[PATCH 1/1] drm/amdgpu: Pin to gtt before cpu accesses dma buf.
Li, Samuel
Samuel.Li at amd.com
Wed Nov 29 16:52:15 UTC 2017
To explain the operations a little bit, the tests include repeated testing of the following sequences,
91 const int sequences[4][4] = {
92 { STEP_MMAP, STEP_FAULT, STEP_FLIP, STEP_DRAW },
93 { STEP_MMAP, STEP_FLIP, STEP_DRAW, STEP_SKIP },
94 { STEP_MMAP, STEP_DRAW, STEP_FLIP, STEP_SKIP },
95 { STEP_FLIP, STEP_MMAP, STEP_DRAW, STEP_SKIP },
96 };
Here STEP_MMAP includes prime_mmap() and begin_cpu_access(). STEP_DRAW includes repeated cpu reads/writes.
It looks to me the dma_buf has to be pinned to gtt here, to prevent it being pinned back by STEP_FLIP before drawing.
> I've send out a WIP branch a for this a few weeks ago, but haven't worked on
> this in a while. BTW it is only supported on Carizzo and Raven.
Can you show me the branch? Looks like we need to get if finished.
Sam
> -----Original Message-----
> From: Koenig, Christian
> Sent: Wednesday, November 29, 2017 10:01 AM
> To: Li, Samuel <Samuel.Li at amd.com>; amd-gfx at lists.freedesktop.org
> Subject: Re: [PATCH 1/1] drm/amdgpu: Pin to gtt before cpu accesses dma
> buf.
>
> > What is the concern for scanning out from GTT on APUs?
> It's simply not implemented yet. You need quite a bunch of different setup in
> DC for this to work.
>
> I've send out a WIP branch a for this a few weeks ago, but haven't worked on
> this in a while. BTW it is only supported on Carizzo and Raven.
>
> But even then pinning a BO to GTT for this would still be a no-go. We could
> just avoid the copy on scanout when the BO is already inside GTT because of
> the CPU access.
>
> In general we should rather work on this as Michel described and avoid
> creating the BO in VRAM in the first place if possible.
>
> Regards,
> Christian.
>
> Am 29.11.2017 um 15:56 schrieb Li, Samuel:
> > One major purpose of the ChromeOS mmap_test is to avoid buffer copying.
> What is the concern for scanning out from GTT on APUs?
> >
> > Sam
> >
> >> -----Original Message-----
> >> From: Koenig, Christian
> >> Sent: Wednesday, November 29, 2017 9:54 AM
> >> To: Li, Samuel <Samuel.Li at amd.com>; amd-gfx at lists.freedesktop.org
> >> Subject: Re: [PATCH 1/1] drm/amdgpu: Pin to gtt before cpu accesses
> >> dma buf.
> >>
> >> And exactly that's the reason why it is a no-go.
> >>
> >> Scanning out from GTT isn't supported at the moment.
> >>
> >> Christian.
> >>
> >> Am 29.11.2017 um 15:48 schrieb Li, Samuel:
> >>> Actually it needs to be pinned here, since otherwise page flip will
> >>> pin it into
> >> vram.
> >>> SAm
> >>>
> >>>
> >>>> -----Original Message-----
> >>>> From: Christian König [mailto:ckoenig.leichtzumerken at gmail.com]
> >>>> Sent: Wednesday, November 29, 2017 4:39 AM
> >>>> To: Li, Samuel <Samuel.Li at amd.com>; amd-gfx at lists.freedesktop.org
> >>>> Subject: Re: [PATCH 1/1] drm/amdgpu: Pin to gtt before cpu accesses
> >>>> dma buf.
> >>>>
> >>>> Am 29.11.2017 um 01:20 schrieb Samuel Li:
> >>>>> To improve cpu read performance. This is implemented for APUs
> >> currently.
> >>>> And once more a NAK for this approach.
> >>>>
> >>>> What we could do is migrate the BO to GTT during mmap, but pinning
> >>>> it is out of question.
> >>>>
> >>>> Regards,
> >>>> Christian.
> >>>>
> >>>>> Change-Id: I300a7daed8f2b0ba6be71a43196a6b8617ddc62e
> >>>>> ---
> >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +
> >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_display.c | 10 +-
> >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 2 +-
> >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c | 108
> >>>> ++++++++++++++++++++++
> >>>>> drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 8 +-
> >>>>> 5 files changed, 126 insertions(+), 4 deletions(-)
> >>>>>
> >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>>>> index f8657c3..193db70 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>>>> @@ -417,6 +417,8 @@ amdgpu_gem_prime_import_sg_table(struct
> >>>> drm_device *dev,
> >>>>> struct dma_buf *amdgpu_gem_prime_export(struct drm_device
> *dev,
> >>>>> struct drm_gem_object
> *gobj,
> >>>>> int flags);
> >>>>> +struct drm_gem_object *amdgpu_gem_prime_import(struct
> >> drm_device
> >>>> *dev,
> >>>>> + struct dma_buf *dma_buf);
> >>>>> int amdgpu_gem_prime_pin(struct drm_gem_object *obj);
> >>>>> void amdgpu_gem_prime_unpin(struct drm_gem_object *obj);
> >>>>> struct reservation_object *amdgpu_gem_prime_res_obj(struct
> >>>>> drm_gem_object *); diff --git
> >>>>> a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
> >>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
> >>>>> index d704a45..b5483e4 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
> >>>>> @@ -147,6 +147,7 @@ int amdgpu_crtc_page_flip_target(struct
> >> drm_crtc
> >>>> *crtc,
> >>>>> struct drm_device *dev = crtc->dev;
> >>>>> struct amdgpu_device *adev = dev->dev_private;
> >>>>> struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
> >>>>> + bool gtt_scannable = (adev->asic_type >= CHIP_CARRIZO &&
> adev-
> >>>>> flags
> >>>>> +& AMD_IS_APU);
> >>>>> struct amdgpu_framebuffer *old_amdgpu_fb;
> >>>>> struct amdgpu_framebuffer *new_amdgpu_fb;
> >>>>> struct drm_gem_object *obj;
> >>>>> @@ -190,8 +191,13 @@ int amdgpu_crtc_page_flip_target(struct
> >>>>> drm_crtc *crtc,
> >>>>>
> >>>>> r = amdgpu_bo_pin(new_abo,
> AMDGPU_GEM_DOMAIN_VRAM,
> >>>> &base);
> >>>>> if (unlikely(r != 0)) {
> >>>>> - DRM_ERROR("failed to pin new abo buffer before
> flip\n");
> >>>>> - goto unreserve;
> >>>>> + /* latest APUs support gtt scan out */
> >>>>> + if (gtt_scannable)
> >>>>> + r = amdgpu_bo_pin(new_abo,
> >>>> AMDGPU_GEM_DOMAIN_GTT, &base);
> >>>>> + if (unlikely(r != 0)) {
> >>>>> + DRM_ERROR("failed to pin new abo buffer
> before
> >>>> flip\n");
> >>>>> + goto unreserve;
> >>>>> + }
> >>>>> }
> >>>>>
> >>>>> r = reservation_object_get_fences_rcu(new_abo->tbo.resv,
> >>>>> &work->excl, diff --git
> a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>>>> index 31383e0..df30b08 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>>>> @@ -868,7 +868,7 @@ static struct drm_driver kms_driver = {
> >>>>> .prime_handle_to_fd = drm_gem_prime_handle_to_fd,
> >>>>> .prime_fd_to_handle = drm_gem_prime_fd_to_handle,
> >>>>> .gem_prime_export = amdgpu_gem_prime_export,
> >>>>> - .gem_prime_import = drm_gem_prime_import,
> >>>>> + .gem_prime_import = amdgpu_gem_prime_import,
> >>>>> .gem_prime_pin = amdgpu_gem_prime_pin,
> >>>>> .gem_prime_unpin = amdgpu_gem_prime_unpin,
> >>>>> .gem_prime_res_obj = amdgpu_gem_prime_res_obj, diff --
> git
> >>>>> a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
> >>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
> >>>>> index ae9c106..9e1424d 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
> >>>>> @@ -164,6 +164,82 @@ struct reservation_object
> >>>> *amdgpu_gem_prime_res_obj(struct drm_gem_object *obj)
> >>>>> return bo->tbo.resv;
> >>>>> }
> >>>>>
> >>>>> +static int amdgpu_gem_begin_cpu_access(struct dma_buf
> *dma_buf,
> >>>> enum
> >>>>> +dma_data_direction direction) {
> >>>>> + struct amdgpu_bo *bo = gem_to_amdgpu_bo(dma_buf-
> >priv);
> >>>>> + struct amdgpu_device *adev = amdgpu_ttm_adev(bo-
> >tbo.bdev);
> >>>>> + long i, ret = 0;
> >>>>> + unsigned old_count;
> >>>>> + bool reads = (direction == DMA_BIDIRECTIONAL || direction
> ==
> >>>> DMA_FROM_DEVICE);
> >>>>> + bool gtt_scannable = (adev->asic_type >= CHIP_CARRIZO &&
> adev-
> >>>>> flags & AMD_IS_APU);
> >>>>> + u32 domain;
> >>>>> +
> >>>>> + if (!reads || !gtt_scannable)
> >>>>> + return 0;
> >>>>> +
> >>>>> + ret = amdgpu_bo_reserve(bo, false);
> >>>>> + if (unlikely(ret != 0))
> >>>>> + return ret;
> >>>>> +
> >>>>> + /*
> >>>>> + * Wait for all shared fences to complete before we switch to
> future
> >>>>> + * use of exclusive fence on this prime shared bo.
> >>>>> + */
> >>>>> + ret = reservation_object_wait_timeout_rcu(bo->tbo.resv,
> true, false,
> >>>>> +
> MAX_SCHEDULE_TIMEOUT);
> >>>>> +
> >>>>> + if (unlikely(ret < 0)) {
> >>>>> + DRM_DEBUG_PRIME("Fence wait failed: %li\n", ret);
> >>>>> + amdgpu_bo_unreserve(bo);
> >>>>> + return ret;
> >>>>> + }
> >>>>> +
> >>>>> + ret = 0;
> >>>>> + /* Pin to gtt */
> >>>>> + domain = amdgpu_mem_type_to_domain(bo-
> >>>>> tbo.mem.mem_type);
> >>>>> + if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
> >>>>> + old_count = bo->pin_count;
> >>>>> + for (i = 0; i < old_count; i++)
> >>>>> + amdgpu_bo_unpin(bo);
> >>>>> + for (i = 0; i < old_count; i++) {
> >>>>> + ret = amdgpu_bo_pin(bo,
> >>>> AMDGPU_GEM_DOMAIN_GTT, NULL);
> >>>>> + if (unlikely(ret != 0))
> >>>>> + break;
> >>>>> + }
> >>>>> + }
> >>>>> + if (ret == 0)
> >>>>> + ret = amdgpu_bo_pin(bo,
> AMDGPU_GEM_DOMAIN_GTT,
> >>>> NULL);
> >>>>> +
> >>>>> + amdgpu_bo_unreserve(bo);
> >>>>> + return ret;
> >>>>> +}
> >>>>> +
> >>>>> +static int amdgpu_gem_end_cpu_access(struct dma_buf *dma_buf,
> >>>> enum
> >>>>> +dma_data_direction direction) {
> >>>>> + struct amdgpu_bo *bo = gem_to_amdgpu_bo(dma_buf-
> >priv);
> >>>>> + int ret = 0;
> >>>>> + bool reads = (direction == DMA_BIDIRECTIONAL || direction
> ==
> >>>> DMA_FROM_DEVICE);
> >>>>> + struct amdgpu_device *adev = amdgpu_ttm_adev(bo-
> >tbo.bdev);
> >>>>> + bool gtt_scannable = (adev->asic_type >= CHIP_CARRIZO &&
> adev-
> >>>>> flags
> >>>>> +& AMD_IS_APU);
> >>>>> +
> >>>>> + if (!reads || !gtt_scannable)
> >>>>> + return 0;
> >>>>> +
> >>>>> + mb();
> >>>>> + ret = amdgpu_bo_reserve(bo, true);
> >>>>> + if (unlikely(ret != 0))
> >>>>> + return ret;
> >>>>> +
> >>>>> + amdgpu_bo_unpin(bo);
> >>>>> +
> >>>>> + amdgpu_bo_unreserve(bo);
> >>>>> +
> >>>>> + return 0;
> >>>>> +}
> >>>>> +
> >>>>> +static struct dma_buf_ops amdgpu_dmabuf_ops; static atomic_t
> >>>>> +aops_lock;
> >>>>> +
> >>>>> struct dma_buf *amdgpu_gem_prime_export(struct drm_device
> *dev,
> >>>>> struct drm_gem_object
> *gobj,
> >>>>> int flags)
> >>>>> @@ -178,5 +254,37 @@ struct dma_buf
> >>>> *amdgpu_gem_prime_export(struct drm_device *dev,
> >>>>> buf = drm_gem_prime_export(dev, gobj, flags);
> >>>>> if (!IS_ERR(buf))
> >>>>> buf->file->f_mapping = dev->anon_inode-
> >i_mapping;
> >>>>> +
> >>>>> + while (amdgpu_dmabuf_ops.begin_cpu_access !=
> >>>> amdgpu_gem_begin_cpu_access ||
> >>>>> + amdgpu_dmabuf_ops.end_cpu_access !=
> >>>> amdgpu_gem_end_cpu_access )
> >>>>> + {
> >>>>> + if (!atomic_cmpxchg(&aops_lock, 0, 1)) {
> >>>>> + amdgpu_dmabuf_ops = *(buf->ops);
> >>>>> + amdgpu_dmabuf_ops.begin_cpu_access =
> >>>> amdgpu_gem_begin_cpu_access;
> >>>>> + amdgpu_dmabuf_ops.end_cpu_access =
> >>>> amdgpu_gem_end_cpu_access;
> >>>>> + }
> >>>>> + }
> >>>>> + buf->ops = &amdgpu_dmabuf_ops;
> >>>>> +
> >>>>> return buf;
> >>>>> }
> >>>>> +
> >>>>> +struct drm_gem_object *amdgpu_gem_prime_import(struct
> >> drm_device
> >>>> *dev,
> >>>>> + struct dma_buf *dma_buf)
> {
> >>>>> + struct drm_gem_object *obj;
> >>>>> +
> >>>>> + if (dma_buf->ops == &amdgpu_dmabuf_ops) {
> >>>>> + obj = dma_buf->priv;
> >>>>> + if (obj->dev == dev) {
> >>>>> + /*
> >>>>> + * Importing dmabuf exported from out own
> gem
> >>>> increases
> >>>>> + * refcount on gem itself instead of f_count
> of
> >>>> dmabuf.
> >>>>> + */
> >>>>> + drm_gem_object_get(obj);
> >>>>> + return obj;
> >>>>> + }
> >>>>> + }
> >>>>> +
> >>>>> + return drm_gem_prime_import(dev, dma_buf); }
> >>>>> diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> >>>>> b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> >>>>> index a3bf021..f25b830 100644
> >>>>> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> >>>>> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> >>>>> @@ -2987,6 +2987,8 @@ static int
> dm_plane_helper_prepare_fb(struct
> >>>> drm_plane *plane,
> >>>>> int r;
> >>>>> struct dm_plane_state *dm_plane_state_new,
> >>>> *dm_plane_state_old;
> >>>>> unsigned int awidth;
> >>>>> + struct amdgpu_device *adev = plane->dev->dev_private;
> >>>>> + bool gtt_scannable = (adev->asic_type >= CHIP_CARRIZO &&
> adev-
> >>>>> flags
> >>>>> +& AMD_IS_APU);
> >>>>>
> >>>>> dm_plane_state_old = to_dm_plane_state(plane->state);
> >>>>> dm_plane_state_new = to_dm_plane_state(new_state);
> @@ -
> >>>> 3005,7
> >>>>> +3007,11 @@ static int dm_plane_helper_prepare_fb(struct
> drm_plane
> >>>> *plane,
> >>>>> return r;
> >>>>>
> >>>>> r = amdgpu_bo_pin(rbo, AMDGPU_GEM_DOMAIN_VRAM,
> &afb-
> >> address);
> >>>>> -
> >>>>> + if (unlikely(r != 0)) {
> >>>>> + /* latest APUs support gtt scan out */
> >>>>> + if (gtt_scannable)
> >>>>> + r = amdgpu_bo_pin(rbo,
> >>>> AMDGPU_GEM_DOMAIN_GTT, &afb->address);
> >>>>> + }
> >>>>>
> >>>>> amdgpu_bo_unreserve(rbo);
> >>>>>
More information about the amd-gfx
mailing list