_iommu_secure_ptbl_size(struct device *dev, u32 spare,
> diff --git a/drivers/firmware/qcom_scm.c b/drivers/firmware/qcom_scm.c
> index 7e285ff3961d..27c1d98a34e6 100644
> --- a/drivers/firmware/qcom_scm.c
> +++ b/drivers/firmware/qcom_scm.c
> @@ -367,6 +367,19 @@ static const struct reset_control_ops qcom_scm_pas_reset_ops = {
>  	.deassert = qcom_scm_pas_reset_deassert,
>  };
>  
> +/**
> + * qcom_scm_restore_sec_cfg_available() - Check if secure environment
> + * supports restore security config interface.
> + *
> + * Return true if restore-cfg interface is supported, false if not.
> + */
> +bool qcom_scm_restore_sec_cfg_available(void)
> +{
> +	return __qcom_scm_is_call_available(__scm->dev, QCOM_SCM_SVC_MP,
> +					    QCOM_SCM_RESTORE_SEC_CFG);
> +}
> +EXPORT_SYMBOL(qcom_scm_restore_sec_cfg_available);
> +
>  int qcom_scm_restore_sec_cfg(u32 device_id, u32 spare)
>  {
>  	return __qcom_scm_restore_sec_cfg(__scm->dev, device_id, spare);
> diff --git a/include/linux/qcom_scm.h b/include/linux/qcom_scm.h
> index b49b734d662c..04382e1798e4 100644
> --- a/include/linux/qcom_scm.h
> +++ b/include/linux/qcom_scm.h
> @@ -34,6 +34,16 @@ enum qcom_scm_ocmem_client {
>  	QCOM_SCM_OCMEM_DEBUG_ID,
>  };
>  
> +enum qcom_scm_sec_dev_id {
> +	QCOM_SCM_MDSS_DEV_ID    = 1,
> +	QCOM_SCM_OCMEM_DEV_ID   = 5,
> +	QCOM_SCM_PCIE0_DEV_ID   = 11,
> +	QCOM_SCM_PCIE1_DEV_ID   = 12,
> +	QCOM_SCM_GFX_DEV_ID     = 18,
> +	QCOM_SCM_UFS_DEV_ID     = 19,
> +	QCOM_SCM_ICE_DEV_ID     = 20,
> +};
> +
>  #define QCOM_SCM_VMID_HLOS       0x3
>  #define QCOM_SCM_VMID_MSS_MSA    0xF
>  #define QCOM_SCM_VMID_WLAN       0x18
> @@ -70,6 +80,7 @@ extern int qcom_scm_assign_mem(phys_addr_t mem_addr, size_t mem_sz,
>  extern void qcom_scm_cpu_power_down(u32 flags);
>  extern u32 qcom_scm_get_version(void);
>  extern int qcom_scm_set_remote_state(u32 state, u32 id);
> +extern bool qcom_scm_restore_sec_cfg_available(void);
>  extern int qcom_scm_restore_sec_cfg(u32 device_id, u32 spare);
>  extern int qcom_scm_iommu_secure_ptbl_size(u32 spare, size_t *size);
>  extern int qcom_scm_iommu_secure_ptbl_init(u64 addr, u32 size, u32 spare);
> -- 
> 2.21.0
> 

Re: [PATCH v6 2/7] dt-bindings: display: msm: gmu: add optional ocmem property

[Rob Herring]

On Thu, Aug 22, 2019 at 9:37 AM Brian Masney <masneyb@onstation.org> wrote:
>
> Some A3xx and A4xx Adreno GPUs do not have GMEM inside the GPU core and
> must use the On Chip MEMory (OCMEM) in order to be functional. Add the
> optional ocmem property to the Adreno Graphics Management Unit bindings.
>
> Signed-off-by: Brian Masney <masneyb@onstation.org>
> ---
> Changes since v5:
> - rename ocmem property to sram to match what TI currently has.
>
> Changes since v4:
> - None
>
> Changes since v3:
> - correct link to qcom,ocmem.yaml
>
> Changes since v2:
> - Add a3xx example with OCMEM
>
> Changes since v1:
> - None
>
>  .../devicetree/bindings/display/msm/gmu.txt   | 50 +++++++++++++++++++
>  1 file changed, 50 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/display/msm/gmu.txt b/Documentation/devicetree/bindings/display/msm/gmu.txt
> index 90af5b0a56a9..2305a2aede5a 100644
> --- a/Documentation/devicetree/bindings/display/msm/gmu.txt
> +++ b/Documentation/devicetree/bindings/display/msm/gmu.txt
> @@ -31,6 +31,10 @@ Required properties:
>  - iommus: phandle to the adreno iommu
>  - operating-points-v2: phandle to the OPP operating points
>
> +Optional properties:
> +- sram: phandle to the On Chip Memory (OCMEM) that's present on some Snapdragon
> +        SoCs. See Documentation/devicetree/bindings/sram/qcom,ocmem.yaml.
> +
>  Example:
>
>  / {
> @@ -63,3 +67,49 @@ Example:
>                 operating-points-v2 = <&gmu_opp_table>;
>         };
>  };
> +
> +a3xx example with OCMEM support:
> +
> +/ {
> +       ...
> +
> +       gpu: adreno@fdb00000 {
> +               compatible = "qcom,adreno-330.2",
> +                            "qcom,adreno";
> +               reg = <0xfdb00000 0x10000>;
> +               reg-names = "kgsl_3d0_reg_memory";
> +               interrupts = <GIC_SPI 33 IRQ_TYPE_LEVEL_HIGH>;
> +               interrupt-names = "kgsl_3d0_irq";
> +               clock-names = "core",
> +                             "iface",
> +                             "mem_iface";
> +               clocks = <&mmcc OXILI_GFX3D_CLK>,
> +                        <&mmcc OXILICX_AHB_CLK>,
> +                        <&mmcc OXILICX_AXI_CLK>;
> +               sram = <&ocmem>;

Shouldn't this point to gmu-sram@0? You can always get the parent from
the child which is a bit easier than the other way around.

> +               power-domains = <&mmcc OXILICX_GDSC>;
> +               operating-points-v2 = <&gpu_opp_table>;
> +               iommus = <&gpu_iommu 0>;
> +       };
> +
> +       ocmem: ocmem@fdd00000 {
> +               compatible = "qcom,msm8974-ocmem";
> +
> +               reg = <0xfdd00000 0x2000>,
> +                     <0xfec00000 0x180000>;
> +               reg-names = "ctrl",
> +                            "mem";
> +
> +               clocks = <&rpmcc RPM_SMD_OCMEMGX_CLK>,
> +                        <&mmcc OCMEMCX_OCMEMNOC_CLK>;
> +               clock-names = "core",
> +                             "iface";
> +
> +               #address-cells = <1>;
> +               #size-cells = <1>;
> +
> +               gmu-sram@0 {
> +                       reg = <0x0 0x100000>;
> +               };
> +       };
> +};
> --
> 2.21.0
>