[PATCH] drm/i915/blt: Remove recursive vma->lock

[PATCH] drm/i915/blt: Remove recursive vma->lock

[Chris Wilson]

As we have already plugged the w->dma into the reservation_object, and
have set ourselves up to automatically signal the request and w->dma on
completion, we do not need to export the rq->fence directly and just use
the w->dma fence.

This avoids having to take the reservation_lock inside the worker which
cross-release lockdep would complain about. :)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
---
Patch based on the i915_active overhaul, but on the face of it should be
safe without. But let's pretend it does need the overhaul as this is one
of the motivations :)
-Chris
---
 drivers/gpu/drm/i915/gem/i915_gem_client_blt.c        | 10 ++++++----
 .../gpu/drm/i915/gem/selftests/i915_gem_client_blt.c  | 11 -----------
 2 files changed, 6 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c b/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c
index f253ec5765ad..83b5c5d13b93 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c
@@ -193,10 +193,12 @@ static void clear_pages_worker(struct work_struct *work)
 			goto out_request;
 	}
 
-	/* XXX: more feverish nightmares await */
-	i915_vma_lock(vma);
-	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
-	i915_vma_unlock(vma);
+	/*
+	 * w->dma is already exported via (vma|obj)->resv we need only
+	 * keep track of the GPU activity within this vma/request, and
+	 * propagate the signal from the request to w->dma.
+	 */
+	err = i915_active_ref(&vma->active, rq->fence.context, &rq->fence);
 	if (err)
 		goto out_request;
 
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
index f3a5eb807c1c..855481252bda 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
@@ -63,17 +63,6 @@ static int igt_client_fill(void *arg)
 		if (err)
 			goto err_unpin;
 
-		/*
-		 * XXX: For now do the wait without the object resv lock to
-		 * ensure we don't deadlock.
-		 */
-		err = i915_gem_object_wait(obj,
-					   I915_WAIT_INTERRUPTIBLE |
-					   I915_WAIT_ALL,
-					   MAX_SCHEDULE_TIMEOUT);
-		if (err)
-			goto err_unpin;
-
 		i915_gem_object_lock(obj);
 		err = i915_gem_object_set_to_cpu_domain(obj, false);
 		i915_gem_object_unlock(obj);
-- 
2.20.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✗ Fi.CI.BAT: failure for drm/i915/blt: Remove recursive vma->lock

[Patchwork]

== Series Details ==

Series: drm/i915/blt: Remove recursive vma->lock
URL   : https://patchwork.freedesktop.org/series/62290/
State : failure

== Summary ==

CALL    scripts/checksyscalls.sh
  CALL    scripts/atomic/check-atomics.sh
  DESCEND  objtool
  CHK     include/generated/compile.h
  AR      drivers/gpu/drm/i915/built-in.a
  CC [M]  drivers/gpu/drm/i915/header_test_i915_active_types.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_debugfs.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_drv.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_irq.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_params.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_priolist_types.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_reg.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_scheduler_types.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_timeline_types.o
  CC [M]  drivers/gpu/drm/i915/header_test_i915_utils.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_csr.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_drv.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_pm.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_runtime_pm.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_sideband.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_uncore.o
  CC [M]  drivers/gpu/drm/i915/header_test_intel_wakeref.o
  CC [M]  drivers/gpu/drm/i915/gem/i915_gem_client_blt.o
drivers/gpu/drm/i915/gem/i915_gem_client_blt.c: In function ‘clear_pages_worker’:
drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:201:57: error: passing argument 3 of ‘i915_active_ref’ from incompatible pointer type [-Werror=incompatible-pointer-types]
  err = i915_active_ref(&vma->active, rq->fence.context, &rq->fence);
                                                         ^
In file included from drivers/gpu/drm/i915/gem/i915_gem_object_types.h:14:0,
                 from drivers/gpu/drm/i915/gem/i915_gem_object.h:16,
                 from drivers/gpu/drm/i915/display/intel_frontbuffer.h:27,
                 from drivers/gpu/drm/i915/i915_drv.h:70,
                 from drivers/gpu/drm/i915/intel_drv.h:45,
                 from drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:8:
drivers/gpu/drm/i915/i915_active.h:376:5: note: expected ‘struct i915_request *’ but argument is of type ‘struct dma_fence *’
 int i915_active_ref(struct i915_active *ref,
     ^~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
scripts/Makefile.build:278: recipe for target 'drivers/gpu/drm/i915/gem/i915_gem_client_blt.o' failed
make[4]: *** [drivers/gpu/drm/i915/gem/i915_gem_client_blt.o] Error 1
scripts/Makefile.build:489: recipe for target 'drivers/gpu/drm/i915' failed
make[3]: *** [drivers/gpu/drm/i915] Error 2
scripts/Makefile.build:489: recipe for target 'drivers/gpu/drm' failed
make[2]: *** [drivers/gpu/drm] Error 2
scripts/Makefile.build:489: recipe for target 'drivers/gpu' failed
make[1]: *** [drivers/gpu] Error 2
Makefile:1071: recipe for target 'drivers' failed
make: *** [drivers] Error 2

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915/blt: Remove recursive vma->lock

[Matthew Auld]

On Tue, 18 Jun 2019 at 10:53, Chris Wilson <chris@chris-wilson.co.uk> wrote:
>
> As we have already plugged the w->dma into the reservation_object, and
> have set ourselves up to automatically signal the request and w->dma on
> completion, we do not need to export the rq->fence directly and just use
> the w->dma fence.
>
> This avoids having to take the reservation_lock inside the worker which
> cross-release lockdep would complain about. :)
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Matthew Auld <matthew.auld@intel.com>
> ---
> Patch based on the i915_active overhaul, but on the face of it should be
> safe without. But let's pretend it does need the overhaul as this is one
> of the motivations :)
> -Chris
> ---
>  drivers/gpu/drm/i915/gem/i915_gem_client_blt.c        | 10 ++++++----
>  .../gpu/drm/i915/gem/selftests/i915_gem_client_blt.c  | 11 -----------
>  2 files changed, 6 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c b/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c
> index f253ec5765ad..83b5c5d13b93 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_client_blt.c
> @@ -193,10 +193,12 @@ static void clear_pages_worker(struct work_struct *work)
>                         goto out_request;
>         }
>
> -       /* XXX: more feverish nightmares await */
> -       i915_vma_lock(vma);
> -       err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
> -       i915_vma_unlock(vma);
> +       /*
> +        * w->dma is already exported via (vma|obj)->resv we need only
> +        * keep track of the GPU activity within this vma/request, and
> +        * propagate the signal from the request to w->dma.
> +        */
> +       err = i915_active_ref(&vma->active, rq->fence.context, &rq->fence);
>         if (err)
>                 goto out_request;

Oh, wow ~o~

Yeah, seems to work locally with the extra ref + active count
tracking, so I guess should fit nicely with your i915_active.acquire,
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915/blt: Remove recursive vma->lock

[kbuild test robot]

Hi Chris,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on drm-intel/for-linux-next]
[also build test WARNING on next-20190618]
[cannot apply to v5.2-rc5]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Chris-Wilson/drm-i915-blt-Remove-recursive-vma-lock/20190618-194749
base:   git://anongit.freedesktop.org/drm-intel for-linux-next
reproduce:
        # apt-get install sparse
        # sparse version: v0.6.1-rc1-7-g2b96cd8-dirty
        make ARCH=x86_64 allmodconfig
        make C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__'

If you fix the issue, kindly add following tag
Reported-by: kbuild test robot <lkp@intel.com>


sparse warnings: (new ones prefixed by >>)

>> drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:201:65: sparse: sparse: incorrect type in argument 3 (different base types) @@    expected struct i915_request *rq @@    got uct i915_request *rq @@
>> drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:201:65: sparse:    expected struct i915_request *rq
>> drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:201:65: sparse:    got struct dma_fence *
   include/linux/reservation.h:220:20: sparse: sparse: dereference of noderef expression
   include/linux/reservation.h:220:45: sparse: sparse: dereference of noderef expression
   include/linux/reservation.h:220:20: sparse: sparse: dereference of noderef expression
   include/linux/reservation.h:220:45: sparse: sparse: dereference of noderef expression

vim +201 drivers/gpu/drm/i915/gem/i915_gem_client_blt.c

   153	
   154	static void clear_pages_worker(struct work_struct *work)
   155	{
   156		struct clear_pages_work *w = container_of(work, typeof(*w), work);
   157		struct drm_i915_private *i915 = w->ce->gem_context->i915;
   158		struct drm_i915_gem_object *obj = w->sleeve->obj;
   159		struct i915_vma *vma = w->sleeve->vma;
   160		struct i915_request *rq;
   161		int err = w->dma.error;
   162	
   163		if (unlikely(err))
   164			goto out_signal;
   165	
   166		if (obj->cache_dirty) {
   167			obj->write_domain = 0;
   168			if (i915_gem_object_has_struct_page(obj))
   169				drm_clflush_sg(w->sleeve->pages);
   170			obj->cache_dirty = false;
   171		}
   172	
   173		/* XXX: we need to kill this */
   174		mutex_lock(&i915->drm.struct_mutex);
   175		err = i915_vma_pin(vma, 0, 0, PIN_USER);
   176		if (unlikely(err))
   177			goto out_unlock;
   178	
   179		rq = i915_request_create(w->ce);
   180		if (IS_ERR(rq)) {
   181			err = PTR_ERR(rq);
   182			goto out_unpin;
   183		}
   184	
   185		/* There's no way the fence has signalled */
   186		if (dma_fence_add_callback(&rq->fence, &w->cb,
   187					   clear_pages_dma_fence_cb))
   188			GEM_BUG_ON(1);
   189	
   190		if (w->ce->engine->emit_init_breadcrumb) {
   191			err = w->ce->engine->emit_init_breadcrumb(rq);
   192			if (unlikely(err))
   193				goto out_request;
   194		}
   195	
   196		/*
   197		 * w->dma is already exported via (vma|obj)->resv we need only
   198		 * keep track of the GPU activity within this vma/request, and
   199		 * propagate the signal from the request to w->dma.
   200		 */
 > 201		err = i915_active_ref(&vma->active, rq->fence.context, &rq->fence);
   202		if (err)
   203			goto out_request;
   204	
   205		err = intel_emit_vma_fill_blt(rq, vma, w->value);
   206	out_request:
   207		if (unlikely(err)) {
   208			i915_request_skip(rq, err);
   209			err = 0;
   210		}
   211	
   212		i915_request_add(rq);
   213	out_unpin:
   214		i915_vma_unpin(vma);
   215	out_unlock:
   216		mutex_unlock(&i915->drm.struct_mutex);
   217	out_signal:
   218		if (unlikely(err)) {
   219			dma_fence_set_error(&w->dma, err);
   220			dma_fence_signal(&w->dma);
   221			dma_fence_put(&w->dma);
   222		}
   223	}
   224	

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915/blt: Remove recursive vma->lock

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 4216 bytes --]

Hi Chris,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on drm-intel/for-linux-next]
[also build test ERROR on next-20190618]
[cannot apply to v5.2-rc5]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Chris-Wilson/drm-i915-blt-Remove-recursive-vma-lock/20190618-194749
base:   git://anongit.freedesktop.org/drm-intel for-linux-next
config: x86_64-rhel (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=x86_64 

If you fix the issue, kindly add following tag
Reported-by: kbuild test robot <lkp@intel.com>

All errors (new ones prefixed by >>):

   drivers/gpu/drm/i915/gem/i915_gem_client_blt.c: In function 'clear_pages_worker':
>> drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:201:57: error: passing argument 3 of 'i915_active_ref' from incompatible pointer type [-Werror=incompatible-pointer-types]
     err = i915_active_ref(&vma->active, rq->fence.context, &rq->fence);
                                                            ^
   In file included from drivers/gpu/drm/i915/i915_timeline.h:30:0,
                    from drivers/gpu/drm/i915/gt/intel_engine.h:17,
                    from drivers/gpu/drm/i915/i915_drv.h:67,
                    from drivers/gpu/drm/i915/intel_drv.h:45,
                    from drivers/gpu/drm/i915/gem/i915_gem_client_blt.c:8:
   drivers/gpu/drm/i915/i915_active.h:376:5: note: expected 'struct i915_request *' but argument is of type 'struct dma_fence *'
    int i915_active_ref(struct i915_active *ref,
        ^~~~~~~~~~~~~~~
   cc1: some warnings being treated as errors

vim +/i915_active_ref +201 drivers/gpu/drm/i915/gem/i915_gem_client_blt.c

   153	
   154	static void clear_pages_worker(struct work_struct *work)
   155	{
   156		struct clear_pages_work *w = container_of(work, typeof(*w), work);
   157		struct drm_i915_private *i915 = w->ce->gem_context->i915;
   158		struct drm_i915_gem_object *obj = w->sleeve->obj;
   159		struct i915_vma *vma = w->sleeve->vma;
   160		struct i915_request *rq;
   161		int err = w->dma.error;
   162	
   163		if (unlikely(err))
   164			goto out_signal;
   165	
   166		if (obj->cache_dirty) {
   167			obj->write_domain = 0;
   168			if (i915_gem_object_has_struct_page(obj))
   169				drm_clflush_sg(w->sleeve->pages);
   170			obj->cache_dirty = false;
   171		}
   172	
   173		/* XXX: we need to kill this */
   174		mutex_lock(&i915->drm.struct_mutex);
   175		err = i915_vma_pin(vma, 0, 0, PIN_USER);
   176		if (unlikely(err))
   177			goto out_unlock;
   178	
   179		rq = i915_request_create(w->ce);
   180		if (IS_ERR(rq)) {
   181			err = PTR_ERR(rq);
   182			goto out_unpin;
   183		}
   184	
   185		/* There's no way the fence has signalled */
   186		if (dma_fence_add_callback(&rq->fence, &w->cb,
   187					   clear_pages_dma_fence_cb))
   188			GEM_BUG_ON(1);
   189	
   190		if (w->ce->engine->emit_init_breadcrumb) {
   191			err = w->ce->engine->emit_init_breadcrumb(rq);
   192			if (unlikely(err))
   193				goto out_request;
   194		}
   195	
   196		/*
   197		 * w->dma is already exported via (vma|obj)->resv we need only
   198		 * keep track of the GPU activity within this vma/request, and
   199		 * propagate the signal from the request to w->dma.
   200		 */
 > 201		err = i915_active_ref(&vma->active, rq->fence.context, &rq->fence);
   202		if (err)
   203			goto out_request;
   204	
   205		err = intel_emit_vma_fill_blt(rq, vma, w->value);
   206	out_request:
   207		if (unlikely(err)) {
   208			i915_request_skip(rq, err);
   209			err = 0;
   210		}
   211	
   212		i915_request_add(rq);
   213	out_unpin:
   214		i915_vma_unpin(vma);
   215	out_unlock:
   216		mutex_unlock(&i915->drm.struct_mutex);
   217	out_signal:
   218		if (unlikely(err)) {
   219			dma_fence_set_error(&w->dma, err);
   220			dma_fence_signal(&w->dma);
   221			dma_fence_put(&w->dma);
   222		}
   223	}
   224	

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 42816 bytes --]

[-- Attachment #3: Type: text/plain, Size: 159 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx