diff mbox series

[RFC,3/4] accel/ivpu: Remove support for uncached buffers

Message ID 20230901164842.178654-4-stanislaw.gruszka@linux.intel.com (mailing list archive)
State New, archived
Headers show
Series accel/ivpu: Use GEM shmem | expand

Commit Message

Stanislaw Gruszka Sept. 1, 2023, 4:48 p.m. UTC 
From: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>

Usages of DRM_IVPU_BO_UNCACHED should be replaced by DRM_IVPU_BO_WC.
There is no functional benefit from DRM_IVPU_BO_UNCACHED if these
buffers are never mapped to host VM.

This allows to cut the buffer handling code in the kernel driver
by half.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
---
 drivers/accel/ivpu/ivpu_fw.c  | 2 +-
 drivers/accel/ivpu/ivpu_gem.c | 3 ---
 include/uapi/drm/ivpu_accel.h | 2 +-
 3 files changed, 2 insertions(+), 5 deletions(-)

Comments

Jeffrey Hugo Sept. 11, 2023, 3:24 p.m. UTC | #1 
On 9/1/2023 10:48 AM, Stanislaw Gruszka wrote:
> From: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
> 
> Usages of DRM_IVPU_BO_UNCACHED should be replaced by DRM_IVPU_BO_WC.
> There is no functional benefit from DRM_IVPU_BO_UNCACHED if these
> buffers are never mapped to host VM.
> 
> This allows to cut the buffer handling code in the kernel driver
> by half.
> 
> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
> Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
> ---
>   drivers/accel/ivpu/ivpu_fw.c  | 2 +-
>   drivers/accel/ivpu/ivpu_gem.c | 3 ---
>   include/uapi/drm/ivpu_accel.h | 2 +-
>   3 files changed, 2 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c
> index 2fef9fe154aa..8ab0f3225205 100644
> --- a/drivers/accel/ivpu/ivpu_fw.c
> +++ b/drivers/accel/ivpu/ivpu_fw.c
> @@ -248,7 +248,7 @@ static int ivpu_fw_mem_init(struct ivpu_device *vdev)
>   
>   	if (fw->shave_nn_size) {
>   		fw->mem_shave_nn = ivpu_bo_alloc_internal(vdev, vdev->hw->ranges.shave.start,
> -							  fw->shave_nn_size, DRM_IVPU_BO_UNCACHED);
> +							  fw->shave_nn_size, DRM_IVPU_BO_WC);
>   		if (!fw->mem_shave_nn) {
>   			ivpu_err(vdev, "Failed to allocate shavenn buffer\n");
>   			ret = -ENOMEM;
> diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c
> index 915c53d7bb97..2a91eb1e3627 100644
> --- a/drivers/accel/ivpu/ivpu_gem.c
> +++ b/drivers/accel/ivpu/ivpu_gem.c
> @@ -89,8 +89,6 @@ static int __must_check shmem_alloc_pages_locked(struct ivpu_bo *bo)
>   
>   	if (bo->flags & DRM_IVPU_BO_WC)
>   		set_pages_array_wc(pages, npages);
> -	else if (bo->flags & DRM_IVPU_BO_UNCACHED)
> -		set_pages_array_uc(pages, npages);
>   
>   	bo->pages = pages;
>   	return 0;
> @@ -366,7 +364,6 @@ ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, const struct ivpu_b
>   
>   	switch (flags & DRM_IVPU_BO_CACHE_MASK) {
>   	case DRM_IVPU_BO_CACHED:
> -	case DRM_IVPU_BO_UNCACHED:
>   	case DRM_IVPU_BO_WC:
>   		break;
>   	default:
> diff --git a/include/uapi/drm/ivpu_accel.h b/include/uapi/drm/ivpu_accel.h
> index 262db0c3beee..de1944e42c65 100644
> --- a/include/uapi/drm/ivpu_accel.h
> +++ b/include/uapi/drm/ivpu_accel.h
> @@ -196,7 +196,7 @@ struct drm_ivpu_bo_create {
>   	 *
>   	 * %DRM_IVPU_BO_UNCACHED:
>   	 *
> -	 * Allocated BO will not be cached on host side nor snooped on the VPU side.
> +	 * Not supported. Use DRM_IVPU_BO_WC instead.
>   	 *
>   	 * %DRM_IVPU_BO_WC:
>   	 *

Feels like this will break existing userspace.  You could see if 
userspace specified UNCACHED and change it to WC before processing the 
request.  Maybe also use WARN_ONCE to indicate that userspace should be 
updated.

Or is it the case that userspace never actually used this?
Stanislaw Gruszka Sept. 19, 2023, 9:49 a.m. UTC | #2 
On Mon, Sep 11, 2023 at 09:24:42AM -0600, Jeffrey Hugo wrote:
> On 9/1/2023 10:48 AM, Stanislaw Gruszka wrote:
> > From: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
> > 
> > Usages of DRM_IVPU_BO_UNCACHED should be replaced by DRM_IVPU_BO_WC.
> > There is no functional benefit from DRM_IVPU_BO_UNCACHED if these
> > buffers are never mapped to host VM.
> > 
> > This allows to cut the buffer handling code in the kernel driver
> > by half.
> > 
> > Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
> > Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
> > ---
> >   drivers/accel/ivpu/ivpu_fw.c  | 2 +-
> >   drivers/accel/ivpu/ivpu_gem.c | 3 ---
> >   include/uapi/drm/ivpu_accel.h | 2 +-
> >   3 files changed, 2 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c
> > index 2fef9fe154aa..8ab0f3225205 100644
> > --- a/drivers/accel/ivpu/ivpu_fw.c
> > +++ b/drivers/accel/ivpu/ivpu_fw.c
> > @@ -248,7 +248,7 @@ static int ivpu_fw_mem_init(struct ivpu_device *vdev)
> >   	if (fw->shave_nn_size) {
> >   		fw->mem_shave_nn = ivpu_bo_alloc_internal(vdev, vdev->hw->ranges.shave.start,
> > -							  fw->shave_nn_size, DRM_IVPU_BO_UNCACHED);
> > +							  fw->shave_nn_size, DRM_IVPU_BO_WC);
> >   		if (!fw->mem_shave_nn) {
> >   			ivpu_err(vdev, "Failed to allocate shavenn buffer\n");
> >   			ret = -ENOMEM;
> > diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c
> > index 915c53d7bb97..2a91eb1e3627 100644
> > --- a/drivers/accel/ivpu/ivpu_gem.c
> > +++ b/drivers/accel/ivpu/ivpu_gem.c
> > @@ -89,8 +89,6 @@ static int __must_check shmem_alloc_pages_locked(struct ivpu_bo *bo)
> >   	if (bo->flags & DRM_IVPU_BO_WC)
> >   		set_pages_array_wc(pages, npages);
> > -	else if (bo->flags & DRM_IVPU_BO_UNCACHED)
> > -		set_pages_array_uc(pages, npages);
> >   	bo->pages = pages;
> >   	return 0;
> > @@ -366,7 +364,6 @@ ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, const struct ivpu_b
> >   	switch (flags & DRM_IVPU_BO_CACHE_MASK) {
> >   	case DRM_IVPU_BO_CACHED:
> > -	case DRM_IVPU_BO_UNCACHED:
> >   	case DRM_IVPU_BO_WC:
> >   		break;
> >   	default:
> > diff --git a/include/uapi/drm/ivpu_accel.h b/include/uapi/drm/ivpu_accel.h
> > index 262db0c3beee..de1944e42c65 100644
> > --- a/include/uapi/drm/ivpu_accel.h
> > +++ b/include/uapi/drm/ivpu_accel.h
> > @@ -196,7 +196,7 @@ struct drm_ivpu_bo_create {
> >   	 *
> >   	 * %DRM_IVPU_BO_UNCACHED:
> >   	 *
> > -	 * Allocated BO will not be cached on host side nor snooped on the VPU side.
> > +	 * Not supported. Use DRM_IVPU_BO_WC instead.
> >   	 *
> >   	 * %DRM_IVPU_BO_WC:
> >   	 *
> 
> Feels like this will break existing userspace.  You could see if userspace
> specified UNCACHED and change it to WC before processing the request.  Maybe
> also use WARN_ONCE to indicate that userspace should be updated.
> 
> Or is it the case that userspace never actually used this?

Usage of those buffers was removed some time ago:
https://github.com/intel/linux-vpu-driver/commit/c473c9826cb28fa3dcf8883fc804b1e84c6b5fb1

And will not be part of first user-mode driver release. I think we can
safely do the change.

Regards
Stanislaw
Jeffrey Hugo Sept. 22, 2023, 3:21 p.m. UTC | #3 
On 9/19/2023 3:49 AM, Stanislaw Gruszka wrote:
> On Mon, Sep 11, 2023 at 09:24:42AM -0600, Jeffrey Hugo wrote:
>> On 9/1/2023 10:48 AM, Stanislaw Gruszka wrote:
>>> From: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
>>>
>>> Usages of DRM_IVPU_BO_UNCACHED should be replaced by DRM_IVPU_BO_WC.
>>> There is no functional benefit from DRM_IVPU_BO_UNCACHED if these
>>> buffers are never mapped to host VM.
>>>
>>> This allows to cut the buffer handling code in the kernel driver
>>> by half.
>>>
>>> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
>>> Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
>>> ---
>>>    drivers/accel/ivpu/ivpu_fw.c  | 2 +-
>>>    drivers/accel/ivpu/ivpu_gem.c | 3 ---
>>>    include/uapi/drm/ivpu_accel.h | 2 +-
>>>    3 files changed, 2 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c
>>> index 2fef9fe154aa..8ab0f3225205 100644
>>> --- a/drivers/accel/ivpu/ivpu_fw.c
>>> +++ b/drivers/accel/ivpu/ivpu_fw.c
>>> @@ -248,7 +248,7 @@ static int ivpu_fw_mem_init(struct ivpu_device *vdev)
>>>    	if (fw->shave_nn_size) {
>>>    		fw->mem_shave_nn = ivpu_bo_alloc_internal(vdev, vdev->hw->ranges.shave.start,
>>> -							  fw->shave_nn_size, DRM_IVPU_BO_UNCACHED);
>>> +							  fw->shave_nn_size, DRM_IVPU_BO_WC);
>>>    		if (!fw->mem_shave_nn) {
>>>    			ivpu_err(vdev, "Failed to allocate shavenn buffer\n");
>>>    			ret = -ENOMEM;
>>> diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c
>>> index 915c53d7bb97..2a91eb1e3627 100644
>>> --- a/drivers/accel/ivpu/ivpu_gem.c
>>> +++ b/drivers/accel/ivpu/ivpu_gem.c
>>> @@ -89,8 +89,6 @@ static int __must_check shmem_alloc_pages_locked(struct ivpu_bo *bo)
>>>    	if (bo->flags & DRM_IVPU_BO_WC)
>>>    		set_pages_array_wc(pages, npages);
>>> -	else if (bo->flags & DRM_IVPU_BO_UNCACHED)
>>> -		set_pages_array_uc(pages, npages);
>>>    	bo->pages = pages;
>>>    	return 0;
>>> @@ -366,7 +364,6 @@ ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, const struct ivpu_b
>>>    	switch (flags & DRM_IVPU_BO_CACHE_MASK) {
>>>    	case DRM_IVPU_BO_CACHED:
>>> -	case DRM_IVPU_BO_UNCACHED:
>>>    	case DRM_IVPU_BO_WC:
>>>    		break;
>>>    	default:
>>> diff --git a/include/uapi/drm/ivpu_accel.h b/include/uapi/drm/ivpu_accel.h
>>> index 262db0c3beee..de1944e42c65 100644
>>> --- a/include/uapi/drm/ivpu_accel.h
>>> +++ b/include/uapi/drm/ivpu_accel.h
>>> @@ -196,7 +196,7 @@ struct drm_ivpu_bo_create {
>>>    	 *
>>>    	 * %DRM_IVPU_BO_UNCACHED:
>>>    	 *
>>> -	 * Allocated BO will not be cached on host side nor snooped on the VPU side.
>>> +	 * Not supported. Use DRM_IVPU_BO_WC instead.
>>>    	 *
>>>    	 * %DRM_IVPU_BO_WC:
>>>    	 *
>>
>> Feels like this will break existing userspace.  You could see if userspace
>> specified UNCACHED and change it to WC before processing the request.  Maybe
>> also use WARN_ONCE to indicate that userspace should be updated.
>>
>> Or is it the case that userspace never actually used this?
> 
> Usage of those buffers was removed some time ago:
> https://github.com/intel/linux-vpu-driver/commit/c473c9826cb28fa3dcf8883fc804b1e84c6b5fb1
> 
> And will not be part of first user-mode driver release. I think we can
> safely do the change.

Fair enough.  Thanks for the clarification.

Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
diff mbox series

Patch

diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c
index 2fef9fe154aa..8ab0f3225205 100644
--- a/drivers/accel/ivpu/ivpu_fw.c
+++ b/drivers/accel/ivpu/ivpu_fw.c
@@ -248,7 +248,7 @@  static int ivpu_fw_mem_init(struct ivpu_device *vdev)
 
 	if (fw->shave_nn_size) {
 		fw->mem_shave_nn = ivpu_bo_alloc_internal(vdev, vdev->hw->ranges.shave.start,
-							  fw->shave_nn_size, DRM_IVPU_BO_UNCACHED);
+							  fw->shave_nn_size, DRM_IVPU_BO_WC);
 		if (!fw->mem_shave_nn) {
 			ivpu_err(vdev, "Failed to allocate shavenn buffer\n");
 			ret = -ENOMEM;
diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c
index 915c53d7bb97..2a91eb1e3627 100644
--- a/drivers/accel/ivpu/ivpu_gem.c
+++ b/drivers/accel/ivpu/ivpu_gem.c
@@ -89,8 +89,6 @@  static int __must_check shmem_alloc_pages_locked(struct ivpu_bo *bo)
 
 	if (bo->flags & DRM_IVPU_BO_WC)
 		set_pages_array_wc(pages, npages);
-	else if (bo->flags & DRM_IVPU_BO_UNCACHED)
-		set_pages_array_uc(pages, npages);
 
 	bo->pages = pages;
 	return 0;
@@ -366,7 +364,6 @@  ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, const struct ivpu_b
 
 	switch (flags & DRM_IVPU_BO_CACHE_MASK) {
 	case DRM_IVPU_BO_CACHED:
-	case DRM_IVPU_BO_UNCACHED:
 	case DRM_IVPU_BO_WC:
 		break;
 	default:
diff --git a/include/uapi/drm/ivpu_accel.h b/include/uapi/drm/ivpu_accel.h
index 262db0c3beee..de1944e42c65 100644
--- a/include/uapi/drm/ivpu_accel.h
+++ b/include/uapi/drm/ivpu_accel.h
@@ -196,7 +196,7 @@  struct drm_ivpu_bo_create {
 	 *
 	 * %DRM_IVPU_BO_UNCACHED:
 	 *
-	 * Allocated BO will not be cached on host side nor snooped on the VPU side.
+	 * Not supported. Use DRM_IVPU_BO_WC instead.
 	 *
 	 * %DRM_IVPU_BO_WC:
 	 *