Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes:
> Be consistent, and even when we know we had used a WC, flush the mapped
> object after writing into it. The flush understands the mapping type and
> will only clflush if !I915_MAP_WC, but will always insert a wmb [sfence]
> so that we can be sure that all writes are visible.
>
> v2: Add the unconditional wmb so we are know that we always flush the
> writes to memory/HW at that point.
>
> Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> Cc: Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx>
> ---
> drivers/gpu/drm/i915/gem/i915_gem_object_blt.c | 8 ++++++--
> drivers/gpu/drm/i915/gem/i915_gem_pages.c | 1 +
> drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c | 2 ++
> drivers/gpu/drm/i915/gem/selftests/igt_gem_utils.c | 2 ++
> drivers/gpu/drm/i915/gt/selftest_ring_submission.c | 2 ++
> drivers/gpu/drm/i915/gt/selftest_rps.c | 2 ++
> drivers/gpu/drm/i915/selftests/i915_request.c | 9 +++++++--
> 7 files changed, 22 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_blt.c b/drivers/gpu/drm/i915/gem/i915_gem_object_blt.c
> index 2fc7737ef5f4..f457d7130491 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_object_blt.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_blt.c
> @@ -78,10 +78,12 @@ struct i915_vma *intel_emit_vma_fill_blt(struct intel_context *ce,
> } while (rem);
>
> *cmd = MI_BATCH_BUFFER_END;
> - intel_gt_chipset_flush(ce->vm->gt);
>
> + i915_gem_object_flush_map(pool->obj);
> i915_gem_object_unpin_map(pool->obj);
Ok, so there is explicit wmb now in flush_map even
if bo is marketed as coherent. That wmb will then
make the writes to flushed past that point...
>
> + intel_gt_chipset_flush(ce->vm->gt);
and then the do the sfence for wcb. This pattern feels
solid for me.
If we would mark map for writes, we could do everything
in unpin_map.
Reviewed-by: Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx>
> +
> batch = i915_vma_instance(pool->obj, ce->vm, NULL);
> if (IS_ERR(batch)) {
> err = PTR_ERR(batch);
> @@ -289,10 +291,12 @@ struct i915_vma *intel_emit_vma_copy_blt(struct intel_context *ce,
> } while (rem);
>
> *cmd = MI_BATCH_BUFFER_END;
> - intel_gt_chipset_flush(ce->vm->gt);
>
> + i915_gem_object_flush_map(pool->obj);
> i915_gem_object_unpin_map(pool->obj);
>
> + intel_gt_chipset_flush(ce->vm->gt);
> +
> batch = i915_vma_instance(pool->obj, ce->vm, NULL);
> if (IS_ERR(batch)) {
> err = PTR_ERR(batch);
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> index 5d855fcd5c0f..189efcd58942 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> @@ -391,6 +391,7 @@ void __i915_gem_object_flush_map(struct drm_i915_gem_object *obj,
> GEM_BUG_ON(range_overflows_t(typeof(obj->base.size),
> offset, size, obj->base.size));
>
> + wmb(); /* let all previous writes be visible to HW */
> obj->mm.dirty = true;
>
> if (obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE)
> diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
> index 3f6079e1dfb6..87d7d8aa080f 100644
> --- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
> +++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
> @@ -158,6 +158,8 @@ static int wc_set(struct context *ctx, unsigned long offset, u32 v)
> return PTR_ERR(map);
>
> map[offset / sizeof(*map)] = v;
> +
> + __i915_gem_object_flush_map(ctx->obj, offset, sizeof(*map));
> i915_gem_object_unpin_map(ctx->obj);
>
> return 0;
> diff --git a/drivers/gpu/drm/i915/gem/selftests/igt_gem_utils.c b/drivers/gpu/drm/i915/gem/selftests/igt_gem_utils.c
> index 772d8cba7da9..226b5fa9b430 100644
> --- a/drivers/gpu/drm/i915/gem/selftests/igt_gem_utils.c
> +++ b/drivers/gpu/drm/i915/gem/selftests/igt_gem_utils.c
> @@ -83,6 +83,8 @@ igt_emit_store_dw(struct i915_vma *vma,
> offset += PAGE_SIZE;
> }
> *cmd = MI_BATCH_BUFFER_END;
> +
> + i915_gem_object_flush_map(obj);
> i915_gem_object_unpin_map(obj);
>
> intel_gt_chipset_flush(vma->vm->gt);
> diff --git a/drivers/gpu/drm/i915/gt/selftest_ring_submission.c b/drivers/gpu/drm/i915/gt/selftest_ring_submission.c
> index 9995faadd7e8..3350e7c995bc 100644
> --- a/drivers/gpu/drm/i915/gt/selftest_ring_submission.c
> +++ b/drivers/gpu/drm/i915/gt/selftest_ring_submission.c
> @@ -54,6 +54,8 @@ static struct i915_vma *create_wally(struct intel_engine_cs *engine)
> *cs++ = STACK_MAGIC;
>
> *cs++ = MI_BATCH_BUFFER_END;
> +
> + i915_gem_object_flush_map(obj);
> i915_gem_object_unpin_map(obj);
>
> vma->private = intel_context_create(engine); /* dummy residuals */
> diff --git a/drivers/gpu/drm/i915/gt/selftest_rps.c b/drivers/gpu/drm/i915/gt/selftest_rps.c
> index bfa1a15564f7..6275d69aa9cc 100644
> --- a/drivers/gpu/drm/i915/gt/selftest_rps.c
> +++ b/drivers/gpu/drm/i915/gt/selftest_rps.c
> @@ -727,6 +727,7 @@ int live_rps_frequency_cs(void *arg)
>
> err_vma:
> *cancel = MI_BATCH_BUFFER_END;
> + i915_gem_object_flush_map(vma->obj);
> i915_gem_object_unpin_map(vma->obj);
> i915_vma_unpin(vma);
> i915_vma_put(vma);
> @@ -868,6 +869,7 @@ int live_rps_frequency_srm(void *arg)
>
> err_vma:
> *cancel = MI_BATCH_BUFFER_END;
> + i915_gem_object_flush_map(vma->obj);
> i915_gem_object_unpin_map(vma->obj);
> i915_vma_unpin(vma);
> i915_vma_put(vma);
> diff --git a/drivers/gpu/drm/i915/selftests/i915_request.c b/drivers/gpu/drm/i915/selftests/i915_request.c
> index 15b1ca9f7a01..c191976e1d15 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_request.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_request.c
> @@ -816,10 +816,12 @@ static int recursive_batch_resolve(struct i915_vma *batch)
> return PTR_ERR(cmd);
>
> *cmd = MI_BATCH_BUFFER_END;
> - intel_gt_chipset_flush(batch->vm->gt);
>
> + __i915_gem_object_flush_map(batch->obj, 0, sizeof(*cmd));
> i915_gem_object_unpin_map(batch->obj);
>
> + intel_gt_chipset_flush(batch->vm->gt);
> +
> return 0;
> }
>
> @@ -1060,9 +1062,12 @@ static int live_sequential_engines(void *arg)
> I915_MAP_WC);
> if (!IS_ERR(cmd)) {
> *cmd = MI_BATCH_BUFFER_END;
> - intel_gt_chipset_flush(engine->gt);
>
> + __i915_gem_object_flush_map(request[idx]->batch->obj,
> + 0, sizeof(*cmd));
> i915_gem_object_unpin_map(request[idx]->batch->obj);
> +
> + intel_gt_chipset_flush(engine->gt);
> }
>
> i915_vma_put(request[idx]->batch);
> --
> 2.20.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
