Without PM support, gcc warns about two unused functions:
platform/qcom/venus/core.c:146:13: error: 'venus_clks_disable' defined but not used [-Werror=unused-function]
platform/qcom/venus/core.c:126:12: error: 'venus_clks_enable' defined but not used [-Werror=unused-function]
The problem as usual are incorrect #ifdefs, so the easiest fix
is to do away with the #ifdef completely and mark the suspend/resume
handlers as __maybe_unused, which they are.
Fixes: af2c3834c8ca ("[media] media: venus: adding core part and helper functions")
Signed-off-by: Arnd Bergmann <[email protected]>
---
drivers/media/platform/qcom/venus/core.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
index d8cbe8549d97..47f79637938c 100644
--- a/drivers/media/platform/qcom/venus/core.c
+++ b/drivers/media/platform/qcom/venus/core.c
@@ -270,8 +270,7 @@ static int venus_remove(struct platform_device *pdev)
return ret;
}
-#ifdef CONFIG_PM
-static int venus_runtime_suspend(struct device *dev)
+static __maybe_unused int venus_runtime_suspend(struct device *dev)
{
struct venus_core *core = dev_get_drvdata(dev);
int ret;
@@ -283,7 +282,7 @@ static int venus_runtime_suspend(struct device *dev)
return ret;
}
-static int venus_runtime_resume(struct device *dev)
+static __maybe_unused int venus_runtime_resume(struct device *dev)
{
struct venus_core *core = dev_get_drvdata(dev);
int ret;
@@ -302,7 +301,6 @@ static int venus_runtime_resume(struct device *dev)
venus_clks_disable(core);
return ret;
}
-#endif
static const struct dev_pm_ops venus_pm_ops = {
SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
--
2.9.0
If QCOM_MDT_LOADER is enabled, but ARCH_QCOM is not, we run into
a build error:
ERROR: "qcom_mdt_load" [drivers/media/platform/qcom/venus/venus-core.ko] undefined!
ERROR: "qcom_mdt_get_size" [drivers/media/platform/qcom/venus/venus-core.ko] undefined!
This changes the 'select' statement again, so we only try to enable
those symbols when the drivers will actually get built.
Fixes: 76724b30f222 ("[media] media: venus: enable building with COMPILE_TEST")
Signed-off-by: Arnd Bergmann <[email protected]>
---
drivers/media/platform/Kconfig | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/media/platform/Kconfig b/drivers/media/platform/Kconfig
index cb2f31cd0088..635c53e61f8a 100644
--- a/drivers/media/platform/Kconfig
+++ b/drivers/media/platform/Kconfig
@@ -475,8 +475,8 @@ config VIDEO_QCOM_VENUS
tristate "Qualcomm Venus V4L2 encoder/decoder driver"
depends on VIDEO_DEV && VIDEO_V4L2 && HAS_DMA
depends on (ARCH_QCOM && IOMMU_DMA) || COMPILE_TEST
- select QCOM_MDT_LOADER if (ARM || ARM64)
- select QCOM_SCM if (ARM || ARM64)
+ select QCOM_MDT_LOADER if ARCH_QCOM
+ select QCOM_SCM if ARCH_QCOM
select VIDEOBUF2_DMA_SG
select V4L2_MEM2MEM_DEV
---help---
--
2.9.0
In venus_boot(), we pass a pointer to a phys_addr_t
into dmam_alloc_coherent, which the compiler warns about:
platform/qcom/venus/firmware.c: In function 'venus_boot':
platform/qcom/venus/firmware.c:63:49: error: passing argument 3 of 'dmam_alloc_coherent' from incompatible pointer type [-Werror=incompatible-pointer-types]
The returned DMA address is later passed on to a function that
takes a phys_addr_t, so it's clearly wrong to use the DMA
mapping interface here: the memory may be uncached, or the
address may be completely wrong if there is an IOMMU connected
to the device.
My interpretation is that using dmam_alloc_coherent() had two
purposes:
a) get a chunk of consecutive memory that may be larger than
the limit for kmalloc()
b) use the devres infrastructure to simplify the unwinding
in the error case.
I think ideally we'd use a devres-based version of
alloc_pages_exact() here, but since that doesn't exist,
let's use devm_get_free_pages() instead. This wastes a little
memory as the size gets rounded up to a power of two, but
is otherwise harmless. If we want to save memory here, calling
devm_free_pages() to release the memory once it is no longer
needed is probably better anyway.
Fixes: af2c3834c8ca ("[media] media: venus: adding core part and helper functions")
Signed-off-by: Arnd Bergmann <[email protected]>
---
The same problem exists in the drm driver, as of commit 7c65817e6d38
("drm/msm: gpu: Enable zap shader for A5XX"), and I submitted the
same patch for that already.
---
drivers/media/platform/qcom/venus/firmware.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/media/platform/qcom/venus/firmware.c b/drivers/media/platform/qcom/venus/firmware.c
index 1b1a4f355918..76edb9f60311 100644
--- a/drivers/media/platform/qcom/venus/firmware.c
+++ b/drivers/media/platform/qcom/venus/firmware.c
@@ -60,11 +60,13 @@ int venus_boot(struct device *parent, struct device *fw_dev, const char *fwname)
mem_size = VENUS_FW_MEM_SIZE;
- mem_va = dmam_alloc_coherent(fw_dev, mem_size, &mem_phys, GFP_KERNEL);
+ mem_va = (void *)devm_get_free_pages(parent, GFP_KERNEL,
+ get_order(mem_size));
if (!mem_va) {
ret = -ENOMEM;
goto err_unreg_device;
}
+ mem_phys = virt_to_phys(mem_va);
ret = request_firmware(&mdt, fwname, fw_dev);
if (ret < 0)
--
2.9.0
Hi Arnd,
On 27.06.2017 18:02, Arnd Bergmann wrote:
> In venus_boot(), we pass a pointer to a phys_addr_t
> into dmam_alloc_coherent, which the compiler warns about:
>
> platform/qcom/venus/firmware.c: In function 'venus_boot':
> platform/qcom/venus/firmware.c:63:49: error: passing argument 3 of 'dmam_alloc_coherent' from incompatible pointer type [-Werror=incompatible-pointer-types]
>
> The returned DMA address is later passed on to a function that
> takes a phys_addr_t, so it's clearly wrong to use the DMA
> mapping interface here: the memory may be uncached, or the
> address may be completely wrong if there is an IOMMU connected
> to the device.
>
> My interpretation is that using dmam_alloc_coherent() had two
> purposes:
>
> a) get a chunk of consecutive memory that may be larger than
> the limit for kmalloc()
>
> b) use the devres infrastructure to simplify the unwinding
> in the error case.
The intension here is to use per-device memory which is removed from
kernel allocator, that memory is used by remote processor (Venus) for
its code section and system memory, the memory must not be mapped to
kernel to avoid any cache issues.
As the memory in subject is reserved per-device memory the only legal
way to allocate it is by dmam_alloc_coherent() ->
dma_alloc_from_coherent().
For me the confusion comes from phys_addr_t which is passed to
qcom_mdt_load() and then the address passed to qcom_scm_pas_mem_setup()
which probably protects that physical memory. And the tz really expects
physical address.
The only solution I see is by casting dma_addr_t to phys_addr_t. Yes it
is ugly but what is proper solution then?
>
> I think ideally we'd use a devres-based version of
> alloc_pages_exact() here, but since that doesn't exist,
> let's use devm_get_free_pages() instead. This wastes a little
> memory as the size gets rounded up to a power of two, but
> is otherwise harmless. If we want to save memory here, calling
> devm_free_pages() to release the memory once it is no longer
> needed is probably better anyway.
>
> Fixes: af2c3834c8ca ("[media] media: venus: adding core part and helper functions")
> Signed-off-by: Arnd Bergmann <[email protected]>
> ---
> The same problem exists in the drm driver, as of commit 7c65817e6d38
> ("drm/msm: gpu: Enable zap shader for A5XX"), and I submitted the
> same patch for that already.
> ---
> drivers/media/platform/qcom/venus/firmware.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/media/platform/qcom/venus/firmware.c b/drivers/media/platform/qcom/venus/firmware.c
> index 1b1a4f355918..76edb9f60311 100644
> --- a/drivers/media/platform/qcom/venus/firmware.c
> +++ b/drivers/media/platform/qcom/venus/firmware.c
> @@ -60,11 +60,13 @@ int venus_boot(struct device *parent, struct device *fw_dev, const char *fwname)
>
> mem_size = VENUS_FW_MEM_SIZE;
>
> - mem_va = dmam_alloc_coherent(fw_dev, mem_size, &mem_phys, GFP_KERNEL);
> + mem_va = (void *)devm_get_free_pages(parent, GFP_KERNEL,
> + get_order(mem_size));
> if (!mem_va) {
> ret = -ENOMEM;
> goto err_unreg_device;
> }
> + mem_phys = virt_to_phys(mem_va);
>
> ret = request_firmware(&mdt, fwname, fw_dev);
> if (ret < 0)
>
regards,
Stan
Hi Arnd,
Thanks for the catch!
On 27.06.2017 18:02, Arnd Bergmann wrote:
> Without PM support, gcc warns about two unused functions:
>
> platform/qcom/venus/core.c:146:13: error: 'venus_clks_disable' defined but not used [-Werror=unused-function]
> platform/qcom/venus/core.c:126:12: error: 'venus_clks_enable' defined but not used [-Werror=unused-function]
>
> The problem as usual are incorrect #ifdefs, so the easiest fix
> is to do away with the #ifdef completely and mark the suspend/resume
> handlers as __maybe_unused, which they are.
>
> Fixes: af2c3834c8ca ("[media] media: venus: adding core part and helper functions")
> Signed-off-by: Arnd Bergmann <[email protected]>
Reviewed-by: Stanimir Varbanov <[email protected]>
> ---
> drivers/media/platform/qcom/venus/core.c | 6 ++----
> 1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
> index d8cbe8549d97..47f79637938c 100644
> --- a/drivers/media/platform/qcom/venus/core.c
> +++ b/drivers/media/platform/qcom/venus/core.c
> @@ -270,8 +270,7 @@ static int venus_remove(struct platform_device *pdev)
> return ret;
> }
>
> -#ifdef CONFIG_PM
> -static int venus_runtime_suspend(struct device *dev)
> +static __maybe_unused int venus_runtime_suspend(struct device *dev)
> {
> struct venus_core *core = dev_get_drvdata(dev);
> int ret;
> @@ -283,7 +282,7 @@ static int venus_runtime_suspend(struct device *dev)
> return ret;
> }
>
> -static int venus_runtime_resume(struct device *dev)
> +static __maybe_unused int venus_runtime_resume(struct device *dev)
> {
> struct venus_core *core = dev_get_drvdata(dev);
> int ret;
> @@ -302,7 +301,6 @@ static int venus_runtime_resume(struct device *dev)
> venus_clks_disable(core);
> return ret;
> }
> -#endif
>
> static const struct dev_pm_ops venus_pm_ops = {
> SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
>
Hi Arnd,
On 27.06.2017 18:02, Arnd Bergmann wrote:
> If QCOM_MDT_LOADER is enabled, but ARCH_QCOM is not, we run into
> a build error:
>
> ERROR: "qcom_mdt_load" [drivers/media/platform/qcom/venus/venus-core.ko] undefined!
> ERROR: "qcom_mdt_get_size" [drivers/media/platform/qcom/venus/venus-core.ko] undefined!
Ahh, thanks for the fix, these two will also pursuing me in my dreams.
>
> This changes the 'select' statement again, so we only try to enable
> those symbols when the drivers will actually get built.
>
> Fixes: 76724b30f222 ("[media] media: venus: enable building with COMPILE_TEST")
> Signed-off-by: Arnd Bergmann <[email protected]>
Reviewed-by: Stanimir Varbanov <[email protected]>
> ---
> drivers/media/platform/Kconfig | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/media/platform/Kconfig b/drivers/media/platform/Kconfig
> index cb2f31cd0088..635c53e61f8a 100644
> --- a/drivers/media/platform/Kconfig
> +++ b/drivers/media/platform/Kconfig
> @@ -475,8 +475,8 @@ config VIDEO_QCOM_VENUS
> tristate "Qualcomm Venus V4L2 encoder/decoder driver"
> depends on VIDEO_DEV && VIDEO_V4L2 && HAS_DMA
> depends on (ARCH_QCOM && IOMMU_DMA) || COMPILE_TEST
> - select QCOM_MDT_LOADER if (ARM || ARM64)
> - select QCOM_SCM if (ARM || ARM64)
> + select QCOM_MDT_LOADER if ARCH_QCOM
> + select QCOM_SCM if ARCH_QCOM
> select VIDEOBUF2_DMA_SG
> select V4L2_MEM2MEM_DEV
> ---help---
>
regards,
Stan
On Tue, Jun 27, 2017 at 9:39 PM, Stanimir Varbanov
<[email protected]> wrote:
> Hi Arnd,
>
> On 27.06.2017 18:02, Arnd Bergmann wrote:
>>
>> In venus_boot(), we pass a pointer to a phys_addr_t
>> into dmam_alloc_coherent, which the compiler warns about:
>>
>> platform/qcom/venus/firmware.c: In function 'venus_boot':
>> platform/qcom/venus/firmware.c:63:49: error: passing argument 3 of
>> 'dmam_alloc_coherent' from incompatible pointer type
>> [-Werror=incompatible-pointer-types]
>>
>> The returned DMA address is later passed on to a function that
>> takes a phys_addr_t, so it's clearly wrong to use the DMA
>> mapping interface here: the memory may be uncached, or the
>> address may be completely wrong if there is an IOMMU connected
>> to the device.
>>
>> My interpretation is that using dmam_alloc_coherent() had two
>> purposes:
>>
>> a) get a chunk of consecutive memory that may be larger than
>> the limit for kmalloc()
>>
>> b) use the devres infrastructure to simplify the unwinding
>> in the error case.
>
>
> The intension here is to use per-device memory which is removed from kernel
> allocator, that memory is used by remote processor (Venus) for its code
> section and system memory, the memory must not be mapped to kernel to avoid
> any cache issues.
>
> As the memory in subject is reserved per-device memory the only legal way to
> allocate it is by dmam_alloc_coherent() -> dma_alloc_from_coherent().
>
> For me the confusion comes from phys_addr_t which is passed to
> qcom_mdt_load() and then the address passed to qcom_scm_pas_mem_setup()
> which probably protects that physical memory. And the tz really expects
> physical address.
>
> The only solution I see is by casting dma_addr_t to phys_addr_t. Yes it is
> ugly but what is proper solution then?
If you actually have a separate remote processor that accesses this memory,
then qcom_mdt_load() is the wrong interface, as it takes a physical address,
and we need to introduce another interface that can take a DMA address
relative to a particular device.
You cannot cast between the two types because phys_addr_t is an address
as seen from the CPU, and dma_addr_t is seen by a particular device,
and can only be used together with that device pointer.
It looks like the pointer gets passed down to
qcom_scm_call(dev, QCOM_SCM_SVC_PIL,
QCOM_SCM_PAS_MEM_SETUP_CMD, ...), which in turn takes
a 32-bit address, suggesting that this is indeed a dma address for that
device (possibly going through an IOMMU), so maybe it just needs to
all be changed to dma_addr_t.
Is there any official documentation for qcom_scm_call() that clarifies
what address space the arguments are in?
Arnd
On Tue, Jun 27, 2017 at 9:45 PM, Stanimir Varbanov
<[email protected]> wrote:
> Hi Arnd,
>
> On 27.06.2017 18:02, Arnd Bergmann wrote:
>>
>> If QCOM_MDT_LOADER is enabled, but ARCH_QCOM is not, we run into
>> a build error:
>>
>> ERROR: "qcom_mdt_load" [drivers/media/platform/qcom/venus/venus-core.ko]
>> undefined!
>> ERROR: "qcom_mdt_get_size"
>> [drivers/media/platform/qcom/venus/venus-core.ko] undefined!
>
>
> Ahh, thanks for the fix, these two will also pursuing me in my dreams.
I just came after me as well, as I hit another corner case, we need this
fixup on top, I'll send a replacement:
Subject: [PATCH] fixup! [media] venus: fix compile-test build on
non-qcom ARM platform
Signed-off-by: Arnd Bergmann <[email protected]>
diff --git a/drivers/media/platform/qcom/venus/firmware.c
b/drivers/media/platform/qcom/venus/firmware.c
index 76edb9f60311..3794b9e3250b 100644
--- a/drivers/media/platform/qcom/venus/firmware.c
+++ b/drivers/media/platform/qcom/venus/firmware.c
@@ -40,7 +40,7 @@ int venus_boot(struct device *parent, struct device
*fw_dev, const char *fwname)
void *mem_va;
int ret;
- if (!qcom_scm_is_available())
+ if (!IS_ENABLED(CONFIG_QCOM_MDT_LOADER) || !qcom_scm_is_available())
return -EPROBE_DEFER;
fw_dev->parent = parent;