diff options
Diffstat (limited to 'drivers/gpu/drm/i915/gem')
| -rw-r--r-- | drivers/gpu/drm/i915/gem/i915_gem_context.c | 24 | ||||
| -rw-r--r-- | drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 59 | ||||
| -rw-r--r-- | drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/i915/gem/selftests/huge_pages.c | 8 |
4 files changed, 71 insertions, 22 deletions
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c index 7f2831efc798..6250de9b9196 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c @@ -1688,6 +1688,10 @@ void i915_gem_init__contexts(struct drm_i915_private *i915) init_contexts(&i915->gem.contexts); } +/* + * Note that this implicitly consumes the ctx reference, by placing + * the ctx in the context_xa. + */ static void gem_context_register(struct i915_gem_context *ctx, struct drm_i915_file_private *fpriv, u32 id) @@ -1703,10 +1707,6 @@ static void gem_context_register(struct i915_gem_context *ctx, snprintf(ctx->name, sizeof(ctx->name), "%s[%d]", current->comm, pid_nr(ctx->pid)); - /* And finally expose ourselves to userspace via the idr */ - old = xa_store(&fpriv->context_xa, id, ctx, GFP_KERNEL); - WARN_ON(old); - spin_lock(&ctx->client->ctx_lock); list_add_tail_rcu(&ctx->client_link, &ctx->client->ctx_list); spin_unlock(&ctx->client->ctx_lock); @@ -1714,6 +1714,10 @@ static void gem_context_register(struct i915_gem_context *ctx, spin_lock(&i915->gem.contexts.lock); list_add_tail(&ctx->link, &i915->gem.contexts.list); spin_unlock(&i915->gem.contexts.lock); + + /* And finally expose ourselves to userspace via the idr */ + old = xa_store(&fpriv->context_xa, id, ctx, GFP_KERNEL); + WARN_ON(old); } int i915_gem_context_open(struct drm_i915_private *i915, @@ -2199,14 +2203,22 @@ finalize_create_context_locked(struct drm_i915_file_private *file_priv, if (IS_ERR(ctx)) return ctx; + /* + * One for the xarray and one for the caller. We need to grab + * the reference *prior* to making the ctx visble to userspace + * in gem_context_register(), as at any point after that + * userspace can try to race us with another thread destroying + * the context under our feet. + */ + i915_gem_context_get(ctx); + gem_context_register(ctx, file_priv, id); old = xa_erase(&file_priv->proto_context_xa, id); GEM_BUG_ON(old != pc); proto_context_close(file_priv->dev_priv, pc); - /* One for the xarray and one for the caller */ - return i915_gem_context_get(ctx); + return ctx; } struct i915_gem_context * diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c index da09767fda07..f266b68cf012 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c @@ -730,32 +730,69 @@ static int eb_reserve(struct i915_execbuffer *eb) bool unpinned; /* - * Attempt to pin all of the buffers into the GTT. - * This is done in 2 phases: + * We have one more buffers that we couldn't bind, which could be due to + * various reasons. To resolve this we have 4 passes, with every next + * level turning the screws tighter: * - * 1. Unbind all objects that do not match the GTT constraints for - * the execbuffer (fenceable, mappable, alignment etc). - * 2. Bind new objects. + * 0. Unbind all objects that do not match the GTT constraints for the + * execbuffer (fenceable, mappable, alignment etc). Bind all new + * objects. This avoids unnecessary unbinding of later objects in order + * to make room for the earlier objects *unless* we need to defragment. * - * This avoid unnecessary unbinding of later objects in order to make - * room for the earlier objects *unless* we need to defragment. + * 1. Reorder the buffers, where objects with the most restrictive + * placement requirements go first (ignoring fixed location buffers for + * now). For example, objects needing the mappable aperture (the first + * 256M of GTT), should go first vs objects that can be placed just + * about anywhere. Repeat the previous pass. * - * Defragmenting is skipped if all objects are pinned at a fixed location. + * 2. Consider buffers that are pinned at a fixed location. Also try to + * evict the entire VM this time, leaving only objects that we were + * unable to lock. Try again to bind the buffers. (still using the new + * buffer order). + * + * 3. We likely have object lock contention for one or more stubborn + * objects in the VM, for which we need to evict to make forward + * progress (perhaps we are fighting the shrinker?). When evicting the + * VM this time around, anything that we can't lock we now track using + * the busy_bo, using the full lock (after dropping the vm->mutex to + * prevent deadlocks), instead of trylock. We then continue to evict the + * VM, this time with the stubborn object locked, which we can now + * hopefully unbind (if still bound in the VM). Repeat until the VM is + * evicted. Finally we should be able bind everything. */ - for (pass = 0; pass <= 2; pass++) { + for (pass = 0; pass <= 3; pass++) { int pin_flags = PIN_USER | PIN_VALIDATE; if (pass == 0) pin_flags |= PIN_NONBLOCK; if (pass >= 1) - unpinned = eb_unbind(eb, pass == 2); + unpinned = eb_unbind(eb, pass >= 2); if (pass == 2) { err = mutex_lock_interruptible(&eb->context->vm->mutex); if (!err) { - err = i915_gem_evict_vm(eb->context->vm, &eb->ww); + err = i915_gem_evict_vm(eb->context->vm, &eb->ww, NULL); + mutex_unlock(&eb->context->vm->mutex); + } + if (err) + return err; + } + + if (pass == 3) { +retry: + err = mutex_lock_interruptible(&eb->context->vm->mutex); + if (!err) { + struct drm_i915_gem_object *busy_bo = NULL; + + err = i915_gem_evict_vm(eb->context->vm, &eb->ww, &busy_bo); mutex_unlock(&eb->context->vm->mutex); + if (err && busy_bo) { + err = i915_gem_object_lock(busy_bo, &eb->ww); + i915_gem_object_put(busy_bo); + if (!err) + goto retry; + } } if (err) return err; diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c index c29efdef8313..0ad44f3868de 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c @@ -369,7 +369,7 @@ retry: if (vma == ERR_PTR(-ENOSPC)) { ret = mutex_lock_interruptible(&ggtt->vm.mutex); if (!ret) { - ret = i915_gem_evict_vm(&ggtt->vm, &ww); + ret = i915_gem_evict_vm(&ggtt->vm, &ww, NULL); mutex_unlock(&ggtt->vm.mutex); } if (ret) diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c index beaf27e09e8a..977dead10ab5 100644 --- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c +++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c @@ -1847,7 +1847,7 @@ static int igt_shrink_thp(void *arg) I915_SHRINK_ACTIVE); i915_vma_unpin(vma); if (err) - goto out_put; + goto out_wf; /* * Now that the pages are *unpinned* shrinking should invoke @@ -1863,19 +1863,19 @@ static int igt_shrink_thp(void *arg) pr_err("unexpected pages mismatch, should_swap=%s\n", str_yes_no(should_swap)); err = -EINVAL; - goto out_put; + goto out_wf; } if (should_swap == (obj->mm.page_sizes.sg || obj->mm.page_sizes.phys)) { pr_err("unexpected residual page-size bits, should_swap=%s\n", str_yes_no(should_swap)); err = -EINVAL; - goto out_put; + goto out_wf; } err = i915_vma_pin(vma, 0, 0, flags); if (err) - goto out_put; + goto out_wf; while (n--) { err = cpu_check(obj, n, 0xdeadbeaf); |