* [PATCH v3 0/4] remoteproc/mediatek: support L1TCM for MT8192 SCP
@ 2021-01-27 8:31 ` Tzung-Bi Shih
0 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: devicetree, mathieu.poirier, linux-remoteproc, tzungbi,
linux-mediatek, matthias.bgg
The series applies after [1].
The series supports L1TCM which is a high performance memory region in
MT8192 SCP.
The 1st patch replaces platform_get_resource_byname() and
devm_ioremap_resource() pairs per [2] suggested.
The 2nd patch enables MPU for all memory regions. The patch was
independent but merged to this series per [3] suggested.
The 3rd patch adds a new reg-name "l1tcm" for L1TCM.
The 4th patch supports L1TCM in the firmware loader. Note that MT8192
SCP is still under development. The patch breaks early MT8192 SCP
firmware which should only break our own development environment.
Changes from v2[4]:
- Rebases the series to [1].
Changes from v1[5]:
- Adds 2 patches at beginning of the series.
[1]: https://patchwork.kernel.org/project/linux-remoteproc/list/?series=422635
[2]: https://patchwork.kernel.org/project/linux-remoteproc/patch/20201214050521.845396-3-tzungbi@google.com/#23879113
[3]: https://patchwork.kernel.org/project/linux-remoteproc/patch/20210107023020.3224002-1-tzungbi@google.com/#23879623
[4]: https://patchwork.kernel.org/project/linux-remoteproc/list/?series=411065
[5]: https://patchwork.kernel.org/project/linux-remoteproc/list/?series=401287
Tzung-Bi Shih (4):
remoteproc/mediatek: use devm_platform_ioremap_resource_byname
remoteproc/mediatek: enable MPU for all memory regions in MT8192 SCP
dt-bindings: remoteproc: mediatek: add L1TCM memory region
remoteproc/mediatek: support L1TCM
.../bindings/remoteproc/mtk,scp.txt | 8 +--
drivers/remoteproc/mtk_common.h | 6 ++
drivers/remoteproc/mtk_scp.c | 62 +++++++++++++++++--
3 files changed, 68 insertions(+), 8 deletions(-)
--
2.30.0.280.ga3ce27912f-goog
_______________________________________________
Linux-mediatek mailing list
Linux-mediatek@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-mediatek
^ permalink raw reply [flat|nested] 11+ messages in thread
* [PATCH v3 1/4] remoteproc/mediatek: use devm_platform_ioremap_resource_byname
2021-01-27 8:31 ` Tzung-Bi Shih
@ 2021-01-27 8:31 ` Tzung-Bi Shih
-1 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: linux-remoteproc, matthias.bgg, linux-mediatek, mathieu.poirier,
devicetree, tzungbi
Replaces platform_get_resource_byname() and devm_ioremap_resource()
pairs to devm_platform_ioremap_resource_byname().
Note that, not every pairs are applicable to replace. Especially when
it needs to access the resource struct from
platform_get_resource_byname().
For example:
scp->sram_size = resource_size(res);
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
drivers/remoteproc/mtk_scp.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index eba825b46696..d83e1164f02f 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -721,8 +721,7 @@ static int scp_probe(struct platform_device *pdev)
for (i = 0; i < SCP_IPI_MAX; i++)
mutex_init(&scp->ipi_desc[i].lock);
- res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
- scp->reg_base = devm_ioremap_resource(dev, res);
+ scp->reg_base = devm_platform_ioremap_resource_byname(pdev, "cfg");
if (IS_ERR((__force void *)scp->reg_base)) {
dev_err(dev, "Failed to parse and map cfg memory\n");
ret = PTR_ERR((__force void *)scp->reg_base);
--
2.30.0.280.ga3ce27912f-goog
^ permalink raw reply related [flat|nested] 11+ messages in thread* [PATCH v3 1/4] remoteproc/mediatek: use devm_platform_ioremap_resource_byname
@ 2021-01-27 8:31 ` Tzung-Bi Shih
0 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: devicetree, mathieu.poirier, linux-remoteproc, tzungbi,
linux-mediatek, matthias.bgg
Replaces platform_get_resource_byname() and devm_ioremap_resource()
pairs to devm_platform_ioremap_resource_byname().
Note that, not every pairs are applicable to replace. Especially when
it needs to access the resource struct from
platform_get_resource_byname().
For example:
scp->sram_size = resource_size(res);
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
drivers/remoteproc/mtk_scp.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index eba825b46696..d83e1164f02f 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -721,8 +721,7 @@ static int scp_probe(struct platform_device *pdev)
for (i = 0; i < SCP_IPI_MAX; i++)
mutex_init(&scp->ipi_desc[i].lock);
- res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
- scp->reg_base = devm_ioremap_resource(dev, res);
+ scp->reg_base = devm_platform_ioremap_resource_byname(pdev, "cfg");
if (IS_ERR((__force void *)scp->reg_base)) {
dev_err(dev, "Failed to parse and map cfg memory\n");
ret = PTR_ERR((__force void *)scp->reg_base);
--
2.30.0.280.ga3ce27912f-goog
_______________________________________________
Linux-mediatek mailing list
Linux-mediatek@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-mediatek
^ permalink raw reply related [flat|nested] 11+ messages in thread
* [PATCH v3 2/4] remoteproc/mediatek: enable MPU for all memory regions in MT8192 SCP
2021-01-27 8:31 ` Tzung-Bi Shih
@ 2021-01-27 8:31 ` Tzung-Bi Shih
-1 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: linux-remoteproc, matthias.bgg, linux-mediatek, mathieu.poirier,
devicetree, tzungbi
The register MT8192_CORE0_MEM_ATT_PREDEF contains attributes for each
memory region. It defines whether a memory region can be managed by MPU
or not.
In the past, due to the default settings in the register, MT8192 SCP
works luckily. After enabling L1TCM, SCP starts to access memory region
that is not included in the default settings. As a result, SCP hangs.
Enables MPU for all memory regions in MT8192 SCP.
Note that the register is read only once when SCP resets. Thus, it must
be set from kernel side.
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
drivers/remoteproc/mtk_common.h | 1 +
drivers/remoteproc/mtk_scp.c | 3 +++
2 files changed, 4 insertions(+)
diff --git a/drivers/remoteproc/mtk_common.h b/drivers/remoteproc/mtk_common.h
index bcab38511bf3..204691138677 100644
--- a/drivers/remoteproc/mtk_common.h
+++ b/drivers/remoteproc/mtk_common.h
@@ -47,6 +47,7 @@
#define MT8192_CORE0_SW_RSTN_CLR 0x10000
#define MT8192_CORE0_SW_RSTN_SET 0x10004
+#define MT8192_CORE0_MEM_ATT_PREDEF 0x10008
#define MT8192_CORE0_WDT_IRQ 0x10030
#define MT8192_CORE0_WDT_CFG 0x10034
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index d83e1164f02f..05b157689121 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -371,6 +371,9 @@ static int mt8192_scp_before_load(struct mtk_scp *scp)
mt8192_power_on_sram(scp->reg_base + MT8192_L1TCM_SRAM_PDN);
mt8192_power_on_sram(scp->reg_base + MT8192_CPU0_SRAM_PD);
+ /* enable MPU for all memory regions */
+ writel(0xff, scp->reg_base + MT8192_CORE0_MEM_ATT_PREDEF);
+
return 0;
}
--
2.30.0.280.ga3ce27912f-goog
^ permalink raw reply related [flat|nested] 11+ messages in thread
* [PATCH v3 2/4] remoteproc/mediatek: enable MPU for all memory regions in MT8192 SCP
@ 2021-01-27 8:31 ` Tzung-Bi Shih
0 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: devicetree, mathieu.poirier, linux-remoteproc, tzungbi,
linux-mediatek, matthias.bgg
The register MT8192_CORE0_MEM_ATT_PREDEF contains attributes for each
memory region. It defines whether a memory region can be managed by MPU
or not.
In the past, due to the default settings in the register, MT8192 SCP
works luckily. After enabling L1TCM, SCP starts to access memory region
that is not included in the default settings. As a result, SCP hangs.
Enables MPU for all memory regions in MT8192 SCP.
Note that the register is read only once when SCP resets. Thus, it must
be set from kernel side.
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
drivers/remoteproc/mtk_common.h | 1 +
drivers/remoteproc/mtk_scp.c | 3 +++
2 files changed, 4 insertions(+)
diff --git a/drivers/remoteproc/mtk_common.h b/drivers/remoteproc/mtk_common.h
index bcab38511bf3..204691138677 100644
--- a/drivers/remoteproc/mtk_common.h
+++ b/drivers/remoteproc/mtk_common.h
@@ -47,6 +47,7 @@
#define MT8192_CORE0_SW_RSTN_CLR 0x10000
#define MT8192_CORE0_SW_RSTN_SET 0x10004
+#define MT8192_CORE0_MEM_ATT_PREDEF 0x10008
#define MT8192_CORE0_WDT_IRQ 0x10030
#define MT8192_CORE0_WDT_CFG 0x10034
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index d83e1164f02f..05b157689121 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -371,6 +371,9 @@ static int mt8192_scp_before_load(struct mtk_scp *scp)
mt8192_power_on_sram(scp->reg_base + MT8192_L1TCM_SRAM_PDN);
mt8192_power_on_sram(scp->reg_base + MT8192_CPU0_SRAM_PD);
+ /* enable MPU for all memory regions */
+ writel(0xff, scp->reg_base + MT8192_CORE0_MEM_ATT_PREDEF);
+
return 0;
}
--
2.30.0.280.ga3ce27912f-goog
_______________________________________________
Linux-mediatek mailing list
Linux-mediatek@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-mediatek
^ permalink raw reply related [flat|nested] 11+ messages in thread
* [PATCH v3 3/4] dt-bindings: remoteproc: mediatek: add L1TCM memory region
2021-01-27 8:31 ` Tzung-Bi Shih
@ 2021-01-27 8:31 ` Tzung-Bi Shih
-1 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: linux-remoteproc, matthias.bgg, linux-mediatek, mathieu.poirier,
devicetree, tzungbi, Rob Herring
Adds L1TCM memory region. The reg-name is "l1tcm".
Acked-by: Rob Herring <robh@kernel.org>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
Documentation/devicetree/bindings/remoteproc/mtk,scp.txt | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt b/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt
index 3ba668bab14b..3f5f78764b60 100644
--- a/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt
+++ b/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt
@@ -6,10 +6,10 @@ Mediatek SoCs.
Required properties:
- compatible Should be "mediatek,mt8183-scp"
-- reg Should contain the address ranges for the two memory
- regions, SRAM and CFG.
-- reg-names Contains the corresponding names for the two memory
- regions. These should be named "sram" & "cfg".
+- reg Should contain the address ranges for memory regions:
+ SRAM, CFG, and L1TCM.
+- reg-names Contains the corresponding names for the memory regions:
+ "sram", "cfg", and "l1tcm".
- clocks Clock for co-processor (See: ../clock/clock-bindings.txt)
- clock-names Contains the corresponding name for the clock. This
should be named "main".
--
2.30.0.280.ga3ce27912f-goog
^ permalink raw reply related [flat|nested] 11+ messages in thread
* [PATCH v3 3/4] dt-bindings: remoteproc: mediatek: add L1TCM memory region
@ 2021-01-27 8:31 ` Tzung-Bi Shih
0 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: devicetree, mathieu.poirier, Rob Herring, linux-remoteproc,
tzungbi, linux-mediatek, matthias.bgg
Adds L1TCM memory region. The reg-name is "l1tcm".
Acked-by: Rob Herring <robh@kernel.org>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
Documentation/devicetree/bindings/remoteproc/mtk,scp.txt | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt b/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt
index 3ba668bab14b..3f5f78764b60 100644
--- a/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt
+++ b/Documentation/devicetree/bindings/remoteproc/mtk,scp.txt
@@ -6,10 +6,10 @@ Mediatek SoCs.
Required properties:
- compatible Should be "mediatek,mt8183-scp"
-- reg Should contain the address ranges for the two memory
- regions, SRAM and CFG.
-- reg-names Contains the corresponding names for the two memory
- regions. These should be named "sram" & "cfg".
+- reg Should contain the address ranges for memory regions:
+ SRAM, CFG, and L1TCM.
+- reg-names Contains the corresponding names for the memory regions:
+ "sram", "cfg", and "l1tcm".
- clocks Clock for co-processor (See: ../clock/clock-bindings.txt)
- clock-names Contains the corresponding name for the clock. This
should be named "main".
--
2.30.0.280.ga3ce27912f-goog
_______________________________________________
Linux-mediatek mailing list
Linux-mediatek@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-mediatek
^ permalink raw reply related [flat|nested] 11+ messages in thread
* [PATCH v3 4/4] remoteproc/mediatek: support L1TCM
2021-01-27 8:31 ` Tzung-Bi Shih
@ 2021-01-27 8:31 ` Tzung-Bi Shih
-1 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: linux-remoteproc, matthias.bgg, linux-mediatek, mathieu.poirier,
devicetree, tzungbi
L1TCM is a high performance memory region in MT8192 SCP.
Reads L1TCM memory region from DTS to determine if the machine supports.
Loads L1TCM memory region to SCP sys if the firmware provides.
Starts from MT8192 SCP, the firmware contains physical addresses for
each memory region, for instance:
Program Headers:
Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align
LOAD 0xXXXXXX 0xXXXXXXXX 0x10500000 0xXXXXX 0xXXXXX XXX 0xXXXX
LOAD 0xXXXXXX 0xXXXXXXXX 0x10700000 0xXXXXX 0xXXXXX XXX 0xXXXX
LOAD 0xXXXXXX 0xXXXXXXXX 0x50000000 0xXXXXX 0xXXXXX XXX 0xXXXX
Kernel driver can use the "PhysAddr" (i.e. da in the da_to_va callbacks)
to know the ELF segment belongs to which region.
To backward compatible to MT8183 SCP, separates the da_to_va callbacks
for new and legacy version.
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
drivers/remoteproc/mtk_common.h | 5 +++
drivers/remoteproc/mtk_scp.c | 56 +++++++++++++++++++++++++++++++--
2 files changed, 59 insertions(+), 2 deletions(-)
diff --git a/drivers/remoteproc/mtk_common.h b/drivers/remoteproc/mtk_common.h
index 204691138677..61901f5efa05 100644
--- a/drivers/remoteproc/mtk_common.h
+++ b/drivers/remoteproc/mtk_common.h
@@ -77,6 +77,7 @@ struct mtk_scp_of_data {
void (*scp_reset_assert)(struct mtk_scp *scp);
void (*scp_reset_deassert)(struct mtk_scp *scp);
void (*scp_stop)(struct mtk_scp *scp);
+ void *(*scp_da_to_va)(struct mtk_scp *scp, u64 da, size_t len);
u32 host_to_scp_reg;
u32 host_to_scp_int_bit;
@@ -91,6 +92,10 @@ struct mtk_scp {
void __iomem *reg_base;
void __iomem *sram_base;
size_t sram_size;
+ phys_addr_t sram_phys;
+ void __iomem *l1tcm_base;
+ size_t l1tcm_size;
+ phys_addr_t l1tcm_phys;
const struct mtk_scp_of_data *data;
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index 05b157689121..ce727598c41c 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -463,9 +463,8 @@ static int scp_start(struct rproc *rproc)
return ret;
}
-static void *scp_da_to_va(struct rproc *rproc, u64 da, size_t len)
+static void *mt8183_scp_da_to_va(struct mtk_scp *scp, u64 da, size_t len)
{
- struct mtk_scp *scp = (struct mtk_scp *)rproc->priv;
int offset;
if (da < scp->sram_size) {
@@ -481,6 +480,42 @@ static void *scp_da_to_va(struct rproc *rproc, u64 da, size_t len)
return NULL;
}
+static void *mt8192_scp_da_to_va(struct mtk_scp *scp, u64 da, size_t len)
+{
+ int offset;
+
+ if (da >= scp->sram_phys &&
+ (da + len) <= scp->sram_phys + scp->sram_size) {
+ offset = da - scp->sram_phys;
+ return (void __force *)scp->sram_base + offset;
+ }
+
+ /* optional memory region */
+ if (scp->l1tcm_size &&
+ da >= scp->l1tcm_phys &&
+ (da + len) <= scp->l1tcm_phys + scp->l1tcm_size) {
+ offset = da - scp->l1tcm_phys;
+ return (void __force *)scp->l1tcm_base + offset;
+ }
+
+ /* optional memory region */
+ if (scp->dram_size &&
+ da >= scp->dma_addr &&
+ (da + len) <= scp->dma_addr + scp->dram_size) {
+ offset = da - scp->dma_addr;
+ return scp->cpu_addr + offset;
+ }
+
+ return NULL;
+}
+
+static void *scp_da_to_va(struct rproc *rproc, u64 da, size_t len)
+{
+ struct mtk_scp *scp = (struct mtk_scp *)rproc->priv;
+
+ return scp->data->scp_da_to_va(scp, da, len);
+}
+
static void mt8183_scp_stop(struct mtk_scp *scp)
{
/* Disable SCP watchdog */
@@ -719,6 +754,21 @@ static int scp_probe(struct platform_device *pdev)
goto free_rproc;
}
scp->sram_size = resource_size(res);
+ scp->sram_phys = res->start;
+
+ /* l1tcm is an optional memory region */
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "l1tcm");
+ scp->l1tcm_base = devm_ioremap_resource(dev, res);
+ if (IS_ERR((__force void *)scp->l1tcm_base)) {
+ ret = PTR_ERR((__force void *)scp->l1tcm_base);
+ if (ret != -EINVAL) {
+ dev_err(dev, "Failed to map l1tcm memory\n");
+ goto free_rproc;
+ }
+ } else {
+ scp->l1tcm_size = resource_size(res);
+ scp->l1tcm_phys = res->start;
+ }
mutex_init(&scp->send_lock);
for (i = 0; i < SCP_IPI_MAX; i++)
@@ -807,6 +857,7 @@ static const struct mtk_scp_of_data mt8183_of_data = {
.scp_reset_assert = mt8183_scp_reset_assert,
.scp_reset_deassert = mt8183_scp_reset_deassert,
.scp_stop = mt8183_scp_stop,
+ .scp_da_to_va = mt8183_scp_da_to_va,
.host_to_scp_reg = MT8183_HOST_TO_SCP,
.host_to_scp_int_bit = MT8183_HOST_IPC_INT_BIT,
.ipi_buf_offset = 0x7bdb0,
@@ -818,6 +869,7 @@ static const struct mtk_scp_of_data mt8192_of_data = {
.scp_reset_assert = mt8192_scp_reset_assert,
.scp_reset_deassert = mt8192_scp_reset_deassert,
.scp_stop = mt8192_scp_stop,
+ .scp_da_to_va = mt8192_scp_da_to_va,
.host_to_scp_reg = MT8192_GIPC_IN_SET,
.host_to_scp_int_bit = MT8192_HOST_IPC_INT_BIT,
};
--
2.30.0.280.ga3ce27912f-goog
^ permalink raw reply related [flat|nested] 11+ messages in thread* [PATCH v3 4/4] remoteproc/mediatek: support L1TCM
@ 2021-01-27 8:31 ` Tzung-Bi Shih
0 siblings, 0 replies; 11+ messages in thread
From: Tzung-Bi Shih @ 2021-01-27 8:31 UTC (permalink / raw)
To: ohad, bjorn.andersson, robh+dt
Cc: devicetree, mathieu.poirier, linux-remoteproc, tzungbi,
linux-mediatek, matthias.bgg
L1TCM is a high performance memory region in MT8192 SCP.
Reads L1TCM memory region from DTS to determine if the machine supports.
Loads L1TCM memory region to SCP sys if the firmware provides.
Starts from MT8192 SCP, the firmware contains physical addresses for
each memory region, for instance:
Program Headers:
Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align
LOAD 0xXXXXXX 0xXXXXXXXX 0x10500000 0xXXXXX 0xXXXXX XXX 0xXXXX
LOAD 0xXXXXXX 0xXXXXXXXX 0x10700000 0xXXXXX 0xXXXXX XXX 0xXXXX
LOAD 0xXXXXXX 0xXXXXXXXX 0x50000000 0xXXXXX 0xXXXXX XXX 0xXXXX
Kernel driver can use the "PhysAddr" (i.e. da in the da_to_va callbacks)
to know the ELF segment belongs to which region.
To backward compatible to MT8183 SCP, separates the da_to_va callbacks
for new and legacy version.
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Tzung-Bi Shih <tzungbi@google.com>
---
drivers/remoteproc/mtk_common.h | 5 +++
drivers/remoteproc/mtk_scp.c | 56 +++++++++++++++++++++++++++++++--
2 files changed, 59 insertions(+), 2 deletions(-)
diff --git a/drivers/remoteproc/mtk_common.h b/drivers/remoteproc/mtk_common.h
index 204691138677..61901f5efa05 100644
--- a/drivers/remoteproc/mtk_common.h
+++ b/drivers/remoteproc/mtk_common.h
@@ -77,6 +77,7 @@ struct mtk_scp_of_data {
void (*scp_reset_assert)(struct mtk_scp *scp);
void (*scp_reset_deassert)(struct mtk_scp *scp);
void (*scp_stop)(struct mtk_scp *scp);
+ void *(*scp_da_to_va)(struct mtk_scp *scp, u64 da, size_t len);
u32 host_to_scp_reg;
u32 host_to_scp_int_bit;
@@ -91,6 +92,10 @@ struct mtk_scp {
void __iomem *reg_base;
void __iomem *sram_base;
size_t sram_size;
+ phys_addr_t sram_phys;
+ void __iomem *l1tcm_base;
+ size_t l1tcm_size;
+ phys_addr_t l1tcm_phys;
const struct mtk_scp_of_data *data;
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index 05b157689121..ce727598c41c 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -463,9 +463,8 @@ static int scp_start(struct rproc *rproc)
return ret;
}
-static void *scp_da_to_va(struct rproc *rproc, u64 da, size_t len)
+static void *mt8183_scp_da_to_va(struct mtk_scp *scp, u64 da, size_t len)
{
- struct mtk_scp *scp = (struct mtk_scp *)rproc->priv;
int offset;
if (da < scp->sram_size) {
@@ -481,6 +480,42 @@ static void *scp_da_to_va(struct rproc *rproc, u64 da, size_t len)
return NULL;
}
+static void *mt8192_scp_da_to_va(struct mtk_scp *scp, u64 da, size_t len)
+{
+ int offset;
+
+ if (da >= scp->sram_phys &&
+ (da + len) <= scp->sram_phys + scp->sram_size) {
+ offset = da - scp->sram_phys;
+ return (void __force *)scp->sram_base + offset;
+ }
+
+ /* optional memory region */
+ if (scp->l1tcm_size &&
+ da >= scp->l1tcm_phys &&
+ (da + len) <= scp->l1tcm_phys + scp->l1tcm_size) {
+ offset = da - scp->l1tcm_phys;
+ return (void __force *)scp->l1tcm_base + offset;
+ }
+
+ /* optional memory region */
+ if (scp->dram_size &&
+ da >= scp->dma_addr &&
+ (da + len) <= scp->dma_addr + scp->dram_size) {
+ offset = da - scp->dma_addr;
+ return scp->cpu_addr + offset;
+ }
+
+ return NULL;
+}
+
+static void *scp_da_to_va(struct rproc *rproc, u64 da, size_t len)
+{
+ struct mtk_scp *scp = (struct mtk_scp *)rproc->priv;
+
+ return scp->data->scp_da_to_va(scp, da, len);
+}
+
static void mt8183_scp_stop(struct mtk_scp *scp)
{
/* Disable SCP watchdog */
@@ -719,6 +754,21 @@ static int scp_probe(struct platform_device *pdev)
goto free_rproc;
}
scp->sram_size = resource_size(res);
+ scp->sram_phys = res->start;
+
+ /* l1tcm is an optional memory region */
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "l1tcm");
+ scp->l1tcm_base = devm_ioremap_resource(dev, res);
+ if (IS_ERR((__force void *)scp->l1tcm_base)) {
+ ret = PTR_ERR((__force void *)scp->l1tcm_base);
+ if (ret != -EINVAL) {
+ dev_err(dev, "Failed to map l1tcm memory\n");
+ goto free_rproc;
+ }
+ } else {
+ scp->l1tcm_size = resource_size(res);
+ scp->l1tcm_phys = res->start;
+ }
mutex_init(&scp->send_lock);
for (i = 0; i < SCP_IPI_MAX; i++)
@@ -807,6 +857,7 @@ static const struct mtk_scp_of_data mt8183_of_data = {
.scp_reset_assert = mt8183_scp_reset_assert,
.scp_reset_deassert = mt8183_scp_reset_deassert,
.scp_stop = mt8183_scp_stop,
+ .scp_da_to_va = mt8183_scp_da_to_va,
.host_to_scp_reg = MT8183_HOST_TO_SCP,
.host_to_scp_int_bit = MT8183_HOST_IPC_INT_BIT,
.ipi_buf_offset = 0x7bdb0,
@@ -818,6 +869,7 @@ static const struct mtk_scp_of_data mt8192_of_data = {
.scp_reset_assert = mt8192_scp_reset_assert,
.scp_reset_deassert = mt8192_scp_reset_deassert,
.scp_stop = mt8192_scp_stop,
+ .scp_da_to_va = mt8192_scp_da_to_va,
.host_to_scp_reg = MT8192_GIPC_IN_SET,
.host_to_scp_int_bit = MT8192_HOST_IPC_INT_BIT,
};
--
2.30.0.280.ga3ce27912f-goog
_______________________________________________
Linux-mediatek mailing list
Linux-mediatek@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-mediatek
^ permalink raw reply related [flat|nested] 11+ messages in thread
* Re: [PATCH v3 0/4] remoteproc/mediatek: support L1TCM for MT8192 SCP
2021-01-27 8:31 ` Tzung-Bi Shih
` (4 preceding siblings ...)
(?)
@ 2021-02-09 17:40 ` patchwork-bot+linux-remoteproc
-1 siblings, 0 replies; 11+ messages in thread
From: patchwork-bot+linux-remoteproc @ 2021-02-09 17:40 UTC (permalink / raw)
To: Tzung-Bi Shih; +Cc: linux-remoteproc
Hello:
This series was applied to andersson/remoteproc.git (refs/heads/for-next):
On Wed, 27 Jan 2021 16:31:32 +0800 you wrote:
> The series applies after [1].
>
> The series supports L1TCM which is a high performance memory region in
> MT8192 SCP.
>
> The 1st patch replaces platform_get_resource_byname() and
> devm_ioremap_resource() pairs per [2] suggested.
>
> [...]
Here is the summary with links:
- [v3,1/4] remoteproc/mediatek: use devm_platform_ioremap_resource_byname
https://git.kernel.org/andersson/remoteproc/c/2e88e8fcdfcd
- [v3,2/4] remoteproc/mediatek: enable MPU for all memory regions in MT8192 SCP
https://git.kernel.org/andersson/remoteproc/c/ff3ea536023e
- [v3,3/4] dt-bindings: remoteproc: mediatek: add L1TCM memory region
https://git.kernel.org/andersson/remoteproc/c/503c64cc42f1
- [v3,4/4] remoteproc/mediatek: support L1TCM
https://git.kernel.org/andersson/remoteproc/c/ca23ecfdbd44
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply [flat|nested] 11+ messages in thread