From 6911a9b8ae8b2a1dab4dfda9c2bd20f7ca2961d6 Mon Sep 17 00:00:00 2001 From: Ben Gamari Date: Thu, 2 Apr 2009 11:24:54 -0700 Subject: drm/i915: Implement batch and ring buffer dumping We create a debugfs node (i915_ringbuffer_data) to expose a hex dump of the ring buffer itself. We also expose another debugfs node (i915_ringbuffer_info) with information on the state (i.e. head, tail addresses) of the ringbuffer. For batchbuffer dumping, we look at the device's active_list, dumping each object which has I915_GEM_DOMAIN_COMMAND in its read domains. This is all exposed through the dri/i915_batchbuffers debugfs file with a header for each object (giving the objects gtt_offset so that it can be matched against the offset given in the BATCH_BUFFER_START command. Signed-off-by: Ben Gamari Signed-off-by: Carl Worth Signed-off-by: Eric Anholt --- drivers/gpu/drm/i915/i915_gem.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) (limited to 'drivers/gpu/drm/i915/i915_gem.c') diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 1449b452cc63..33ab07b0d712 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -43,8 +43,6 @@ static int i915_gem_object_set_cpu_read_domain_range(struct drm_gem_object *obj, uint64_t offset, uint64_t size); static void i915_gem_object_set_to_full_cpu_read_domain(struct drm_gem_object *obj); -static int i915_gem_object_get_pages(struct drm_gem_object *obj); -static void i915_gem_object_put_pages(struct drm_gem_object *obj); static int i915_gem_object_wait_rendering(struct drm_gem_object *obj); static int i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment); @@ -1285,7 +1283,7 @@ i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data, return 0; } -static void +void i915_gem_object_put_pages(struct drm_gem_object *obj) { struct drm_i915_gem_object *obj_priv = obj->driver_private; @@ -1884,7 +1882,7 @@ i915_gem_evict_everything(struct drm_device *dev) return ret; } -static int +int i915_gem_object_get_pages(struct drm_gem_object *obj) { struct drm_i915_gem_object *obj_priv = obj->driver_private; @@ -3243,7 +3241,7 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, exec_offset = exec_list[args->buffer_count - 1].offset; #if WATCH_EXEC - i915_gem_dump_object(object_list[args->buffer_count - 1], + i915_gem_dump_object(batch_obj, args->batch_len, __func__, ~0); -- cgit v1.2.3 From 2bc43b5cf5158a26fa1328234795abed2dff5275 Mon Sep 17 00:00:00 2001 From: Florian Mickler Date: Mon, 6 Apr 2009 22:55:41 +0200 Subject: drm/i915: Fix use of uninitialized var in 40a5f0de i915_gem_put_relocs_to_user returned an uninitialized value which got returned to userspace. This caused libdrm in my setup to never get out of a do{}while() loop retrying i915_gem_execbuffer. result was hanging X, overheating of cpu and 2-3gb of logfile-spam. This patch adresses the issue by 1. initializing vars in this file where necessary 2. correcting wrongly interpreted return values of copy_[from/to]_user Signed-off-by: Florian Mickler [anholt: cleanups of unnecessary changes, consistency in APIs] Signed-off-by: Eric Anholt --- drivers/gpu/drm/i915/i915_gem.c | 34 ++++++++++++++++++++++------------ 1 file changed, 22 insertions(+), 12 deletions(-) (limited to 'drivers/gpu/drm/i915/i915_gem.c') diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 33ab07b0d712..6f7d0e27036f 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -141,15 +141,18 @@ fast_shmem_read(struct page **pages, int length) { char __iomem *vaddr; - int ret; + int unwritten; vaddr = kmap_atomic(pages[page_base >> PAGE_SHIFT], KM_USER0); if (vaddr == NULL) return -ENOMEM; - ret = __copy_to_user_inatomic(data, vaddr + page_offset, length); + unwritten = __copy_to_user_inatomic(data, vaddr + page_offset, length); kunmap_atomic(vaddr, KM_USER0); - return ret; + if (unwritten) + return -EFAULT; + + return 0; } static inline int @@ -3000,13 +3003,13 @@ i915_gem_get_relocs_from_user(struct drm_i915_gem_exec_object *exec_list, drm_free(*relocs, reloc_count * sizeof(**relocs), DRM_MEM_DRIVER); *relocs = NULL; - return ret; + return -EFAULT; } reloc_index += exec_list[i].relocation_count; } - return ret; + return 0; } static int @@ -3015,23 +3018,28 @@ i915_gem_put_relocs_to_user(struct drm_i915_gem_exec_object *exec_list, struct drm_i915_gem_relocation_entry *relocs) { uint32_t reloc_count = 0, i; - int ret; + int ret = 0; for (i = 0; i < buffer_count; i++) { struct drm_i915_gem_relocation_entry __user *user_relocs; + int unwritten; user_relocs = (void __user *)(uintptr_t)exec_list[i].relocs_ptr; - if (ret == 0) { - ret = copy_to_user(user_relocs, - &relocs[reloc_count], - exec_list[i].relocation_count * - sizeof(*relocs)); + unwritten = copy_to_user(user_relocs, + &relocs[reloc_count], + exec_list[i].relocation_count * + sizeof(*relocs)); + + if (unwritten) { + ret = -EFAULT; + goto err; } reloc_count += exec_list[i].relocation_count; } +err: drm_free(relocs, reloc_count * sizeof(*relocs), DRM_MEM_DRIVER); return ret; @@ -3306,10 +3314,12 @@ err: (uintptr_t) args->buffers_ptr, exec_list, sizeof(*exec_list) * args->buffer_count); - if (ret) + if (ret) { + ret = -EFAULT; DRM_ERROR("failed to copy %d exec entries " "back to user (%d)\n", args->buffer_count, ret); + } } /* Copy the updated relocations out regardless of current error -- cgit v1.2.3 From e5e9ecde63ba365b510df0f4a9cb3b048a0ad785 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Tue, 7 Apr 2009 16:01:22 -0700 Subject: drm/i915: Correctly set the write flag for get_user_pages in pread. Otherwise, the results of our read didn't show up when we were faulting in the page being read into (as happened with a testcase reading into a big stack area). Likely accounts for some conformance test failures. Signed-off-by: Eric Anholt --- drivers/gpu/drm/i915/i915_gem.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/i915/i915_gem.c') diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 6f7d0e27036f..3a1189d94a9a 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -287,7 +287,7 @@ i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj, down_read(&mm->mmap_sem); pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr, - num_pages, 0, 0, user_pages, NULL); + num_pages, 1, 0, user_pages, NULL); up_read(&mm->mmap_sem); if (pinned_pages < num_pages) { ret = -EFAULT; -- cgit v1.2.3 From 280b713b5b0fd84cf2469098aee88acbb5de859c Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Thu, 12 Mar 2009 16:56:27 -0700 Subject: drm/i915: Allow tiling of objects with bit 17 swizzling by the CPU. Save the bit 17 state of the pages when freeing the page list, and reswizzle them if necessary when rebinding the pages (in case they were swapped out). Since we have userland with expectations that the swizzle enums let it pread and pwrite contents accurately, we can't expose a new swizzle enum for bit 17 (which it would have to GTT map to handle), so we handle it down in pread and pwrite by swizzling the copy when bit 17 of the page address is set. Signed-off-by: Eric Anholt --- drivers/gpu/drm/i915/i915_gem.c | 130 ++++++++++++++++++++++++++++++++++++---- 1 file changed, 118 insertions(+), 12 deletions(-) (limited to 'drivers/gpu/drm/i915/i915_gem.c') diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 3a1189d94a9a..6dca9fc7c1db 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -155,6 +155,15 @@ fast_shmem_read(struct page **pages, return 0; } +static int i915_gem_object_needs_bit17_swizzle(struct drm_gem_object *obj) +{ + drm_i915_private_t *dev_priv = obj->dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + return dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_9_10_17 && + obj_priv->tiling_mode != I915_TILING_NONE; +} + static inline int slow_shmem_copy(struct page *dst_page, int dst_offset, @@ -182,6 +191,64 @@ slow_shmem_copy(struct page *dst_page, return 0; } +static inline int +slow_shmem_bit17_copy(struct page *gpu_page, + int gpu_offset, + struct page *cpu_page, + int cpu_offset, + int length, + int is_read) +{ + char *gpu_vaddr, *cpu_vaddr; + + /* Use the unswizzled path if this page isn't affected. */ + if ((page_to_phys(gpu_page) & (1 << 17)) == 0) { + if (is_read) + return slow_shmem_copy(cpu_page, cpu_offset, + gpu_page, gpu_offset, length); + else + return slow_shmem_copy(gpu_page, gpu_offset, + cpu_page, cpu_offset, length); + } + + gpu_vaddr = kmap_atomic(gpu_page, KM_USER0); + if (gpu_vaddr == NULL) + return -ENOMEM; + + cpu_vaddr = kmap_atomic(cpu_page, KM_USER1); + if (cpu_vaddr == NULL) { + kunmap_atomic(gpu_vaddr, KM_USER0); + return -ENOMEM; + } + + /* Copy the data, XORing A6 with A17 (1). The user already knows he's + * XORing with the other bits (A9 for Y, A9 and A10 for X) + */ + while (length > 0) { + int cacheline_end = ALIGN(gpu_offset + 1, 64); + int this_length = min(cacheline_end - gpu_offset, length); + int swizzled_gpu_offset = gpu_offset ^ 64; + + if (is_read) { + memcpy(cpu_vaddr + cpu_offset, + gpu_vaddr + swizzled_gpu_offset, + this_length); + } else { + memcpy(gpu_vaddr + swizzled_gpu_offset, + cpu_vaddr + cpu_offset, + this_length); + } + cpu_offset += this_length; + gpu_offset += this_length; + length -= this_length; + } + + kunmap_atomic(cpu_vaddr, KM_USER1); + kunmap_atomic(gpu_vaddr, KM_USER0); + + return 0; +} + /** * This is the fast shmem pread path, which attempts to copy_from_user directly * from the backing pages of the object to the user's address space. On a @@ -270,6 +337,7 @@ i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj, int page_length; int ret; uint64_t data_ptr = args->data_ptr; + int do_bit17_swizzling; remain = args->size; @@ -294,6 +362,8 @@ i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj, goto fail_put_user_pages; } + do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); + mutex_lock(&dev->struct_mutex); ret = i915_gem_object_get_pages(obj); @@ -328,11 +398,20 @@ i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj, if ((data_page_offset + page_length) > PAGE_SIZE) page_length = PAGE_SIZE - data_page_offset; - ret = slow_shmem_copy(user_pages[data_page_index], - data_page_offset, - obj_priv->pages[shmem_page_index], - shmem_page_offset, - page_length); + if (do_bit17_swizzling) { + ret = slow_shmem_bit17_copy(obj_priv->pages[shmem_page_index], + shmem_page_offset, + user_pages[data_page_index], + data_page_offset, + page_length, + 1); + } else { + ret = slow_shmem_copy(user_pages[data_page_index], + data_page_offset, + obj_priv->pages[shmem_page_index], + shmem_page_offset, + page_length); + } if (ret) goto fail_put_pages; @@ -384,9 +463,14 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data, return -EINVAL; } - ret = i915_gem_shmem_pread_fast(dev, obj, args, file_priv); - if (ret != 0) + if (i915_gem_object_needs_bit17_swizzle(obj)) { ret = i915_gem_shmem_pread_slow(dev, obj, args, file_priv); + } else { + ret = i915_gem_shmem_pread_fast(dev, obj, args, file_priv); + if (ret != 0) + ret = i915_gem_shmem_pread_slow(dev, obj, args, + file_priv); + } drm_gem_object_unreference(obj); @@ -728,6 +812,7 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj, int page_length; int ret; uint64_t data_ptr = args->data_ptr; + int do_bit17_swizzling; remain = args->size; @@ -752,6 +837,8 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj, goto fail_put_user_pages; } + do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); + mutex_lock(&dev->struct_mutex); ret = i915_gem_object_get_pages(obj); @@ -786,11 +873,20 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj, if ((data_page_offset + page_length) > PAGE_SIZE) page_length = PAGE_SIZE - data_page_offset; - ret = slow_shmem_copy(obj_priv->pages[shmem_page_index], - shmem_page_offset, - user_pages[data_page_index], - data_page_offset, - page_length); + if (do_bit17_swizzling) { + ret = slow_shmem_bit17_copy(obj_priv->pages[shmem_page_index], + shmem_page_offset, + user_pages[data_page_index], + data_page_offset, + page_length, + 0); + } else { + ret = slow_shmem_copy(obj_priv->pages[shmem_page_index], + shmem_page_offset, + user_pages[data_page_index], + data_page_offset, + page_length); + } if (ret) goto fail_put_pages; @@ -855,6 +951,8 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, ret = i915_gem_gtt_pwrite_slow(dev, obj, args, file_priv); } + } else if (i915_gem_object_needs_bit17_swizzle(obj)) { + ret = i915_gem_shmem_pwrite_slow(dev, obj, args, file_priv); } else { ret = i915_gem_shmem_pwrite_fast(dev, obj, args, file_priv); if (ret == -EFAULT) { @@ -1298,6 +1396,9 @@ i915_gem_object_put_pages(struct drm_gem_object *obj) if (--obj_priv->pages_refcount != 0) return; + if (obj_priv->tiling_mode != I915_TILING_NONE) + i915_gem_object_save_bit_17_swizzle(obj); + for (i = 0; i < page_count; i++) if (obj_priv->pages[i] != NULL) { if (obj_priv->dirty) @@ -1923,6 +2024,10 @@ i915_gem_object_get_pages(struct drm_gem_object *obj) } obj_priv->pages[i] = page; } + + if (obj_priv->tiling_mode != I915_TILING_NONE) + i915_gem_object_do_bit_17_swizzle(obj); + return 0; } @@ -3601,6 +3706,7 @@ void i915_gem_free_object(struct drm_gem_object *obj) i915_gem_free_mmap_offset(obj); drm_free(obj_priv->page_cpu_valid, 1, DRM_MEM_DRIVER); + kfree(obj_priv->bit_17); drm_free(obj->driver_private, 1, DRM_MEM_DRIVER); } -- cgit v1.2.3 From 68c84342171034120c8a1f6dfb8ef51b14250f11 Mon Sep 17 00:00:00 2001 From: Shaohua Li Date: Wed, 8 Apr 2009 10:58:23 +0800 Subject: drm/i915: fix scheduling while holding the new active list spinlock regression caused by commit 5e118f4139feafe97e913df67b1f7c1e5083e535: i915_gem_object_move_to_inactive() should be called in task context, as it calls fput(); Signed-off-by: Shaohua Li [anholt: Add more detail to the comment about the lock break that's added] Signed-off-by: Eric Anholt --- drivers/gpu/drm/i915/i915_gem.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/i915/i915_gem.c') diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 6dca9fc7c1db..4642115902d6 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -1596,8 +1596,19 @@ i915_gem_retire_request(struct drm_device *dev, if (obj->write_domain != 0) i915_gem_object_move_to_flushing(obj); - else + else { + /* Take a reference on the object so it won't be + * freed while the spinlock is held. The list + * protection for this spinlock is safe when breaking + * the lock like this since the next thing we do + * is just get the head of the list again. + */ + drm_gem_object_reference(obj); i915_gem_object_move_to_inactive(obj); + spin_unlock(&dev_priv->mm.active_list_lock); + drm_gem_object_unreference(obj); + spin_lock(&dev_priv->mm.active_list_lock); + } } out: spin_unlock(&dev_priv->mm.active_list_lock); -- cgit v1.2.3 From d816f6ac4f9877509baa3610be0d25b7afc29c00 Mon Sep 17 00:00:00 2001 From: Wu Fengguang Date: Sat, 18 Apr 2009 10:43:32 +0800 Subject: drm/i915: fix unpaired i915 device mutex on entervt failure. Signed-off-by: Wu Fengguang Signed-off-by: Eric Anholt --- drivers/gpu/drm/i915/i915_gem.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/i915/i915_gem.c') diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 4642115902d6..ee896d91c5bc 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -4087,8 +4087,10 @@ i915_gem_entervt_ioctl(struct drm_device *dev, void *data, dev_priv->mm.suspended = 0; ret = i915_gem_init_ringbuffer(dev); - if (ret != 0) + if (ret != 0) { + mutex_unlock(&dev->struct_mutex); return ret; + } spin_lock(&dev_priv->mm.active_list_lock); BUG_ON(!list_empty(&dev_priv->mm.active_list)); -- cgit v1.2.3