Completion of i915 VMAs v2
Subsystems (2)
cover Completion of i915 VMAs v2
Sliced and diced the mega patch into tiny little pieces on the request of Daniel. Overall, I think it's a big improvement (TBD if it was worth the time and effort though)
Here I also drop vma->active and leave obj->active; also a request from Daniel.
Finally, I moved the virtual function stuff to the end of the series. I think Daniel requested that one too, but I can't remember.
Odds are with all this rebasing, I introduced new bugs. I've been a bit too preoccupied to check each patch thoroughly - but the end result is right, and works.
Ben Widawsky (29):
drm/i915: Create an init vm
drm/i915: Rework drop caches for checkpatch drm/i915: Make proper functions for VMs drm/i915: Use bound list for inactive shrink drm/i915: Add VM to pin
drm/i915: Use ggtt_vm to save some typing drm/i915: Update describe_obj
drm/i915: Rework __i915_gem_shrink drm/i915: thread address space through execbuf drm/i915: make caching operate on all address spaces drm/i915: BUG_ON put_pages later
drm/i915: make reset&hangcheck code VM aware drm/i915: clear domains for all objects on reset drm/i915: Restore PDEs on gtt restore drm/i915: Improve VMA comments
drm/i915: Cleanup more of VMA in destroy drm/i915: plumb VM into bind/unbind code drm/i915: Use new bind/unbind in eviction code drm/i915: turn bound_ggtt checks to bound_any drm/i915: Fix up map and fenceable for VMA drm/i915: mm_list is per VMA
drm/i915: Update error capture for VMs drm/i915: Add vma to list at creation drm/i915: create vmas at execbuf
drm/i915: Convert execbuf code to use vmas drm/i915: Convert active API to VMA drm/i915: Add bind/unbind object functions to VM drm/i915: Use the new vm [un]bind functions drm/i915: eliminate vm->insert_entries()
drivers/gpu/drm/i915/i915_debugfs.c | 68 +++-- drivers/gpu/drm/i915/i915_dma.c | 4 - drivers/gpu/drm/i915/i915_drv.h | 185 +++++++------ drivers/gpu/drm/i915/i915_gem.c | 420 ++++++++++++++++++++--------- drivers/gpu/drm/i915/i915_gem_context.c | 17 +- drivers/gpu/drm/i915/i915_gem_evict.c | 78 +++--- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 360 ++++++++++++++----------- drivers/gpu/drm/i915/i915_gem_gtt.c | 138 ++++++---- drivers/gpu/drm/i915/i915_gem_stolen.c | 10 +- drivers/gpu/drm/i915/i915_gem_tiling.c | 9 +- drivers/gpu/drm/i915/i915_gpu_error.c | 111 +++++--- drivers/gpu/drm/i915/i915_trace.h | 37 +-- drivers/gpu/drm/i915/intel_overlay.c | 2 +- drivers/gpu/drm/i915/intel_pm.c | 2 +- drivers/gpu/drm/i915/intel_ringbuffer.c | 8 +- 15 files changed, 904 insertions(+), 545 deletions(-)
Signature · 1 line
1.8.3.4
PATCH 1/29 drm/i915: Create an init vm
Move all the similar address space (VM) initialization code to one function. Until we have multiple VMs, there should only ever be 1 VM. The aliasing ppgtt is a special case without it's own VM (since it doesn't need it's own address space management).
--- drivers/gpu/drm/i915/i915_dma.c | 4 ---- drivers/gpu/drm/i915/i915_gem.c | 15 +++++++++++++-- 2 files changed, 13 insertions(+), 6 deletions(-)
drivers/gpu/drm/i915/i915_dma.c +0 -4
@@ -1499,10 +1499,6 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)i915_dump_device_info (dev_priv); -INIT_LIST_HEAD (&dev_priv->vm_list); -INIT_LIST_HEAD (&dev_priv->gtt.base.global_link); -list_add (&dev_priv->gtt.base.global_link, &dev_priv->vm_list); -if (i915_get_bridge_dev (dev)) { ret = -EIO ; goto free_priv;
drivers/gpu/drm/i915/i915_gem.c +13 -2
@@ -4319,6 +4319,16 @@ init_ring_lists(struct intel_ring_buffer *ring)INIT_LIST_HEAD (&ring->request_list); } +static void i915_init_vm (struct drm_i915_private *dev_priv, +struct i915_address_space *vm) +{ + vm->dev = dev_priv->dev; +INIT_LIST_HEAD (&vm->active_list); +INIT_LIST_HEAD (&vm->inactive_list); +INIT_LIST_HEAD (&vm->global_link); +list_add (&vm->global_link, &dev_priv->vm_list); +} +void i915_gem_load (struct drm_device *dev) {
@@ -4331,8 +4341,9 @@ i915_gem_load(struct drm_device *dev)SLAB_HWCACHE_ALIGN , NULL); -INIT_LIST_HEAD (&dev_priv->gtt.base.active_list); -INIT_LIST_HEAD (&dev_priv->gtt.base.inactive_list); +INIT_LIST_HEAD (&dev_priv->vm_list); +i915_init_vm (dev_priv, &dev_priv->gtt.base); +INIT_LIST_HEAD (&dev_priv->mm.unbound_list);INIT_LIST_HEAD (&dev_priv->mm.bound_list);INIT_LIST_HEAD (&dev_priv->mm.fence_list);
Signature · 1 line
1.8.3.4
PATCH 2/29 drm/i915: Rework drop caches for checkpatch
In reply to Ben WidawskyWith an upcoming change to bind, to make checkpatch happy and keep the code clean, we need to rework this code a bit.
This should have no functional impact.
--- drivers/gpu/drm/i915/i915_debugfs.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +7 -6
@@ -1781,12 +1781,13 @@ i915_drop_caches_set(void *data, u64 val)if (val &DROP_BOUND ) {list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) -if (obj->pin_count ==0 ) { - ret =i915_gem_object_unbind (obj); -if (ret) - goto unlock; - } + mm_list) { +if (obj->pin_count) +continue ; + ret =i915_gem_object_unbind (obj); +if (ret) + goto unlock; + } }if (val &DROP_UNBOUND ) {
Signature · 1 line
1.8.3.4
Re: [PATCH 02/29] drm/i915: Rework drop caches for checkpatch
Quoted 17 lines · Ben Widawsky
With an upcoming change to bind, to make checkpatch happy and keep the code clean, we need to rework this code a bit.
This should have no functional impact.
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +0
-0@@ -1781,12 +1781,13 @@ i915_drop_caches_set(void *data, u64 val)if (val &DROP_BOUND ) {
drivers/gpu/drm/i915/i915_debugfs.c i915_drop_caches_set(void *data, u64 val) · line 1786
Quoted 10 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-6list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) -if (obj->pin_count ==0 ) { - ret =i915_gem_object_unbind (obj); -if (ret) - goto unlock; - } + mm_list) { +if (obj->pin_count) +continue ;
Give me a newline here, and I'm sold.
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +4
-0+ ret =i915_gem_object_unbind (obj); +if (ret) + goto unlock; + } }
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 02/29] drm/i915: Rework drop caches for checkpatch
Quoted 17 lines · Chris Wilson
Quoted 16 lines · Ben Widawsky
With an upcoming change to bind, to make checkpatch happy and keep the code clean, we need to rework this code a bit.
This should have no functional impact.
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_debugfs.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +0
-0@@ -1781,12 +1781,13 @@ i915_drop_caches_set(void *data, u64 val)
Quoted 14 lines
drivers/gpu/drm/i915/i915_debugfs.c i915_drop_caches_set(void *data, u64 val) · line 1786
Quoted 11 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-6if (val &DROP_BOUND ) {list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) -if (obj->pin_count ==0 ) { - ret =i915_gem_object_unbind (obj); -if (ret) - goto unlock; - } + mm_list) { +if (obj->pin_count) +continue ;Give me a newline here, and I'm sold.
Got it. As you may have noticed, this was already fixed in patch 20 of the series.
Quoted 8 lines
Quoted 5 lines
drivers/gpu/drm/i915/i915_debugfs.c +4
-0+ ret =i915_gem_object_unbind (obj); +if (ret) + goto unlock; + } }Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
PATCH 3/29 drm/i915: Make proper functions for VMs
In reply to Ben WidawskyEarlier in the conversion sequence we attempted to quickly wedge in the transitional interface as static inlines.
Now that we're sure these interfaces are sane, for easier debug and to decrease code size (since many of these functions may be called quite a bit), make them real functions
While at it, kill off the set_color interface. We'll always have the VMA, or easily get to it.
--- drivers/gpu/drm/i915/i915_drv.h | 83 ++++++++++++++++------------------- drivers/gpu/drm/i915/i915_gem.c | 78 ++++++++++++++++++++++++++++++-- drivers/gpu/drm/i915/i915_gem_evict.c | 8 ++-- drivers/gpu/drm/i915/i915_gem_gtt.c | 2 +- 4 files changed, 118 insertions(+), 53 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +37 -46
@@ -1379,52 +1379,6 @@ struct drm_i915_gem_object {#define to_intel_bo (x) container_of(x, struct drm_i915_gem_object, base) -/* This is a temporary define to help transition us to real VMAs. If you see -* this, you're either reviewing code, or bisecting it. */ -static inline struct i915_vma * -__i915_gem_obj_to_vma (struct drm_i915_gem_object *obj) -{ -if (list_empty (&obj->vma_list)) -return NULL ; - return list_first_entry(&obj->vma_list, struct i915_vma, vma_link); -} - -/* Whether or not this object is currently mapped by the translation tables */ -static inline bool -i915_gem_obj_ggtt_bound (struct drm_i915_gem_object *o) -{ -struct i915_vma *vma =__i915_gem_obj_to_vma (o); -if (vma ==NULL ) -return false; -return drm_mm_node_allocated (&vma->node); -} - -/* Offset of the first PTE pointing to this object */ -static inline unsigned long -i915_gem_obj_ggtt_offset (struct drm_i915_gem_object *o) -{ -BUG_ON (list_empty (&o->vma_list)); -return __i915_gem_obj_to_vma (o)->node.start; -} - -/* The size used in the translation tables may be larger than the actual size of -* the object on GEN2/GEN3 because of the way tiling is handled. See -* i915_gem_get_gtt_size() for more details. -*/ -static inline unsigned long -i915_gem_obj_ggtt_size (struct drm_i915_gem_object *o) -{ -BUG_ON (list_empty (&o->vma_list)); -return __i915_gem_obj_to_vma (o)->node.size; -} - -static inline void -i915_gem_obj_ggtt_set_color (struct drm_i915_gem_object *o, -enum i915_cache_level color) -{ -__i915_gem_obj_to_vma (o)->node.color = color; -} -/** * Request queue structure. *
@@ -1886,6 +1840,43 @@ struct dma_buf *i915_gem_prime_export(struct drm_device *dev,void i915_gem_restore_fences (struct drm_device *dev); +unsigned long i915_gem_obj_offset (struct drm_i915_gem_object *o, +struct i915_address_space *vm); +bool i915_gem_obj_bound_any (struct drm_i915_gem_object *o); +bool i915_gem_obj_bound (struct drm_i915_gem_object *o, +struct i915_address_space *vm); +unsigned long i915_gem_obj_size (struct drm_i915_gem_object *o, +struct i915_address_space *vm); +struct i915_vma *i915_gem_obj_to_vma (struct drm_i915_gem_object *obj, +struct i915_address_space *vm); +/* Some GGTT VM helpers */ +#define obj_to_ggtt (obj) \ + (&((struct drm_i915_private *)(obj)->base.dev->dev_private)->gtt.base) +static inline bool i915_is_ggtt (struct i915_address_space *vm) +{ +struct i915_address_space *ggtt = + &((struct drm_i915_private *)(vm)->dev->dev_private)->gtt.base; +return vm == ggtt; +} + +static inline bool i915_gem_obj_ggtt_bound (struct drm_i915_gem_object *obj) +{ +return i915_gem_obj_bound (obj,obj_to_ggtt (obj)); +} + +static inline unsigned long +i915_gem_obj_ggtt_offset (struct drm_i915_gem_object *obj) +{ +return i915_gem_obj_offset (obj,obj_to_ggtt (obj)); +} + +static inline unsigned long +i915_gem_obj_ggtt_size (struct drm_i915_gem_object *obj) +{ +return i915_gem_obj_size (obj,obj_to_ggtt (obj)); +} +#undef obj_to_ggtt +/* i915_gem_context.c */ void i915_gem_context_init (struct drm_device *dev);void i915_gem_context_fini (struct drm_device *dev);
drivers/gpu/drm/i915/i915_gem.c +75 -3
99 lines not on this page · raw message
drivers/gpu/drm/i915/i915_gem_evict.c +5 -3
@@ -34,7 +34,9 @@static bool mark_free (struct drm_i915_gem_object *obj,struct list_head *unwind) { -struct i915_vma *vma =__i915_gem_obj_to_vma (obj); +struct drm_device *dev = obj->base.dev; +struct drm_i915_private *dev_priv = dev->dev_private; +struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base);if (obj->pin_count)return false;
@@ -109,7 +111,7 @@ none: obj =list_first_entry (&unwind_list, struct drm_i915_gem_object, exec_list); - vma =__i915_gem_obj_to_vma (obj); + vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); ret =drm_mm_scan_remove_block (&vma->node);BUG_ON (ret);
@@ -130,7 +132,7 @@ found: obj =list_first_entry (&unwind_list, struct drm_i915_gem_object, exec_list); - vma =__i915_gem_obj_to_vma (obj); + vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base);if (drm_mm_scan_remove_block (&vma->node)) {list_move (&obj->exec_list, &eviction_list);drm_gem_object_reference (&obj->base);
drivers/gpu/drm/i915/i915_gem_gtt.c +1 -1
@@ -662,7 +662,7 @@ void i915_gem_setup_global_gtt(struct drm_device *dev,/* Mark any preallocated objects as occupied */ list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list) { -struct i915_vma *vma =__i915_gem_obj_to_vma (obj); +struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base);int ret;DRM_DEBUG_KMS ("reserving preallocated space: %lx + %zx\n" ,i915_gem_obj_ggtt_offset (obj), obj->base.size);
Signature · 1 line
1.8.3.4
PATCH 4/29 drm/i915: Use bound list for inactive shrink
In reply to Ben WidawskyDo to the move active/inactive lists, it no longer makes sense to use them for shrinking, since shrinking isn't VM specific (such a need may also exist, but doesn't yet).
What we can do instead is use the global bound list to find all objects which aren't active.
--- drivers/gpu/drm/i915/i915_gem.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +6 -2
@@ -4614,7 +4614,6 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) struct drm_i915_private, mm.inactive_shrinker);struct drm_device *dev = dev_priv->dev; -struct i915_address_space *vm = &dev_priv->gtt.base;struct drm_i915_gem_object *obj;int nr_to_scan = sc->nr_to_scan;bool unlock = true;
@@ -4643,9 +4642,14 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc)list_for_each_entry (obj, &dev_priv->mm.unbound_list, global_list)if (obj->pages_pin_count ==0 ) cnt += obj->base.size >>PAGE_SHIFT ; -list_for_each_entry (obj, &vm->inactive_list, mm_list) + +list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list) { +if (obj->active) +continue ; +if (obj->pin_count ==0 && obj->pages_pin_count ==0 ) cnt += obj->base.size >>PAGE_SHIFT ; + }if (unlock)mutex_unlock (&dev->struct_mutex);
Signature · 1 line
1.8.3.4
PATCH 5/29 drm/i915: Add VM to pin
In reply to Ben WidawskyTo verbalize it, one can say, "pin an object into the given address space." The semantics of pinning remain the same otherwise.
Certain objects will always have to be bound into the global GTT. Therefore, global GTT is a special case, and keep a special interface around for it (i915_gem_obj_ggtt_pin).
v2: s/i915_gem_ggtt_pin/i915_gem_obj_ggtt_pin
--- drivers/gpu/drm/i915/i915_drv.h | 11 +++++++++++ drivers/gpu/drm/i915/i915_gem.c | 9 +++++---- drivers/gpu/drm/i915/i915_gem_context.c | 4 ++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 4 +++- drivers/gpu/drm/i915/intel_overlay.c | 2 +- drivers/gpu/drm/i915/intel_pm.c | 2 +- drivers/gpu/drm/i915/intel_ringbuffer.c | 8 ++++---- 7 files changed, 27 insertions(+), 13 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +11 -0
@@ -1690,6 +1690,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma);int __must_check i915_gem_object_pin (struct drm_i915_gem_object *obj, +struct i915_address_space *vm,uint32_t alignment,bool map_and_fenceable,bool nonblocking);
@@ -1875,6 +1876,16 @@ i915_gem_obj_ggtt_size(struct drm_i915_gem_object *obj) {return i915_gem_obj_size (obj,obj_to_ggtt (obj)); } + +static inline int __must_check +i915_gem_obj_ggtt_pin (struct drm_i915_gem_object *obj, +uint32_t alignment, +bool map_and_fenceable, +bool nonblocking) +{ +return i915_gem_object_pin (obj,obj_to_ggtt (obj), alignment, + map_and_fenceable, nonblocking); +}#undef obj_to_ggtt/* i915_gem_context.c */
drivers/gpu/drm/i915/i915_gem.c +5 -4
@@ -578,7 +578,7 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev,char __user *user_data;int page_offset, page_length, ret; - ret =i915_gem_object_pin (obj,0 , true, true); + ret =i915_gem_obj_ggtt_pin (obj,0 , true, true);if (ret) goto out;
@@ -1332,7 +1332,7 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) }/* Now bind it into the GTT if needed */ - ret =i915_gem_object_pin (obj,0 , true, false); + ret =i915_gem_obj_ggtt_pin (obj,0 , true, false);if (ret) goto unlock;
@@ -3472,7 +3472,7 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,* (e.g. libkms for the bootup splash), we have to ensure that we * always use map_and_fenceable for all scanout buffers. */ - ret =i915_gem_object_pin (obj, alignment, true, false); + ret =i915_gem_obj_ggtt_pin (obj, alignment, true, false);if (ret)return ret;
@@ -3615,6 +3615,7 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file)int i915_gem_object_pin (struct drm_i915_gem_object *obj, +struct i915_address_space *vm,uint32_t alignment,bool map_and_fenceable,bool nonblocking)
@@ -3704,7 +3705,7 @@ i915_gem_pin_ioctl(struct drm_device *dev, void *data, }if (obj->user_pin_count ==0 ) { - ret =i915_gem_object_pin (obj, args->alignment, true, false); + ret =i915_gem_obj_ggtt_pin (obj, args->alignment, true, false);if (ret) goto out; }
drivers/gpu/drm/i915/i915_gem_context.c +2 -2
@@ -214,7 +214,7 @@ static int create_default_context(struct drm_i915_private *dev_priv)* default context. */ dev_priv->ring[RCS ].default_context = ctx; - ret =i915_gem_object_pin (ctx->obj,CONTEXT_ALIGN , false, false); + ret =i915_gem_obj_ggtt_pin (ctx->obj,CONTEXT_ALIGN , false, false);if (ret) {DRM_DEBUG_DRIVER ("Couldn't pin %d\n" , ret); goto err_destroy;
@@ -400,7 +400,7 @@ static int do_switch(struct i915_hw_context *to)if (from == to)return 0 ; - ret =i915_gem_object_pin (to->obj,CONTEXT_ALIGN , false, false); + ret =i915_gem_obj_ggtt_pin (to->obj,CONTEXT_ALIGN , false, false);if (ret)return ret;
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3 -1
@@ -409,7 +409,9 @@ i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, obj->tiling_mode !=I915_TILING_NONE ; need_mappable = need_fence ||need_reloc_mappable (obj); - ret =i915_gem_object_pin (obj, entry->alignment, need_mappable, false); +/* FIXME: vm plubming */ + ret =i915_gem_object_pin (obj, &dev_priv->gtt.base, entry->alignment, + need_mappable, false);if (ret)return ret;
drivers/gpu/drm/i915/intel_overlay.c +1 -1
@@ -1350,7 +1350,7 @@ void intel_setup_overlay(struct drm_device *dev) } overlay->flip_addr = reg_bo->phys_obj->handle->busaddr; }else { - ret =i915_gem_object_pin (reg_bo,PAGE_SIZE , true, false); + ret =i915_gem_obj_ggtt_pin (reg_bo,PAGE_SIZE , true, false);if (ret) {DRM_ERROR ("failed to pin overlay register bo\n" ); goto out_free_bo;
drivers/gpu/drm/i915/intel_pm.c +1 -1
@@ -2860,7 +2860,7 @@ intel_alloc_context_page(struct drm_device *dev)return NULL ; } - ret =i915_gem_object_pin (ctx,4096 , true, false); + ret =i915_gem_obj_ggtt_pin (ctx,4096 , true, false);if (ret) {DRM_ERROR ("failed to pin power context: %d\n" , ret); goto err_unref;
drivers/gpu/drm/i915/intel_ringbuffer.c +4 -4
@@ -501,7 +501,7 @@ init_pipe_control(struct intel_ring_buffer *ring)i915_gem_object_set_cache_level (obj,I915_CACHE_LLC ); - ret =i915_gem_object_pin (obj,4096 , true, false); + ret =i915_gem_obj_ggtt_pin (obj,4096 , true, false);if (ret) goto err_unref;
@@ -1224,7 +1224,7 @@ static int init_status_page(struct intel_ring_buffer *ring)i915_gem_object_set_cache_level (obj,I915_CACHE_LLC ); - ret =i915_gem_object_pin (obj,4096 , true, false); + ret =i915_gem_obj_ggtt_pin (obj,4096 , true, false);if (ret !=0 ) { goto err_unref; }
@@ -1307,7 +1307,7 @@ static int intel_init_ring_buffer(struct drm_device *dev, ring->obj = obj; - ret =i915_gem_object_pin (obj,PAGE_SIZE , true, false); + ret =i915_gem_obj_ggtt_pin (obj,PAGE_SIZE , true, false);if (ret) goto err_unref;
@@ -1828,7 +1828,7 @@ int intel_init_render_ring_buffer(struct drm_device *dev)return -ENOMEM ; } - ret =i915_gem_object_pin (obj,0 , true, false); + ret =i915_gem_obj_ggtt_pin (obj,0 , true, false);if (ret !=0 ) {drm_gem_object_unreference (&obj->base);DRM_ERROR ("Failed to ping batch bo\n" );
Signature · 1 line
1.8.3.4
PATCH 6/29 drm/i915: Use ggtt_vm to save some typing
In reply to Ben WidawskyJust some small cleanups, and a rename of vm->ggtt_vm requested by Daniel.
--- drivers/gpu/drm/i915/i915_gem_gtt.c | 19 ++++++++----------- drivers/gpu/drm/i915/i915_gem_stolen.c | 10 +++++----- 2 files changed, 13 insertions(+), 16 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +8 -11
@@ -648,7 +648,8 @@ void i915_gem_setup_global_gtt(struct drm_device *dev,* aperture. One page should be enough to keep any prefetching inside * of the aperture. */ -drm_i915_private_t *dev_priv = dev->dev_private; +struct drm_i915_private *dev_priv = dev->dev_private; +struct i915_address_space *ggtt_vm = &dev_priv->gtt.base;struct drm_mm_node *entry;struct drm_i915_gem_object *obj;unsigned long hole_start, hole_end;
@@ -656,19 +657,19 @@ void i915_gem_setup_global_gtt(struct drm_device *dev,BUG_ON (mappable_end > end);/* Subtract the guard page ... */ -drm_mm_init (&dev_priv->gtt.base.mm, start, end - start -PAGE_SIZE ); +drm_mm_init (&ggtt_vm->mm, start, end - start -PAGE_SIZE );if (!HAS_LLC (dev)) dev_priv->gtt.base.mm.color_adjust = i915_gtt_color_adjust;/* Mark any preallocated objects as occupied */ list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); +struct i915_vma *vma =i915_gem_obj_to_vma (obj, ggtt_vm);int ret;DRM_DEBUG_KMS ("reserving preallocated space: %lx + %zx\n" ,i915_gem_obj_ggtt_offset (obj), obj->base.size);WARN_ON (i915_gem_obj_ggtt_bound (obj)); - ret =drm_mm_reserve_node (&dev_priv->gtt.base.mm, &vma->node); + ret =drm_mm_reserve_node (&ggtt_vm->mm, &vma->node);if (ret)DRM_DEBUG_KMS ("Reservation failed\n" ); obj->has_global_gtt_mapping =1 ;
@@ -679,19 +680,15 @@ void i915_gem_setup_global_gtt(struct drm_device *dev, dev_priv->gtt.base.total = end - start;/* Clear any non-preallocated blocks */ -drm_mm_for_each_hole (entry, &dev_priv->gtt.base.mm, - hole_start, hole_end) { +drm_mm_for_each_hole (entry, &ggtt_vm->mm, hole_start, hole_end) {const unsigned long count = (hole_end - hole_start) /PAGE_SIZE ;DRM_DEBUG_KMS ("clearing unused GTT space: [%lx, %lx]\n" , hole_start, hole_end); - dev_priv->gtt.base.clear_range (&dev_priv->gtt.base, - hole_start /PAGE_SIZE , - count); + ggtt_vm->clear_range (ggtt_vm, hole_start /PAGE_SIZE , count); }/* And finally clear the reserved guard page */ - dev_priv->gtt.base.clear_range (&dev_priv->gtt.base, - end /PAGE_SIZE -1 ,1 ); + ggtt_vm->clear_range (ggtt_vm, end /PAGE_SIZE -1 ,1 ); } static bool
drivers/gpu/drm/i915/i915_gem_stolen.c +5 -5
@@ -351,7 +351,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, u32 size) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *ggtt = &dev_priv->gtt.base;struct drm_i915_gem_object *obj;struct drm_mm_node *stolen;struct i915_vma *vma;
@@ -394,7 +394,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev,if (gtt_offset ==I915_GTT_OFFSET_NONE )return obj; - vma =i915_gem_vma_create (obj, &dev_priv->gtt.base); + vma =i915_gem_vma_create (obj, ggtt);if (IS_ERR (vma)) { ret =PTR_ERR (vma); goto err_out;
@@ -407,8 +407,8 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev,*/ vma->node.start = gtt_offset; vma->node.size = size; -if (drm_mm_initialized (&dev_priv->gtt.base.mm)) { - ret =drm_mm_reserve_node (&dev_priv->gtt.base.mm, &vma->node); +if (drm_mm_initialized (&ggtt->mm)) { + ret =drm_mm_reserve_node (&ggtt->mm, &vma->node);if (ret) {DRM_DEBUG_KMS ("failed to allocate stolen GTT space\n" );i915_gem_vma_destroy (vma);
@@ -419,7 +419,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, obj->has_global_gtt_mapping =1 ;list_add_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &vm->inactive_list); +list_add_tail (&obj->mm_list, &ggtt->inactive_list);return obj;
Signature · 1 line
1.8.3.4
PATCH 7/29 drm/i915: Update describe_obj
In reply to Ben WidawskyMake it aware of which domain it is bound into GGTT, or PPGTT.
While modifying the function, add a global gtt flag to the object description. Global is more interesting than aliasing since aliasing is the default.
v2: Access VMA directly for start/size instead of helpers (Daniel)
--- drivers/gpu/drm/i915/i915_debugfs.c | 20 ++++++++++++++++---- 1 file changed, 16 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +16 -4
@@ -89,13 +89,20 @@ static const char *get_tiling_flag(struct drm_i915_gem_object *obj) } } +static inline const char *get_global_flag (struct drm_i915_gem_object *obj) +{ +return obj->has_global_gtt_mapping ?"g" :" " ; +} +static void describe_obj (struct seq_file *m,struct drm_i915_gem_object *obj) { - seq_printf(m, "%pK: %s%s %8zdKiB %02x %02x %d %d %d%s%s%s", +struct i915_vma *vma; + seq_printf(m, "%pK: %s%s%s %8zdKiB %02x %02x %d %d %d%s%s%s", &obj->base, get_pin_flag(obj), get_tiling_flag(obj), + get_global_flag(obj), obj->base.size / 1024, obj->base.read_domains, obj->base.write_domain,
@@ -111,9 +118,14 @@ describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)seq_printf (m," (pinned x %d)" , obj->pin_count);if (obj->fence_reg !=I915_FENCE_REG_NONE )seq_printf (m," (fence: %d)" , obj->fence_reg); -if (i915_gem_obj_ggtt_bound (obj)) -seq_printf (m," (gtt offset: %08lx, size: %08x)" , -i915_gem_obj_ggtt_offset (obj), (unsigned int )i915_gem_obj_ggtt_size (obj)); +list_for_each_entry (vma, &obj->vma_list, vma_link) { +if (!i915_is_ggtt (vma->vm)) +seq_puts (m," (pp" ); +else +seq_puts (m," (g" ); +seq_printf (m,"gtt offset: %08lx, size: %08lx)" , + vma->node.start, vma->node.size); + }if (obj->stolen)seq_printf (m," (stolen: %08lx)" , obj->stolen->start);if (obj->pin_mappable || obj->fault_mappable) {
Signature · 1 line
1.8.3.4
PATCH 8/29 drm/i915: Rework __i915_gem_shrink
In reply to Ben WidawskyIn order to do this for all VMs, it's convenient to rework the logic a bit. This should have no functional impact.
--- drivers/gpu/drm/i915/i915_gem.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +8 -3
@@ -1693,9 +1693,14 @@ __i915_gem_shrink(struct drm_i915_private *dev_priv, long target, }list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) { -if ((i915_gem_object_is_purgeable (obj) || !purgeable_only) && -i915_gem_object_unbind (obj) ==0 && -i915_gem_object_put_pages (obj) ==0 ) { + +if (!i915_gem_object_is_purgeable (obj) && purgeable_only) +continue ; + +if (i915_gem_object_unbind (obj)) +continue ; + +if (!i915_gem_object_put_pages (obj)) { count += obj->base.size >>PAGE_SHIFT ;if (count >= target)return count;
Signature · 1 line
1.8.3.4
Re: [PATCH 08/29] drm/i915: Rework __i915_gem_shrink
Quoted 4 lines · Ben Widawsky
In order to do this for all VMs, it's convenient to rework the logic a bit. This should have no functional impact.
Signed-off-by Ben Widawsky <[email protected]>
I didn't read ahead but I guess the actual shrink code here will move to the unbound list like the count code beforehand? Just for symmetry I think that'd be worth it. Patch applied since it looks neater ;-) -Daniel
Quoted 33 lines
--- drivers/gpu/drm/i915/i915_gem.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +8
-3@@ -1693,9 +1693,14 @@ __i915_gem_shrink(struct drm_i915_private *dev_priv, long target, }list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) { -if ((i915_gem_object_is_purgeable (obj) || !purgeable_only) && -i915_gem_object_unbind (obj) ==0 && -i915_gem_object_put_pages (obj) ==0 ) { + +if (!i915_gem_object_is_purgeable (obj) && purgeable_only) +continue ; + +if (i915_gem_object_unbind (obj)) +continue ; + +if (!i915_gem_object_put_pages (obj)) { count += obj->base.size >>PAGE_SHIFT ;if (count >= target)return count;Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 9/29 drm/i915: thread address space through execbuf
In reply to Ben WidawskyThis represents the first half of hooking up VMs to execbuf. Here we basically pass an address space all around to the different internal functions. It should be much more readable, and have less risk than the second half, which begins switching over to using VMAs instead of an obj,vm.
The overall series echoes this style of, "add a VM, then make it smart later"
--- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 77 +++++++++++++++++++----------- 1 file changed, 49 insertions(+), 28 deletions(-)
drivers/gpu/drm/i915/i915_gem_execbuffer.c +49 -28
@@ -174,7 +174,8 @@ static inline int use_cpu_reloc(struct drm_i915_gem_object *obj)static int i915_gem_execbuffer_relocate_entry (struct drm_i915_gem_object *obj,struct eb_objects *eb, -struct drm_i915_gem_relocation_entry *reloc) +struct drm_i915_gem_relocation_entry *reloc, +struct i915_address_space *vm) {struct drm_device *dev = obj->base.dev;struct drm_gem_object *target_obj;
@@ -297,7 +298,8 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,static int i915_gem_execbuffer_relocate_object (struct drm_i915_gem_object *obj, -struct eb_objects *eb) +struct eb_objects *eb, +struct i915_address_space *vm) {#define N_RELOC (x) ((x) / sizeof(struct drm_i915_gem_relocation_entry))struct drm_i915_gem_relocation_entry stack_reloc[N_RELOC (512 )];
@@ -321,7 +323,8 @@ i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj, do { u64 offset = r->presumed_offset; - ret = i915_gem_execbuffer_relocate_entry(obj, eb, r); + ret = i915_gem_execbuffer_relocate_entry(obj, eb, r, + vm); if (ret) return ret;
@@ -344,13 +347,15 @@ i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj,static int i915_gem_execbuffer_relocate_object_slow (struct drm_i915_gem_object *obj,struct eb_objects *eb, -struct drm_i915_gem_relocation_entry *relocs) +struct drm_i915_gem_relocation_entry *relocs, +struct i915_address_space *vm) {const struct drm_i915_gem_exec_object2 *entry = obj->exec_entry;int i, ret;for (i =0 ; i < entry->relocation_count; i++) { - ret =i915_gem_execbuffer_relocate_entry (obj, eb, &relocs[i]); + ret =i915_gem_execbuffer_relocate_entry (obj, eb, &relocs[i], + vm);if (ret)return ret; }
@@ -359,7 +364,8 @@ i915_gem_execbuffer_relocate_object_slow(struct drm_i915_gem_object *obj, }static int -i915_gem_execbuffer_relocate (struct eb_objects *eb) +i915_gem_execbuffer_relocate (struct eb_objects *eb, +struct i915_address_space *vm) {struct drm_i915_gem_object *obj;int ret =0 ;
@@ -373,7 +379,7 @@ i915_gem_execbuffer_relocate(struct eb_objects *eb)*/ pagefault_disable ();list_for_each_entry (obj, &eb->objects, exec_list) { - ret =i915_gem_execbuffer_relocate_object (obj, eb); + ret =i915_gem_execbuffer_relocate_object (obj, eb, vm);if (ret)break ; }
@@ -395,6 +401,7 @@ need_reloc_mappable(struct drm_i915_gem_object *obj)static int i915_gem_execbuffer_reserve_object (struct drm_i915_gem_object *obj,struct intel_ring_buffer *ring, +struct i915_address_space *vm,bool *need_reloc) {struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
@@ -409,9 +416,8 @@ i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, obj->tiling_mode !=I915_TILING_NONE ; need_mappable = need_fence ||need_reloc_mappable (obj); -/* FIXME: vm plubming */ - ret =i915_gem_object_pin (obj, &dev_priv->gtt.base, entry->alignment, - need_mappable, false); + ret =i915_gem_object_pin (obj, vm, entry->alignment, need_mappable, + false);if (ret)return ret;
@@ -438,8 +444,8 @@ i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, obj->has_aliasing_ppgtt_mapping =1 ; } -if (entry->offset !=i915_gem_obj_ggtt_offset (obj)) { - entry->offset =i915_gem_obj_ggtt_offset (obj); +if (entry->offset !=i915_gem_obj_offset (obj, vm)) { + entry->offset =i915_gem_obj_offset (obj, vm); *need_reloc = true; }
@@ -477,6 +483,7 @@ i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj)static int i915_gem_execbuffer_reserve (struct intel_ring_buffer *ring,struct list_head *objects, +struct i915_address_space *vm,bool *need_relocs) {struct drm_i915_gem_object *obj;
@@ -531,32 +538,37 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,list_for_each_entry (obj, objects, exec_list) {struct drm_i915_gem_exec_object2 *entry = obj->exec_entry;bool need_fence, need_mappable; +u32 obj_offset; -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound (obj, vm))continue ; + obj_offset =i915_gem_obj_offset (obj, vm); need_fence = has_fenced_gpu_access && entry->flags &EXEC_OBJECT_NEEDS_FENCE && obj->tiling_mode !=I915_TILING_NONE ; need_mappable = need_fence ||need_reloc_mappable (obj); +BUG_ON ((need_mappable || need_fence) && + !i915_is_ggtt (vm)); +if ((entry->alignment && -i915_gem_obj_ggtt_offset (obj) & (entry->alignment -1 )) || + obj_offset & (entry->alignment -1 )) || (need_mappable && !obj->map_and_fenceable)) ret =i915_gem_object_unbind (obj);else - ret =i915_gem_execbuffer_reserve_object (obj, ring, need_relocs); + ret =i915_gem_execbuffer_reserve_object (obj, ring, vm, need_relocs);if (ret) goto err; }/* Bind fresh objects */ list_for_each_entry (obj, objects, exec_list) { -if (i915_gem_obj_ggtt_bound (obj)) +if (i915_gem_obj_bound (obj, vm))continue ; - ret =i915_gem_execbuffer_reserve_object (obj, ring, need_relocs); + ret =i915_gem_execbuffer_reserve_object (obj, ring, vm, need_relocs);if (ret) goto err; }
@@ -580,7 +592,8 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,struct drm_file *file, struct intel_ring_buffer *ring, struct eb_objects *eb, - struct drm_i915_gem_exec_object2 *exec) + struct drm_i915_gem_exec_object2 *exec, + struct i915_address_space *vm) { struct drm_i915_gem_relocation_entry *reloc;struct drm_i915_gem_object *obj;
@@ -664,14 +677,15 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev, goto err; need_relocs = (args->flags &I915_EXEC_NO_RELOC ) ==0 ; - ret =i915_gem_execbuffer_reserve (ring, &eb->objects, &need_relocs); + ret =i915_gem_execbuffer_reserve (ring, &eb->objects, vm, &need_relocs);if (ret) goto err;list_for_each_entry (obj, &eb->objects, exec_list) {int offset = obj->exec_entry - exec; ret =i915_gem_execbuffer_relocate_object_slow (obj, eb, - reloc + reloc_offset[offset]); + reloc + reloc_offset[offset], + vm);if (ret) goto err; }
@@ -772,6 +786,7 @@ validate_exec_list(struct drm_i915_gem_exec_object2 *exec,static void i915_gem_execbuffer_move_to_active (struct list_head *objects, +struct i915_address_space *vm,struct intel_ring_buffer *ring) {struct drm_i915_gem_object *obj;
@@ -840,7 +855,8 @@ static int i915_gem_do_execbuffer(struct drm_device *dev, void *data, struct drm_file *file, struct drm_i915_gem_execbuffer2 *args, - struct drm_i915_gem_exec_object2 *exec) + struct drm_i915_gem_exec_object2 *exec, + struct i915_address_space *vm) { drm_i915_private_t *dev_priv = dev->dev_private;struct eb_objects *eb;
@@ -1002,17 +1018,17 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,/* Move the objects en-masse into the GTT, evicting if necessary. */ need_relocs = (args->flags &I915_EXEC_NO_RELOC ) ==0 ; - ret =i915_gem_execbuffer_reserve (ring, &eb->objects, &need_relocs); + ret =i915_gem_execbuffer_reserve (ring, &eb->objects, vm, &need_relocs);if (ret) goto err;/* The objects are in their final locations, apply the relocations. */ if (need_relocs) - ret =i915_gem_execbuffer_relocate (eb); + ret =i915_gem_execbuffer_relocate (eb, vm);if (ret) {if (ret == -EFAULT ) { ret =i915_gem_execbuffer_relocate_slow (dev, args, file, ring, - eb, exec); + eb, exec, vm);BUG_ON (!mutex_is_locked (&dev->struct_mutex)); }if (ret)
@@ -1063,7 +1079,8 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, goto err; } - exec_start =i915_gem_obj_ggtt_offset (batch_obj) + args->batch_start_offset; + exec_start =i915_gem_obj_offset (batch_obj, vm) + + args->batch_start_offset; exec_len = args->batch_len;if (cliprects) {for (i =0 ; i < args->num_cliprects; i++) {
@@ -1088,7 +1105,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,trace_i915_gem_ring_dispatch (ring,intel_ring_get_seqno (ring), flags); -i915_gem_execbuffer_move_to_active (&eb->objects, ring); +i915_gem_execbuffer_move_to_active (&eb->objects, vm, ring);i915_gem_execbuffer_retire_commands (dev, file, ring, batch_obj); err:
@@ -1109,6 +1126,7 @@ int i915_gem_execbuffer(struct drm_device *dev, void *data, struct drm_file *file) { + struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_execbuffer *args = data;struct drm_i915_gem_execbuffer2 exec2;struct drm_i915_gem_exec_object *exec_list =NULL ;
@@ -1164,7 +1182,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, exec2.flags =I915_EXEC_RENDER ;i915_execbuffer2_set_context_id (exec2,0 ); - ret =i915_gem_do_execbuffer (dev, data, file, &exec2, exec2_list); + ret =i915_gem_do_execbuffer (dev, data, file, &exec2, exec2_list, + &dev_priv->gtt.base);if (!ret) {/* Copy the new buffer offsets back to the user's exec list. */ for (i =0 ; i < args->buffer_count; i++)
@@ -1190,6 +1209,7 @@ int i915_gem_execbuffer2(struct drm_device *dev, void *data, struct drm_file *file) { + struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_execbuffer2 *args = data;struct drm_i915_gem_exec_object2 *exec2_list =NULL ;int ret;
@@ -1220,7 +1240,8 @@ i915_gem_execbuffer2(struct drm_device *dev, void *data,return -EFAULT ; } - ret =i915_gem_do_execbuffer (dev, data, file, args, exec2_list); + ret =i915_gem_do_execbuffer (dev, data, file, args, exec2_list, + &dev_priv->gtt.base);if (!ret) {/* Copy the new buffer offsets back to the user's exec list. */ ret = copy_to_user(to_user_ptr(args->buffers_ptr),
Signature · 1 line
1.8.3.4
Re: [PATCH 09/29] drm/i915: thread address space through execbuf
Quoted 11 lines · Ben Widawsky
This represents the first half of hooking up VMs to execbuf. Here we basically pass an address space all around to the different internal functions. It should be much more readable, and have less risk than the second half, which begins switching over to using VMAs instead of an obj,vm.
The overall series echoes this style of, "add a VM, then make it smart later"
Signed-off-by Ben Widawsky <[email protected]>---
[snip]
Quoted 17 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3
-1@@ -477,6 +483,7 @@ i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj)static int i915_gem_execbuffer_reserve (struct intel_ring_buffer *ring,struct list_head *objects, +struct i915_address_space *vm,bool *need_relocs) {struct drm_i915_gem_object *obj;@@ -531,32 +538,37 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,list_for_each_entry (obj, objects, exec_list) {struct drm_i915_gem_exec_object2 *entry = obj->exec_entry;bool need_fence, need_mappable; +u32 obj_offset; -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound (obj, vm))continue ;
drivers/gpu/drm/i915/i915_gem_execbuffer.c i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring, · line 554
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +3
-0+ obj_offset =i915_gem_obj_offset (obj, vm); need_fence = has_fenced_gpu_access && entry->flags &EXEC_OBJECT_NEEDS_FENCE && obj->tiling_mode !=I915_TILING_NONE ; need_mappable = need_fence ||need_reloc_mappable (obj); +BUG_ON ((need_mappable || need_fence) && + !i915_is_ggtt (vm));
No BUG_ON if the error isn't fatal, please. I've fixed this up to a WARN_ON while applying.
I know that you prefer BUG_ON for developing, but when we hit one of these in the wild it's a royal pain to debug. But please either switch to WARNs when submitting the patches or (imo preferred) get into the habit of grepping dmesg for backtraces when testing.
Cheers, Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 10/29 drm/i915: make caching operate on all address spaces
In reply to Ben WidawskyFor now, objects will maintain the same cache levels amongst all address spaces. This is to limit the risk of bugs, as playing with cacheability in the different domains can be very error prone.
In the future, it may be optimal to allow setting domains per VMA (ie. an object bound into an address space).
--- drivers/gpu/drm/i915/i915_gem.c | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +12 -8
@@ -3308,7 +3308,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); +struct i915_vma *vma;int ret;if (obj->cache_level == cache_level)
@@ -3319,13 +3319,17 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return -EBUSY ; } -if (vma && !i915_gem_valid_gtt_space (dev, &vma->node, cache_level)) { - ret =i915_gem_object_unbind (obj); -if (ret) -return ret; +list_for_each_entry (vma, &obj->vma_list, vma_link) { +if (!i915_gem_valid_gtt_space (dev, &vma->node, cache_level)) { + ret =i915_gem_object_unbind (obj); +if (ret) +return ret; + +break ; + } } -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_finish_gpu (obj);if (ret)return ret;
@@ -3347,8 +3351,6 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,if (obj->has_aliasing_ppgtt_mapping)i915_ppgtt_bind_object (dev_priv->mm.aliasing_ppgtt, obj, cache_level); - -i915_gem_obj_to_vma (obj, &dev_priv->gtt.base)->node.color = cache_level; }if (cache_level ==I915_CACHE_NONE ) {
@@ -3374,6 +3376,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, old_write_domain); } +list_for_each_entry (vma, &obj->vma_list, vma_link) + vma->node.color = cache_level; obj->cache_level = cache_level;i915_gem_verify_gtt (dev);return 0 ;
Signature · 1 line
1.8.3.4
Re: [PATCH 10/29] drm/i915: make caching operate on all address spaces
Quoted 35 lines · Ben Widawsky
For now, objects will maintain the same cache levels amongst all address spaces. This is to limit the risk of bugs, as playing with cacheability in the different domains can be very error prone.
In the future, it may be optimal to allow setting domains per VMA (ie. an object bound into an address space).
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-)drivers/gpu/drm/i915/i915_gem.c +3
-5@@ -3308,7 +3308,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); +struct i915_vma *vma;int ret;if (obj->cache_level == cache_level)@@ -3319,13 +3319,17 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return -EBUSY ; } -if (vma && !i915_gem_valid_gtt_space (dev, &vma->node, cache_level)) { - ret =i915_gem_object_unbind (obj); -if (ret) -return ret; +list_for_each_entry (vma, &obj->vma_list, vma_link) { +if (!i915_gem_valid_gtt_space (dev, &vma->node, cache_level)) {
drivers/gpu/drm/i915/i915_gem.c int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, · line 3332
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +7
-1+ ret =i915_gem_object_unbind (obj); +if (ret) +return ret; + +break ; + } } -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) {
Hm, I guess this will change later on to a for_each_vma loop?
Patch applied meanwhile.
-Daniel
Quoted 28 lines
drivers/gpu/drm/i915/i915_gem.c +2
-2ret =i915_gem_object_finish_gpu (obj);if (ret)return ret;@@ -3347,8 +3351,6 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,if (obj->has_aliasing_ppgtt_mapping)i915_ppgtt_bind_object (dev_priv->mm.aliasing_ppgtt, obj, cache_level); - -i915_gem_obj_to_vma (obj, &dev_priv->gtt.base)->node.color = cache_level; }if (cache_level ==I915_CACHE_NONE ) {@@ -3374,6 +3376,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, old_write_domain); } +list_for_each_entry (vma, &obj->vma_list, vma_link) + vma->node.color = cache_level; obj->cache_level = cache_level;i915_gem_verify_gtt (dev);return 0 ;Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 11/29 drm/i915: BUG_ON put_pages later
In reply to Ben WidawskyWith multiple VMs, the eviction code benefits from being able to blindly put pages without needing to know if there are any entities still holding on to those pages. As such it's preferable to return the -EBUSY before the BUG.
Eviction code is the only user for now, but overall it makes sense anyway, IMO.
--- drivers/gpu/drm/i915/i915_gem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +2 -2
@@ -1654,11 +1654,11 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj)if (obj->pages ==NULL )return 0 ; -BUG_ON (i915_gem_obj_ggtt_bound (obj)); -if (obj->pages_pin_count)return -EBUSY ; +BUG_ON (i915_gem_obj_ggtt_bound (obj)); +/* ->put_pages might need to allocate memory for the bit17 swizzle * array, hence protect them from being reaped by removing them from gtt * lists early. */
Signature · 1 line
1.8.3.4
Re: [PATCH 11/29] drm/i915: BUG_ON put_pages later
Quoted 18 lines · Ben Widawsky
With multiple VMs, the eviction code benefits from being able to blindly put pages without needing to know if there are any entities still holding on to those pages. As such it's preferable to return the
-EBUSYbefore the BUG.Eviction code is the only user for now, but overall it makes sense anyway, IMO.
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)drivers/gpu/drm/i915/i915_gem.c +0
-0
drivers/gpu/drm/i915/i915_gem.c i915_gem_object_put_pages(struct drm_i915_gem_object *obj) · line 1660
Quoted 9 lines
drivers/gpu/drm/i915/i915_gem.c +1
-2@@ -1654,11 +1654,11 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj)if (obj->pages ==NULL )return 0 ; -BUG_ON (i915_gem_obj_ggtt_bound (obj)); -if (obj->pages_pin_count)return -EBUSY ; +BUG_ON (i915_gem_obj_ggtt_bound (obj));
Hm, shouldn't this be a bound_any eventually? Again I'm too layz to check the end result, just noting my thoughs here ;-)
-Daniel
Quoted 11 lines
drivers/gpu/drm/i915/i915_gem.c +1
-0+/* ->put_pages might need to allocate memory for the bit17 swizzle * array, hence protect them from being reaped by removing them from gtt * lists early. */ Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 12/29 drm/i915: make reset&hangcheck code VM aware
In reply to Ben WidawskyHangcheck, and some of the recent reset code for guilty batches need to know which address space the object was in at the time of a hangcheck. This is because we use offsets in the (PP|G)GTT to determine this information, and those offsets can differ depending on which VM they are bound into.
Since we still only have 1 VM ever, this code shouldn't yet any any impact.
--- drivers/gpu/drm/i915/i915_gem.c | 30 +++++++++++++++++++++++------- 1 file changed, 23 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +23 -7
@@ -2110,10 +2110,11 @@ i915_gem_request_remove_from_client(struct drm_i915_gem_request *request)spin_unlock (&file_priv->mm.lock); } -static bool i915_head_inside_object (u32 acthd,struct drm_i915_gem_object *obj) +static bool i915_head_inside_object (u32 acthd,struct drm_i915_gem_object *obj, +struct i915_address_space *vm) { -if (acthd >=i915_gem_obj_ggtt_offset (obj) && - acthd <i915_gem_obj_ggtt_offset (obj) + obj->base.size) +if (acthd >=i915_gem_obj_offset (obj, vm) && + acthd <i915_gem_obj_offset (obj, vm) + obj->base.size)return true;return false;
@@ -2136,6 +2137,17 @@ static bool i915_head_inside_request(const u32 acthd_unmasked,return false; } +static struct i915_address_space * +request_to_vm (struct drm_i915_gem_request *request) +{ +struct drm_i915_private *dev_priv = request->ring->dev->dev_private; +struct i915_address_space *vm; + + vm = &dev_priv->gtt.base; + +return vm; +} +static bool i915_request_guilty (struct drm_i915_gem_request *request,const u32 acthd,bool *inside) {
@@ -2143,9 +2155,9 @@ static bool i915_request_guilty(struct drm_i915_gem_request *request,* pointing inside the ring, matches the batch_obj address range. * However this is extremely unlikely. */ -if (request->batch_obj) { -if (i915_head_inside_object (acthd, request->batch_obj)) { +if (i915_head_inside_object (acthd, request->batch_obj, +request_to_vm (request))) { *inside = true;return true; }
@@ -2165,17 +2177,21 @@ static void i915_set_reset_status(struct intel_ring_buffer *ring, {struct i915_ctx_hang_stats *hs =NULL ;bool inside, guilty; +unsigned long offset =0 ;/* Innocent until proven guilty */ guilty = false; +if (request->batch_obj) + offset =i915_gem_obj_offset (request->batch_obj, +request_to_vm (request)); +if (ring->hangcheck.action != wait &&i915_request_guilty (request, acthd, &inside)) {DRM_ERROR ("%s hung %s bo (0x%lx ctx %d) at 0x%x\n" , ring->name, inside ?"inside" :"flushing" , - request->batch_obj ? -i915_gem_obj_ggtt_offset (request->batch_obj) :0 , + offset, request->ctx ? request->ctx->id :0 , acthd);
Signature · 1 line
1.8.3.4
PATCH 13/29 drm/i915: clear domains for all objects on reset
In reply to Ben WidawskySimply iterating over 1 inactive list is insufficient for the way we now track inactive (1 list per address space). We could alternatively do this with bound + unbound lists, and an inactive check. To me, this way is a bit easier to understand.
--- drivers/gpu/drm/i915/i915_gem.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +4 -3
@@ -2282,7 +2282,7 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm;struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring;int i;
@@ -2293,8 +2293,9 @@ void i915_gem_reset(struct drm_device *dev)/* Move everything out of the GPU domains to ensure we do any * necessary invalidation upon reuse. */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) +list_for_each_entry (obj, &vm->inactive_list, mm_list) + obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;i915_gem_restore_fences (dev); }
Signature · 1 line
1.8.3.4
Re: [PATCH 13/29] drm/i915: clear domains for all objects on reset
Quoted 24 lines · Ben Widawsky
Simply iterating over 1 inactive list is insufficient for the way we now track inactive (1 list per address space). We could alternatively do this with bound + unbound lists, and an inactive check. To me, this way is a bit easier to understand.
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -2282,7 +2282,7 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm;struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring;int i;
drivers/gpu/drm/i915/i915_gem.c void i915_gem_reset(struct drm_device *dev) · line 2298
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c +3
-2@@ -2293,8 +2293,9 @@ void i915_gem_reset(struct drm_device *dev)/* Move everything out of the GPU domains to ensure we do any * necessary invalidation upon reuse. */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) +list_for_each_entry (obj, &vm->inactive_list, mm_list) + obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;
This code is dead. Just remove it rather than port it to vma. -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 13/29] drm/i915: clear domains for all objects on reset
Quoted 28 lines · Chris Wilson
Quoted 27 lines · Ben Widawsky
Simply iterating over 1 inactive list is insufficient for the way we now track inactive (1 list per address space). We could alternatively do this with bound + unbound lists, and an inactive check. To me, this way is a bit easier to understand.
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -2282,7 +2282,7 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm;struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring;int i;@@ -2293,8 +2293,9 @@ void i915_gem_reset(struct drm_device *dev)/* Move everything out of the GPU domains to ensure we do any * necessary invalidation upon reuse. */
Quoted 11 lines
drivers/gpu/drm/i915/i915_gem.c void i915_gem_reset(struct drm_device *dev) · line 2298
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem.c +3
-2-list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) +list_for_each_entry (obj, &vm->inactive_list, mm_list) + obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;This code is dead. Just remove it rather than port it to vma. -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Got it, and moved to the front of the series.
commit 8472f08863da69159aa0a7555836ca0511754877
Author: Ben Widawsky <[email protected]> Date: Sat Aug 3 15:22:17 2013 -0700
drm/i915: eliminate dead domain clearing on reset
The code itself is no longer accurate without updating once we have multiple address space since clearing the domains of every object requires scanning the inactive list for all VMs.
"This code is dead. Just remove it rather than port it to vma." - Chris Wilson
drivers/gpu/drm/i915/i915_gem.c +0 -6
@@ -2277,12 +2277,6 @@ void i915_gem_reset(struct drm_device *dev)for_each_ring (ring, dev_priv, i)i915_gem_reset_ring_lists (dev_priv, ring); -/* Move everything out of the GPU domains to ensure we do any -* necessary invalidation upon reuse. -*/ -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; -i915_gem_restore_fences (dev); }
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 13/29] drm/i915: clear domains for all objects on reset
Quoted 53 lines · Ben Widawsky
Quoted 28 lines · Chris Wilson
Quoted 27 lines · Ben Widawsky
Simply iterating over 1 inactive list is insufficient for the way we now track inactive (1 list per address space). We could alternatively do this with bound + unbound lists, and an inactive check. To me, this way is a bit easier to understand.
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -2282,7 +2282,7 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm;struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring;int i;@@ -2293,8 +2293,9 @@ void i915_gem_reset(struct drm_device *dev)/* Move everything out of the GPU domains to ensure we do any * necessary invalidation upon reuse. */ Quoted 11 lines
drivers/gpu/drm/i915/i915_gem.c void i915_gem_reset(struct drm_device *dev) · line 2298
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem.c +3
-2-list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) +list_for_each_entry (obj, &vm->inactive_list, mm_list) + obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;This code is dead. Just remove it rather than port it to vma. -Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology CentreGot it, and moved to the front of the series.
commit
8472f08863da69159aa0a7555836ca0511754877Author: Ben Widawsky <[email protected]> Date: Sat Aug 3 15:22:17 2013 -0700drm/i915: eliminate dead domain clearing on reset
The code itself is no longer accurate without updating once we have multiple address space since clearing the domains of every object requires scanning the inactive list for all VMs.
Quoted 5 lines
"This code is dead. Just remove it rather than port it to vma." - Chris Wilson
Recommended-by Chris Wilson <[email protected]>Signed-off-by Ben Widawsky <[email protected]>
That's not a properly formatted patch, so I've stopped merging for now.
/me loves cheap excuses
But overall I _really_ like what the series looks now, I can dwell in the cozy feeling that I actually understand what's going on. So if the name of the game is to keep your maintainer happy I think the goal is unlocked ;-)
Cheers, Daniel
Quoted 26 lines
drivers/gpu/drm/i915/i915_gem.c +0
-6@@ -2277,12 +2277,6 @@ void i915_gem_reset(struct drm_device *dev)for_each_ring (ring, dev_priv, i)i915_gem_reset_ring_lists (dev_priv, ring); -/* Move everything out of the GPU domains to ensure we do any -* necessary invalidation upon reuse. -*/ -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; -i915_gem_restore_fences (dev); }Signature · 6 lines
-- Ben Widawsky, Intel Open Source Technology Center _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 13/29 drm/i915: eliminate dead domain clearing on reset
In reply to Ben WidawskyThe code itself is no longer accurate without updating once we have multiple address space since clearing the domains of every object requires scanning the inactive list for all VMs.
"This code is dead. Just remove it rather than port it to vma." - Chris Wilson
--- drivers/gpu/drm/i915/i915_gem.c | 8 -------- 1 file changed, 8 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +0 -8
@@ -2290,20 +2290,12 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring;int i;for_each_ring (ring, dev_priv, i)i915_gem_reset_ring_lists (dev_priv, ring); -/* Move everything out of the GPU domains to ensure we do any -* necessary invalidation upon reuse. -*/ -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; -i915_gem_restore_fences (dev); }
Signature · 1 line
1.8.3.4
Re: [PATCH 13/29] drm/i915: eliminate dead domain clearing on reset
Quoted 9 lines · Ben Widawsky
The code itself is no longer accurate without updating once we have multiple address space since clearing the domains of every object requires scanning the inactive list for all VMs.
"This code is dead. Just remove it rather than port it to vma." - Chris Wilson
Recommended-by Chris Wilson <[email protected]>Signed-off-by Ben Widawsky <[email protected]>
Thanks for resend, slurped into dinq. I'd like to take a little break here now and resume merging tomorrow - I've merge over 50 patches just today ;-)
-Daniel
Quoted 36 lines
--- drivers/gpu/drm/i915/i915_gem.c | 8 -------- 1 file changed, 8 deletions(-)drivers/gpu/drm/i915/i915_gem.c +0
-8@@ -2290,20 +2290,12 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring;int i;for_each_ring (ring, dev_priv, i)i915_gem_reset_ring_lists (dev_priv, ring); -/* Move everything out of the GPU domains to ensure we do any -* necessary invalidation upon reuse. -*/ -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; -i915_gem_restore_fences (dev); }Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 14/29 drm/i915: Restore PDEs on gtt restore
In reply to Ben WidawskyI can't remember why I added this initially.
TODO: Throw out if not necessary
--- drivers/gpu/drm/i915/i915_gem_gtt.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/gpu/drm/i915/i915_gem_gtt.c +3 -0
@@ -470,6 +470,9 @@ void i915_gem_restore_gtt_mappings(struct drm_device *dev) dev_priv->gtt.base.start /PAGE_SIZE , dev_priv->gtt.base.total / PAGE_SIZE); +if (dev_priv->mm.aliasing_ppgtt) +gen6_write_pdes (dev_priv->mm.aliasing_ppgtt); +list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list) {i915_gem_clflush_object (obj);i915_gem_gtt_bind_object (obj, obj->cache_level);
Signature · 1 line
1.8.3.4
Re: [PATCH 14/29] drm/i915: Restore PDEs on gtt restore
Quoted 5 lines · Ben Widawsky
I can't remember why I added this initially.
TODO: Throw out if not necessary
Signed-off-by Ben Widawsky <[email protected]>
Oops. I vaguely remember that there was some shuffling around with where the ppgtt pdes are written once we have real ppgtt. Since the aliasing ppgtt isn't a real ppgtt it could be that the write_pdes call in gen6_ppgtt_enable isn't executed any more and hence we need to do this manually.
Otoh it's looks like we should just call gen6_ppgtt_enable in such a case. So I can't really help with a reason why we need this. I'm pretty sure that we don't need it right now, so I'll punt for now. But please keep it in mind when rebasing the real ppgtt stuff on top, once the VMA patches are merged to dinq.
-Daniel
Quoted 25 lines
--- drivers/gpu/drm/i915/i915_gem_gtt.c | 3 +++ 1 file changed, 3 insertions(+)drivers/gpu/drm/i915/i915_gem_gtt.c +3
-0@@ -470,6 +470,9 @@ void i915_gem_restore_gtt_mappings(struct drm_device *dev) dev_priv->gtt.base.start /PAGE_SIZE , dev_priv->gtt.base.total / PAGE_SIZE); +if (dev_priv->mm.aliasing_ppgtt) +gen6_write_pdes (dev_priv->mm.aliasing_ppgtt); +list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list) {i915_gem_clflush_object (obj);i915_gem_gtt_bind_object (obj, obj->cache_level);Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 15/29 drm/i915: Improve VMA comments
In reply to Ben Widawsky--- drivers/gpu/drm/i915/i915_drv.h | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_drv.h +6 -1
@@ -533,7 +533,12 @@ struct i915_hw_ppgtt {int (*enable)(struct drm_device *dev); }; -/* To make things as simple as possible (ie. no refcounting), a VMA's lifetime +/** +* A VMA represents a GEM BO that is bound into an address space. Therefore, a +* VMA's presence cannot be guaranteed before binding, or after unbinding the +* object into/from the address space. +* +* To make things as simple as possible (ie. no refcounting), a VMA's lifetime * will always be <= an objects lifetime. So object refcounting should cover us. */ struct i915_vma {
Signature · 1 line
1.8.3.4
PATCH 16/29 drm/i915: Cleanup more of VMA in destroy
In reply to Ben WidawskyRemove the VMA from the object's list, and remove the VMA's node from the allocator. This just helps avoid duplication, since we always want those two things to occur on destroy, and at least for now, we only do those two actions on destroy anyway.
--- drivers/gpu/drm/i915/i915_gem.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +2 -4
@@ -2640,8 +2640,6 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj) obj->map_and_fenceable = true; vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); -list_del (&vma->vma_link); -drm_mm_remove_node (&vma->node);i915_gem_vma_destroy (vma);/* Since the unbound list is global, only move to that list if
@@ -3176,7 +3174,6 @@ search_free:return 0 ; err_out: -drm_mm_remove_node (&vma->node);i915_gem_vma_destroy (vma);i915_gem_object_unpin_pages (obj);return ret;
@@ -4020,7 +4017,8 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma) { -WARN_ON (vma->node.allocated); +list_del_init (&vma->vma_link); +drm_mm_remove_node (&vma->node);kfree (vma); }
Signature · 1 line
1.8.3.4
PATCH 17/29 drm/i915: plumb VM into bind/unbind code
In reply to Ben WidawskyAs alluded to in several patches, and it will be reiterated later... A VMA is an abstraction for a GEM BO bound into an address space. Therefore it stands to reason, that the existing bind, and unbind are the ones which will be the most impacted. This patch implements this, and updates all callers which weren't already updated in the series (because it was too messy).
This patch represents the bulk of an earlier, larger patch. I've pulled out a bunch of things by the request of Daniel. The history is preserved for posterity with the email convention of ">" One big change from the original patch aside from a bunch of cropping is I've created an i915_vma_unbind() function. That is because we always have the VMA anyway, and doing an extra lookup is useful. There is a caveat, we retain an i915_gem_object_ggtt_unbind, for the global cases which might not talk in VMAs.
Quoted 34 lines
drm/i915: plumb VM into object operations
This patch was formerly known as:
"drm/i915: Create VMAs (part 3) - plumbing"This patch adds a VM argument, bind/unbind, and the object offset/size/color getters/setters. It preserves the old ggtt helper functions because things still need, and will continue to need them.
Some code will still need to be ported over after this.
v2: Fix purge to pick an object and unbind all vmas This was doable because of the global bound list change.
v3: With the commit to actually pin/unpin pages in place, there is no longer a need to check if unbind succeeded before calling
put_pages(). Makeput_pagesonlyBUG()after checking pin count.v4: Rebased on top of the new hangcheck work by Mika plumbed
eb_destroyalso
Many checkpatch related fixesv5: Very large rebase
v6:
ChangeBUG_ONtoWARN_ON(Daniel)
Rename vm to ggtt in preallocate stolen, since it is always ggtt when dealing with stolen memory. (Daniel)list_for_eachwill short-circuit already (Daniel)
remove superflous space (Daniel)
Use per object list of vmas (Daniel)
Makeobj_bound_any()useobj_boundfor each vm (Ben) s/bind_to_gtt/bind_to_vm/ (Ben)
Quoted 7 lines
Fixed up the inactive shrinker. As Daniel noticed the code could potentially count the same object multiple times. While it's not possible in the current case, since 1 object can only ever be bound into 1 address space thus far - we may as well try to get something more future proof in place now. With a prep patch before this to switch over to using the bound list + inactive check, we're now able to carry that forward for every address space an object is bound into.
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 3 +- drivers/gpu/drm/i915/i915_gem.c | 134 +++++++++++++++++++---------- drivers/gpu/drm/i915/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- drivers/gpu/drm/i915/i915_gem_tiling.c | 9 +- drivers/gpu/drm/i915/i915_trace.h | 37 ++++---- 7 files changed, 120 insertions(+), 71 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +1 -1
@@ -1796,7 +1796,7 @@ i915_drop_caches_set(void *data, u64 val) mm_list) {if (obj->pin_count)continue ; - ret =i915_gem_object_unbind (obj); + ret =i915_gem_object_ggtt_unbind (obj);if (ret) goto unlock; }
drivers/gpu/drm/i915/i915_drv.h +2 -1
@@ -1700,7 +1700,8 @@ int __must_check i915_gem_object_pin(struct drm_i915_gem_object *obj,bool map_and_fenceable, bool nonblocking);void i915_gem_object_unpin (struct drm_i915_gem_object *obj); -int __must_check i915_gem_object_unbind (struct drm_i915_gem_object *obj); +int __must_check i915_vma_unbind (struct i915_vma *vma); +int __must_check i915_gem_object_ggtt_unbind (struct drm_i915_gem_object *obj);int i915_gem_object_put_pages (struct drm_i915_gem_object *obj);void i915_gem_release_mmap (struct drm_i915_gem_object *obj);void i915_gem_lastclose (struct drm_device *dev);
drivers/gpu/drm/i915/i915_gem.c +88 -46
267 lines not on this page · raw message
drivers/gpu/drm/i915/i915_gem_evict.c +2 -2
@@ -147,7 +147,7 @@ found: struct drm_i915_gem_object, exec_list);if (ret ==0 ) - ret =i915_gem_object_unbind (obj); + ret =i915_gem_object_ggtt_unbind (obj);list_del_init (&obj->exec_list);drm_gem_object_unreference (&obj->base);
@@ -185,7 +185,7 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list)if (obj->pin_count ==0 ) -WARN_ON (i915_gem_object_unbind (obj)); +WARN_ON (i915_gem_object_ggtt_unbind (obj));return 0 ; }
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -1
@@ -556,7 +556,7 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,if ((entry->alignment && obj_offset & (entry->alignment -1 )) || (need_mappable && !obj->map_and_fenceable)) - ret =i915_gem_object_unbind (obj); + ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm));else ret =i915_gem_execbuffer_reserve_object (obj, ring, vm, need_relocs);if (ret)
drivers/gpu/drm/i915/i915_gem_tiling.c +5 -4
@@ -360,17 +360,18 @@ i915_gem_set_tiling(struct drm_device *dev, void *data, obj->map_and_fenceable = !i915_gem_obj_ggtt_bound (obj) || - (i915_gem_obj_ggtt_offset (obj) + obj->base.size <= dev_priv->gtt.mappable_end && + (i915_gem_obj_ggtt_offset (obj) + + obj->base.size <= dev_priv->gtt.mappable_end &&i915_gem_object_fence_ok (obj, args->tiling_mode));/* Rebind if we need a change of alignment */ if (!obj->map_and_fenceable) { -u32 unfenced_alignment = +u32 unfenced_align =i915_gem_get_gtt_alignment (dev, obj->base.size, args->tiling_mode, false); -if (i915_gem_obj_ggtt_offset (obj) & (unfenced_alignment -1 )) - ret =i915_gem_object_unbind (obj); +if (i915_gem_obj_ggtt_offset (obj) & (unfenced_align -1 )) + ret =i915_gem_object_ggtt_unbind (obj); }if (ret ==0 ) {
drivers/gpu/drm/i915/i915_trace.h +21 -16
@@ -33,47 +33,52 @@ TRACE_EVENT(i915_gem_object_create,TP_printk ("obj=%p, size=%u" , __entry->obj, __entry->size) ); -TRACE_EVENT (i915_gem_object_bind, - TP_PROTO(struct drm_i915_gem_object *obj, bool mappable), -TP_ARGS (obj, mappable), +TRACE_EVENT (i915_vma_bind, + TP_PROTO(struct i915_vma *vma, bool mappable), +TP_ARGS (vma, mappable),TP_STRUCT__entry ( __field(struct drm_i915_gem_object *, obj) + __field(struct i915_address_space *, vm) __field(u32, offset) __field(u32, size)__field (bool, mappable) ),TP_fast_assign ( - __entry->obj = obj; - __entry->offset = i915_gem_obj_ggtt_offset(obj); - __entry->size = i915_gem_obj_ggtt_size(obj); + __entry->obj = vma->obj; + __entry->vm = vma->vm; + __entry->offset = vma->node.start; + __entry->size = vma->node.size; __entry->mappable = mappable; ), -TP_printk ("obj=%p, offset=%08x size=%x%s" , +TP_printk ("obj=%p, offset=%08x size=%x%s vm=%p" , __entry->obj, __entry->offset, __entry->size, - __entry->mappable ?", mappable" :"" ) + __entry->mappable ?", mappable" :"" , + __entry->vm) ); -TRACE_EVENT (i915_gem_object_unbind, - TP_PROTO(struct drm_i915_gem_object *obj), -TP_ARGS (obj), +TRACE_EVENT (i915_vma_unbind, + TP_PROTO(struct i915_vma *vma), +TP_ARGS (vma),TP_STRUCT__entry ( __field(struct drm_i915_gem_object *, obj) + __field(struct i915_address_space *, vm) __field(u32, offset)__field (u32, size) ),TP_fast_assign ( - __entry->obj = obj; - __entry->offset = i915_gem_obj_ggtt_offset(obj); - __entry->size = i915_gem_obj_ggtt_size(obj); + __entry->obj = vma->obj; + __entry->vm = vma->vm; + __entry->offset = vma->node.start; + __entry->size = vma->node.size; ), -TP_printk ("obj=%p, offset=%08x size=%x" , - __entry->obj, __entry->offset, __entry->size) +TP_printk ("obj=%p, offset=%08x size=%x vm=%p" , + __entry->obj, __entry->offset, __entry->size, __entry->vm) ); TRACE_EVENT(i915_gem_object_change_domain,
Signature · 1 line
1.8.3.4
Re: [PATCH 17/29] drm/i915: plumb VM into bind/unbind code
Quoted 58 lines · Ben Widawsky
As alluded to in several patches, and it will be reiterated later... A VMA is an abstraction for a GEM BO bound into an address space. Therefore it stands to reason, that the existing bind, and unbind are the ones which will be the most impacted. This patch implements this, and updates all callers which weren't already updated in the series (because it was too messy).
This patch represents the bulk of an earlier, larger patch. I've pulled out a bunch of things by the request of Daniel. The history is preserved for posterity with the email convention of ">" One big change from the original patch aside from a bunch of cropping is I've created an
i915_vma_unbind()function. That is because we always have the VMA anyway, and doing an extra lookup is useful. There is a caveat, we retain ani915_gem_object_ggtt_unbind, for the global cases which might not talk in VMAs.Quoted 34 lines
drm/i915: plumb VM into object operations
This patch was formerly known as:
"drm/i915: Create VMAs (part 3) - plumbing"This patch adds a VM argument, bind/unbind, and the object offset/size/color getters/setters. It preserves the old ggtt helper functions because things still need, and will continue to need them.
Some code will still need to be ported over after this.
v2: Fix purge to pick an object and unbind all vmas This was doable because of the global bound list change.
v3: With the commit to actually pin/unpin pages in place, there is no longer a need to check if unbind succeeded before calling
put_pages(). Makeput_pagesonlyBUG()after checking pin count.v4: Rebased on top of the new hangcheck work by Mika plumbed
eb_destroyalso
Many checkpatch related fixesv5: Very large rebase
v6:
ChangeBUG_ONtoWARN_ON(Daniel)
Rename vm to ggtt in preallocate stolen, since it is always ggtt when dealing with stolen memory. (Daniel)list_for_eachwill short-circuit already (Daniel)
remove superflous space (Daniel)
Use per object list of vmas (Daniel)
Makeobj_bound_any()useobj_boundfor each vm (Ben) s/bind_to_gtt/bind_to_vm/ (Ben)Quoted 7 lines
Fixed up the inactive shrinker. As Daniel noticed the code could potentially count the same object multiple times. While it's not possible in the current case, since 1 object can only ever be bound into 1 address space thus far - we may as well try to get something more future proof in place now. With a prep patch before this to switch over to using the bound list + inactive check, we're now able to carry that forward for every address space an object is bound into.
Quoted 1 line
Signed-off-by Ben Widawsky <[email protected]>
A bunch of comments below. Overall that patch is imo really easy to review and the comments can all be addressed in follow-up patches (if at all). But I think I've spotted a leak in object_pin.
-Daniel
Quoted 95 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 3 +- drivers/gpu/drm/i915/i915_gem.c | 134 +++++++++++++++++++---------- drivers/gpu/drm/i915/i915_gem_evict.c | 4 +- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- drivers/gpu/drm/i915/i915_gem_tiling.c | 9 +- drivers/gpu/drm/i915/i915_trace.h | 37 ++++---- 7 files changed, 120 insertions(+), 71 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +1
-1@@ -1796,7 +1796,7 @@ i915_drop_caches_set(void *data, u64 val) mm_list) {if (obj->pin_count)continue ; - ret =i915_gem_object_unbind (obj); + ret =i915_gem_object_ggtt_unbind (obj);if (ret) goto unlock; }drivers/gpu/drm/i915/i915_drv.h +2
-1@@ -1700,7 +1700,8 @@ int __must_check i915_gem_object_pin(struct drm_i915_gem_object *obj,bool map_and_fenceable, bool nonblocking);void i915_gem_object_unpin (struct drm_i915_gem_object *obj); -int __must_check i915_gem_object_unbind (struct drm_i915_gem_object *obj); +int __must_check i915_vma_unbind (struct i915_vma *vma); +int __must_check i915_gem_object_ggtt_unbind (struct drm_i915_gem_object *obj);int i915_gem_object_put_pages (struct drm_i915_gem_object *obj);void i915_gem_release_mmap (struct drm_i915_gem_object *obj);void i915_gem_lastclose (struct drm_device *dev);drivers/gpu/drm/i915/i915_gem.c +12
-13@@ -38,10 +38,12 @@static void i915_gem_object_flush_gtt_write_domain (struct drm_i915_gem_object *obj);static void i915_gem_object_flush_cpu_write_domain (struct drm_i915_gem_object *obj); -static __must_check int i915_gem_object_bind_to_gtt (struct drm_i915_gem_object *obj, -unsigned alignment, -bool map_and_fenceable, -bool nonblocking); +static __must_check int +i915_gem_object_bind_to_vm (struct drm_i915_gem_object *obj, +struct i915_address_space *vm, +unsigned alignment, +bool map_and_fenceable, +bool nonblocking); static int i915_gem_phys_pwrite(struct drm_device *dev, struct drm_i915_gem_object *obj, struct drm_i915_gem_pwrite *args,@@ -1678,7 +1680,6 @@ __i915_gem_shrink(struct drm_i915_private *dev_priv, long target, bool purgeable_only) {struct drm_i915_gem_object *obj, *next; -struct i915_address_space *vm = &dev_priv->gtt.base;long count =0 ; list_for_each_entry_safe(obj, next,@@ -1692,13 +1693,16 @@ __i915_gem_shrink(struct drm_i915_private *dev_priv, long target, } } -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) { +list_for_each_entry_safe (obj, next, &dev_priv->mm.bound_list, + global_list) { +struct i915_vma *vma, *v;if (!i915_gem_object_is_purgeable (obj) && purgeable_only)continue ; -if (i915_gem_object_unbind (obj)) -continue ; +list_for_each_entry_safe (vma, v, &obj->vma_list, vma_link) +if (i915_vma_unbind (vma)) +break ;if (!i915_gem_object_put_pages (obj)) { count += obj->base.size >>PAGE_SHIFT ;@@ -2591,17 +2595,13 @@ static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj) old_write_domain); } -/** -* Unbinds an object from the GTT aperture. -*/ -int -i915_gem_object_unbind (struct drm_i915_gem_object *obj)
drivers/gpu/drm/i915/i915_gem.c static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj) · line 2605
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +3
-2+int i915_vma_unbind (struct i915_vma *vma) { +struct drm_i915_gem_object *obj = vma->obj;drm_i915_private_t *dev_priv = obj->base.dev->dev_private; -struct i915_vma *vma;int ret; -if (!i915_gem_obj_ggtt_bound (obj)) +if (list_empty (&vma->vma_link))return 0 ;
This smells like something which should never be the case. Add a WARN_ON in a follow-up patch?
Quoted 73 lines
drivers/gpu/drm/i915/i915_gem.c +28
-9if (obj->pin_count)@@ -2624,7 +2624,7 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)if (ret)return ret; -trace_i915_gem_object_unbind (obj); +trace_i915_vma_unbind (vma);if (obj->has_global_gtt_mapping)i915_gem_gtt_unbind_object (obj);@@ -2639,7 +2639,6 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)/* Avoid an unnecessary call to unbind on rebind. */ obj->map_and_fenceable = true; - vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base);i915_gem_vma_destroy (vma);/* Since the unbound list is global, only move to that list if @@ -2652,6 +2651,26 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)return 0 ; } +/** +* Unbinds an object from the global GTT aperture. +*/ +int +i915_gem_object_ggtt_unbind (struct drm_i915_gem_object *obj) +{ +struct drm_i915_private *dev_priv = obj->base.dev->dev_private; +struct i915_address_space *ggtt = &dev_priv->gtt.base; + +if (!i915_gem_obj_ggtt_bound (obj)); +return 0 ; + +if (obj->pin_count) +return -EBUSY ; + +BUG_ON (obj->pages ==NULL ); + +return i915_vma_unbind (i915_gem_obj_to_vma (obj, ggtt)); +} +int i915_gpu_idle (struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private;@@ -3069,18 +3088,18 @@ static void i915_gem_verify_gtt(struct drm_device *dev)* Finds free space in the GTT aperture and binds the object there. */ static int -i915_gem_object_bind_to_gtt (struct drm_i915_gem_object *obj, -unsigned alignment, -bool map_and_fenceable, -bool nonblocking) +i915_gem_object_bind_to_vm (struct drm_i915_gem_object *obj, +struct i915_address_space *vm, +unsigned alignment, +bool map_and_fenceable, +bool nonblocking) {struct drm_device *dev = obj->base.dev;drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;u32 size, fence_size, fence_alignment, unfenced_alignment;bool mappable, fenceable; -size_t gtt_max = map_and_fenceable ? - dev_priv->gtt.mappable_end : dev_priv->gtt.base.total; +size_t gtt_max = + map_and_fenceable ? dev_priv->gtt.mappable_end : vm->total;struct i915_vma *vma;int ret;
drivers/gpu/drm/i915/i915_gem.c i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj, · line 3148
Quoted 6 lines
drivers/gpu/drm/i915/i915_gem.c +2
-1@@ -3125,15 +3144,18 @@ i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,i915_gem_object_pin_pages (obj); - vma =i915_gem_vma_create (obj, &dev_priv->gtt.base); +/* FIXME: For now we only ever use 1 VMA per object */ +BUG_ON (!i915_is_ggtt (vm));
I'll change this to a WARN_ON for now ... My upfront apologies for the rebase conflict.
Quoted 74 lines
drivers/gpu/drm/i915/i915_gem.c +24
-9+WARN_ON (!list_empty (&obj->vma_list)); + + vma =i915_gem_vma_create (obj, vm);if (IS_ERR (vma)) {i915_gem_object_unpin_pages (obj);return PTR_ERR (vma); } search_free: - ret =drm_mm_insert_node_in_range_generic (&dev_priv->gtt.base.mm, - &vma->node, + ret =drm_mm_insert_node_in_range_generic (&vm->mm, &vma->node, size, alignment, obj->cache_level,0 , gtt_max);if (ret) {@@ -3158,18 +3180,25 @@ search_free:list_move_tail (&obj->global_list, &dev_priv->mm.bound_list);list_add_tail (&obj->mm_list, &vm->inactive_list); -list_add (&vma->vma_link, &obj->vma_list); + +/* Keep GGTT vmas first to make debug easier */ +if (i915_is_ggtt (vm)) +list_add (&vma->vma_link, &obj->vma_list); +else +list_add_tail (&vma->vma_link, &obj->vma_list); fenceable = +i915_is_ggtt (vm) &&i915_gem_obj_ggtt_size (obj) == fence_size && (i915_gem_obj_ggtt_offset (obj) & (fence_alignment -1 )) ==0 ; - mappable =i915_gem_obj_ggtt_offset (obj) + obj->base.size <= - dev_priv->gtt.mappable_end; + mappable = +i915_is_ggtt (vm) && + vma->node.start + obj->base.size <= dev_priv->gtt.mappable_end; obj->map_and_fenceable = mappable && fenceable; -trace_i915_gem_object_bind (obj, map_and_fenceable); +trace_i915_vma_bind (vma, map_and_fenceable);i915_gem_verify_gtt (dev);return 0 ;@@ -3335,7 +3364,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,list_for_each_entry (vma, &obj->vma_list, vma_link) {if (!i915_gem_valid_gtt_space (dev, &vma->node, cache_level)) { - ret =i915_gem_object_unbind (obj); + ret =i915_vma_unbind (vma);if (ret)return ret;@@ -3643,33 +3672,39 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj, bool map_and_fenceable, bool nonblocking) { +struct i915_vma *vma;int ret;if (WARN_ON (obj->pin_count ==DRM_I915_GEM_OBJECT_MAX_PIN_COUNT ))return -EBUSY ; -if (i915_gem_obj_ggtt_bound (obj)) { -if ((alignment &&i915_gem_obj_ggtt_offset (obj) & (alignment -1 )) || +WARN_ON (map_and_fenceable && !i915_is_ggtt (vm)); + + vma =i915_gem_obj_to_vma (obj, vm); + +if (vma) { +if ((alignment && + vma->node.start & (alignment -1 )) || (map_and_fenceable && !obj->map_and_fenceable)) {
drivers/gpu/drm/i915/i915_gem.c i915_gem_object_pin(struct drm_i915_gem_object *obj, · line 3696
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +2
-2WARN (obj->pin_count,"bo is already pinned with incorrect alignment:" " offset=%lx, req.alignment=%x, req.map_and_fenceable=%d," " obj->map_and_fenceable=%d\n" , -i915_gem_obj_ggtt_offset (obj), alignment, +i915_gem_obj_offset (obj, vm), alignment, map_and_fenceable, obj->map_and_fenceable); - ret =i915_gem_object_unbind (obj); + ret =i915_vma_unbind (vma);
If I read this correctly then we wont' call i915_gem_vma_destroy anymore and so will leak the vma. Is that correct? If so I guess a new slab for vmas could be handy to easily detect such bugs.
Quoted 41 lines
drivers/gpu/drm/i915/i915_gem.c +12
-7if (ret)return ret; } } -if (!i915_gem_obj_ggtt_bound (obj)) { +if (!i915_gem_obj_bound (obj, vm)) {struct drm_i915_private *dev_priv = obj->base.dev->dev_private; - ret =i915_gem_object_bind_to_gtt (obj, alignment, - map_and_fenceable, - nonblocking); + ret =i915_gem_object_bind_to_vm (obj, vm, alignment, + map_and_fenceable, + nonblocking);if (ret)return ret;@@ -3961,6 +3996,7 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)struct drm_i915_gem_object *obj =to_intel_bo (gem_obj);struct drm_device *dev = obj->base.dev;drm_i915_private_t *dev_priv = dev->dev_private; +struct i915_vma *vma, *next;trace_i915_gem_object_destroy (obj);@@ -3968,15 +4004,21 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)i915_gem_detach_phys_object (dev, obj); obj->pin_count =0 ; -if (WARN_ON (i915_gem_object_unbind (obj) == -ERESTARTSYS )) { -bool was_interruptible; +/* NB: 0 or 1 elements */ +WARN_ON (!list_empty (&obj->vma_list) && + !list_is_singular (&obj->vma_list)); +list_for_each_entry_safe (vma, next, &obj->vma_list, vma_link) { +int ret =i915_vma_unbind (vma); +if (WARN_ON (ret == -ERESTARTSYS )) { +bool was_interruptible; - was_interruptible = dev_priv->mm.interruptible;
drivers/gpu/drm/i915/i915_gem.c void i915_gem_free_object(struct drm_gem_object *gem_obj) · line 4021
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +5
-3- dev_priv->mm.interruptible = false; + was_interruptible = dev_priv->mm.interruptible; + dev_priv->mm.interruptible = false; -WARN_ON (i915_gem_object_unbind (obj)); +WARN_ON (i915_vma_unbind (vma)); - dev_priv->mm.interruptible = was_interruptible; + dev_priv->mm.interruptible = was_interruptible; + }
Hm, I think we've released sufficient amounts of kernels and never hit the above WARN_ON. Can I volunteer you for a follow-up patch to rip out the code here (but keep the WARN_ON ofc)?
Quoted 145 lines
drivers/gpu/drm/i915/i915_gem.c +0
-0}/* Stolen objects don't hold a ref, but do hold pin count. Fix that up drivers/gpu/drm/i915/i915_gem_evict.c +2
-2@@ -147,7 +147,7 @@ found: struct drm_i915_gem_object, exec_list);if (ret ==0 ) - ret =i915_gem_object_unbind (obj); + ret =i915_gem_object_ggtt_unbind (obj);list_del_init (&obj->exec_list);drm_gem_object_unreference (&obj->base);@@ -185,7 +185,7 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list)if (obj->pin_count ==0 ) -WARN_ON (i915_gem_object_unbind (obj)); +WARN_ON (i915_gem_object_ggtt_unbind (obj));return 0 ; }drivers/gpu/drm/i915/i915_gem_execbuffer.c +1
-1@@ -556,7 +556,7 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,if ((entry->alignment && obj_offset & (entry->alignment -1 )) || (need_mappable && !obj->map_and_fenceable)) - ret =i915_gem_object_unbind (obj); + ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm));else ret =i915_gem_execbuffer_reserve_object (obj, ring, vm, need_relocs);if (ret)drivers/gpu/drm/i915/i915_gem_tiling.c +5
-4@@ -360,17 +360,18 @@ i915_gem_set_tiling(struct drm_device *dev, void *data, obj->map_and_fenceable = !i915_gem_obj_ggtt_bound (obj) || - (i915_gem_obj_ggtt_offset (obj) + obj->base.size <= dev_priv->gtt.mappable_end && + (i915_gem_obj_ggtt_offset (obj) + + obj->base.size <= dev_priv->gtt.mappable_end &&i915_gem_object_fence_ok (obj, args->tiling_mode));/* Rebind if we need a change of alignment */ if (!obj->map_and_fenceable) { -u32 unfenced_alignment = +u32 unfenced_align =i915_gem_get_gtt_alignment (dev, obj->base.size, args->tiling_mode, false); -if (i915_gem_obj_ggtt_offset (obj) & (unfenced_alignment -1 )) - ret =i915_gem_object_unbind (obj); +if (i915_gem_obj_ggtt_offset (obj) & (unfenced_align -1 )) + ret =i915_gem_object_ggtt_unbind (obj); }if (ret ==0 ) {drivers/gpu/drm/i915/i915_trace.h +21
-16@@ -33,47 +33,52 @@ TRACE_EVENT(i915_gem_object_create,TP_printk ("obj=%p, size=%u" , __entry->obj, __entry->size) ); -TRACE_EVENT (i915_gem_object_bind, - TP_PROTO(struct drm_i915_gem_object *obj, bool mappable), -TP_ARGS (obj, mappable), +TRACE_EVENT (i915_vma_bind, + TP_PROTO(struct i915_vma *vma, bool mappable), +TP_ARGS (vma, mappable),TP_STRUCT__entry ( __field(struct drm_i915_gem_object *, obj) + __field(struct i915_address_space *, vm) __field(u32, offset) __field(u32, size)__field (bool, mappable) ),TP_fast_assign ( - __entry->obj = obj; - __entry->offset = i915_gem_obj_ggtt_offset(obj); - __entry->size = i915_gem_obj_ggtt_size(obj); + __entry->obj = vma->obj; + __entry->vm = vma->vm; + __entry->offset = vma->node.start; + __entry->size = vma->node.size; __entry->mappable = mappable; ), -TP_printk ("obj=%p, offset=%08x size=%x%s" , +TP_printk ("obj=%p, offset=%08x size=%x%s vm=%p" , __entry->obj, __entry->offset, __entry->size, - __entry->mappable ?", mappable" :"" ) + __entry->mappable ?", mappable" :"" , + __entry->vm) ); -TRACE_EVENT (i915_gem_object_unbind, - TP_PROTO(struct drm_i915_gem_object *obj), -TP_ARGS (obj), +TRACE_EVENT (i915_vma_unbind, + TP_PROTO(struct i915_vma *vma), +TP_ARGS (vma),TP_STRUCT__entry ( __field(struct drm_i915_gem_object *, obj) + __field(struct i915_address_space *, vm) __field(u32, offset)__field (u32, size) ),TP_fast_assign ( - __entry->obj = obj; - __entry->offset = i915_gem_obj_ggtt_offset(obj); - __entry->size = i915_gem_obj_ggtt_size(obj); + __entry->obj = vma->obj; + __entry->vm = vma->vm; + __entry->offset = vma->node.start; + __entry->size = vma->node.size; ), -TP_printk ("obj=%p, offset=%08x size=%x" , - __entry->obj, __entry->offset, __entry->size) +TP_printk ("obj=%p, offset=%08x size=%x vm=%p" , + __entry->obj, __entry->offset, __entry->size, __entry->vm) ); TRACE_EVENT(i915_gem_object_change_domain,Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 17/29] drm/i915: plumb VM into bind/unbind code
Quoted 1 line · Daniel Vetter
On Wed, Jul 31, 2013 at 05:00:10PM -0700, Ben Widawsky wrote:
[snip]
Quoted 26 lines
Quoted 26 lines
drivers/gpu/drm/i915/i915_gem.c +9
-3@@ -3643,33 +3672,39 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj, bool map_and_fenceable, bool nonblocking) { +struct i915_vma *vma;int ret;if (WARN_ON (obj->pin_count ==DRM_I915_GEM_OBJECT_MAX_PIN_COUNT ))return -EBUSY ; -if (i915_gem_obj_ggtt_bound (obj)) { -if ((alignment &&i915_gem_obj_ggtt_offset (obj) & (alignment -1 )) || +WARN_ON (map_and_fenceable && !i915_is_ggtt (vm)); + + vma =i915_gem_obj_to_vma (obj, vm); + +if (vma) { +if ((alignment && + vma->node.start & (alignment -1 )) || (map_and_fenceable && !obj->map_and_fenceable)) { WARN(obj->pin_count, "bo is already pinned with incorrect alignment:" " offset=%lx, req.alignment=%x, req.map_and_fenceable=%d," " obj->map_and_fenceable=%d\n", - i915_gem_obj_ggtt_offset(obj), alignment, + i915_gem_obj_offset(obj, vm), alignment,
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c i915_gem_object_pin(struct drm_i915_gem_object *obj, · line 3696
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem.c +1
-1map_and_fenceable, obj->map_and_fenceable); - ret =i915_gem_object_unbind (obj); + ret =i915_vma_unbind (vma);If I read this correctly then we wont' call
i915_gem_vma_destroyanymore and so will leak the vma. Is that correct? If so I guess a new slab for vmas could be handy to easily detect such bugs.
On re-reading all seems to be fine here since object_unbind was converted to vma_unbind and so inherited the call to vma_destroy. So no leak here. The other stuff isn't really critical, so I'll merge this patch (and the next one).
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 18/29 drm/i915: Use new bind/unbind in eviction code
In reply to Ben WidawskyEviction code, like the rest of the converted code needs to be aware of the address space for which it is evicting (or the everything case, all addresses). With the updated bind/unbind interfaces of the last patch, we can now safely move the eviction code over.
--- drivers/gpu/drm/i915/i915_drv.h | 4 ++- drivers/gpu/drm/i915/i915_gem.c | 2 +- drivers/gpu/drm/i915/i915_gem_evict.c | 53 +++++++++++++++++++---------------- 3 files changed, 33 insertions(+), 26 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +3 -1
@@ -1946,7 +1946,9 @@ static inline void i915_gem_chipset_flush(struct drm_device *dev)/* i915_gem_evict.c */ -int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size, +int __must_check i915_gem_evict_something(struct drm_device *dev, + struct i915_address_space *vm, + int min_size, unsigned alignment, unsigned cache_level, bool mappable,
drivers/gpu/drm/i915/i915_gem.c +1 -1
@@ -3159,7 +3159,7 @@ search_free: size, alignment, obj->cache_level, 0, gtt_max);if (ret) { - ret =i915_gem_evict_something (dev, size, alignment, + ret =i915_gem_evict_something (dev, vm, size, alignment, obj->cache_level, map_and_fenceable, nonblocking);
drivers/gpu/drm/i915/i915_gem_evict.c +29 -24
121 lines not on this page · raw message
Signature · 1 line
1.8.3.4
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 6 lines · Ben Widawsky
Eviction code, like the rest of the converted code needs to be aware of the address space for which it is evicting (or the everything case, all addresses). With the updated bind/unbind interfaces of the last patch, we can now safely move the eviction code over.
Signed-off-by Ben Widawsky <[email protected]>
Two comments below.
-Daniel
Quoted 121 lines
--- drivers/gpu/drm/i915/i915_drv.h | 4 ++- drivers/gpu/drm/i915/i915_gem.c | 2 +- drivers/gpu/drm/i915/i915_gem_evict.c | 53 +++++++++++++++++++---------------- 3 files changed, 33 insertions(+), 26 deletions(-)drivers/gpu/drm/i915/i915_drv.h +3
-1@@ -1946,7 +1946,9 @@ static inline void i915_gem_chipset_flush(struct drm_device *dev)/* i915_gem_evict.c */ -int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size, +int __must_check i915_gem_evict_something(struct drm_device *dev, + struct i915_address_space *vm, + int min_size, unsigned alignment, unsigned cache_level, bool mappable,drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -3159,7 +3159,7 @@ search_free: size, alignment, obj->cache_level, 0, gtt_max);if (ret) { - ret =i915_gem_evict_something (dev, size, alignment, + ret =i915_gem_evict_something (dev, vm, size, alignment, obj->cache_level, map_and_fenceable, nonblocking);drivers/gpu/drm/i915/i915_gem_evict.c +14
-16@@ -32,26 +32,21 @@#include "i915_trace.h" static bool -mark_free (struct drm_i915_gem_object *obj,struct list_head *unwind) +mark_free (struct i915_vma *vma,struct list_head *unwind) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); - -if (obj->pin_count) +if (vma->obj->pin_count)return false; -list_add (&obj->exec_list, unwind); +list_add (&vma->obj->exec_list, unwind);return drm_mm_scan_add_block (&vma->node); }int -i915_gem_evict_something (struct drm_device *dev,int min_size, -unsigned alignment,unsigned cache_level, +i915_gem_evict_something (struct drm_device *dev,struct i915_address_space *vm, +int min_size,unsigned alignment,unsigned cache_level,bool mappable,bool nonblocking) {drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;struct list_head eviction_list, unwind_list;struct i915_vma *vma;struct drm_i915_gem_object *obj;@@ -83,16 +78,18 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,*/ INIT_LIST_HEAD (&unwind_list); -if (mappable) +if (mappable) { +BUG_ON (!i915_is_ggtt (vm));drm_mm_init_scan_with_range (&vm->mm, min_size, alignment, cache_level,0 , dev_priv->gtt.mappable_end); -else + }else drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ list_for_each_entry (obj, &vm->inactive_list, mm_list) { -if (mark_free (obj, &unwind_list)) +struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +if (mark_free (vma, &unwind_list)) goto found; }@@ -101,7 +98,8 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,/* Now merge in the soon-to-be-expired objects... */ list_for_each_entry (obj, &vm->active_list, mm_list) { -if (mark_free (obj, &unwind_list)) +struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +if (mark_free (vma, &unwind_list)) goto found; }@@ -111,7 +109,7 @@ none: obj =list_first_entry (&unwind_list, struct drm_i915_gem_object, exec_list); - vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); + vma =i915_gem_obj_to_vma (obj, vm); ret =drm_mm_scan_remove_block (&vma->node);BUG_ON (ret);@@ -132,7 +130,7 @@ found: obj =list_first_entry (&unwind_list, struct drm_i915_gem_object, exec_list); - vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); + vma =i915_gem_obj_to_vma (obj, vm);if (drm_mm_scan_remove_block (&vma->node)) {list_move (&obj->exec_list, &eviction_list);drm_gem_object_reference (&obj->base);
drivers/gpu/drm/i915/i915_gem_evict.c found: · line 148
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_evict.c +1
-1@@ -147,7 +145,7 @@ found: struct drm_i915_gem_object, exec_list);if (ret ==0 ) - ret =i915_gem_object_ggtt_unbind (obj); + ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm));
Again I think the ggtt_unbind->vma_unbind conversion seems to leak the vma. It feels like vma_unbind should call vma_destroy?
Quoted 28 lines
drivers/gpu/drm/i915/i915_gem_evict.c +9
-4list_del_init (&obj->exec_list);drm_gem_object_unreference (&obj->base);@@ -160,13 +158,18 @@ int i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm;struct drm_i915_gem_object *obj, *next; -bool lists_empty; +bool lists_empty = true;int ret; - lists_empty = (list_empty (&vm->inactive_list) && -list_empty (&vm->active_list)); +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { + lists_empty = (list_empty (&vm->inactive_list) && +list_empty (&vm->active_list)); +if (!lists_empty) + lists_empty = false; + } +if (lists_empty)return -ENOSPC ;@@ -183,9 +186,11 @@ i915_gem_evict_everything(struct drm_device *dev)i915_gem_retire_requests (dev);
drivers/gpu/drm/i915/i915_gem_evict.c i915_gem_evict_everything(struct drm_device *dev) · line 193
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_evict.c +5
-3/* Having flushed everything, unbind() should never raise an error */ -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_gem_object_ggtt_unbind (obj)); +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) +if (obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); + }
The conversion of evict_everything looks a bit strange. Essentially we have tree callers:
- ums+gem support code in leavevt to rid the gtt of all gem objects when the userspace X ums ddx stops controlling the hw.
- When we seriously ran out of memory in, shrink_all.
- In execbuf when we've fragmented the gtt address space so badly that we need to start over completely fresh.
With this it imo would make sense to just loop over the global bound object lists. But for the execbuf caller adding a vm parameter (and only evicting from that special vm, skipping all others) would make sense. Other callers would pass NULL since they want everything to get evicted. Volunteered for that follow-up?
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_evict.c +0
-0return 0 ; }Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 132 lines · Daniel Vetter
Quoted 6 lines · Ben Widawsky
Eviction code, like the rest of the converted code needs to be aware of the address space for which it is evicting (or the everything case, all addresses). With the updated bind/unbind interfaces of the last patch, we can now safely move the eviction code over.
Signed-off-by Ben Widawsky <[email protected]>Two comments below.
-DanielQuoted 121 lines
--- drivers/gpu/drm/i915/i915_drv.h | 4 ++- drivers/gpu/drm/i915/i915_gem.c | 2 +- drivers/gpu/drm/i915/i915_gem_evict.c | 53 +++++++++++++++++++---------------- 3 files changed, 33 insertions(+), 26 deletions(-)drivers/gpu/drm/i915/i915_drv.h +3
-1@@ -1946,7 +1946,9 @@ static inline void i915_gem_chipset_flush(struct drm_device *dev)/* i915_gem_evict.c */ -int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size, +int __must_check i915_gem_evict_something(struct drm_device *dev, + struct i915_address_space *vm, + int min_size, unsigned alignment, unsigned cache_level, bool mappable,drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -3159,7 +3159,7 @@ search_free: size, alignment, obj->cache_level, 0, gtt_max);if (ret) { - ret =i915_gem_evict_something (dev, size, alignment, + ret =i915_gem_evict_something (dev, vm, size, alignment, obj->cache_level, map_and_fenceable, nonblocking);drivers/gpu/drm/i915/i915_gem_evict.c +14
-16@@ -32,26 +32,21 @@#include "i915_trace.h" static bool -mark_free (struct drm_i915_gem_object *obj,struct list_head *unwind) +mark_free (struct i915_vma *vma,struct list_head *unwind) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_vma *vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); - -if (obj->pin_count) +if (vma->obj->pin_count)return false; -list_add (&obj->exec_list, unwind); +list_add (&vma->obj->exec_list, unwind);return drm_mm_scan_add_block (&vma->node); }int -i915_gem_evict_something (struct drm_device *dev,int min_size, -unsigned alignment,unsigned cache_level, +i915_gem_evict_something (struct drm_device *dev,struct i915_address_space *vm, +int min_size,unsigned alignment,unsigned cache_level,bool mappable,bool nonblocking) {drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;struct list_head eviction_list, unwind_list;struct i915_vma *vma;struct drm_i915_gem_object *obj;@@ -83,16 +78,18 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,*/ INIT_LIST_HEAD (&unwind_list); -if (mappable) +if (mappable) { +BUG_ON (!i915_is_ggtt (vm));drm_mm_init_scan_with_range (&vm->mm, min_size, alignment, cache_level,0 , dev_priv->gtt.mappable_end); -else + }else drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ list_for_each_entry (obj, &vm->inactive_list, mm_list) { -if (mark_free (obj, &unwind_list)) +struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +if (mark_free (vma, &unwind_list)) goto found; }@@ -101,7 +98,8 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,/* Now merge in the soon-to-be-expired objects... */ list_for_each_entry (obj, &vm->active_list, mm_list) { -if (mark_free (obj, &unwind_list)) +struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +if (mark_free (vma, &unwind_list)) goto found; }@@ -111,7 +109,7 @@ none: obj =list_first_entry (&unwind_list, struct drm_i915_gem_object, exec_list); - vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); + vma =i915_gem_obj_to_vma (obj, vm); ret =drm_mm_scan_remove_block (&vma->node);BUG_ON (ret);@@ -132,7 +130,7 @@ found: obj =list_first_entry (&unwind_list, struct drm_i915_gem_object, exec_list); - vma =i915_gem_obj_to_vma (obj, &dev_priv->gtt.base); + vma =i915_gem_obj_to_vma (obj, vm);if (drm_mm_scan_remove_block (&vma->node)) {list_move (&obj->exec_list, &eviction_list);drm_gem_object_reference (&obj->base);@@ -147,7 +145,7 @@ found:
Quoted 9 lines
drivers/gpu/drm/i915/i915_gem_evict.c found: · line 148
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_evict.c +1
-1struct drm_i915_gem_object, exec_list);if (ret ==0 ) - ret =i915_gem_object_ggtt_unbind (obj); + ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm));Again I think the
ggtt_unbind->vma_unbindconversion seems to leak the vma. It feels likevma_unbindshould callvma_destroy?
The VMA should get cleaned up when an object backing the vmas is destroyed also. I agree that at present, unbind and destroy have little distinction, but I can foresee that changing. Proper faulting is one such case OTTOMH
Anyway, let me know if your leak concern is addressed.
Quoted 47 lines
Quoted 28 lines
drivers/gpu/drm/i915/i915_gem_evict.c +9
-4list_del_init (&obj->exec_list);drm_gem_object_unreference (&obj->base);@@ -160,13 +158,18 @@ int i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm;struct drm_i915_gem_object *obj, *next; -bool lists_empty; +bool lists_empty = true;int ret; - lists_empty = (list_empty (&vm->inactive_list) && -list_empty (&vm->active_list)); +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { + lists_empty = (list_empty (&vm->inactive_list) && +list_empty (&vm->active_list)); +if (!lists_empty) + lists_empty = false; + } +if (lists_empty)return -ENOSPC ;@@ -183,9 +186,11 @@ i915_gem_evict_everything(struct drm_device *dev)i915_gem_retire_requests (dev);drivers/gpu/drm/i915/i915_gem_evict.c i915_gem_evict_everything(struct drm_device *dev) · line 193
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_evict.c +5
-3/* Having flushed everything, unbind() should never raise an error */ -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_gem_object_ggtt_unbind (obj)); +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) +if (obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); + }The conversion of
evict_everythinglooks a bit strange. Essentially we have tree callers:
- ums+gem support code in leavevt to rid the gtt of all gem objects when the userspace X ums ddx stops controlling the hw.
- When we seriously ran out of memory in,shrink_all.
- In execbuf when we've fragmented the gtt address space so badly that we need to start over completely fresh.
Quoted 6 lines
With this it imo would make sense to just loop over the global bound object lists. But for the execbuf caller adding a vm parameter (and only evicting from that special vm, skipping all others) would make sense. Other callers would pass NULL since they want everything to get evicted. Volunteered for that follow-up?
We need evict_everything for #1. For #2, we call evict_something already for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?
Quoted 3 lines
Quoted 3 lines
drivers/gpu/drm/i915/i915_gem_evict.c +0
-0return 0 ; }
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 30 lines · Ben Widawsky
Quoted 22 lines · Daniel Vetter
drivers/gpu/drm/i915/i915_gem_evict.c i915_gem_evict_everything(struct drm_device *dev) · line 193
Quoted 12 lines · Ben Widawsky
drivers/gpu/drm/i915/i915_gem_evict.c +5
-3@@ -183,9 +186,11 @@ i915_gem_evict_everything(struct drm_device *dev)i915_gem_retire_requests (dev);/* Having flushed everything, unbind() should never raise an error */ -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_gem_object_ggtt_unbind (obj)); +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) +if (obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); + }The conversion of
evict_everythinglooks a bit strange. Essentially we have tree callers:
- ums+gem support code in leavevt to rid the gtt of all gem objects when the userspace X ums ddx stops controlling the hw.
- When we seriously ran out of memory in,shrink_all.
- In execbuf when we've fragmented the gtt address space so badly that we need to start over completely fresh.Quoted 6 lines
With this it imo would make sense to just loop over the global bound object lists. But for the execbuf caller adding a vm parameter (and only evicting from that special vm, skipping all others) would make sense. Other callers would pass NULL since they want everything to get evicted. Volunteered for that follow-up?
Quoted 6 lines
We need
evict_everythingfor #1. For #2, we callevict_somethingalready for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?
Yes, when we've terminally fragmented the gtt we kick out everything and start over. That's the 3rd usecase.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 38 lines · Daniel Vetter
Quoted 30 lines · Ben Widawsky
Quoted 22 lines · Daniel Vetter
drivers/gpu/drm/i915/i915_gem_evict.c i915_gem_evict_everything(struct drm_device *dev) · line 193
Quoted 12 lines · Ben Widawsky
drivers/gpu/drm/i915/i915_gem_evict.c +5
-3@@ -183,9 +186,11 @@ i915_gem_evict_everything(struct drm_device *dev)i915_gem_retire_requests (dev);/* Having flushed everything, unbind() should never raise an error */ -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_gem_object_ggtt_unbind (obj)); +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) +if (obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); + }The conversion of
evict_everythinglooks a bit strange. Essentially we have tree callers:
- ums+gem support code in leavevt to rid the gtt of all gem objects when the userspace X ums ddx stops controlling the hw.
- When we seriously ran out of memory in,shrink_all.
- In execbuf when we've fragmented the gtt address space so badly that we need to start over completely fresh.Quoted 6 lines
With this it imo would make sense to just loop over the global bound object lists. But for the execbuf caller adding a vm parameter (and only evicting from that special vm, skipping all others) would make sense. Other callers would pass NULL since they want everything to get evicted. Volunteered for that follow-up?
Quoted 6 lines
We need
evict_everythingfor #1. For #2, we callevict_somethingalready for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?
Quoted 3 lines
Yes, when we've terminally fragmented the gtt we kick out everything and start over. That's the 3rd usecase.
-Daniel
I am not seeing it. To me evict_something is what you want, and the fix for wherever the 3rd usecase is (please point it out, I'm dense) is it should call evict_something, not evict_everything.
If by GTT you mean the aperture... that's kind of a different can of worms completely. In that case I don't think you want to do anything per VM, though potentially you can do it that way and be a little fairer.
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 7 lines · Ben Widawsky
Quoted 7 lines
Quoted 6 lines
We need
evict_everythingfor #1. For #2, we callevict_somethingalready for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?
Quoted 6 lines
Quoted 2 lines
Yes, when we've terminally fragmented the gtt we kick out everything and start over. That's the 3rd usecase.
I am not seeing it. To me
evict_somethingis what you want, and the fix for wherever the 3rd usecase is (please point it out, I'm dense) is it should callevict_something, notevict_everything.
See the call to evict_everything in
i915_gem_execbuffer.c:i915_gem_execbuffer_reserve
Cheers, Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 11 lines · Daniel Vetter
Quoted 10 lines · Ben Widawsky
Quoted 9 lines
Quoted 6 lines
We need
evict_everythingfor #1. For #2, we callevict_somethingalready for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?Yes, when we've terminally fragmented the gtt we kick out everything and start over. That's the 3rd usecase.
Quoted 7 lines
Quoted 3 lines
I am not seeing it. To me
evict_somethingis what you want, and the fix for wherever the 3rd usecase is (please point it out, I'm dense) is it should callevict_something, notevict_everything.See the call to
evict_everythingin
i915_gem_execbuffer.c:i915_gem_execbuffer_reserve
As I was saying in the first response - you only hit this if evict_something() for a vm fails, right? That's the way ret == ENOSPC AFAICT.
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 16 lines · Ben Widawsky
Quoted 15 lines · Daniel Vetter
Quoted 13 lines · Ben Widawsky
Quoted 9 lines
Quoted 6 lines
We need
evict_everythingfor #1. For #2, we callevict_somethingalready for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?Yes, when we've terminally fragmented the gtt we kick out everything and start over. That's the 3rd usecase.
I am not seeing it. To me
evict_somethingis what you want, and the fix for wherever the 3rd usecase is (please point it out, I'm dense) is it should callevict_something, notevict_everything.
Quoted 7 lines
Quoted 3 lines
See the call to
evict_everythingin
i915_gem_execbuffer.c:i915_gem_execbuffer_reserveAs I was saying in the first response - you only hit this if
evict_something()for a vm fails, right? That's the way ret == ENOSPC AFAICT.
Like I've said if we can't fit a batch we do a last ditch effort of evicting everything and starting over anew. That's also what the retry logic in there is for. This happens after evict_something failed. Dunno what exactly isn't clear or what's confusing ... -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 18/29] drm/i915: Use new bind/unbind in eviction code
Quoted 25 lines · Daniel Vetter
Quoted 16 lines · Ben Widawsky
Quoted 15 lines · Daniel Vetter
Quoted 13 lines · Ben Widawsky
Quoted 9 lines
Quoted 6 lines
We need
evict_everythingfor #1. For #2, we callevict_somethingalready for the vm when we go through the out of space path. If that failed, evicting everything for a specific VM is just the same operation. But maybe I've glossed over something in the details. Please correct if I'm wrong. Is there a case where evict something might fail with ENOSPC, and evict everything in a VM would help?Yes, when we've terminally fragmented the gtt we kick out everything and start over. That's the 3rd usecase.
I am not seeing it. To me
evict_somethingis what you want, and the fix for wherever the 3rd usecase is (please point it out, I'm dense) is it should callevict_something, notevict_everything.Quoted 7 lines
Quoted 3 lines
See the call to
evict_everythingin
i915_gem_execbuffer.c:i915_gem_execbuffer_reserveAs I was saying in the first response - you only hit this if
evict_something()for a vm fails, right? That's the way ret == ENOSPC AFAICT.
Quoted 5 lines
Like I've said if we can't fit a batch we do a last ditch effort of evicting everything and starting over anew. That's also what the retry logic in there is for. This happens after
evict_somethingfailed. Dunno what exactly isn't clear or what's confusing ... -Daniel
Okay, sorted this out on IRC. You'll get a new patch as described with a new function for per vm eviction (which will just idle, and call evict_something() with proper args)
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
PATCH 19/29 drm/i915: turn bound_ggtt checks to bound_any
In reply to Ben WidawskyIn some places, we want to know if an object is bound in any address space, and not just the global GTT. This often applies when there is a single global resource (object, pages, etc.)
function | reason
--------------------------------------------------
i915_gem_object_is_inactive | global object i915_gem_object_put_pages | object's pages 915_gem_object_unpin | global object i915_gem_execbuffer_unreserve_object | temporary until we plumb vma pread/pread | object's domain
--- drivers/gpu/drm/i915/i915_gem.c | 12 ++++++------ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- 2 files changed, 7 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +6 -6
@@ -122,7 +122,7 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)static inline bool i915_gem_object_is_inactive (struct drm_i915_gem_object *obj) { -return i915_gem_obj_ggtt_bound (obj) && !obj->active; +return i915_gem_obj_bound_any (obj) && !obj->active; }int
@@ -408,7 +408,7 @@ i915_gem_shmem_pread(struct drm_device *dev,* anyway again before the next pread happens. */ if (obj->cache_level ==I915_CACHE_NONE ) needs_clflush =1 ; -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_set_to_gtt_domain (obj, false);if (ret)return ret;
@@ -725,7 +725,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev,* right away and we therefore have to clflush anyway. */ if (obj->cache_level ==I915_CACHE_NONE ) needs_clflush_after =1 ; -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_set_to_gtt_domain (obj, true);if (ret)return ret;
@@ -1659,7 +1659,7 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj)if (obj->pages_pin_count)return -EBUSY ; -BUG_ON (i915_gem_obj_ggtt_bound (obj)); +BUG_ON (i915_gem_obj_bound_any (obj));/* ->put_pages might need to allocate memory for the bit17 swizzle * array, hence protect them from being reaped by removing them from gtt
@@ -3301,7 +3301,7 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)int ret;/* Not valid to be called on unbound objects. */ -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound_any (obj))return -EINVAL ;if (obj->base.write_domain ==I915_GEM_DOMAIN_GTT )
@@ -3725,7 +3725,7 @@ void i915_gem_object_unpin(struct drm_i915_gem_object *obj) {BUG_ON (obj->pin_count ==0 ); -BUG_ON (!i915_gem_obj_ggtt_bound (obj)); +BUG_ON (!i915_gem_obj_bound_any (obj));if (--obj->pin_count ==0 ) obj->pin_mappable = false;
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -1
@@ -466,7 +466,7 @@ i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj) {struct drm_i915_gem_exec_object2 *entry; -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound_any (obj))return ; entry = obj->exec_entry;
Signature · 1 line
1.8.3.4
Re: [PATCH 19/29] drm/i915: turn bound_ggtt checks to bound_any
Quoted 11 lines · Ben Widawsky
In some places, we want to know if an object is bound in any address space, and not just the global GTT. This often applies when there is a single global resource (object, pages, etc.)
function | reason
--------------------------------------------------i915_gem_object_is_inactive | global object i915_gem_object_put_pages | object's pages 915_gem_object_unpin | global object i915_gem_execbuffer_unreserve_object | temporary until we plumb vma pread/pread | object's domain
Most of these we want to operate on ggtt, so this change looks confused at best.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH 19/29] drm/i915: turn bound_ggtt checks to bound_any
Quoted 13 lines · Chris Wilson
Quoted 11 lines · Ben Widawsky
In some places, we want to know if an object is bound in any address space, and not just the global GTT. This often applies when there is a single global resource (object, pages, etc.)
function | reason
--------------------------------------------------i915_gem_object_is_inactive | global object i915_gem_object_put_pages | object's pages 915_gem_object_unpin | global object i915_gem_execbuffer_unreserve_object | temporary until we plumb vma pread/pread | object's domain
Quoted 4 lines
Most of these we want to operate on ggtt, so this change looks confused at best.
-Chris
And at worst? I guess I should have explained it a bit differently - yes the operations may very well be on the global GTT, but for the given reason (in the table), we need to make sure we unbind (or assert the object isn't bound) in another address space first.
Possibly I've misunderstood what you'd like me to do however.
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 19/29] drm/i915: turn bound_ggtt checks to bound_any
In reply to Ben WidawskyQuoted 11 lines · Ben Widawsky
In some places, we want to know if an object is bound in any address space, and not just the global GTT. This often applies when there is a single global resource (object, pages, etc.)
function | reason
--------------------------------------------------i915_gem_object_is_inactive | global object i915_gem_object_put_pages | object's pages 915_gem_object_unpin | global object i915_gem_execbuffer_unreserve_object | temporary until we plumb vma pread/pread | object's domain
pread/pwrite isn't about the object's domain at all, but purely about synchronizing for outstanding rendering. Replacing the call to set_to_gtt_domain with a wait_rendering would imo improve code readability. Furthermore we could pimp pread to only block for outstanding writes and not for reads.
Since you're not the first one to trip over this: Can I volunteer you for a follow-up patch to fix this?
Otherwise patch looks good.
-Daniel
Quoted 85 lines
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 12 ++++++------ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- 2 files changed, 7 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/i915_gem.c +6
-6@@ -122,7 +122,7 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)static inline bool i915_gem_object_is_inactive (struct drm_i915_gem_object *obj) { -return i915_gem_obj_ggtt_bound (obj) && !obj->active; +return i915_gem_obj_bound_any (obj) && !obj->active; }int @@ -408,7 +408,7 @@ i915_gem_shmem_pread(struct drm_device *dev,* anyway again before the next pread happens. */ if (obj->cache_level ==I915_CACHE_NONE ) needs_clflush =1 ; -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_set_to_gtt_domain (obj, false);if (ret)return ret;@@ -725,7 +725,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev,* right away and we therefore have to clflush anyway. */ if (obj->cache_level ==I915_CACHE_NONE ) needs_clflush_after =1 ; -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_set_to_gtt_domain (obj, true);if (ret)return ret;@@ -1659,7 +1659,7 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj)if (obj->pages_pin_count)return -EBUSY ; -BUG_ON (i915_gem_obj_ggtt_bound (obj)); +BUG_ON (i915_gem_obj_bound_any (obj));/* ->put_pages might need to allocate memory for the bit17 swizzle * array, hence protect them from being reaped by removing them from gtt @@ -3301,7 +3301,7 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)int ret;/* Not valid to be called on unbound objects. */ -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound_any (obj))return -EINVAL ;if (obj->base.write_domain ==I915_GEM_DOMAIN_GTT )@@ -3725,7 +3725,7 @@ void i915_gem_object_unpin(struct drm_i915_gem_object *obj) {BUG_ON (obj->pin_count ==0 ); -BUG_ON (!i915_gem_obj_ggtt_bound (obj)); +BUG_ON (!i915_gem_obj_bound_any (obj));if (--obj->pin_count ==0 ) obj->pin_mappable = false;drivers/gpu/drm/i915/i915_gem_execbuffer.c +1
-1@@ -466,7 +466,7 @@ i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj) {struct drm_i915_gem_exec_object2 *entry; -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound_any (obj))return ; entry = obj->exec_entry;Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 19/29] drm/i915: turn bound_ggtt checks to bound_any
Quoted 13 lines · Daniel Vetter
Quoted 11 lines · Ben Widawsky
In some places, we want to know if an object is bound in any address space, and not just the global GTT. This often applies when there is a single global resource (object, pages, etc.)
function | reason
--------------------------------------------------i915_gem_object_is_inactive | global object i915_gem_object_put_pages | object's pages 915_gem_object_unpin | global object i915_gem_execbuffer_unreserve_object | temporary until we plumb vma pread/pread | object's domain
Quoted 5 lines
pread/pwrite isn't about the object's domain at all, but purely about synchronizing for outstanding rendering. Replacing the call to
set_to_gtt_domainwith await_renderingwould imo improve code readability. Furthermore we could pimp pread to only block for outstanding writes and not for reads.
I might have been the first to trip over it, but this isn't my first instance ;-).
Quoted 3 lines
Since you're not the first one to trip over this: Can I volunteer you for a follow-up patch to fix this?
Working on it now.
Quoted 94 lines
Otherwise patch looks good.
-DanielQuoted 85 lines
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 12 ++++++------ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- 2 files changed, 7 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/i915_gem.c +6
-6@@ -122,7 +122,7 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)static inline bool i915_gem_object_is_inactive (struct drm_i915_gem_object *obj) { -return i915_gem_obj_ggtt_bound (obj) && !obj->active; +return i915_gem_obj_bound_any (obj) && !obj->active; }int @@ -408,7 +408,7 @@ i915_gem_shmem_pread(struct drm_device *dev,* anyway again before the next pread happens. */ if (obj->cache_level ==I915_CACHE_NONE ) needs_clflush =1 ; -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_set_to_gtt_domain (obj, false);if (ret)return ret;@@ -725,7 +725,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev,* right away and we therefore have to clflush anyway. */ if (obj->cache_level ==I915_CACHE_NONE ) needs_clflush_after =1 ; -if (i915_gem_obj_ggtt_bound (obj)) { +if (i915_gem_obj_bound_any (obj)) { ret =i915_gem_object_set_to_gtt_domain (obj, true);if (ret)return ret;@@ -1659,7 +1659,7 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj)if (obj->pages_pin_count)return -EBUSY ; -BUG_ON (i915_gem_obj_ggtt_bound (obj)); +BUG_ON (i915_gem_obj_bound_any (obj));/* ->put_pages might need to allocate memory for the bit17 swizzle * array, hence protect them from being reaped by removing them from gtt @@ -3301,7 +3301,7 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)int ret;/* Not valid to be called on unbound objects. */ -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound_any (obj))return -EINVAL ;if (obj->base.write_domain ==I915_GEM_DOMAIN_GTT )@@ -3725,7 +3725,7 @@ void i915_gem_object_unpin(struct drm_i915_gem_object *obj) {BUG_ON (obj->pin_count ==0 ); -BUG_ON (!i915_gem_obj_ggtt_bound (obj)); +BUG_ON (!i915_gem_obj_bound_any (obj));if (--obj->pin_count ==0 ) obj->pin_mappable = false;drivers/gpu/drm/i915/i915_gem_execbuffer.c +1
-1@@ -466,7 +466,7 @@ i915_gem_execbuffer_unreserve_object(struct drm_i915_gem_object *obj) {struct drm_i915_gem_exec_object2 *entry; -if (!i915_gem_obj_ggtt_bound (obj)) +if (!i915_gem_obj_bound_any (obj))return ; entry = obj->exec_entry;Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfxSignature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
PATCH 20/29 drm/i915: Fix up map and fenceable for VMA
In reply to Ben Widawskyformerly: "drm/i915: Create VMAs (part 3.5) - map and fenceable tracking"
The map_and_fenceable tracking is per object. GTT mapping, and fences only apply to global GTT. As such, object operations which are not performed on the global GTT should not effect mappable or fenceable characteristics.
Functionally, this commit could very well be squashed in to a previous patch which updated object operations to take a VM argument. This commit is split out because it's a bit tricky (or at least it was for me).
--- drivers/gpu/drm/i915/i915_gem.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +6 -3
@@ -2626,7 +2626,7 @@ int i915_vma_unbind(struct i915_vma *vma)trace_i915_vma_unbind (vma); -if (obj->has_global_gtt_mapping) +if (obj->has_global_gtt_mapping &&i915_is_ggtt (vma->vm))i915_gem_gtt_unbind_object (obj);if (obj->has_aliasing_ppgtt_mapping) {i915_ppgtt_unbind_object (dev_priv->mm.aliasing_ppgtt, obj);
@@ -2637,7 +2637,8 @@ int i915_vma_unbind(struct i915_vma *vma)list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ - obj->map_and_fenceable = true; +if (i915_is_ggtt (vma->vm)) + obj->map_and_fenceable = true;i915_gem_vma_destroy (vma);
@@ -3196,7 +3197,9 @@ search_free:i915_is_ggtt (vm) && vma->node.start + obj->base.size <= dev_priv->gtt.mappable_end; - obj->map_and_fenceable = mappable && fenceable; +/* Map and fenceable only changes if the VM is the global GGTT */ +if (i915_is_ggtt (vm)) + obj->map_and_fenceable = mappable && fenceable;trace_i915_vma_bind (vma, map_and_fenceable);i915_gem_verify_gtt (dev);
Signature · 1 line
1.8.3.4
Re: [PATCH 20/29] drm/i915: Fix up map and fenceable for VMA
Quoted 23 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 3.5) - map and fenceable tracking"
The
map_and_fenceabletracking is per object. GTT mapping, and fences only apply to global GTT. As such, object operations which are not performed on the global GTT should not effect mappable or fenceable characteristics.Functionally, this commit could very well be squashed in to a previous patch which updated object operations to take a VM argument. This commit is split out because it's a bit tricky (or at least it was for me).
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +0
-0
drivers/gpu/drm/i915/i915_gem.c int i915_vma_unbind(struct i915_vma *vma) · line 2632
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -2626,7 +2626,7 @@ int i915_vma_unbind(struct i915_vma *vma)trace_i915_vma_unbind (vma); -if (obj->has_global_gtt_mapping) +if (obj->has_global_gtt_mapping &&i915_is_ggtt (vma->vm))i915_gem_gtt_unbind_object (obj);if (obj->has_aliasing_ppgtt_mapping) {i915_ppgtt_unbind_object (dev_priv->mm.aliasing_ppgtt, obj);
Hm, shouldn't we do the is_ggtt check for both? After all only the global ggtt can be aliased ever ... This would also be more symmetric with some of the other global gtt checks I've spotted. You're take or will that run afoul of your Great Plan?
-Daniel
Quoted 28 lines
drivers/gpu/drm/i915/i915_gem.c +5
-2@@ -2637,7 +2637,8 @@ int i915_vma_unbind(struct i915_vma *vma)list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ - obj->map_and_fenceable = true; +if (i915_is_ggtt (vma->vm)) + obj->map_and_fenceable = true;i915_gem_vma_destroy (vma);@@ -3196,7 +3197,9 @@ search_free:i915_is_ggtt (vm) && vma->node.start + obj->base.size <= dev_priv->gtt.mappable_end; - obj->map_and_fenceable = mappable && fenceable; +/* Map and fenceable only changes if the VM is the global GGTT */ +if (i915_is_ggtt (vm)) + obj->map_and_fenceable = mappable && fenceable;trace_i915_vma_bind (vma, map_and_fenceable);i915_gem_verify_gtt (dev);Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 20/29] drm/i915: Fix up map and fenceable for VMA
Quoted 27 lines · Daniel Vetter
Quoted 26 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 3.5) - map and fenceable tracking"
The
map_and_fenceabletracking is per object. GTT mapping, and fences only apply to global GTT. As such, object operations which are not performed on the global GTT should not effect mappable or fenceable characteristics.Functionally, this commit could very well be squashed in to a previous patch which updated object operations to take a VM argument. This commit is split out because it's a bit tricky (or at least it was for me).
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +0
-0@@ -2626,7 +2626,7 @@ int i915_vma_unbind(struct i915_vma *vma)trace_i915_vma_unbind (vma);
Quoted 12 lines
drivers/gpu/drm/i915/i915_gem.c int i915_vma_unbind(struct i915_vma *vma) · line 2632
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem.c +1
-1-if (obj->has_global_gtt_mapping) +if (obj->has_global_gtt_mapping &&i915_is_ggtt (vma->vm))i915_gem_gtt_unbind_object (obj);if (obj->has_aliasing_ppgtt_mapping) {i915_ppgtt_unbind_object (dev_priv->mm.aliasing_ppgtt, obj);Hm, shouldn't we do the
is_ggttcheck for both? After all only the global ggtt can be aliased ever ... This would also be more symmetric with some of the other global gtt checks I've spotted. You're take or will that run afoul of your Great Plan?
-Daniel
You're right. The check makes sense for both cases. In both the original series, and n a few patches, this code turns into:vma->vm->unbind_vma(vma);
This ugliness is a result of bad rebasing on my part.
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 20/29] drm/i915: Fix up map and fenceable for VMA
Quoted 41 lines · Ben Widawsky
Quoted 27 lines · Daniel Vetter
Quoted 26 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 3.5) - map and fenceable tracking"
The
map_and_fenceabletracking is per object. GTT mapping, and fences only apply to global GTT. As such, object operations which are not performed on the global GTT should not effect mappable or fenceable characteristics.Functionally, this commit could very well be squashed in to a previous patch which updated object operations to take a VM argument. This commit is split out because it's a bit tricky (or at least it was for me).
Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)drivers/gpu/drm/i915/i915_gem.c +0
-0@@ -2626,7 +2626,7 @@ int i915_vma_unbind(struct i915_vma *vma)trace_i915_vma_unbind (vma);Quoted 12 lines
drivers/gpu/drm/i915/i915_gem.c int i915_vma_unbind(struct i915_vma *vma) · line 2632
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem.c +1
-1-if (obj->has_global_gtt_mapping) +if (obj->has_global_gtt_mapping &&i915_is_ggtt (vma->vm))i915_gem_gtt_unbind_object (obj);if (obj->has_aliasing_ppgtt_mapping) {i915_ppgtt_unbind_object (dev_priv->mm.aliasing_ppgtt, obj);Hm, shouldn't we do the
is_ggttcheck for both? After all only the global ggtt can be aliased ever ... This would also be more symmetric with some of the other global gtt checks I've spotted. You're take or will that run afoul of your Great Plan?
-Daniel
Quoted 3 lines
You're right. The check makes sense for both cases. In both the original series, and n a few patches, this code turns into:
vma->vm->unbind_vma(vma);
Ok, I've killed it and merged the patch.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 21/29 drm/i915: mm_list is per VMA
In reply to Ben Widawskyformerly: "drm/i915: Create VMAs (part 5) - move mm_list"
The mm_list is used for the active/inactive LRUs. Since those LRUs are per address space, the link should be per VMx .
Because we'll only ever have 1 VMA before this point, it's not incorrect to defer this change until this point in the patch series, and doing it here makes the change much easier to understand.
Shamelessly manipulated out of Daniel:
"active/inactive stuff is used by eviction when we run out of address space, so needs to be per-vma and per-address space. Bound/unbound otoh is used by the shrinker which only cares about the amount of memory used and not one bit about in which address space this memory is all used in. Of course to actual kick out an object we need to unbind it from every address space, but for that we have the per-object list of vmas."
v2: only bump GGTT LRU in i915_gem_object_set_to_gtt_domain (Chris)
v3: Moved earlier in the series
v4: Add dropped message from v3
--- drivers/gpu/drm/i915/i915_debugfs.c | 53 ++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_drv.h | 5 +-- drivers/gpu/drm/i915/i915_gem.c | 37 +++++++++++---------- drivers/gpu/drm/i915/i915_gem_context.c | 3 ++ drivers/gpu/drm/i915/i915_gem_evict.c | 14 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 ++ drivers/gpu/drm/i915/i915_gem_stolen.c | 2 +- drivers/gpu/drm/i915/i915_gpu_error.c | 37 ++++++++++++--------- 8 files changed, 91 insertions(+), 62 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +35 -18
@@ -149,7 +149,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj; +struct i915_vma *vma;size_t total_obj_size, total_gtt_size;int count, ret;
@@ -157,6 +157,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)if (ret)return ret; +/* FIXME: the user of this interface might want more than just GGTT */ switch (list) {case ACTIVE_LIST :seq_puts (m,"Active:\n" );
@@ -172,12 +173,12 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data) } total_obj_size = total_gtt_size = count =0 ; -list_for_each_entry (obj, head, mm_list) { -seq_puts (m," " ); -describe_obj (m, obj); -seq_putc (m,'\n' ); - total_obj_size += obj->base.size; - total_gtt_size +=i915_gem_obj_ggtt_size (obj); +list_for_each_entry (vma, head, mm_list) { +seq_printf (m," " ); +describe_obj (m, vma->obj); +seq_printf (m,"\n" ); + total_obj_size += vma->obj->base.size; + total_gtt_size +=i915_gem_obj_size (vma->obj, vma->vm); count++; }mutex_unlock (&dev->struct_mutex);
@@ -224,7 +225,18 @@ static int per_file_stats(int id, void *ptr, void *data)return 0 ; } -static int i915_gem_object_info (struct seq_file *m,void *data) +#define count_vmas (list, member) do { \ + list_for_each_entry(vma, list, member) { \ + size += i915_gem_obj_ggtt_size(vma->obj); \ + ++count; \ + if (vma->obj->map_and_fenceable) { \ + mappable_size += i915_gem_obj_ggtt_size(vma->obj); \ + ++mappable_count; \ + } \ + } \ +} while (0) + +static int i915_gem_object_info (struct seq_file *m,void * data) {struct drm_info_node *node = (struct drm_info_node *) m->private;struct drm_device *dev = node->minor->dev;
@@ -234,6 +246,7 @@ static int i915_gem_object_info(struct seq_file *m, void *data)struct drm_i915_gem_object *obj;struct i915_address_space *vm = &dev_priv->gtt.base;struct drm_file *file; +struct i915_vma *vma;int ret; ret =mutex_lock_interruptible (&dev->struct_mutex);
@@ -253,12 +266,12 @@ static int i915_gem_object_info(struct seq_file *m, void *data) count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->active_list, mm_list); +count_vmas (&vm->active_list, mm_list);seq_printf (m," %u [%u] active objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->inactive_list, mm_list); +count_vmas (&vm->inactive_list, mm_list);seq_printf (m," %u [%u] inactive objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size);
@@ -1771,7 +1784,8 @@ i915_drop_caches_set(void *data, u64 val)struct drm_device *dev = data;struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_object *obj, *next; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm; +struct i915_vma *vma, *x;int ret;DRM_DEBUG_DRIVER ("Dropping caches: 0x%08llx\n" , val);
@@ -1792,13 +1806,16 @@ i915_drop_caches_set(void *data, u64 val)i915_gem_retire_requests (dev);if (val &DROP_BOUND ) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) { -if (obj->pin_count) -continue ; - ret =i915_gem_object_ggtt_unbind (obj); -if (ret) - goto unlock; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (vma, x, &vm->inactive_list, + mm_list) { +if (vma->obj->pin_count) +continue ; + + ret =i915_vma_unbind (vma); +if (ret) + goto unlock; + } } }
drivers/gpu/drm/i915/i915_drv.h +3 -2
@@ -546,6 +546,9 @@ struct i915_vma {struct drm_i915_gem_object *obj;struct i915_address_space *vm; +/** This object's place on the active/inactive lists */ +struct list_head mm_list; +struct list_head vma_link;/* Link in the object's VMA list */ };
@@ -1263,9 +1266,7 @@ struct drm_i915_gem_object {struct drm_mm_node *stolen;struct list_head global_list; -/** This object's place on the active/inactive lists */ struct list_head ring_list; -struct list_head mm_list;/** This object's place in the batchbuffer or on the eviction list */ struct list_head exec_list;
drivers/gpu/drm/i915/i915_gem.c +20 -17
@@ -1872,7 +1872,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;u32 seqno =intel_ring_get_seqno (ring);BUG_ON (ring ==NULL );
@@ -1888,8 +1887,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, obj->active =1 ; } -/* Move from whatever list we were on to the tail of execution. */ -list_move_tail (&obj->mm_list, &vm->active_list);list_move_tail (&obj->ring_list, &ring->active_list); obj->last_read_seqno = seqno;
@@ -1911,14 +1908,14 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = obj->base.dev->dev_private; +struct i915_address_space *ggtt_vm = &dev_priv->gtt.base; +struct i915_vma *vma =i915_gem_obj_to_vma (obj, ggtt_vm);BUG_ON (obj->base.write_domain & ~I915_GEM_GPU_DOMAINS );BUG_ON (!obj->active); -list_move_tail (&obj->mm_list, &vm->inactive_list); +list_move_tail (&vma->mm_list, &ggtt_vm->inactive_list);list_del_init (&obj->ring_list); obj->ring =NULL ;
@@ -2286,9 +2283,9 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring; +struct i915_address_space *vm; +struct i915_vma *vma;int i;for_each_ring (ring, dev_priv, i)
@@ -2298,8 +2295,8 @@ void i915_gem_reset(struct drm_device *dev)* necessary invalidation upon reuse. */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vma, &vm->inactive_list, mm_list) + vma->obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;i915_gem_restore_fences (dev); }
@@ -2353,6 +2350,7 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +BUG_ON (!obj->active);i915_gem_object_move_to_inactive (obj); }
@@ -2635,7 +2633,6 @@ int i915_vma_unbind(struct i915_vma *vma)i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj); -list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ if (i915_is_ggtt (vma->vm)) obj->map_and_fenceable = true;
@@ -3180,7 +3177,7 @@ search_free: goto err_out;list_move_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &vm->inactive_list); +list_add_tail (&vma->mm_list, &vm->inactive_list);/* Keep GGTT vmas first to make debug easier */ if (i915_is_ggtt (vm))
@@ -3342,9 +3339,14 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain);/* And bump the LRU for this access */ -if (i915_gem_object_is_inactive (obj)) -list_move_tail (&obj->mm_list, - &dev_priv->gtt.base.inactive_list); +if (i915_gem_object_is_inactive (obj)) { +struct i915_vma *vma =i915_gem_obj_to_vma (obj, + &dev_priv->gtt.base); +if (vma) +list_move_tail (&vma->mm_list, + &dev_priv->gtt.base.inactive_list); + + }return 0 ; }
@@ -3917,7 +3919,6 @@ unlock:void i915_gem_object_init (struct drm_i915_gem_object *obj,const struct drm_i915_gem_object_ops *ops) { -INIT_LIST_HEAD (&obj->mm_list);INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->exec_list);
@@ -4054,6 +4055,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,return ERR_PTR (-ENOMEM );INIT_LIST_HEAD (&vma->vma_link); +INIT_LIST_HEAD (&vma->mm_list); vma->vm = vm; vma->obj = obj;
@@ -4063,6 +4065,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma) {list_del_init (&vma->vma_link); +list_del (&vma->mm_list);drm_mm_remove_node (&vma->node);kfree (vma); }
drivers/gpu/drm/i915/i915_gem_context.c +3 -0
@@ -436,7 +436,10 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) { +struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; +struct i915_address_space *ggtt = &dev_priv->gtt.base; from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; +list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list);i915_gem_object_move_to_active (from->obj, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the
drivers/gpu/drm/i915/i915_gem_evict.c +6 -8
@@ -87,8 +87,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->inactive_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }
@@ -97,8 +96,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, goto none;/* Now merge in the soon-to-be-expired objects... */ -list_for_each_entry (obj, &vm->active_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->active_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }
@@ -159,7 +157,7 @@ i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private;struct i915_address_space *vm; -struct drm_i915_gem_object *obj, *next; +struct i915_vma *vma, *next;bool lists_empty = true;int ret;
@@ -187,9 +185,9 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); +list_for_each_entry_safe (vma, next, &vm->inactive_list, mm_list) +if (vma->obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (vma)); }return 0 ;
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2 -0
@@ -801,6 +801,8 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; +/* FIXME: This lookup gets fixed later <-- danvet */ +list_move_tail (&i915_gem_obj_to_vma (obj, vm)->mm_list, &vm->active_list);i915_gem_object_move_to_active (obj, ring);if (obj->base.write_domain) { obj->dirty =1 ;
drivers/gpu/drm/i915/i915_gem_stolen.c +1 -1
@@ -419,7 +419,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, obj->has_global_gtt_mapping =1 ;list_add_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &ggtt->inactive_list); +list_add_tail (&vma->mm_list, &ggtt->inactive_list);return obj;
drivers/gpu/drm/i915/i915_gpu_error.c +21 -16
@@ -556,11 +556,11 @@ static void capture_bo(struct drm_i915_error_buffer *err,static u32 capture_active_bo (struct drm_i915_error_buffer *err,int count,struct list_head *head) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;int i =0 ; -list_for_each_entry (obj, head, mm_list) { -capture_bo (err++, obj); +list_for_each_entry (vma, head, mm_list) { +capture_bo (err++, vma->obj);if (++i == count)break ; }
@@ -622,7 +622,8 @@ static struct drm_i915_error_object * i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, struct intel_ring_buffer *ring) { - struct i915_address_space *vm = &dev_priv->gtt.base; + struct i915_address_space *vm; +struct i915_vma *vma;struct drm_i915_gem_object *obj;u32 seqno;
@@ -642,20 +643,23 @@ i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, } seqno = ring->get_seqno (ring, false); -list_for_each_entry (obj, &vm->active_list, mm_list) { -if (obj->ring != ring) -continue ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry (vma, &vm->active_list, mm_list) { + obj = vma->obj; +if (obj->ring != ring) +continue ; -if (i915_seqno_passed (seqno, obj->last_read_seqno)) -continue ; +if (i915_seqno_passed (seqno, obj->last_read_seqno)) +continue ; -if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) -continue ; +if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) +continue ; -/* We need to copy these to an anonymous buffer as the simplest -* method to avoid being overwritten by userspace. -*/ -return i915_error_object_create (dev_priv, obj); +/* We need to copy these to an anonymous buffer as the simplest +* method to avoid being overwritten by userspace. +*/ +return i915_error_object_create (dev_priv, obj); + } }return NULL ;
@@ -775,11 +779,12 @@ static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv, struct drm_i915_error_state *error) {struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_vma *vma;struct drm_i915_gem_object *obj;int i; i =0 ; -list_for_each_entry (obj, &vm->active_list, mm_list) +list_for_each_entry (vma, &vm->active_list, mm_list) i++; error->active_bo_count = i;list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list)
Signature · 1 line
1.8.3.4
Re: [PATCH 21/29] drm/i915: mm_list is per VMA
Quoted 17 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 5) - move
mm_list"The
mm_listis used for the active/inactive LRUs. Since those LRUs are per address space, the link should be per VMx .Because we'll only ever have 1 VMA before this point, it's not incorrect to defer this change until this point in the patch series, and doing it here makes the change much easier to understand.
Shamelessly manipulated out of Daniel:
"active/inactive stuff is used by eviction when we run out of address space, so needs to be per-vma and per-address space. Bound/unbound otoh is used by the shrinker which only cares about the amount of memory used and not one bit about in which address space this memory is all used in. Of course to actual kick out an object we need to unbind it from every address space, but for that we have the per-object list of vmas."
Quoted 7 lines
v2: only bump GGTT LRU in
i915_gem_object_set_to_gtt_domain(Chris)v3: Moved earlier in the series
v4: Add dropped message from v3
Signed-off-by Ben Widawsky <[email protected]>
Some comments below for this one. The lru changes look a bit strange so I'll wait for your confirmation that the do_switch hunk has the same reasons s the one in execbuf with the FIXME comment.
-Daniel
Quoted 38 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 53 ++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_drv.h | 5 +-- drivers/gpu/drm/i915/i915_gem.c | 37 +++++++++++---------- drivers/gpu/drm/i915/i915_gem_context.c | 3 ++ drivers/gpu/drm/i915/i915_gem_evict.c | 14 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 ++ drivers/gpu/drm/i915/i915_gem_stolen.c | 2 +- drivers/gpu/drm/i915/i915_gpu_error.c | 37 ++++++++++++--------- 8 files changed, 91 insertions(+), 62 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +2
-3@@ -149,7 +149,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj; +struct i915_vma *vma;size_t total_obj_size, total_gtt_size;int count, ret;@@ -157,6 +157,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)if (ret)return ret; +/* FIXME: the user of this interface might want more than just GGTT */ switch (list) {case ACTIVE_LIST :seq_puts (m,"Active:\n" );@@ -172,12 +173,12 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data) } total_obj_size = total_gtt_size = count =0 ; -list_for_each_entry (obj, head, mm_list) { -seq_puts (m," " );
drivers/gpu/drm/i915/i915_debugfs.c static int i915_gem_object_list_info(struct seq_file *m, void *data) · line 181
Quoted 10 lines
drivers/gpu/drm/i915/i915_debugfs.c +6
-4-describe_obj (m, obj); -seq_putc (m,'\n' ); - total_obj_size += obj->base.size; - total_gtt_size +=i915_gem_obj_ggtt_size (obj); +list_for_each_entry (vma, head, mm_list) { +seq_printf (m," " ); +describe_obj (m, vma->obj); +seq_printf (m,"\n" ); + total_obj_size += vma->obj->base.size; + total_gtt_size +=i915_gem_obj_size (vma->obj, vma->vm);
Why not use vma->node.size? If you don't disagree I'll bikeshed this while applying.
Quoted 60 lines
drivers/gpu/drm/i915/i915_debugfs.c +17
-4count++; }mutex_unlock (&dev->struct_mutex);@@ -224,7 +225,18 @@ static int per_file_stats(int id, void *ptr, void *data)return 0 ; } -static int i915_gem_object_info (struct seq_file *m,void *data) +#define count_vmas (list, member) do { \ + list_for_each_entry(vma, list, member) { \ + size += i915_gem_obj_ggtt_size(vma->obj); \ + ++count; \ + if (vma->obj->map_and_fenceable) { \ + mappable_size += i915_gem_obj_ggtt_size(vma->obj); \ + ++mappable_count; \ + } \ + } \ +} while (0) + +static int i915_gem_object_info (struct seq_file *m,void * data) {struct drm_info_node *node = (struct drm_info_node *) m->private;struct drm_device *dev = node->minor->dev;@@ -234,6 +246,7 @@ static int i915_gem_object_info(struct seq_file *m, void *data)struct drm_i915_gem_object *obj;struct i915_address_space *vm = &dev_priv->gtt.base;struct drm_file *file; +struct i915_vma *vma;int ret; ret =mutex_lock_interruptible (&dev->struct_mutex);@@ -253,12 +266,12 @@ static int i915_gem_object_info(struct seq_file *m, void *data) count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->active_list, mm_list); +count_vmas (&vm->active_list, mm_list);seq_printf (m," %u [%u] active objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->inactive_list, mm_list); +count_vmas (&vm->inactive_list, mm_list);seq_printf (m," %u [%u] inactive objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size);@@ -1771,7 +1784,8 @@ i915_drop_caches_set(void *data, u64 val)struct drm_device *dev = data;struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_object *obj, *next; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm; +struct i915_vma *vma, *x;int ret;DRM_DEBUG_DRIVER ("Dropping caches: 0x%08llx\n" , val);@@ -1792,13 +1806,16 @@ i915_drop_caches_set(void *data, u64 val)i915_gem_retire_requests (dev);if (val &DROP_BOUND ) {
drivers/gpu/drm/i915/i915_debugfs.c i915_drop_caches_set(void *data, u64 val) · line 1811
Quoted 10 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-7-list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) { -if (obj->pin_count) -continue ; - ret =i915_gem_object_ggtt_unbind (obj); -if (ret) - goto unlock; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (vma, x, &vm->inactive_list, + mm_list) {
Imo the double-loop is a bit funny, looping over the global bound list and skipping all active objects is imo the more straightfoward logic. But I agree that this is the more straightforward conversion, so I'm ok with a follow-up fixup patch.
Quoted 169 lines
drivers/gpu/drm/i915/i915_debugfs.c +7
-0+if (vma->obj->pin_count) +continue ; + + ret =i915_vma_unbind (vma); +if (ret) + goto unlock; + } } }drivers/gpu/drm/i915/i915_drv.h +3
-2@@ -546,6 +546,9 @@ struct i915_vma {struct drm_i915_gem_object *obj;struct i915_address_space *vm; +/** This object's place on the active/inactive lists */ +struct list_head mm_list; +struct list_head vma_link;/* Link in the object's VMA list */ };@@ -1263,9 +1266,7 @@ struct drm_i915_gem_object {struct drm_mm_node *stolen;struct list_head global_list; -/** This object's place on the active/inactive lists */ struct list_head ring_list; -struct list_head mm_list;/** This object's place in the batchbuffer or on the eviction list */ struct list_head exec_list;drivers/gpu/drm/i915/i915_gem.c +20
-17@@ -1872,7 +1872,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;u32 seqno =intel_ring_get_seqno (ring);BUG_ON (ring ==NULL );@@ -1888,8 +1887,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, obj->active =1 ; } -/* Move from whatever list we were on to the tail of execution. */ -list_move_tail (&obj->mm_list, &vm->active_list);list_move_tail (&obj->ring_list, &ring->active_list); obj->last_read_seqno = seqno;@@ -1911,14 +1908,14 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = obj->base.dev->dev_private; +struct i915_address_space *ggtt_vm = &dev_priv->gtt.base; +struct i915_vma *vma =i915_gem_obj_to_vma (obj, ggtt_vm);BUG_ON (obj->base.write_domain & ~I915_GEM_GPU_DOMAINS );BUG_ON (!obj->active); -list_move_tail (&obj->mm_list, &vm->inactive_list); +list_move_tail (&vma->mm_list, &ggtt_vm->inactive_list);list_del_init (&obj->ring_list); obj->ring =NULL ;@@ -2286,9 +2283,9 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring; +struct i915_address_space *vm; +struct i915_vma *vma;int i;for_each_ring (ring, dev_priv, i)@@ -2298,8 +2295,8 @@ void i915_gem_reset(struct drm_device *dev)* necessary invalidation upon reuse. */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vma, &vm->inactive_list, mm_list) + vma->obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;i915_gem_restore_fences (dev); }@@ -2353,6 +2350,7 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +BUG_ON (!obj->active);i915_gem_object_move_to_inactive (obj); }@@ -2635,7 +2633,6 @@ int i915_vma_unbind(struct i915_vma *vma)i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj); -list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ if (i915_is_ggtt (vma->vm)) obj->map_and_fenceable = true;@@ -3180,7 +3177,7 @@ search_free: goto err_out;list_move_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &vm->inactive_list); +list_add_tail (&vma->mm_list, &vm->inactive_list);/* Keep GGTT vmas first to make debug easier */ if (i915_is_ggtt (vm))@@ -3342,9 +3339,14 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain);/* And bump the LRU for this access */ -if (i915_gem_object_is_inactive (obj)) -list_move_tail (&obj->mm_list, - &dev_priv->gtt.base.inactive_list); +if (i915_gem_object_is_inactive (obj)) { +struct i915_vma *vma =i915_gem_obj_to_vma (obj, + &dev_priv->gtt.base); +if (vma) +list_move_tail (&vma->mm_list, + &dev_priv->gtt.base.inactive_list); + + }return 0 ; }@@ -3917,7 +3919,6 @@ unlock:void i915_gem_object_init (struct drm_i915_gem_object *obj,const struct drm_i915_gem_object_ops *ops) { -INIT_LIST_HEAD (&obj->mm_list);INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->exec_list);@@ -4054,6 +4055,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,return ERR_PTR (-ENOMEM );INIT_LIST_HEAD (&vma->vma_link); +INIT_LIST_HEAD (&vma->mm_list); vma->vm = vm; vma->obj = obj;@@ -4063,6 +4065,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma) {list_del_init (&vma->vma_link); +list_del (&vma->mm_list);drm_mm_remove_node (&vma->node);kfree (vma); }drivers/gpu/drm/i915/i915_gem_context.c +0
-0
drivers/gpu/drm/i915/i915_gem_context.c static int do_switch(struct i915_hw_context *to) · line 442
Quoted 7 lines
drivers/gpu/drm/i915/i915_gem_context.c +3
-0@@ -436,7 +436,10 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) { +struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; +struct i915_address_space *ggtt = &dev_priv->gtt.base; from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; +list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list);
I don't really see a reason to add this here ... shouldn't move_to_active take care of this? Obviously not in this patch here but later on when it's converted over.
Quoted 54 lines
drivers/gpu/drm/i915/i915_gem_context.c +0
-0i915_gem_object_move_to_active (from->obj, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the drivers/gpu/drm/i915/i915_gem_evict.c +6
-8@@ -87,8 +87,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->inactive_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -97,8 +96,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, goto none;/* Now merge in the soon-to-be-expired objects... */ -list_for_each_entry (obj, &vm->active_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->active_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -159,7 +157,7 @@ i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private;struct i915_address_space *vm; -struct drm_i915_gem_object *obj, *next; +struct i915_vma *vma, *next;bool lists_empty = true;int ret;@@ -187,9 +185,9 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); +list_for_each_entry_safe (vma, next, &vm->inactive_list, mm_list) +if (vma->obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (vma)); }return 0 ;drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0
drivers/gpu/drm/i915/i915_gem_execbuffer.c i915_gem_execbuffer_move_to_active(struct list_head *objects, · line 805
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2
-0@@ -801,6 +801,8 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; +/* FIXME: This lookup gets fixed later <-- danvet */ +list_move_tail (&i915_gem_obj_to_vma (obj, vm)->mm_list, &vm->active_list);
Ah, I guess the same comment applies to the lru frobbing in do_switch?
Quoted 46 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0i915_gem_object_move_to_active (obj, ring);if (obj->base.write_domain) { obj->dirty =1 ;drivers/gpu/drm/i915/i915_gem_stolen.c +1
-1@@ -419,7 +419,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, obj->has_global_gtt_mapping =1 ;list_add_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &ggtt->inactive_list); +list_add_tail (&vma->mm_list, &ggtt->inactive_list);return obj;drivers/gpu/drm/i915/i915_gpu_error.c +5
-4@@ -556,11 +556,11 @@ static void capture_bo(struct drm_i915_error_buffer *err,static u32 capture_active_bo (struct drm_i915_error_buffer *err,int count,struct list_head *head) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;int i =0 ; -list_for_each_entry (obj, head, mm_list) { -capture_bo (err++, obj); +list_for_each_entry (vma, head, mm_list) { +capture_bo (err++, vma->obj);if (++i == count)break ; }@@ -622,7 +622,8 @@ static struct drm_i915_error_object * i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, struct intel_ring_buffer *ring) { - struct i915_address_space *vm = &dev_priv->gtt.base; + struct i915_address_space *vm; +struct i915_vma *vma;struct drm_i915_gem_object *obj;u32 seqno;
drivers/gpu/drm/i915/i915_gpu_error.c i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, · line 647
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +2
-3@@ -642,20 +643,23 @@ i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, } seqno = ring->get_seqno (ring, false); -list_for_each_entry (obj, &vm->active_list, mm_list) { -if (obj->ring != ring) -continue ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry (vma, &vm->active_list, mm_list) {
We could instead loop over the bound list and check for ->active. But this is ok, too albeit a bit convoluted imo.
Quoted 47 lines
drivers/gpu/drm/i915/i915_gpu_error.c +14
-9+ obj = vma->obj; +if (obj->ring != ring) +continue ; -if (i915_seqno_passed (seqno, obj->last_read_seqno)) -continue ; +if (i915_seqno_passed (seqno, obj->last_read_seqno)) +continue ; -if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) -continue ; +if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) +continue ; -/* We need to copy these to an anonymous buffer as the simplest -* method to avoid being overwritten by userspace. -*/ -return i915_error_object_create (dev_priv, obj); +/* We need to copy these to an anonymous buffer as the simplest +* method to avoid being overwritten by userspace. +*/ +return i915_error_object_create (dev_priv, obj); + } }return NULL ;@@ -775,11 +779,12 @@ static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv, struct drm_i915_error_state *error) {struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_vma *vma;struct drm_i915_gem_object *obj;int i; i =0 ; -list_for_each_entry (obj, &vm->active_list, mm_list) +list_for_each_entry (vma, &vm->active_list, mm_list) i++; error->active_bo_count = i;list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list)Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 21/29] drm/i915: mm_list is per VMA
Quoted 66 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 5) - move
mm_list"The
mm_listis used for the active/inactive LRUs. Since those LRUs are per address space, the link should be per VMx .Because we'll only ever have 1 VMA before this point, it's not incorrect to defer this change until this point in the patch series, and doing it here makes the change much easier to understand.
Shamelessly manipulated out of Daniel:
"active/inactive stuff is used by eviction when we run out of address space, so needs to be per-vma and per-address space. Bound/unbound otoh is used by the shrinker which only cares about the amount of memory used and not one bit about in which address space this memory is all used in. Of course to actual kick out an object we need to unbind it from every address space, but for that we have the per-object list of vmas."Quoted 7 lines
v2: only bump GGTT LRU in
i915_gem_object_set_to_gtt_domain(Chris)v3: Moved earlier in the series
v4: Add dropped message from v3
Signed-off-by Ben Widawsky <[email protected]>Some comments below for this one. The lru changes look a bit strange so I'll wait for your confirmation that the
do_switchhunk has the same reasons s the one in execbuf with the FIXME comment.
-DanielQuoted 35 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 53 ++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_drv.h | 5 +-- drivers/gpu/drm/i915/i915_gem.c | 37 +++++++++++---------- drivers/gpu/drm/i915/i915_gem_context.c | 3 ++ drivers/gpu/drm/i915/i915_gem_evict.c | 14 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 ++ drivers/gpu/drm/i915/i915_gem_stolen.c | 2 +- drivers/gpu/drm/i915/i915_gpu_error.c | 37 ++++++++++++--------- 8 files changed, 91 insertions(+), 62 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +2
-1@@ -149,7 +149,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj; +struct i915_vma *vma;size_t total_obj_size, total_gtt_size;int count, ret;@@ -157,6 +157,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)if (ret)return ret; +/* FIXME: the user of this interface might want more than just GGTT */ switch (list) {case ACTIVE_LIST :seq_puts (m,"Active:\n" );@@ -172,12 +173,12 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data) }
Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c static int i915_gem_object_list_info(struct seq_file *m, void *data) · line 181
Quoted 13 lines
drivers/gpu/drm/i915/i915_debugfs.c +6
-6total_obj_size = total_gtt_size = count =0 ; -list_for_each_entry (obj, head, mm_list) { -seq_puts (m," " ); -describe_obj (m, obj); -seq_putc (m,'\n' ); - total_obj_size += obj->base.size; - total_gtt_size +=i915_gem_obj_ggtt_size (obj); +list_for_each_entry (vma, head, mm_list) { +seq_printf (m," " ); +describe_obj (m, vma->obj); +seq_printf (m,"\n" ); + total_obj_size += vma->obj->base.size; + total_gtt_size +=i915_gem_obj_size (vma->obj, vma->vm);Why not use
vma->node.size? If you don't disagree I'll bikeshed this while applying.
I think in terms of the diff, it's more logical to do it how I did. The result should damn well be the same though, so go right ahead. When I set about writing the series, I really didn't want to use node.size/start directly as much as possible - so we can sneak things into the helpers as needed.
Quoted 59 lines
Quoted 59 lines
drivers/gpu/drm/i915/i915_debugfs.c +17
-4count++; }mutex_unlock (&dev->struct_mutex);@@ -224,7 +225,18 @@ static int per_file_stats(int id, void *ptr, void *data)return 0 ; } -static int i915_gem_object_info (struct seq_file *m,void *data) +#define count_vmas (list, member) do { \ + list_for_each_entry(vma, list, member) { \ + size += i915_gem_obj_ggtt_size(vma->obj); \ + ++count; \ + if (vma->obj->map_and_fenceable) { \ + mappable_size += i915_gem_obj_ggtt_size(vma->obj); \ + ++mappable_count; \ + } \ + } \ +} while (0) + +static int i915_gem_object_info (struct seq_file *m,void * data) {struct drm_info_node *node = (struct drm_info_node *) m->private;struct drm_device *dev = node->minor->dev;@@ -234,6 +246,7 @@ static int i915_gem_object_info(struct seq_file *m, void *data)struct drm_i915_gem_object *obj;struct i915_address_space *vm = &dev_priv->gtt.base;struct drm_file *file; +struct i915_vma *vma;int ret; ret =mutex_lock_interruptible (&dev->struct_mutex);@@ -253,12 +266,12 @@ static int i915_gem_object_info(struct seq_file *m, void *data) count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->active_list, mm_list); +count_vmas (&vm->active_list, mm_list);seq_printf (m," %u [%u] active objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->inactive_list, mm_list); +count_vmas (&vm->inactive_list, mm_list);seq_printf (m," %u [%u] inactive objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size);@@ -1771,7 +1784,8 @@ i915_drop_caches_set(void *data, u64 val)struct drm_device *dev = data;struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_object *obj, *next; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm; +struct i915_vma *vma, *x;int ret;DRM_DEBUG_DRIVER ("Dropping caches: 0x%08llx\n" , val);@@ -1792,13 +1806,16 @@ i915_drop_caches_set(void *data, u64 val)i915_gem_retire_requests (dev);
Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c i915_drop_caches_set(void *data, u64 val) · line 1811
Quoted 11 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-7if (val &DROP_BOUND ) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) { -if (obj->pin_count) -continue ; - ret =i915_gem_object_ggtt_unbind (obj); -if (ret) - goto unlock; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (vma, x, &vm->inactive_list, + mm_list) {Imo the double-loop is a bit funny, looping over the global bound list and skipping all active objects is imo the more straightfoward logic. But I agree that this is the more straightforward conversion, so I'm ok with a follow-up fixup patch.
I guess we have a lot of such conversions. I don't really mind the change, just a bit worried that it's less tested than what I've already done. I'm also not yet convinced the result will be a huge improvement for readability, but I've been starting at these lists for so long, my opinion is quite biased.
I guess we'll have to see. I've made a note to myself to look into converting all these types of loops over, but as we should see little to no functional impact from the change, I'd like to hold off until we get the rest merged.
Quoted 172 lines
Quoted 172 lines
drivers/gpu/drm/i915/i915_debugfs.c +7
-0+if (vma->obj->pin_count) +continue ; + + ret =i915_vma_unbind (vma); +if (ret) + goto unlock; + } } }drivers/gpu/drm/i915/i915_drv.h +3
-2@@ -546,6 +546,9 @@ struct i915_vma {struct drm_i915_gem_object *obj;struct i915_address_space *vm; +/** This object's place on the active/inactive lists */ +struct list_head mm_list; +struct list_head vma_link;/* Link in the object's VMA list */ };@@ -1263,9 +1266,7 @@ struct drm_i915_gem_object {struct drm_mm_node *stolen;struct list_head global_list; -/** This object's place on the active/inactive lists */ struct list_head ring_list; -struct list_head mm_list;/** This object's place in the batchbuffer or on the eviction list */ struct list_head exec_list;drivers/gpu/drm/i915/i915_gem.c +20
-17@@ -1872,7 +1872,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;u32 seqno =intel_ring_get_seqno (ring);BUG_ON (ring ==NULL );@@ -1888,8 +1887,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, obj->active =1 ; } -/* Move from whatever list we were on to the tail of execution. */ -list_move_tail (&obj->mm_list, &vm->active_list);list_move_tail (&obj->ring_list, &ring->active_list); obj->last_read_seqno = seqno;@@ -1911,14 +1908,14 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = obj->base.dev->dev_private; +struct i915_address_space *ggtt_vm = &dev_priv->gtt.base; +struct i915_vma *vma =i915_gem_obj_to_vma (obj, ggtt_vm);BUG_ON (obj->base.write_domain & ~I915_GEM_GPU_DOMAINS );BUG_ON (!obj->active); -list_move_tail (&obj->mm_list, &vm->inactive_list); +list_move_tail (&vma->mm_list, &ggtt_vm->inactive_list);list_del_init (&obj->ring_list); obj->ring =NULL ;@@ -2286,9 +2283,9 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring; +struct i915_address_space *vm; +struct i915_vma *vma;int i;for_each_ring (ring, dev_priv, i)@@ -2298,8 +2295,8 @@ void i915_gem_reset(struct drm_device *dev)* necessary invalidation upon reuse. */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vma, &vm->inactive_list, mm_list) + vma->obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;i915_gem_restore_fences (dev); }@@ -2353,6 +2350,7 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +BUG_ON (!obj->active);i915_gem_object_move_to_inactive (obj); }@@ -2635,7 +2633,6 @@ int i915_vma_unbind(struct i915_vma *vma)i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj); -list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ if (i915_is_ggtt (vma->vm)) obj->map_and_fenceable = true;@@ -3180,7 +3177,7 @@ search_free: goto err_out;list_move_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &vm->inactive_list); +list_add_tail (&vma->mm_list, &vm->inactive_list);/* Keep GGTT vmas first to make debug easier */ if (i915_is_ggtt (vm))@@ -3342,9 +3339,14 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain);/* And bump the LRU for this access */ -if (i915_gem_object_is_inactive (obj)) -list_move_tail (&obj->mm_list, - &dev_priv->gtt.base.inactive_list); +if (i915_gem_object_is_inactive (obj)) { +struct i915_vma *vma =i915_gem_obj_to_vma (obj, + &dev_priv->gtt.base); +if (vma) +list_move_tail (&vma->mm_list, + &dev_priv->gtt.base.inactive_list); + + }return 0 ; }@@ -3917,7 +3919,6 @@ unlock:void i915_gem_object_init (struct drm_i915_gem_object *obj,const struct drm_i915_gem_object_ops *ops) { -INIT_LIST_HEAD (&obj->mm_list);INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->exec_list);@@ -4054,6 +4055,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,return ERR_PTR (-ENOMEM );INIT_LIST_HEAD (&vma->vma_link); +INIT_LIST_HEAD (&vma->mm_list); vma->vm = vm; vma->obj = obj;@@ -4063,6 +4065,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma) {list_del_init (&vma->vma_link); +list_del (&vma->mm_list);drm_mm_remove_node (&vma->node);kfree (vma); }drivers/gpu/drm/i915/i915_gem_context.c +0
-0@@ -436,7 +436,10 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) {
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_context.c static int do_switch(struct i915_hw_context *to) · line 442
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem_context.c +3
-0+struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; +struct i915_address_space *ggtt = &dev_priv->gtt.base; from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; +list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list);I don't really see a reason to add this here ... shouldn't
move_to_activetake care of this? Obviously not in this patch here but later on when it's converted over.
Yes. You're right - it's sort of an ugly intermediate artifact.
Quoted 161 lines
Quoted 55 lines
drivers/gpu/drm/i915/i915_gem_context.c +0
-0i915_gem_object_move_to_active (from->obj, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the drivers/gpu/drm/i915/i915_gem_evict.c +6
-8@@ -87,8 +87,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->inactive_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -97,8 +96,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, goto none;/* Now merge in the soon-to-be-expired objects... */ -list_for_each_entry (obj, &vm->active_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->active_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -159,7 +157,7 @@ i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private;struct i915_address_space *vm; -struct drm_i915_gem_object *obj, *next; +struct i915_vma *vma, *next;bool lists_empty = true;int ret;@@ -187,9 +185,9 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); +list_for_each_entry_safe (vma, next, &vm->inactive_list, mm_list) +if (vma->obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (vma)); }return 0 ;drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0drivers/gpu/drm/i915/i915_gem_execbuffer.c i915_gem_execbuffer_move_to_active(struct list_head *objects, · line 805
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2
-0@@ -801,6 +801,8 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; +/* FIXME: This lookup gets fixed later <-- danvet */ +list_move_tail (&i915_gem_obj_to_vma (obj, vm)->mm_list, &vm->active_list);Ah, I guess the same comment applies to the lru frobbing in
do_switch?Quoted 46 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0i915_gem_object_move_to_active (obj, ring);if (obj->base.write_domain) { obj->dirty =1 ;drivers/gpu/drm/i915/i915_gem_stolen.c +1
-1@@ -419,7 +419,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, obj->has_global_gtt_mapping =1 ;list_add_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &ggtt->inactive_list); +list_add_tail (&vma->mm_list, &ggtt->inactive_list);return obj;drivers/gpu/drm/i915/i915_gpu_error.c +5
-4@@ -556,11 +556,11 @@ static void capture_bo(struct drm_i915_error_buffer *err,static u32 capture_active_bo (struct drm_i915_error_buffer *err,int count,struct list_head *head) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;int i =0 ; -list_for_each_entry (obj, head, mm_list) { -capture_bo (err++, obj); +list_for_each_entry (vma, head, mm_list) { +capture_bo (err++, vma->obj);if (++i == count)break ; }@@ -622,7 +622,8 @@ static struct drm_i915_error_object * i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, struct intel_ring_buffer *ring) { - struct i915_address_space *vm = &dev_priv->gtt.base; + struct i915_address_space *vm; +struct i915_vma *vma;struct drm_i915_gem_object *obj;u32 seqno;drivers/gpu/drm/i915/i915_gpu_error.c i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, · line 647
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +2
-3@@ -642,20 +643,23 @@ i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, } seqno = ring->get_seqno (ring, false); -list_for_each_entry (obj, &vm->active_list, mm_list) { -if (obj->ring != ring) -continue ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry (vma, &vm->active_list, mm_list) {We could instead loop over the bound list and check for ->active. But this is ok, too albeit a bit convoluted imo.
Quoted 40 lines
drivers/gpu/drm/i915/i915_gpu_error.c +14
-9+ obj = vma->obj; +if (obj->ring != ring) +continue ; -if (i915_seqno_passed (seqno, obj->last_read_seqno)) -continue ; +if (i915_seqno_passed (seqno, obj->last_read_seqno)) +continue ; -if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) -continue ; +if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) +continue ; -/* We need to copy these to an anonymous buffer as the simplest -* method to avoid being overwritten by userspace. -*/ -return i915_error_object_create (dev_priv, obj); +/* We need to copy these to an anonymous buffer as the simplest +* method to avoid being overwritten by userspace. +*/ +return i915_error_object_create (dev_priv, obj); + } }return NULL ;@@ -775,11 +779,12 @@ static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv, struct drm_i915_error_state *error) {struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_vma *vma;struct drm_i915_gem_object *obj;int i; i =0 ; -list_for_each_entry (obj, &vm->active_list, mm_list) +list_for_each_entry (vma, &vm->active_list, mm_list) i++; error->active_bo_count = i;list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list)
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 21/29] drm/i915: mm_list is per VMA
Quoted 85 lines · Ben Widawsky
Quoted 66 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 5) - move
mm_list"The
mm_listis used for the active/inactive LRUs. Since those LRUs are per address space, the link should be per VMx .Because we'll only ever have 1 VMA before this point, it's not incorrect to defer this change until this point in the patch series, and doing it here makes the change much easier to understand.
Shamelessly manipulated out of Daniel:
"active/inactive stuff is used by eviction when we run out of address space, so needs to be per-vma and per-address space. Bound/unbound otoh is used by the shrinker which only cares about the amount of memory used and not one bit about in which address space this memory is all used in. Of course to actual kick out an object we need to unbind it from every address space, but for that we have the per-object list of vmas."Quoted 7 lines
v2: only bump GGTT LRU in
i915_gem_object_set_to_gtt_domain(Chris)v3: Moved earlier in the series
v4: Add dropped message from v3
Signed-off-by Ben Widawsky <[email protected]>Some comments below for this one. The lru changes look a bit strange so I'll wait for your confirmation that the
do_switchhunk has the same reasons s the one in execbuf with the FIXME comment.
-DanielQuoted 35 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 53 ++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_drv.h | 5 +-- drivers/gpu/drm/i915/i915_gem.c | 37 +++++++++++---------- drivers/gpu/drm/i915/i915_gem_context.c | 3 ++ drivers/gpu/drm/i915/i915_gem_evict.c | 14 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 ++ drivers/gpu/drm/i915/i915_gem_stolen.c | 2 +- drivers/gpu/drm/i915/i915_gpu_error.c | 37 ++++++++++++--------- 8 files changed, 91 insertions(+), 62 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +2
-1@@ -149,7 +149,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj; +struct i915_vma *vma;size_t total_obj_size, total_gtt_size;int count, ret;@@ -157,6 +157,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)if (ret)return ret; +/* FIXME: the user of this interface might want more than just GGTT */ switch (list) {case ACTIVE_LIST :seq_puts (m,"Active:\n" );@@ -172,12 +173,12 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data) }Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c static int i915_gem_object_list_info(struct seq_file *m, void *data) · line 181
Quoted 13 lines
drivers/gpu/drm/i915/i915_debugfs.c +6
-6total_obj_size = total_gtt_size = count =0 ; -list_for_each_entry (obj, head, mm_list) { -seq_puts (m," " ); -describe_obj (m, obj); -seq_putc (m,'\n' ); - total_obj_size += obj->base.size; - total_gtt_size +=i915_gem_obj_ggtt_size (obj); +list_for_each_entry (vma, head, mm_list) { +seq_printf (m," " ); +describe_obj (m, vma->obj); +seq_printf (m,"\n" ); + total_obj_size += vma->obj->base.size; + total_gtt_size +=i915_gem_obj_size (vma->obj, vma->vm);Why not use
vma->node.size? If you don't disagree I'll bikeshed this while applying.
Quoted 5 lines
I think in terms of the diff, it's more logical to do it how I did. The result should damn well be the same though, so go right ahead. When I set about writing the series, I really didn't want to use node.size/start directly as much as possible - so we can sneak things into the helpers as needed.
I've applied this bikeshed, but the patch required some wiggling in and conflict resolution. I've checked with your branch and that seems to be due to the removel of the inactive list walking to adjust the gpu domains in i915_gem_reset. Please check that I didn't botch the patch rebasing with your tree.
-Daniel
Quoted 436 lines
Quoted 61 lines
Quoted 61 lines
drivers/gpu/drm/i915/i915_debugfs.c +17
-4count++; }mutex_unlock (&dev->struct_mutex);@@ -224,7 +225,18 @@ static int per_file_stats(int id, void *ptr, void *data)return 0 ; } -static int i915_gem_object_info (struct seq_file *m,void *data) +#define count_vmas (list, member) do { \ + list_for_each_entry(vma, list, member) { \ + size += i915_gem_obj_ggtt_size(vma->obj); \ + ++count; \ + if (vma->obj->map_and_fenceable) { \ + mappable_size += i915_gem_obj_ggtt_size(vma->obj); \ + ++mappable_count; \ + } \ + } \ +} while (0) + +static int i915_gem_object_info (struct seq_file *m,void * data) {struct drm_info_node *node = (struct drm_info_node *) m->private;struct drm_device *dev = node->minor->dev;@@ -234,6 +246,7 @@ static int i915_gem_object_info(struct seq_file *m, void *data)struct drm_i915_gem_object *obj;struct i915_address_space *vm = &dev_priv->gtt.base;struct drm_file *file; +struct i915_vma *vma;int ret; ret =mutex_lock_interruptible (&dev->struct_mutex);@@ -253,12 +266,12 @@ static int i915_gem_object_info(struct seq_file *m, void *data) count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->active_list, mm_list); +count_vmas (&vm->active_list, mm_list);seq_printf (m," %u [%u] active objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->inactive_list, mm_list); +count_vmas (&vm->inactive_list, mm_list);seq_printf (m," %u [%u] inactive objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size);@@ -1771,7 +1784,8 @@ i915_drop_caches_set(void *data, u64 val)struct drm_device *dev = data;struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_object *obj, *next; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm; +struct i915_vma *vma, *x;int ret;DRM_DEBUG_DRIVER ("Dropping caches: 0x%08llx\n" , val);@@ -1792,13 +1806,16 @@ i915_drop_caches_set(void *data, u64 val)i915_gem_retire_requests (dev);Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c i915_drop_caches_set(void *data, u64 val) · line 1811
Quoted 11 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-7if (val &DROP_BOUND ) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) { -if (obj->pin_count) -continue ; - ret =i915_gem_object_ggtt_unbind (obj); -if (ret) - goto unlock; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (vma, x, &vm->inactive_list, + mm_list) {Imo the double-loop is a bit funny, looping over the global bound list and skipping all active objects is imo the more straightfoward logic. But I agree that this is the more straightforward conversion, so I'm ok with a follow-up fixup patch.
I guess we have a lot of such conversions. I don't really mind the change, just a bit worried that it's less tested than what I've already done. I'm also not yet convinced the result will be a huge improvement for readability, but I've been starting at these lists for so long, my opinion is quite biased.
I guess we'll have to see. I've made a note to myself to look into converting all these types of loops over, but as we should see little to no functional impact from the change, I'd like to hold off until we get the rest merged.
Quoted 172 lines
Quoted 172 lines
drivers/gpu/drm/i915/i915_debugfs.c +7
-0+if (vma->obj->pin_count) +continue ; + + ret =i915_vma_unbind (vma); +if (ret) + goto unlock; + } } }drivers/gpu/drm/i915/i915_drv.h +3
-2@@ -546,6 +546,9 @@ struct i915_vma {struct drm_i915_gem_object *obj;struct i915_address_space *vm; +/** This object's place on the active/inactive lists */ +struct list_head mm_list; +struct list_head vma_link;/* Link in the object's VMA list */ };@@ -1263,9 +1266,7 @@ struct drm_i915_gem_object {struct drm_mm_node *stolen;struct list_head global_list; -/** This object's place on the active/inactive lists */ struct list_head ring_list; -struct list_head mm_list;/** This object's place in the batchbuffer or on the eviction list */ struct list_head exec_list;drivers/gpu/drm/i915/i915_gem.c +20
-17@@ -1872,7 +1872,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;u32 seqno =intel_ring_get_seqno (ring);BUG_ON (ring ==NULL );@@ -1888,8 +1887,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, obj->active =1 ; } -/* Move from whatever list we were on to the tail of execution. */ -list_move_tail (&obj->mm_list, &vm->active_list);list_move_tail (&obj->ring_list, &ring->active_list); obj->last_read_seqno = seqno;@@ -1911,14 +1908,14 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = obj->base.dev->dev_private; +struct i915_address_space *ggtt_vm = &dev_priv->gtt.base; +struct i915_vma *vma =i915_gem_obj_to_vma (obj, ggtt_vm);BUG_ON (obj->base.write_domain & ~I915_GEM_GPU_DOMAINS );BUG_ON (!obj->active); -list_move_tail (&obj->mm_list, &vm->inactive_list); +list_move_tail (&vma->mm_list, &ggtt_vm->inactive_list);list_del_init (&obj->ring_list); obj->ring =NULL ;@@ -2286,9 +2283,9 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring; +struct i915_address_space *vm; +struct i915_vma *vma;int i;for_each_ring (ring, dev_priv, i)@@ -2298,8 +2295,8 @@ void i915_gem_reset(struct drm_device *dev)* necessary invalidation upon reuse. */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vma, &vm->inactive_list, mm_list) + vma->obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;i915_gem_restore_fences (dev); }@@ -2353,6 +2350,7 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +BUG_ON (!obj->active);i915_gem_object_move_to_inactive (obj); }@@ -2635,7 +2633,6 @@ int i915_vma_unbind(struct i915_vma *vma)i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj); -list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ if (i915_is_ggtt (vma->vm)) obj->map_and_fenceable = true;@@ -3180,7 +3177,7 @@ search_free: goto err_out;list_move_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &vm->inactive_list); +list_add_tail (&vma->mm_list, &vm->inactive_list);/* Keep GGTT vmas first to make debug easier */ if (i915_is_ggtt (vm))@@ -3342,9 +3339,14 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain);/* And bump the LRU for this access */ -if (i915_gem_object_is_inactive (obj)) -list_move_tail (&obj->mm_list, - &dev_priv->gtt.base.inactive_list); +if (i915_gem_object_is_inactive (obj)) { +struct i915_vma *vma =i915_gem_obj_to_vma (obj, + &dev_priv->gtt.base); +if (vma) +list_move_tail (&vma->mm_list, + &dev_priv->gtt.base.inactive_list); + + }return 0 ; }@@ -3917,7 +3919,6 @@ unlock:void i915_gem_object_init (struct drm_i915_gem_object *obj,const struct drm_i915_gem_object_ops *ops) { -INIT_LIST_HEAD (&obj->mm_list);INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->exec_list);@@ -4054,6 +4055,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,return ERR_PTR (-ENOMEM );INIT_LIST_HEAD (&vma->vma_link); +INIT_LIST_HEAD (&vma->mm_list); vma->vm = vm; vma->obj = obj;@@ -4063,6 +4065,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma) {list_del_init (&vma->vma_link); +list_del (&vma->mm_list);drm_mm_remove_node (&vma->node);kfree (vma); }drivers/gpu/drm/i915/i915_gem_context.c +0
-0@@ -436,7 +436,10 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) {Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_context.c static int do_switch(struct i915_hw_context *to) · line 442
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem_context.c +3
-0+struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; +struct i915_address_space *ggtt = &dev_priv->gtt.base; from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; +list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list);I don't really see a reason to add this here ... shouldn't
move_to_activetake care of this? Obviously not in this patch here but later on when it's converted over.Yes. You're right - it's sort of an ugly intermediate artifact.
Quoted 161 lines
Quoted 55 lines
drivers/gpu/drm/i915/i915_gem_context.c +0
-0i915_gem_object_move_to_active (from->obj, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the drivers/gpu/drm/i915/i915_gem_evict.c +6
-8@@ -87,8 +87,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->inactive_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -97,8 +96,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, goto none;/* Now merge in the soon-to-be-expired objects... */ -list_for_each_entry (obj, &vm->active_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->active_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -159,7 +157,7 @@ i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private;struct i915_address_space *vm; -struct drm_i915_gem_object *obj, *next; +struct i915_vma *vma, *next;bool lists_empty = true;int ret;@@ -187,9 +185,9 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); +list_for_each_entry_safe (vma, next, &vm->inactive_list, mm_list) +if (vma->obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (vma)); }return 0 ;drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0drivers/gpu/drm/i915/i915_gem_execbuffer.c i915_gem_execbuffer_move_to_active(struct list_head *objects, · line 805
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2
-0@@ -801,6 +801,8 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; +/* FIXME: This lookup gets fixed later <-- danvet */ +list_move_tail (&i915_gem_obj_to_vma (obj, vm)->mm_list, &vm->active_list);Ah, I guess the same comment applies to the lru frobbing in
do_switch?Quoted 46 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0i915_gem_object_move_to_active (obj, ring);if (obj->base.write_domain) { obj->dirty =1 ;drivers/gpu/drm/i915/i915_gem_stolen.c +1
-1@@ -419,7 +419,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, obj->has_global_gtt_mapping =1 ;list_add_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &ggtt->inactive_list); +list_add_tail (&vma->mm_list, &ggtt->inactive_list);return obj;drivers/gpu/drm/i915/i915_gpu_error.c +5
-4@@ -556,11 +556,11 @@ static void capture_bo(struct drm_i915_error_buffer *err,static u32 capture_active_bo (struct drm_i915_error_buffer *err,int count,struct list_head *head) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;int i =0 ; -list_for_each_entry (obj, head, mm_list) { -capture_bo (err++, obj); +list_for_each_entry (vma, head, mm_list) { +capture_bo (err++, vma->obj);if (++i == count)break ; }@@ -622,7 +622,8 @@ static struct drm_i915_error_object * i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, struct intel_ring_buffer *ring) { - struct i915_address_space *vm = &dev_priv->gtt.base; + struct i915_address_space *vm; +struct i915_vma *vma;struct drm_i915_gem_object *obj;u32 seqno;drivers/gpu/drm/i915/i915_gpu_error.c i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, · line 647
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +2
-3@@ -642,20 +643,23 @@ i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, } seqno = ring->get_seqno (ring, false); -list_for_each_entry (obj, &vm->active_list, mm_list) { -if (obj->ring != ring) -continue ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry (vma, &vm->active_list, mm_list) {We could instead loop over the bound list and check for ->active. But this is ok, too albeit a bit convoluted imo.
Quoted 40 lines
drivers/gpu/drm/i915/i915_gpu_error.c +14
-9+ obj = vma->obj; +if (obj->ring != ring) +continue ; -if (i915_seqno_passed (seqno, obj->last_read_seqno)) -continue ; +if (i915_seqno_passed (seqno, obj->last_read_seqno)) +continue ; -if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) -continue ; +if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) +continue ; -/* We need to copy these to an anonymous buffer as the simplest -* method to avoid being overwritten by userspace. -*/ -return i915_error_object_create (dev_priv, obj); +/* We need to copy these to an anonymous buffer as the simplest +* method to avoid being overwritten by userspace. +*/ +return i915_error_object_create (dev_priv, obj); + } }return NULL ;@@ -775,11 +779,12 @@ static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv, struct drm_i915_error_state *error) {struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_vma *vma;struct drm_i915_gem_object *obj;int i; i =0 ; -list_for_each_entry (obj, &vm->active_list, mm_list) +list_for_each_entry (vma, &vm->active_list, mm_list) i++; error->active_bo_count = i;list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list)Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 21/29] drm/i915: mm_list is per VMA
Quoted 92 lines · Daniel Vetter
Quoted 85 lines · Ben Widawsky
Quoted 66 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky
formerly: "drm/i915: Create VMAs (part 5) - move
mm_list"The
mm_listis used for the active/inactive LRUs. Since those LRUs are per address space, the link should be per VMx .Because we'll only ever have 1 VMA before this point, it's not incorrect to defer this change until this point in the patch series, and doing it here makes the change much easier to understand.
Shamelessly manipulated out of Daniel:
"active/inactive stuff is used by eviction when we run out of address space, so needs to be per-vma and per-address space. Bound/unbound otoh is used by the shrinker which only cares about the amount of memory used and not one bit about in which address space this memory is all used in. Of course to actual kick out an object we need to unbind it from every address space, but for that we have the per-object list of vmas."Quoted 7 lines
v2: only bump GGTT LRU in
i915_gem_object_set_to_gtt_domain(Chris)v3: Moved earlier in the series
v4: Add dropped message from v3
Signed-off-by Ben Widawsky <[email protected]>Some comments below for this one. The lru changes look a bit strange so I'll wait for your confirmation that the
do_switchhunk has the same reasons s the one in execbuf with the FIXME comment.
-DanielQuoted 35 lines
--- drivers/gpu/drm/i915/i915_debugfs.c | 53 ++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_drv.h | 5 +-- drivers/gpu/drm/i915/i915_gem.c | 37 +++++++++++---------- drivers/gpu/drm/i915/i915_gem_context.c | 3 ++ drivers/gpu/drm/i915/i915_gem_evict.c | 14 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 ++ drivers/gpu/drm/i915/i915_gem_stolen.c | 2 +- drivers/gpu/drm/i915/i915_gpu_error.c | 37 ++++++++++++--------- 8 files changed, 91 insertions(+), 62 deletions(-)drivers/gpu/drm/i915/i915_debugfs.c +2
-1@@ -149,7 +149,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)struct drm_device *dev = node->minor->dev;struct drm_i915_private *dev_priv = dev->dev_private;struct i915_address_space *vm = &dev_priv->gtt.base; -struct drm_i915_gem_object *obj; +struct i915_vma *vma;size_t total_obj_size, total_gtt_size;int count, ret;@@ -157,6 +157,7 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)if (ret)return ret; +/* FIXME: the user of this interface might want more than just GGTT */ switch (list) {case ACTIVE_LIST :seq_puts (m,"Active:\n" );@@ -172,12 +173,12 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data) }Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c static int i915_gem_object_list_info(struct seq_file *m, void *data) · line 181
Quoted 13 lines
drivers/gpu/drm/i915/i915_debugfs.c +6
-6total_obj_size = total_gtt_size = count =0 ; -list_for_each_entry (obj, head, mm_list) { -seq_puts (m," " ); -describe_obj (m, obj); -seq_putc (m,'\n' ); - total_obj_size += obj->base.size; - total_gtt_size +=i915_gem_obj_ggtt_size (obj); +list_for_each_entry (vma, head, mm_list) { +seq_printf (m," " ); +describe_obj (m, vma->obj); +seq_printf (m,"\n" ); + total_obj_size += vma->obj->base.size; + total_gtt_size +=i915_gem_obj_size (vma->obj, vma->vm);Why not use
vma->node.size? If you don't disagree I'll bikeshed this while applying.Quoted 5 lines
I think in terms of the diff, it's more logical to do it how I did. The result should damn well be the same though, so go right ahead. When I set about writing the series, I really didn't want to use node.size/start directly as much as possible - so we can sneak things into the helpers as needed.
Quoted 7 lines
I've applied this bikeshed, but the patch required some wiggling in and conflict resolution. I've checked with your branch and that seems to be due to the removel of the inactive list walking to adjust the gpu domains in
i915_gem_reset. Please check that I didn't botch the patch rebasing with your tree.
-Daniel
You killed a BUG in i915_gem_retire_requests_ring, shouldn't that be a WARN or are you in the business of completely killing assertions now :p?
Otherwise, it looks good to me. There are enough diffs because of some other patches you merged (like watermarks) - that I may have well missed something in the noise; ie. no promises.
Quoted 441 lines
Quoted 436 lines
Quoted 61 lines
Quoted 61 lines
drivers/gpu/drm/i915/i915_debugfs.c +17
-4count++; }mutex_unlock (&dev->struct_mutex);@@ -224,7 +225,18 @@ static int per_file_stats(int id, void *ptr, void *data)return 0 ; } -static int i915_gem_object_info (struct seq_file *m,void *data) +#define count_vmas (list, member) do { \ + list_for_each_entry(vma, list, member) { \ + size += i915_gem_obj_ggtt_size(vma->obj); \ + ++count; \ + if (vma->obj->map_and_fenceable) { \ + mappable_size += i915_gem_obj_ggtt_size(vma->obj); \ + ++mappable_count; \ + } \ + } \ +} while (0) + +static int i915_gem_object_info (struct seq_file *m,void * data) {struct drm_info_node *node = (struct drm_info_node *) m->private;struct drm_device *dev = node->minor->dev;@@ -234,6 +246,7 @@ static int i915_gem_object_info(struct seq_file *m, void *data)struct drm_i915_gem_object *obj;struct i915_address_space *vm = &dev_priv->gtt.base;struct drm_file *file; +struct i915_vma *vma;int ret; ret =mutex_lock_interruptible (&dev->struct_mutex);@@ -253,12 +266,12 @@ static int i915_gem_object_info(struct seq_file *m, void *data) count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->active_list, mm_list); +count_vmas (&vm->active_list, mm_list);seq_printf (m," %u [%u] active objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size); size = count = mappable_size = mappable_count =0 ; -count_objects (&vm->inactive_list, mm_list); +count_vmas (&vm->inactive_list, mm_list);seq_printf (m," %u [%u] inactive objects, %zu [%zu] bytes\n" , count, mappable_count, size, mappable_size);@@ -1771,7 +1784,8 @@ i915_drop_caches_set(void *data, u64 val)struct drm_device *dev = data;struct drm_i915_private *dev_priv = dev->dev_private;struct drm_i915_gem_object *obj, *next; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_address_space *vm; +struct i915_vma *vma, *x;int ret;DRM_DEBUG_DRIVER ("Dropping caches: 0x%08llx\n" , val);@@ -1792,13 +1806,16 @@ i915_drop_caches_set(void *data, u64 val)i915_gem_retire_requests (dev);Quoted 17 lines
drivers/gpu/drm/i915/i915_debugfs.c i915_drop_caches_set(void *data, u64 val) · line 1811
Quoted 11 lines
drivers/gpu/drm/i915/i915_debugfs.c +3
-7if (val &DROP_BOUND ) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, - mm_list) { -if (obj->pin_count) -continue ; - ret =i915_gem_object_ggtt_unbind (obj); -if (ret) - goto unlock; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry_safe (vma, x, &vm->inactive_list, + mm_list) {Imo the double-loop is a bit funny, looping over the global bound list and skipping all active objects is imo the more straightfoward logic. But I agree that this is the more straightforward conversion, so I'm ok with a follow-up fixup patch.
I guess we have a lot of such conversions. I don't really mind the change, just a bit worried that it's less tested than what I've already done. I'm also not yet convinced the result will be a huge improvement for readability, but I've been starting at these lists for so long, my opinion is quite biased.
I guess we'll have to see. I've made a note to myself to look into converting all these types of loops over, but as we should see little to no functional impact from the change, I'd like to hold off until we get the rest merged.
Quoted 172 lines
Quoted 172 lines
drivers/gpu/drm/i915/i915_debugfs.c +7
-0+if (vma->obj->pin_count) +continue ; + + ret =i915_vma_unbind (vma); +if (ret) + goto unlock; + } } }drivers/gpu/drm/i915/i915_drv.h +3
-2@@ -546,6 +546,9 @@ struct i915_vma {struct drm_i915_gem_object *obj;struct i915_address_space *vm; +/** This object's place on the active/inactive lists */ +struct list_head mm_list; +struct list_head vma_link;/* Link in the object's VMA list */ };@@ -1263,9 +1266,7 @@ struct drm_i915_gem_object {struct drm_mm_node *stolen;struct list_head global_list; -/** This object's place on the active/inactive lists */ struct list_head ring_list; -struct list_head mm_list;/** This object's place in the batchbuffer or on the eviction list */ struct list_head exec_list;drivers/gpu/drm/i915/i915_gem.c +20
-17@@ -1872,7 +1872,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, {struct drm_device *dev = obj->base.dev;struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base;u32 seqno =intel_ring_get_seqno (ring);BUG_ON (ring ==NULL );@@ -1888,8 +1887,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, obj->active =1 ; } -/* Move from whatever list we were on to the tail of execution. */ -list_move_tail (&obj->mm_list, &vm->active_list);list_move_tail (&obj->ring_list, &ring->active_list); obj->last_read_seqno = seqno;@@ -1911,14 +1908,14 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) { -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = obj->base.dev->dev_private; +struct i915_address_space *ggtt_vm = &dev_priv->gtt.base; +struct i915_vma *vma =i915_gem_obj_to_vma (obj, ggtt_vm);BUG_ON (obj->base.write_domain & ~I915_GEM_GPU_DOMAINS );BUG_ON (!obj->active); -list_move_tail (&obj->mm_list, &vm->inactive_list); +list_move_tail (&vma->mm_list, &ggtt_vm->inactive_list);list_del_init (&obj->ring_list); obj->ring =NULL ;@@ -2286,9 +2283,9 @@ void i915_gem_restore_fences(struct drm_device *dev)void i915_gem_reset (struct drm_device *dev) {struct drm_i915_private *dev_priv = dev->dev_private; -struct i915_address_space *vm; -struct drm_i915_gem_object *obj;struct intel_ring_buffer *ring; +struct i915_address_space *vm; +struct i915_vma *vma;int i;for_each_ring (ring, dev_priv, i)@@ -2298,8 +2295,8 @@ void i915_gem_reset(struct drm_device *dev)* necessary invalidation upon reuse. */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) -list_for_each_entry (obj, &vm->inactive_list, mm_list) - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ; +list_for_each_entry (vma, &vm->inactive_list, mm_list) + vma->obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS ;i915_gem_restore_fences (dev); }@@ -2353,6 +2350,7 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +BUG_ON (!obj->active);i915_gem_object_move_to_inactive (obj); }@@ -2635,7 +2633,6 @@ int i915_vma_unbind(struct i915_vma *vma)i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj); -list_del (&obj->mm_list);/* Avoid an unnecessary call to unbind on rebind. */ if (i915_is_ggtt (vma->vm)) obj->map_and_fenceable = true;@@ -3180,7 +3177,7 @@ search_free: goto err_out;list_move_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &vm->inactive_list); +list_add_tail (&vma->mm_list, &vm->inactive_list);/* Keep GGTT vmas first to make debug easier */ if (i915_is_ggtt (vm))@@ -3342,9 +3339,14 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain);/* And bump the LRU for this access */ -if (i915_gem_object_is_inactive (obj)) -list_move_tail (&obj->mm_list, - &dev_priv->gtt.base.inactive_list); +if (i915_gem_object_is_inactive (obj)) { +struct i915_vma *vma =i915_gem_obj_to_vma (obj, + &dev_priv->gtt.base); +if (vma) +list_move_tail (&vma->mm_list, + &dev_priv->gtt.base.inactive_list); + + }return 0 ; }@@ -3917,7 +3919,6 @@ unlock:void i915_gem_object_init (struct drm_i915_gem_object *obj,const struct drm_i915_gem_object_ops *ops) { -INIT_LIST_HEAD (&obj->mm_list);INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->exec_list);@@ -4054,6 +4055,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,return ERR_PTR (-ENOMEM );INIT_LIST_HEAD (&vma->vma_link); +INIT_LIST_HEAD (&vma->mm_list); vma->vm = vm; vma->obj = obj;@@ -4063,6 +4065,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,void i915_gem_vma_destroy (struct i915_vma *vma) {list_del_init (&vma->vma_link); +list_del (&vma->mm_list);drm_mm_remove_node (&vma->node);kfree (vma); }drivers/gpu/drm/i915/i915_gem_context.c +0
-0@@ -436,7 +436,10 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) {Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_context.c static int do_switch(struct i915_hw_context *to) · line 442
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem_context.c +3
-0+struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; +struct i915_address_space *ggtt = &dev_priv->gtt.base; from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; +list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list);I don't really see a reason to add this here ... shouldn't
move_to_activetake care of this? Obviously not in this patch here but later on when it's converted over.Yes. You're right - it's sort of an ugly intermediate artifact.
Quoted 161 lines
Quoted 55 lines
drivers/gpu/drm/i915/i915_gem_context.c +0
-0i915_gem_object_move_to_active (from->obj, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the drivers/gpu/drm/i915/i915_gem_evict.c +6
-8@@ -87,8 +87,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_mm_init_scan (&vm->mm, min_size, alignment, cache_level);/* First see if there is a large enough contiguous idle region... */ -list_for_each_entry (obj, &vm->inactive_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->inactive_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -97,8 +96,7 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, goto none;/* Now merge in the soon-to-be-expired objects... */ -list_for_each_entry (obj, &vm->active_list, mm_list) { -struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); +list_for_each_entry (vma, &vm->active_list, mm_list) {if (mark_free (vma, &unwind_list)) goto found; }@@ -159,7 +157,7 @@ i915_gem_evict_everything(struct drm_device *dev) {drm_i915_private_t *dev_priv = dev->dev_private;struct i915_address_space *vm; -struct drm_i915_gem_object *obj, *next; +struct i915_vma *vma, *next;bool lists_empty = true;int ret;@@ -187,9 +185,9 @@ i915_gem_evict_everything(struct drm_device *dev)/* Having flushed everything, unbind() should never raise an error */ list_for_each_entry (vm, &dev_priv->vm_list, global_link) { -list_for_each_entry_safe (obj, next, &vm->inactive_list, mm_list) -if (obj->pin_count ==0 ) -WARN_ON (i915_vma_unbind (i915_gem_obj_to_vma (obj, vm))); +list_for_each_entry_safe (vma, next, &vm->inactive_list, mm_list) +if (vma->obj->pin_count ==0 ) +WARN_ON (i915_vma_unbind (vma)); }return 0 ;drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0drivers/gpu/drm/i915/i915_gem_execbuffer.c i915_gem_execbuffer_move_to_active(struct list_head *objects, · line 805
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2
-0@@ -801,6 +801,8 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; +/* FIXME: This lookup gets fixed later <-- danvet */ +list_move_tail (&i915_gem_obj_to_vma (obj, vm)->mm_list, &vm->active_list);Ah, I guess the same comment applies to the lru frobbing in
do_switch?Quoted 46 lines
drivers/gpu/drm/i915/i915_gem_execbuffer.c +0
-0i915_gem_object_move_to_active (obj, ring);if (obj->base.write_domain) { obj->dirty =1 ;drivers/gpu/drm/i915/i915_gem_stolen.c +1
-1@@ -419,7 +419,7 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev, obj->has_global_gtt_mapping =1 ;list_add_tail (&obj->global_list, &dev_priv->mm.bound_list); -list_add_tail (&obj->mm_list, &ggtt->inactive_list); +list_add_tail (&vma->mm_list, &ggtt->inactive_list);return obj;drivers/gpu/drm/i915/i915_gpu_error.c +5
-4@@ -556,11 +556,11 @@ static void capture_bo(struct drm_i915_error_buffer *err,static u32 capture_active_bo (struct drm_i915_error_buffer *err,int count,struct list_head *head) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;int i =0 ; -list_for_each_entry (obj, head, mm_list) { -capture_bo (err++, obj); +list_for_each_entry (vma, head, mm_list) { +capture_bo (err++, vma->obj);if (++i == count)break ; }@@ -622,7 +622,8 @@ static struct drm_i915_error_object * i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, struct intel_ring_buffer *ring) { - struct i915_address_space *vm = &dev_priv->gtt.base; + struct i915_address_space *vm; +struct i915_vma *vma;struct drm_i915_gem_object *obj;u32 seqno;drivers/gpu/drm/i915/i915_gpu_error.c i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, · line 647
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +2
-3@@ -642,20 +643,23 @@ i915_error_first_batchbuffer(struct drm_i915_private *dev_priv, } seqno = ring->get_seqno (ring, false); -list_for_each_entry (obj, &vm->active_list, mm_list) { -if (obj->ring != ring) -continue ; +list_for_each_entry (vm, &dev_priv->vm_list, global_link) { +list_for_each_entry (vma, &vm->active_list, mm_list) {We could instead loop over the bound list and check for ->active. But this is ok, too albeit a bit convoluted imo.
Quoted 40 lines
drivers/gpu/drm/i915/i915_gpu_error.c +14
-9+ obj = vma->obj; +if (obj->ring != ring) +continue ; -if (i915_seqno_passed (seqno, obj->last_read_seqno)) -continue ; +if (i915_seqno_passed (seqno, obj->last_read_seqno)) +continue ; -if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) -continue ; +if ((obj->base.read_domains &I915_GEM_DOMAIN_COMMAND ) ==0 ) +continue ; -/* We need to copy these to an anonymous buffer as the simplest -* method to avoid being overwritten by userspace. -*/ -return i915_error_object_create (dev_priv, obj); +/* We need to copy these to an anonymous buffer as the simplest +* method to avoid being overwritten by userspace. +*/ +return i915_error_object_create (dev_priv, obj); + } }return NULL ;@@ -775,11 +779,12 @@ static void i915_gem_capture_buffers(struct drm_i915_private *dev_priv, struct drm_i915_error_state *error) {struct i915_address_space *vm = &dev_priv->gtt.base; +struct i915_vma *vma;struct drm_i915_gem_object *obj;int i; i =0 ; -list_for_each_entry (obj, &vm->active_list, mm_list) +list_for_each_entry (vma, &vm->active_list, mm_list) i++; error->active_bo_count = i;list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list)Signature · 1 line
Ben Widawsky, Intel Open Source Technology CenterSignature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
Re: [PATCH 21/29] drm/i915: mm_list is per VMA
Quoted 1 line · Ben Widawsky
You killed a BUG in
i915_gem_retire_requests_ring, shouldn't that be a WARN or are you in the business of completely killing assertions now :p?
Yeah, and my little commit message annotation even explained that it's fully redundant since the move_to_inactive function called on the next line has the exact same check ;-)
Quoted 3 lines
Otherwise, it looks good to me. There are enough diffs because of some other patches you merged (like watermarks) - that I may have well missed something in the noise; ie. no promises.
Thanks, though stupid me failed to push out the last patch I've merged. But that one applied without fuzz, so I think it should be ok. -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 21/29] drm/i915: mm_list is per VMA
Quoted 6 lines · Daniel Vetter
Quoted 1 line · Ben Widawsky
You killed a BUG in
i915_gem_retire_requests_ring, shouldn't that be a WARN or are you in the business of completely killing assertions now :p?Yeah, and my little commit message annotation even explained that it's fully redundant since the
move_to_inactivefunction called on the next line has the exact same check ;-)
Heh, I realized this as I was dozing off last night... "didn't danvet point that out already." Oh well, at least I passed the test.
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
PATCH 22/29 drm/i915: Update error capture for VMs
In reply to Ben Widawskyformerly: "drm/i915: Create VMAs (part 4) - Error capture"
Since the active/inactive lists are per VM, we need to modify the error capture code to be aware of this, and also extend it to capture the buffers from all the VMs. For now all the code assumes only 1 VM, but it will become more generic over the next few patches.
NOTE: If the number of VMs in a real world system grows significantly we'll have to focus on only capturing the guilty VM, or else it's likely there won't be enough space for error capture.
v2: Squashed in the "part 6" which had dependencies on the mm_list change. Since I've moved the mm_list change to an earlier point in the series, we were able to accomplish it here and now.
v3: Rebased over new error capture
--- drivers/gpu/drm/i915/i915_drv.h | 4 +- drivers/gpu/drm/i915/i915_gpu_error.c | 76 ++++++++++++++++++++++++----------- 2 files changed, 55 insertions(+), 25 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +2 -2
@@ -323,8 +323,8 @@ struct drm_i915_error_state { u32 purgeable:1; s32 ring:4; u32 cache_level:2; - } *active_bo, *pinned_bo; -u32 active_bo_count, pinned_bo_count; + } **active_bo, **pinned_bo; +u32 *active_bo_count, *pinned_bo_count;struct intel_overlay_error_state *overlay;struct intel_display_error_state *display; };
drivers/gpu/drm/i915/i915_gpu_error.c +53 -23
108 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 23/29 drm/i915: Add vma to list at creation
In reply to Ben WidawskyWith the current code there shouldn't be a distinction - however with an upcoming change we intend to allocate a vma much earlier, before it's actually bound anywhere.
To do this we have to check node allocation as well for the _bound() check.
--- drivers/gpu/drm/i915/i915_gem.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +7 -7
@@ -3179,12 +3179,6 @@ search_free:list_move_tail (&obj->global_list, &dev_priv->mm.bound_list);list_add_tail (&vma->mm_list, &vm->inactive_list); -/* Keep GGTT vmas first to make debug easier */ -if (i915_is_ggtt (vm)) -list_add (&vma->vma_link, &obj->vma_list); -else -list_add_tail (&vma->vma_link, &obj->vma_list); - fenceable =i915_is_ggtt (vm) && i915_gem_obj_ggtt_size(obj) == fence_size &&
@@ -4059,6 +4053,12 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj, vma->vm = vm; vma->obj = obj; +/* Keep GGTT vmas first to make debug easier */ +if (i915_is_ggtt (vm)) +list_add (&vma->vma_link, &obj->vma_list); +else +list_add_tail (&vma->vma_link, &obj->vma_list); +return vma; }
@@ -4754,7 +4754,7 @@ bool i915_gem_obj_bound(struct drm_i915_gem_object *o,struct i915_vma *vma;list_for_each_entry (vma, &o->vma_list, vma_link) -if (vma->vm == vm) +if (vma->vm == vm &&drm_mm_node_allocated (&vma->node))return true;return false;
Signature · 1 line
1.8.3.4
PATCH 24/29 drm/i915: create vmas at execbuf
In reply to Ben WidawskyIn order to transition more of our code over to using a VMA instead of an <OBJ, VM> pair - we must have the vma accessible at execbuf time. Up until now, we've only had a VMA when actually binding an object.
The previous patch helped handle the distinction on bound vs. unbound. This patch will help us catch leaks, and other issues before we actually shuffle a bunch of stuff around.
The subsequent patch to fix up the rest of execbuf should be mostly just moving code around, and this is the major functional change.
v2: Release table_lock earlier so vma allocation needn't be atomic. (Chris)
--- drivers/gpu/drm/i915/i915_drv.h | 3 +++ drivers/gpu/drm/i915/i915_gem.c | 25 ++++++++++++++++++------- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 18 +++++++++++++----- 3 files changed, 34 insertions(+), 12 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +3 -0
@@ -1857,6 +1857,9 @@ unsigned long i915_gem_obj_size(struct drm_i915_gem_object *o, struct i915_address_space *vm);struct i915_vma *i915_gem_obj_to_vma (struct drm_i915_gem_object *obj,struct i915_address_space *vm); +struct i915_vma * +i915_gem_obj_lookup_or_create_vma (struct drm_i915_gem_object *obj, +struct i915_address_space *vm);/* Some GGTT VM helpers */ #define obj_to_ggtt (obj) \ (&((struct drm_i915_private *)(obj)->base.dev->dev_private)->gtt.base)
drivers/gpu/drm/i915/i915_gem.c +18 -7
@@ -3101,8 +3101,7 @@ i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,struct i915_vma *vma;int ret; -if (WARN_ON (!list_empty (&obj->vma_list))) -return -EBUSY ; +BUG_ON (!i915_is_ggtt (vm)); fence_size = i915_gem_get_gtt_size(dev, obj->base.size,
@@ -3142,16 +3141,15 @@ i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,i915_gem_object_pin_pages (obj); -/* FIXME: For now we only ever use 1 VMA per object */ -BUG_ON (!i915_is_ggtt (vm)); -WARN_ON (!list_empty (&obj->vma_list)); - - vma =i915_gem_vma_create (obj, vm); + vma =i915_gem_obj_lookup_or_create_vma (obj, vm);if (IS_ERR (vma)) {i915_gem_object_unpin_pages (obj);return PTR_ERR (vma); } +/* For now we only ever use 1 vma per object */ +WARN_ON (!list_is_singular (&obj->vma_list)); + search_free: ret = drm_mm_insert_node_in_range_generic(&vm->mm, &vma->node, size, alignment,
@@ -4800,3 +4798,16 @@ struct i915_vma *i915_gem_obj_to_vma(struct drm_i915_gem_object *obj,return NULL ; } + +struct i915_vma * +i915_gem_obj_lookup_or_create_vma (struct drm_i915_gem_object *obj, +struct i915_address_space *vm) +{ +struct i915_vma *vma; + + vma =i915_gem_obj_to_vma (obj, vm); +if (!vma) + vma =i915_gem_vma_create (obj, vm); + +return vma; +}
drivers/gpu/drm/i915/i915_gem_execbuffer.c +13 -5
@@ -85,14 +85,14 @@ static int eb_lookup_objects(struct eb_objects *eb, struct drm_i915_gem_exec_object2 *exec, const struct drm_i915_gem_execbuffer2 *args, + struct i915_address_space *vm, struct drm_file *file) { + struct drm_i915_gem_object *obj;int i;spin_lock (&file->table_lock);for (i =0 ; i < args->buffer_count; i++) { -struct drm_i915_gem_object *obj; - obj =to_intel_bo (idr_find (&file->object_idr, exec[i].handle));if (obj ==NULL ) {spin_unlock (&file->table_lock);
@@ -110,6 +110,15 @@ eb_lookup_objects(struct eb_objects *eb,drm_gem_object_reference (&obj->base);list_add_tail (&obj->exec_list, &eb->objects); + } +spin_unlock (&file->table_lock); + +list_for_each_entry (obj, &eb->objects, exec_list) { +struct i915_vma *vma; + + vma =i915_gem_obj_lookup_or_create_vma (obj, vm); +if (IS_ERR (vma)) +return PTR_ERR (vma); obj->exec_entry = &exec[i];if (eb->and <0 ) {
@@ -121,7 +130,6 @@ eb_lookup_objects(struct eb_objects *eb, &eb->buckets[handle & eb->and]); } } -spin_unlock (&file->table_lock);return 0 ; }
@@ -672,7 +680,7 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,/* reacquire the objects */ eb_reset (eb); - ret =eb_lookup_objects (eb, exec, args, file); + ret =eb_lookup_objects (eb, exec, args, vm, file);if (ret) goto err;
@@ -1009,7 +1017,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, }/* Look up object handles */ - ret =eb_lookup_objects (eb, exec, args, file); + ret =eb_lookup_objects (eb, exec, args, vm, file);if (ret) goto err;
Signature · 1 line
1.8.3.4
Re: [PATCH 24/29] drm/i915: create vmas at execbuf
Quoted 8 lines · Ben Widawsky
In order to transition more of our code over to using a VMA instead of an <OBJ, VM> pair - we must have the vma accessible at execbuf time. Up until now, we've only had a VMA when actually binding an object.
The previous patch helped handle the distinction on bound vs. unbound. This patch will help us catch leaks, and other issues before we actually shuffle a bunch of stuff around.
Quoted 7 lines
The subsequent patch to fix up the rest of execbuf should be mostly just moving code around, and this is the major functional change.
v2: Release
table_lockearlier so vma allocation needn't be atomic. (Chris)Signed-off-by Ben Widawsky <[email protected]>
Merged up to this patch to dinq, thanks.
-Daniel
Quoted 142 lines
--- drivers/gpu/drm/i915/i915_drv.h | 3 +++ drivers/gpu/drm/i915/i915_gem.c | 25 ++++++++++++++++++------- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 18 +++++++++++++----- 3 files changed, 34 insertions(+), 12 deletions(-)drivers/gpu/drm/i915/i915_drv.h +3
-0@@ -1857,6 +1857,9 @@ unsigned long i915_gem_obj_size(struct drm_i915_gem_object *o, struct i915_address_space *vm);struct i915_vma *i915_gem_obj_to_vma (struct drm_i915_gem_object *obj,struct i915_address_space *vm); +struct i915_vma * +i915_gem_obj_lookup_or_create_vma (struct drm_i915_gem_object *obj, +struct i915_address_space *vm);/* Some GGTT VM helpers */ #define obj_to_ggtt (obj) \ (&((struct drm_i915_private *)(obj)->base.dev->dev_private)->gtt.base)drivers/gpu/drm/i915/i915_gem.c +18
-7@@ -3101,8 +3101,7 @@ i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,struct i915_vma *vma;int ret; -if (WARN_ON (!list_empty (&obj->vma_list))) -return -EBUSY ; +BUG_ON (!i915_is_ggtt (vm)); fence_size = i915_gem_get_gtt_size(dev, obj->base.size,@@ -3142,16 +3141,15 @@ i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,i915_gem_object_pin_pages (obj); -/* FIXME: For now we only ever use 1 VMA per object */ -BUG_ON (!i915_is_ggtt (vm)); -WARN_ON (!list_empty (&obj->vma_list)); - - vma =i915_gem_vma_create (obj, vm); + vma =i915_gem_obj_lookup_or_create_vma (obj, vm);if (IS_ERR (vma)) {i915_gem_object_unpin_pages (obj);return PTR_ERR (vma); } +/* For now we only ever use 1 vma per object */ +WARN_ON (!list_is_singular (&obj->vma_list)); + search_free: ret = drm_mm_insert_node_in_range_generic(&vm->mm, &vma->node, size, alignment,@@ -4800,3 +4798,16 @@ struct i915_vma *i915_gem_obj_to_vma(struct drm_i915_gem_object *obj,return NULL ; } + +struct i915_vma * +i915_gem_obj_lookup_or_create_vma (struct drm_i915_gem_object *obj, +struct i915_address_space *vm) +{ +struct i915_vma *vma; + + vma =i915_gem_obj_to_vma (obj, vm); +if (!vma) + vma =i915_gem_vma_create (obj, vm); + +return vma; +}drivers/gpu/drm/i915/i915_gem_execbuffer.c +13
-5@@ -85,14 +85,14 @@ static int eb_lookup_objects(struct eb_objects *eb, struct drm_i915_gem_exec_object2 *exec, const struct drm_i915_gem_execbuffer2 *args, + struct i915_address_space *vm, struct drm_file *file) { + struct drm_i915_gem_object *obj;int i;spin_lock (&file->table_lock);for (i =0 ; i < args->buffer_count; i++) { -struct drm_i915_gem_object *obj; - obj =to_intel_bo (idr_find (&file->object_idr, exec[i].handle));if (obj ==NULL ) {spin_unlock (&file->table_lock);@@ -110,6 +110,15 @@ eb_lookup_objects(struct eb_objects *eb,drm_gem_object_reference (&obj->base);list_add_tail (&obj->exec_list, &eb->objects); + } +spin_unlock (&file->table_lock); + +list_for_each_entry (obj, &eb->objects, exec_list) { +struct i915_vma *vma; + + vma =i915_gem_obj_lookup_or_create_vma (obj, vm); +if (IS_ERR (vma)) +return PTR_ERR (vma); obj->exec_entry = &exec[i];if (eb->and <0 ) {@@ -121,7 +130,6 @@ eb_lookup_objects(struct eb_objects *eb, &eb->buckets[handle & eb->and]); } } -spin_unlock (&file->table_lock);return 0 ; }@@ -672,7 +680,7 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,/* reacquire the objects */ eb_reset (eb); - ret =eb_lookup_objects (eb, exec, args, file); + ret =eb_lookup_objects (eb, exec, args, vm, file);if (ret) goto err;@@ -1009,7 +1017,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, }/* Look up object handles */ - ret =eb_lookup_objects (eb, exec, args, file); + ret =eb_lookup_objects (eb, exec, args, vm, file);if (ret) goto err;Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 25/29 drm/i915: Convert execbuf code to use vmas
In reply to Ben WidawskyThis attempts to convert all the execbuf code to speak in vmas. Since the execbuf code is very self contained it was a nice isolated conversion.
The meat of the code is about turning eb_objects into eb_vma, and then wiring up the rest of the code to use vmas instead of obj, vm pairs.
Unfortunately, to do this, we must move the exec_list link from the obj structure. This list is reused in the eviction code, so we must also modify the eviction code to make this work.
v2: Release table lock early, and two a 2 phase vma lookup to avoid having to use a GFP_ATOMIC. (Chris)
--- drivers/gpu/drm/i915/i915_drv.h | 22 +- drivers/gpu/drm/i915/i915_gem.c | 3 +- drivers/gpu/drm/i915/i915_gem_evict.c | 31 ++- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 322 +++++++++++++++-------------- 4 files changed, 201 insertions(+), 177 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +13 -9
@@ -550,6 +550,17 @@ struct i915_vma {struct list_head mm_list;struct list_head vma_link;/* Link in the object's VMA list */ + +/** This vma's place in the batchbuffer or on the eviction list */ +struct list_head exec_list; + +/** +* Used for performing relocations during execbuffer insertion. +*/ +struct hlist_node exec_node; +unsigned long exec_handle; +struct drm_i915_gem_exec_object2 *exec_entry; + };struct i915_ctx_hang_stats {
@@ -1267,8 +1278,8 @@ struct drm_i915_gem_object {struct list_head global_list;struct list_head ring_list; -/** This object's place in the batchbuffer or on the eviction list */ -struct list_head exec_list; +/** Used in execbuf to temporarily hold a ref */ +struct list_head obj_exec_list;/** * This is set if the object is on the active lists (has pending
@@ -1353,13 +1364,6 @@ struct drm_i915_gem_object {void *dma_buf_vmapping;int vmapping_count; -/** -* Used for performing relocations during execbuffer insertion. -*/ -struct hlist_node exec_node; -unsigned long exec_handle; -struct drm_i915_gem_exec_object2 *exec_entry; -struct intel_ring_buffer *ring;/** Breadcrumb of last rendering to the buffer. */
drivers/gpu/drm/i915/i915_gem.c +2 -1
@@ -3913,7 +3913,7 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj, {INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list); -INIT_LIST_HEAD (&obj->exec_list); +INIT_LIST_HEAD (&obj->obj_exec_list);INIT_LIST_HEAD (&obj->vma_list); obj->ops = ops;
@@ -4048,6 +4048,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,INIT_LIST_HEAD (&vma->vma_link);INIT_LIST_HEAD (&vma->mm_list); +INIT_LIST_HEAD (&vma->exec_list); vma->vm = vm; vma->obj = obj;
drivers/gpu/drm/i915/i915_gem_evict.c +14 -17
@@ -37,7 +37,7 @@ mark_free(struct i915_vma *vma, struct list_head *unwind)if (vma->obj->pin_count)return false; -list_add (&vma->obj->exec_list, unwind); +list_add (&vma->exec_list, unwind);return drm_mm_scan_add_block (&vma->node); }
@@ -49,7 +49,6 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_i915_private_t *dev_priv = dev->dev_private;struct list_head eviction_list, unwind_list;struct i915_vma *vma; -struct drm_i915_gem_object *obj;int ret =0 ;trace_i915_gem_evict (dev, min_size, alignment, mappable);
@@ -104,14 +103,13 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, none:/* Nothing found, clean up and bail out! */ while (!list_empty (&unwind_list)) { - obj =list_first_entry (&unwind_list, - struct drm_i915_gem_object, + vma =list_first_entry (&unwind_list, + struct i915_vma, exec_list); - vma =i915_gem_obj_to_vma (obj, vm); ret =drm_mm_scan_remove_block (&vma->node);BUG_ON (ret); -list_del_init (&obj->exec_list); +list_del_init (&vma->exec_list); }/* We expect the caller to unpin, evict all and try again, or give up.
@@ -125,28 +123,27 @@ found:* temporary list. */ INIT_LIST_HEAD (&eviction_list);while (!list_empty (&unwind_list)) { - obj =list_first_entry (&unwind_list, - struct drm_i915_gem_object, + vma =list_first_entry (&unwind_list, + struct i915_vma, exec_list); - vma =i915_gem_obj_to_vma (obj, vm);if (drm_mm_scan_remove_block (&vma->node)) { -list_move (&obj->exec_list, &eviction_list); -drm_gem_object_reference (&obj->base); +list_move (&vma->exec_list, &eviction_list); +drm_gem_object_reference (&vma->obj->base);continue ; } -list_del_init (&obj->exec_list); +list_del_init (&vma->exec_list); }/* Unbinding will emit any required flushes */ while (!list_empty (&eviction_list)) { - obj =list_first_entry (&eviction_list, - struct drm_i915_gem_object, + vma =list_first_entry (&eviction_list, + struct i915_vma, exec_list);if (ret ==0 ) - ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm)); + ret =i915_vma_unbind (vma); -list_del_init (&obj->exec_list); -drm_gem_object_unreference (&obj->base); +list_del_init (&vma->exec_list); +drm_gem_object_unreference (&vma->obj->base); }return ret;
drivers/gpu/drm/i915/i915_gem_execbuffer.c +172 -150
666 lines not on this page · raw message
Signature · 1 line
1.8.3.4
Re: [PATCH 25/29] drm/i915: Convert execbuf code to use vmas
Quoted 7 lines · Ben Widawsky
This attempts to convert all the execbuf code to speak in vmas. Since the execbuf code is very self contained it was a nice isolated conversion.
The meat of the code is about turning
eb_objectsintoeb_vma, and then wiring up the rest of the code to use vmas instead of obj, vm pairs.
Quoted 8 lines
Unfortunately, to do this, we must move the
exec_listlink from the obj structure. This list is reused in the eviction code, so we must also modify the eviction code to make this work.v2: Release table lock early, and two a 2 phase vma lookup to avoid having to use a
GFP_ATOMIC. (Chris)Signed-off-by Ben Widawsky <[email protected]>
I think the leaking of preallocated vmas if execbuf fails can blow up:
1. We call lookup_or_create and create new vmas, linked into the vma_link chain.
2. Later on execbuf fails somewhere (for an igt the simplest way is probaly to use more buffers than what would fit into the gtt) and we bail out.
-> Note that at this point we leak vmas which are on the vma_link list but which have no gtt node allocation.
3. Userspace dies in flames (or just quits).
4. All buffers get their final unref and we call vma_unbind on each vma, even the ones that do not have an allocation.
5. hilarity ensues since vma_unbind doesn't bail out if drm_mm_node_allocated(vma->node) == false.
We need broken userspace to actually exercise this bug since all normal ways for execbuf to bail out involve singals and ioclt restarting. If this is a real bug I think we need an igt to exercise it.
-Daniel
Quoted 855 lines
--- drivers/gpu/drm/i915/i915_drv.h | 22 +- drivers/gpu/drm/i915/i915_gem.c | 3 +- drivers/gpu/drm/i915/i915_gem_evict.c | 31 ++- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 322 +++++++++++++++-------------- 4 files changed, 201 insertions(+), 177 deletions(-)drivers/gpu/drm/i915/i915_drv.h +13
-9@@ -550,6 +550,17 @@ struct i915_vma {struct list_head mm_list;struct list_head vma_link;/* Link in the object's VMA list */ + +/** This vma's place in the batchbuffer or on the eviction list */ +struct list_head exec_list; + +/** +* Used for performing relocations during execbuffer insertion. +*/ +struct hlist_node exec_node; +unsigned long exec_handle; +struct drm_i915_gem_exec_object2 *exec_entry; + };struct i915_ctx_hang_stats {@@ -1267,8 +1278,8 @@ struct drm_i915_gem_object {struct list_head global_list;struct list_head ring_list; -/** This object's place in the batchbuffer or on the eviction list */ -struct list_head exec_list; +/** Used in execbuf to temporarily hold a ref */ +struct list_head obj_exec_list;/** * This is set if the object is on the active lists (has pending @@ -1353,13 +1364,6 @@ struct drm_i915_gem_object {void *dma_buf_vmapping;int vmapping_count; -/** -* Used for performing relocations during execbuffer insertion. -*/ -struct hlist_node exec_node; -unsigned long exec_handle; -struct drm_i915_gem_exec_object2 *exec_entry; -struct intel_ring_buffer *ring;/** Breadcrumb of last rendering to the buffer. */ drivers/gpu/drm/i915/i915_gem.c +2
-1@@ -3913,7 +3913,7 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj, {INIT_LIST_HEAD (&obj->global_list);INIT_LIST_HEAD (&obj->ring_list); -INIT_LIST_HEAD (&obj->exec_list); +INIT_LIST_HEAD (&obj->obj_exec_list);INIT_LIST_HEAD (&obj->vma_list); obj->ops = ops;@@ -4048,6 +4048,7 @@ struct i915_vma *i915_gem_vma_create(struct drm_i915_gem_object *obj,INIT_LIST_HEAD (&vma->vma_link);INIT_LIST_HEAD (&vma->mm_list); +INIT_LIST_HEAD (&vma->exec_list); vma->vm = vm; vma->obj = obj;drivers/gpu/drm/i915/i915_gem_evict.c +14
-17@@ -37,7 +37,7 @@ mark_free(struct i915_vma *vma, struct list_head *unwind)if (vma->obj->pin_count)return false; -list_add (&vma->obj->exec_list, unwind); +list_add (&vma->exec_list, unwind);return drm_mm_scan_add_block (&vma->node); }@@ -49,7 +49,6 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,drm_i915_private_t *dev_priv = dev->dev_private;struct list_head eviction_list, unwind_list;struct i915_vma *vma; -struct drm_i915_gem_object *obj;int ret =0 ;trace_i915_gem_evict (dev, min_size, alignment, mappable);@@ -104,14 +103,13 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm, none:/* Nothing found, clean up and bail out! */ while (!list_empty (&unwind_list)) { - obj =list_first_entry (&unwind_list, - struct drm_i915_gem_object, + vma =list_first_entry (&unwind_list, + struct i915_vma, exec_list); - vma =i915_gem_obj_to_vma (obj, vm); ret =drm_mm_scan_remove_block (&vma->node);BUG_ON (ret); -list_del_init (&obj->exec_list); +list_del_init (&vma->exec_list); }/* We expect the caller to unpin, evict all and try again, or give up. @@ -125,28 +123,27 @@ found:* temporary list. */ INIT_LIST_HEAD (&eviction_list);while (!list_empty (&unwind_list)) { - obj =list_first_entry (&unwind_list, - struct drm_i915_gem_object, + vma =list_first_entry (&unwind_list, + struct i915_vma, exec_list); - vma =i915_gem_obj_to_vma (obj, vm);if (drm_mm_scan_remove_block (&vma->node)) { -list_move (&obj->exec_list, &eviction_list); -drm_gem_object_reference (&obj->base); +list_move (&vma->exec_list, &eviction_list); +drm_gem_object_reference (&vma->obj->base);continue ; } -list_del_init (&obj->exec_list); +list_del_init (&vma->exec_list); }/* Unbinding will emit any required flushes */ while (!list_empty (&eviction_list)) { - obj =list_first_entry (&eviction_list, - struct drm_i915_gem_object, + vma =list_first_entry (&eviction_list, + struct i915_vma, exec_list);if (ret ==0 ) - ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm)); + ret =i915_vma_unbind (vma); -list_del_init (&obj->exec_list); -drm_gem_object_unreference (&obj->base); +list_del_init (&vma->exec_list); +drm_gem_object_unreference (&vma->obj->base); }return ret;drivers/gpu/drm/i915/i915_gem_execbuffer.c +172
-150@@ -33,24 +33,24 @@#include "intel_drv.h" #include <linux/dma_remapping.h> -struct eb_objects { -struct list_head objects; +struct eb_vmas { +struct list_head vmas;int and;union { -struct drm_i915_gem_object *lut[0 ]; +struct i915_vma *lut[0 ];struct hlist_head buckets[0 ]; }; }; -static struct eb_objects * -eb_create (struct drm_i915_gem_execbuffer2 *args) +static struct eb_vmas * +eb_create (struct drm_i915_gem_execbuffer2 *args,struct i915_address_space *vm) { -struct eb_objects *eb =NULL ; +struct eb_vmas *eb =NULL ;if (args->flags &I915_EXEC_HANDLE_LUT ) {int size = args->buffer_count; - size *=sizeof (struct drm_i915_gem_object *); - size +=sizeof (struct eb_objects ); + size *=sizeof (struct i915_vma *); + size +=sizeof (struct eb_vmas ); eb =kmalloc (size,GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY); }@@ -61,7 +61,7 @@ eb_create(struct drm_i915_gem_execbuffer2 *args)while (count >2 *size) count >>=1 ; eb =kzalloc (count*sizeof (struct hlist_head ) + -sizeof (struct eb_objects ), +sizeof (struct eb_vmas ),GFP_TEMPORARY );if (eb ==NULL )return eb;@@ -70,72 +70,97 @@ eb_create(struct drm_i915_gem_execbuffer2 *args) }else eb->and = -args->buffer_count; -INIT_LIST_HEAD (&eb->objects); +INIT_LIST_HEAD (&eb->vmas);return eb; }static void -eb_reset (struct eb_objects *eb) +eb_reset (struct eb_vmas *eb) {if (eb->and >=0 )memset (eb->buckets,0 , (eb->and+1 )*sizeof (struct hlist_head )); }static int -eb_lookup_objects (struct eb_objects *eb, -struct drm_i915_gem_exec_object2 *exec, -const struct drm_i915_gem_execbuffer2 *args, -struct i915_address_space *vm, -struct drm_file *file) +eb_lookup_vmas (struct eb_vmas *eb, +struct drm_i915_gem_exec_object2 *exec, +const struct drm_i915_gem_execbuffer2 *args, +struct i915_address_space *vm, +struct drm_file *file) {struct drm_i915_gem_object *obj; -int i; +struct list_head objects; +int i, ret =0 ; +INIT_LIST_HEAD (&objects);spin_lock (&file->table_lock); +/* Grab a reference to the object and release the lock so we can lookup +* or create the VMA without using GFP_ATOMIC */ for (i =0 ; i < args->buffer_count; i++) { obj =to_intel_bo (idr_find (&file->object_idr, exec[i].handle));if (obj ==NULL ) {spin_unlock (&file->table_lock);DRM_DEBUG ("Invalid object handle %d at index %d\n" , exec[i].handle, i); -return -ENOENT ; + ret = -ENOENT ; + goto out; } -if (!list_empty (&obj->exec_list)) { +if (!list_empty (&obj->obj_exec_list)) {spin_unlock (&file->table_lock);DRM_DEBUG ("Object %p [handle %d, index %d] appears more than once in object list\n" , obj, exec[i].handle, i); -return -EINVAL ; + ret = -EINVAL ; + goto out; }drm_gem_object_reference (&obj->base); -list_add_tail (&obj->exec_list, &eb->objects); +list_add_tail (&obj->obj_exec_list, &objects); }spin_unlock (&file->table_lock); -list_for_each_entry (obj, &eb->objects, exec_list) { + i =0 ; +list_for_each_entry (obj, &objects, obj_exec_list) {struct i915_vma *vma; vma =i915_gem_obj_lookup_or_create_vma (obj, vm); -if (IS_ERR (vma)) -return PTR_ERR (vma); +if (IS_ERR (vma)) { +/* XXX: We don't need an error path fro vma because if +* the vma was created just for this execbuf, object +* unreference should kill it off.*/ +DRM_DEBUG ("Failed to lookup VMA\n" ); + ret =PTR_ERR (vma); + goto out; + } + +list_add_tail (&vma->exec_list, &eb->vmas); - obj->exec_entry = &exec[i]; + vma->exec_entry = &exec[i];if (eb->and <0 ) { - eb->lut[i] = obj; + eb->lut[i] = vma; }else {uint32_t handle = args->flags &I915_EXEC_HANDLE_LUT ? i : exec[i].handle; - obj->exec_handle = handle; -hlist_add_head (&obj->exec_node, + vma->exec_handle = handle; +hlist_add_head (&vma->exec_node, &eb->buckets[handle & eb->and]); } + ++i; } -return 0 ; + +out: +while (!list_empty (&objects)) { + obj =list_first_entry (&objects, + struct drm_i915_gem_object, + obj_exec_list); +list_del_init (&obj->obj_exec_list); +if (ret) +drm_gem_object_unreference (&obj->base); + } +return ret; } -static struct drm_i915_gem_object * -eb_get_object (struct eb_objects *eb,unsigned long handle) +static struct i915_vma *eb_get_vma (struct eb_vmas *eb,unsigned long handle) {if (eb->and <0 ) {if (handle >= -eb->and)@@ -147,27 +172,25 @@ eb_get_object(struct eb_objects *eb, unsigned long handle) head = &eb->buckets[handle & eb->and];hlist_for_each (node, head) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma; - obj = hlist_entry(node, struct drm_i915_gem_object, exec_node); -if (obj->exec_handle == handle) -return obj; + vma = hlist_entry(node, struct i915_vma, exec_node); +if (vma->exec_handle == handle) +return vma; }return NULL ; } } -static void -eb_destroy (struct eb_objects *eb) -{ -while (!list_empty (&eb->objects)) { -struct drm_i915_gem_object *obj; +static void eb_destroy (struct eb_vmas *eb) { +while (!list_empty (&eb->vmas)) { +struct i915_vma *vma; - obj =list_first_entry (&eb->objects, - struct drm_i915_gem_object, + vma =list_first_entry (&eb->vmas, + struct i915_vma, exec_list); -list_del_init (&obj->exec_list); -drm_gem_object_unreference (&obj->base); +list_del_init (&vma->exec_list); +drm_gem_object_unreference (&vma->obj->base); }kfree (eb); }@@ -181,22 +204,24 @@ static inline int use_cpu_reloc(struct drm_i915_gem_object *obj)static int i915_gem_execbuffer_relocate_entry (struct drm_i915_gem_object *obj, -struct eb_objects *eb, +struct eb_vmas *eb,struct drm_i915_gem_relocation_entry *reloc,struct i915_address_space *vm) {struct drm_device *dev = obj->base.dev;struct drm_gem_object *target_obj;struct drm_i915_gem_object *target_i915_obj; +struct i915_vma *target_vma;uint32_t target_offset;int ret = -EINVAL ;/* we've already hold a reference to all valid objects */ - target_obj = &eb_get_object (eb, reloc->target_handle)->base; -if (unlikely (target_obj ==NULL )) + target_vma =eb_get_vma (eb, reloc->target_handle); +if (unlikely (target_vma ==NULL ))return -ENOENT ; + target_i915_obj = target_vma->obj; + target_obj = &target_vma->obj->base; - target_i915_obj =to_intel_bo (target_obj); target_offset =i915_gem_obj_ggtt_offset (target_i915_obj);/* Sandybridge PPGTT errata: We need a global gtt mapping for MI and @@ -305,14 +330,13 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj, }static int -i915_gem_execbuffer_relocate_object (struct drm_i915_gem_object *obj, -struct eb_objects *eb, -struct i915_address_space *vm) +i915_gem_execbuffer_relocate_vma (struct i915_vma *vma, +struct eb_vmas *eb) {#define N_RELOC (x) ((x) / sizeof(struct drm_i915_gem_relocation_entry))struct drm_i915_gem_relocation_entry stack_reloc[N_RELOC (512 )];struct drm_i915_gem_relocation_entry __user *user_relocs; -struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; +struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;int remain, ret; user_relocs =to_user_ptr (entry->relocs_ptr);@@ -331,8 +355,8 @@ i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj, do { u64 offset = r->presumed_offset; - ret = i915_gem_execbuffer_relocate_entry(obj, eb, r, - vm); + ret = i915_gem_execbuffer_relocate_entry(vma->obj, eb, r, + vma->vm); if (ret) return ret;@@ -353,17 +377,16 @@ i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj, }static int -i915_gem_execbuffer_relocate_object_slow (struct drm_i915_gem_object *obj, -struct eb_objects *eb, -struct drm_i915_gem_relocation_entry *relocs, -struct i915_address_space *vm) +i915_gem_execbuffer_relocate_vma_slow (struct i915_vma *vma, +struct eb_vmas *eb, +struct drm_i915_gem_relocation_entry *relocs) { -const struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; +const struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;int i, ret;for (i =0 ; i < entry->relocation_count; i++) { - ret =i915_gem_execbuffer_relocate_entry (obj, eb, &relocs[i], - vm); + ret =i915_gem_execbuffer_relocate_entry (vma->obj, eb, &relocs[i], + vma->vm);if (ret)return ret; }@@ -372,10 +395,10 @@ i915_gem_execbuffer_relocate_object_slow(struct drm_i915_gem_object *obj, }static int -i915_gem_execbuffer_relocate (struct eb_objects *eb, +i915_gem_execbuffer_relocate (struct eb_vmas *eb,struct i915_address_space *vm) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;int ret =0 ;/* This is the fast path and we cannot handle a pagefault whilst @@ -386,8 +409,8 @@ i915_gem_execbuffer_relocate(struct eb_objects *eb,* lockdep complains vehemently. */ pagefault_disable (); -list_for_each_entry (obj, &eb->objects, exec_list) { - ret =i915_gem_execbuffer_relocate_object (obj, eb, vm); +list_for_each_entry (vma, &eb->vmas, exec_list) { + ret =i915_gem_execbuffer_relocate_vma (vma, eb);if (ret)break ; }@@ -400,31 +423,31 @@ i915_gem_execbuffer_relocate(struct eb_objects *eb,#define __EXEC_OBJECT_HAS_FENCE (1<<30)static int -need_reloc_mappable (struct drm_i915_gem_object *obj) +need_reloc_mappable (struct i915_vma *vma) { -struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; -return entry->relocation_count && !use_cpu_reloc (obj); +struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; +return entry->relocation_count && !use_cpu_reloc (vma->obj); }static int -i915_gem_execbuffer_reserve_object (struct drm_i915_gem_object *obj, -struct intel_ring_buffer *ring, -struct i915_address_space *vm, -bool *need_reloc) +i915_gem_execbuffer_reserve_vma (struct i915_vma *vma, +struct intel_ring_buffer *ring, +bool *need_reloc) { -struct drm_i915_private *dev_priv = obj->base.dev->dev_private; -struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; +struct drm_i915_private *dev_priv = ring->dev->dev_private; +struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;bool has_fenced_gpu_access =INTEL_INFO (ring->dev)->gen <4 ;bool need_fence, need_mappable; +struct drm_i915_gem_object *obj = vma->obj;int ret; need_fence = has_fenced_gpu_access && entry->flags &EXEC_OBJECT_NEEDS_FENCE && obj->tiling_mode !=I915_TILING_NONE ; - need_mappable = need_fence ||need_reloc_mappable (obj); + need_mappable = need_fence ||need_reloc_mappable (vma); - ret =i915_gem_object_pin (obj, vm, entry->alignment, need_mappable, + ret =i915_gem_object_pin (obj, vma->vm, entry->alignment, need_mappable, false);if (ret)return ret;@@ -452,8 +475,8 @@ i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, obj->has_aliasing_ppgtt_mapping =1 ; } -if (entry->offset !=i915_gem_obj_offset (obj, vm)) { - entry->offset =i915_gem_obj_offset (obj, vm); +if (entry->offset != vma->node.start) { + entry->offset = vma->node.start; *need_reloc = true; }@@ -470,61 +493,60 @@ i915_gem_execbuffer_reserve_object(struct drm_i915_gem_object *obj, }static void -i915_gem_execbuffer_unreserve_object (struct drm_i915_gem_object *obj) +i915_gem_execbuffer_unreserve_vma (struct i915_vma *vma) {struct drm_i915_gem_exec_object2 *entry; -if (!i915_gem_obj_bound_any (obj)) +if (!drm_mm_node_allocated (&vma->node))return ; - entry = obj->exec_entry; + entry = vma->exec_entry;if (entry->flags & __EXEC_OBJECT_HAS_FENCE) -i915_gem_object_unpin_fence (obj); +i915_gem_object_unpin_fence (vma->obj);if (entry->flags & __EXEC_OBJECT_HAS_PIN) -i915_gem_object_unpin (obj); +i915_gem_object_unpin (vma->obj); entry->flags &= ~(__EXEC_OBJECT_HAS_FENCE | __EXEC_OBJECT_HAS_PIN); }static int i915_gem_execbuffer_reserve (struct intel_ring_buffer *ring, -struct list_head *objects, -struct i915_address_space *vm, +struct list_head *vmas,bool *need_relocs) {struct drm_i915_gem_object *obj; -struct list_head ordered_objects; +struct i915_vma *vma; +struct list_head ordered_vmas;bool has_fenced_gpu_access =INTEL_INFO (ring->dev)->gen <4 ;int retry; -INIT_LIST_HEAD (&ordered_objects); -while (!list_empty (objects)) { +INIT_LIST_HEAD (&ordered_vmas); +while (!list_empty (vmas)) {struct drm_i915_gem_exec_object2 *entry;bool need_fence, need_mappable; - obj =list_first_entry (objects, - struct drm_i915_gem_object, - exec_list); - entry = obj->exec_entry; + vma = list_first_entry(vmas, struct i915_vma, exec_list); + obj = vma->obj; + entry = vma->exec_entry; need_fence = has_fenced_gpu_access && entry->flags &EXEC_OBJECT_NEEDS_FENCE && obj->tiling_mode !=I915_TILING_NONE ; - need_mappable = need_fence ||need_reloc_mappable (obj); + need_mappable = need_fence ||need_reloc_mappable (vma);if (need_mappable) -list_move (&obj->exec_list, &ordered_objects); +list_move (&vma->exec_list, &ordered_vmas);else -list_move_tail (&obj->exec_list, &ordered_objects); +list_move_tail (&vma->exec_list, &ordered_vmas); obj->base.pending_read_domains =I915_GEM_GPU_DOMAINS & ~I915_GEM_DOMAIN_COMMAND ; obj->base.pending_write_domain =0 ; obj->pending_fenced_gpu_access = false; } -list_splice (&ordered_objects, objects); +list_splice (&ordered_vmas, vmas);/* Attempt to pin all of the buffers into the GTT. * This is done in 3 phases: @@ -543,47 +565,47 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,int ret =0 ;/* Unbind any ill-fitting objects or pin. */ -list_for_each_entry (obj, objects, exec_list) { -struct drm_i915_gem_exec_object2 *entry = obj->exec_entry; +list_for_each_entry (vma, vmas, exec_list) { +struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;bool need_fence, need_mappable; -u32 obj_offset; -if (!i915_gem_obj_bound (obj, vm)) + obj = vma->obj; + +if (!drm_mm_node_allocated (&vma->node))continue ; - obj_offset =i915_gem_obj_offset (obj, vm); need_fence = has_fenced_gpu_access && entry->flags &EXEC_OBJECT_NEEDS_FENCE && obj->tiling_mode !=I915_TILING_NONE ; - need_mappable = need_fence ||need_reloc_mappable (obj); + need_mappable = need_fence ||need_reloc_mappable (vma);BUG_ON ((need_mappable || need_fence) && - !i915_is_ggtt (vm)); + !i915_is_ggtt (vma->vm));if ((entry->alignment && - obj_offset & (entry->alignment -1 )) || + vma->node.start & (entry->alignment -1 )) || (need_mappable && !obj->map_and_fenceable)) - ret =i915_vma_unbind (i915_gem_obj_to_vma (obj, vm)); + ret =i915_vma_unbind (vma);else - ret =i915_gem_execbuffer_reserve_object (obj, ring, vm, need_relocs); + ret =i915_gem_execbuffer_reserve_vma (vma, ring, need_relocs);if (ret) goto err; }/* Bind fresh objects */ -list_for_each_entry (obj, objects, exec_list) { -if (i915_gem_obj_bound (obj, vm)) +list_for_each_entry (vma, vmas, exec_list) { +if (drm_mm_node_allocated (&vma->node))continue ; - ret =i915_gem_execbuffer_reserve_object (obj, ring, vm, need_relocs); + ret =i915_gem_execbuffer_reserve_vma (vma, ring, need_relocs);if (ret) goto err; } err:/* Decrement pin count for bound objects */ -list_for_each_entry (obj, objects, exec_list) -i915_gem_execbuffer_unreserve_object (obj); +list_for_each_entry (vma, vmas, exec_list) +i915_gem_execbuffer_unreserve_vma (vma);if (ret != -ENOSPC || retry++)return ret;@@ -599,24 +621,27 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,struct drm_i915_gem_execbuffer2 *args, struct drm_file *file, struct intel_ring_buffer *ring, - struct eb_objects *eb, - struct drm_i915_gem_exec_object2 *exec, - struct i915_address_space *vm) + struct eb_vmas *eb, + struct drm_i915_gem_exec_object2 *exec) { struct drm_i915_gem_relocation_entry *reloc; -struct drm_i915_gem_object *obj; +struct i915_address_space *vm; +struct i915_vma *vma;bool need_relocs;int *reloc_offset;int i, total, ret;int count = args->buffer_count; +if (WARN_ON (list_empty (&eb->vmas))) +return 0 ; + + vm = list_first_entry(&eb->vmas, struct i915_vma, exec_list)->vm; +/* We may process another execbuffer during the unlock... */ -while (!list_empty (&eb->objects)) { - obj =list_first_entry (&eb->objects, - struct drm_i915_gem_object, - exec_list); -list_del_init (&obj->exec_list); -drm_gem_object_unreference (&obj->base); +while (!list_empty (&eb->vmas)) { + vma = list_first_entry(&eb->vmas, struct i915_vma, exec_list); +list_del_init (&vma->exec_list); +drm_gem_object_unreference (&vma->obj->base); }mutex_unlock (&dev->struct_mutex);@@ -680,20 +705,19 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,/* reacquire the objects */ eb_reset (eb); - ret =eb_lookup_objects (eb, exec, args, vm, file); + ret =eb_lookup_vmas (eb, exec, args, vm, file);if (ret) goto err; need_relocs = (args->flags &I915_EXEC_NO_RELOC ) ==0 ; - ret =i915_gem_execbuffer_reserve (ring, &eb->objects, vm, &need_relocs); + ret =i915_gem_execbuffer_reserve (ring, &eb->vmas, &need_relocs);if (ret) goto err; -list_for_each_entry (obj, &eb->objects, exec_list) { -int offset = obj->exec_entry - exec; - ret =i915_gem_execbuffer_relocate_object_slow (obj, eb, - reloc + reloc_offset[offset], - vm); +list_for_each_entry (vma, &eb->vmas, exec_list) { +int offset = vma->exec_entry - exec; + ret =i915_gem_execbuffer_relocate_vma_slow (vma, eb, + reloc + reloc_offset[offset]);if (ret) goto err; }@@ -712,21 +736,21 @@ err:static int i915_gem_execbuffer_move_to_gpu (struct intel_ring_buffer *ring, -struct list_head *objects) +struct list_head *vmas) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma;uint32_t flush_domains =0 ;int ret; -list_for_each_entry (obj, objects, exec_list) { - ret =i915_gem_object_sync (obj, ring); +list_for_each_entry (vma, vmas, exec_list) { + ret =i915_gem_object_sync (vma->obj, ring);if (ret)return ret; -if (obj->base.write_domain &I915_GEM_DOMAIN_CPU ) -i915_gem_clflush_object (obj); +if (vma->obj->base.write_domain &I915_GEM_DOMAIN_CPU ) +i915_gem_clflush_object (vma->obj); - flush_domains |= obj->base.write_domain; + flush_domains |= vma->obj->base.write_domain; }if (flush_domains &I915_GEM_DOMAIN_CPU )@@ -793,13 +817,13 @@ validate_exec_list(struct drm_i915_gem_exec_object2 *exec, }static void -i915_gem_execbuffer_move_to_active (struct list_head *objects, -struct i915_address_space *vm, +i915_gem_execbuffer_move_to_active (struct list_head *vmas,struct intel_ring_buffer *ring) { -struct drm_i915_gem_object *obj; +struct i915_vma *vma; -list_for_each_entry (obj, objects, exec_list) { +list_for_each_entry (vma, vmas, exec_list) { +struct drm_i915_gem_object *obj = vma->obj;u32 old_read = obj->base.read_domains;u32 old_write = obj->base.write_domain;@@ -810,7 +834,7 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects, obj->fenced_gpu_access = obj->pending_fenced_gpu_access;/* FIXME: This lookup gets fixed later <-- danvet */ -list_move_tail (&i915_gem_obj_to_vma (obj, vm)->mm_list, &vm->active_list); +list_move_tail (&vma->mm_list, &vma->vm->active_list);i915_gem_object_move_to_active (obj, ring);if (obj->base.write_domain) { obj->dirty =1 ;@@ -869,7 +893,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, struct i915_address_space *vm) {drm_i915_private_t *dev_priv = dev->dev_private; -struct eb_objects *eb; +struct eb_vmas *eb;struct drm_i915_gem_object *batch_obj;struct drm_clip_rect *cliprects =NULL ;struct intel_ring_buffer *ring;@@ -1009,7 +1033,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, goto pre_mutex_err; } - eb =eb_create (args); + eb =eb_create (args, vm);if (eb ==NULL ) {mutex_unlock (&dev->struct_mutex); ret = -ENOMEM ;@@ -1017,18 +1041,16 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, }/* Look up object handles */ - ret =eb_lookup_objects (eb, exec, args, vm, file); + ret =eb_lookup_vmas (eb, exec, args, vm, file);if (ret) goto err;/* take note of the batch buffer before we might reorder the lists */ - batch_obj =list_entry (eb->objects.prev, - struct drm_i915_gem_object, - exec_list); + batch_obj = list_entry(eb->vmas.prev, struct i915_vma, exec_list)->obj;/* Move the objects en-masse into the GTT, evicting if necessary. */ need_relocs = (args->flags &I915_EXEC_NO_RELOC ) ==0 ; - ret =i915_gem_execbuffer_reserve (ring, &eb->objects, vm, &need_relocs); + ret =i915_gem_execbuffer_reserve (ring, &eb->vmas, &need_relocs);if (ret) goto err;@@ -1038,7 +1060,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,if (ret) {if (ret == -EFAULT ) { ret =i915_gem_execbuffer_relocate_slow (dev, args, file, ring, - eb, exec, vm); + eb, exec);BUG_ON (!mutex_is_locked (&dev->struct_mutex)); }if (ret)@@ -1060,7 +1082,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,if (flags &I915_DISPATCH_SECURE && !batch_obj->has_global_gtt_mapping)i915_gem_gtt_bind_object (batch_obj, batch_obj->cache_level); - ret =i915_gem_execbuffer_move_to_gpu (ring, &eb->objects); + ret =i915_gem_execbuffer_move_to_gpu (ring, &eb->vmas);if (ret) goto err;@@ -1115,7 +1137,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,trace_i915_gem_ring_dispatch (ring,intel_ring_get_seqno (ring), flags); -i915_gem_execbuffer_move_to_active (&eb->objects, vm, ring); +i915_gem_execbuffer_move_to_active (&eb->vmas, ring);i915_gem_execbuffer_retire_commands (dev, file, ring, batch_obj); err:Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH 25/29] drm/i915: Convert execbuf code to use vmas
Quoted 31 lines · Daniel Vetter
Quoted 7 lines · Ben Widawsky
This attempts to convert all the execbuf code to speak in vmas. Since the execbuf code is very self contained it was a nice isolated conversion.
The meat of the code is about turning
eb_objectsintoeb_vma, and then wiring up the rest of the code to use vmas instead of obj, vm pairs.Quoted 8 lines
Unfortunately, to do this, we must move the
exec_listlink from the obj structure. This list is reused in the eviction code, so we must also modify the eviction code to make this work.v2: Release table lock early, and two a 2 phase vma lookup to avoid having to use a
GFP_ATOMIC. (Chris)Signed-off-by Ben Widawsky <[email protected]>I think the leaking of preallocated vmas if execbuf fails can blow up:
1. We calllookup_or_createand create new vmas, linked into thevma_linkchain.
2. Later on execbuf fails somewhere (for an igt the simplest way is probaly to use more buffers than what would fit into the gtt) and we bail out.
-> Note that at this point we leak vmas which are on thevma_linklist but which have no gtt node allocation.
3. Userspace dies in flames (or just quits).
4. All buffers get their final unref and we callvma_unbindon each vma, even the ones that do not have an allocation.
5. hilarity ensues sincevma_unbinddoesn't bail out ifdrm_mm_node_allocated(vma->node)== false.
Quoted 3 lines
We need broken userspace to actually exercise this bug since all normal ways for execbuf to bail out involve singals and ioclt restarting. If this is a real bug I think we need an igt to exercise it.
Forgotten to mention: Beside that error path the patch looks good. Same applies to patches 21-24 before.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 26/29 drm/i915: Convert active API to VMA
In reply to Ben WidawskyEven though we track object activeness and not VMA, because we have the active_list be based on the VM, it makes the most sense to use VMAs in the APIs.
NOTE: Daniel intends to eventually rip out active/inactive LRUs, but for now, leave them be.
--- drivers/gpu/drm/i915/i915_drv.h | 5 ++--- drivers/gpu/drm/i915/i915_gem.c | 11 +++++++++-- drivers/gpu/drm/i915/i915_gem_context.c | 8 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 4 +--- 4 files changed, 16 insertions(+), 12 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +2 -3
@@ -1735,9 +1735,8 @@ static inline void i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj)int __must_check i915_mutex_lock_interruptible (struct drm_device *dev);int i915_gem_object_sync (struct drm_i915_gem_object *obj,struct intel_ring_buffer *to); -void i915_gem_object_move_to_active (struct drm_i915_gem_object *obj, -struct intel_ring_buffer *ring); - +void i915_vma_move_to_active (struct i915_vma *vma, +struct intel_ring_buffer *ring);int i915_gem_dumb_create (struct drm_file *file_priv,struct drm_device *dev,struct drm_mode_create_dumb *args);
drivers/gpu/drm/i915/i915_gem.c +9 -2
@@ -1866,11 +1866,11 @@ i915_gem_object_get_pages(struct drm_i915_gem_object *obj)return 0 ; } -void +static void i915_gem_object_move_to_active (struct drm_i915_gem_object *obj,struct intel_ring_buffer *ring) { -struct drm_device *dev = obj->base.dev; +struct drm_device *dev = ring->dev;struct drm_i915_private *dev_priv = dev->dev_private;u32 seqno =intel_ring_get_seqno (ring);
@@ -1905,6 +1905,13 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, } } +void i915_vma_move_to_active (struct i915_vma *vma, +struct intel_ring_buffer *ring) +{ +list_move_tail (&vma->mm_list, &vma->vm->active_list); +return i915_gem_object_move_to_active (vma->obj, ring); +} +static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) {
drivers/gpu/drm/i915/i915_gem_context.c +4 -4
@@ -436,11 +436,11 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) { -struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; -struct i915_address_space *ggtt = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = ring->dev->dev_private; +struct i915_vma *vma = +i915_gem_obj_to_vma (from->obj, &dev_priv->gtt.base); from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; -list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list); -i915_gem_object_move_to_active (from->obj, ring); +i915_vma_move_to_active (vma, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the * object dirty. The only exception is that the context must be
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -3
@@ -833,9 +833,7 @@ i915_gem_execbuffer_move_to_active(struct list_head *vmas, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; -/* FIXME: This lookup gets fixed later <-- danvet */ -list_move_tail (&vma->mm_list, &vma->vm->active_list); -i915_gem_object_move_to_active (obj, ring); +i915_vma_move_to_active (vma, ring);if (obj->base.write_domain) { obj->dirty =1 ; obj->last_write_seqno =intel_ring_get_seqno (ring);
Signature · 1 line
1.8.3.4
Re: [PATCH 26/29] drm/i915: Convert active API to VMA
Quoted 8 lines · Ben Widawsky
Even though we track object activeness and not VMA, because we have the
active_listbe based on the VM, it makes the most sense to use VMAs in the APIs.NOTE: Daniel intends to eventually rip out active/inactive LRUs, but for now, leave them be.
Signed-off-by Ben Widawsky <[email protected]>
Ah, here's the patch that addresses two of my earlier questions. I guess the split was to due the execbuf conversion that needed to happen first. Looks good.
-Daniel
Quoted 97 lines
--- drivers/gpu/drm/i915/i915_drv.h | 5 ++--- drivers/gpu/drm/i915/i915_gem.c | 11 +++++++++-- drivers/gpu/drm/i915/i915_gem_context.c | 8 ++++---- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 4 +--- 4 files changed, 16 insertions(+), 12 deletions(-)drivers/gpu/drm/i915/i915_drv.h +2
-3@@ -1735,9 +1735,8 @@ static inline void i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj)int __must_check i915_mutex_lock_interruptible (struct drm_device *dev);int i915_gem_object_sync (struct drm_i915_gem_object *obj,struct intel_ring_buffer *to); -void i915_gem_object_move_to_active (struct drm_i915_gem_object *obj, -struct intel_ring_buffer *ring); - +void i915_vma_move_to_active (struct i915_vma *vma, +struct intel_ring_buffer *ring);int i915_gem_dumb_create (struct drm_file *file_priv,struct drm_device *dev,struct drm_mode_create_dumb *args);drivers/gpu/drm/i915/i915_gem.c +9
-2@@ -1866,11 +1866,11 @@ i915_gem_object_get_pages(struct drm_i915_gem_object *obj)return 0 ; } -void +static void i915_gem_object_move_to_active (struct drm_i915_gem_object *obj,struct intel_ring_buffer *ring) { -struct drm_device *dev = obj->base.dev; +struct drm_device *dev = ring->dev;struct drm_i915_private *dev_priv = dev->dev_private;u32 seqno =intel_ring_get_seqno (ring);@@ -1905,6 +1905,13 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, } } +void i915_vma_move_to_active (struct i915_vma *vma, +struct intel_ring_buffer *ring) +{ +list_move_tail (&vma->mm_list, &vma->vm->active_list); +return i915_gem_object_move_to_active (vma->obj, ring); +} +static void i915_gem_object_move_to_inactive (struct drm_i915_gem_object *obj) {drivers/gpu/drm/i915/i915_gem_context.c +4
-4@@ -436,11 +436,11 @@ static int do_switch(struct i915_hw_context *to)* MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from !=NULL ) { -struct drm_i915_private *dev_priv = from->obj->base.dev->dev_private; -struct i915_address_space *ggtt = &dev_priv->gtt.base; +struct drm_i915_private *dev_priv = ring->dev->dev_private; +struct i915_vma *vma = +i915_gem_obj_to_vma (from->obj, &dev_priv->gtt.base); from->obj->base.read_domains =I915_GEM_DOMAIN_INSTRUCTION ; -list_move_tail (&i915_gem_obj_to_vma (from->obj, ggtt)->mm_list, &ggtt->active_list); -i915_gem_object_move_to_active (from->obj, ring); +i915_vma_move_to_active (vma, ring);/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the * object dirty. The only exception is that the context must be drivers/gpu/drm/i915/i915_gem_execbuffer.c +1
-3@@ -833,9 +833,7 @@ i915_gem_execbuffer_move_to_active(struct list_head *vmas, obj->base.read_domains = obj->base.pending_read_domains; obj->fenced_gpu_access = obj->pending_fenced_gpu_access; -/* FIXME: This lookup gets fixed later <-- danvet */ -list_move_tail (&vma->mm_list, &vma->vm->active_list); -i915_gem_object_move_to_active (obj, ring); +i915_vma_move_to_active (vma, ring);if (obj->base.write_domain) { obj->dirty =1 ; obj->last_write_seqno =intel_ring_get_seqno (ring);Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 27/29 drm/i915: Add bind/unbind object functions to VM
In reply to Ben WidawskyAs we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT.
Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge.
v2:
Make ggtt handle the quirky aliasing ppgtt
Add flags to bind object to support above
Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this)
Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels
Use VMA for bind/unbind (Daniel, Ben)
--- drivers/gpu/drm/i915/i915_drv.h | 69 +++++++++++++----------- drivers/gpu/drm/i915/i915_gem_gtt.c | 101 ++++++++++++++++++++++++++++++++++++ 2 files changed, 140 insertions(+), 30 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +39 -30
@@ -446,6 +446,36 @@ enum i915_cache_level {typedef uint32_t gen6_gtt_pte_t ; +/** +* A VMA represents a GEM BO that is bound into an address space. Therefore, a +* VMA's presence cannot be guaranteed before binding, or after unbinding the +* object into/from the address space. +* +* To make things as simple as possible (ie. no refcounting), a VMA's lifetime +* will always be <= an objects lifetime. So object refcounting should cover us. +*/ +struct i915_vma { +struct drm_mm_node node; +struct drm_i915_gem_object *obj; +struct i915_address_space *vm; + +/** This object's place on the active/inactive lists */ +struct list_head mm_list; + +struct list_head vma_link;/* Link in the object's VMA list */ + +/** This vma's place in the batchbuffer or on the eviction list */ +struct list_head exec_list; + +/** +* Used for performing relocations during execbuffer insertion. +*/ +struct hlist_node exec_node; +unsigned long exec_handle; +struct drm_i915_gem_exec_object2 *exec_entry; + +}; +struct i915_address_space {struct drm_mm mm;struct drm_device *dev;
@@ -484,9 +514,18 @@ struct i915_address_space {/* FIXME: Need a more generic return type */ gen6_gtt_pte_t (*pte_encode)(dma_addr_t addr,enum i915_cache_level level); + +/** Unmap an object from an address space. This usually consists of +* setting the valid PTE entries to a reserved scratch page. */ +void (*unbind_vma)(struct i915_vma *vma);void (*clear_range)(struct i915_address_space *vm,unsigned int first_entry,unsigned int num_entries); +/* Map an object into an address space with the given cache flags. */ +#define GLOBAL_BIND (1<<0) +void (*bind_vma)(struct i915_vma *vma, +enum i915_cache_level cache_level, +u32 flags); void (*insert_entries)(struct i915_address_space *vm, struct sg_table *st, unsigned int first_entry,
@@ -533,36 +572,6 @@ struct i915_hw_ppgtt {int (*enable)(struct drm_device *dev); }; -/** -* A VMA represents a GEM BO that is bound into an address space. Therefore, a -* VMA's presence cannot be guaranteed before binding, or after unbinding the -* object into/from the address space. -* -* To make things as simple as possible (ie. no refcounting), a VMA's lifetime -* will always be <= an objects lifetime. So object refcounting should cover us. -*/ -struct i915_vma { -struct drm_mm_node node; -struct drm_i915_gem_object *obj; -struct i915_address_space *vm; - -/** This object's place on the active/inactive lists */ -struct list_head mm_list; - -struct list_head vma_link;/* Link in the object's VMA list */ - -/** This vma's place in the batchbuffer or on the eviction list */ -struct list_head exec_list; - -/** -* Used for performing relocations during execbuffer insertion. -*/ -struct hlist_node exec_node; -unsigned long exec_handle; -struct drm_i915_gem_exec_object2 *exec_entry; - -}; -struct i915_ctx_hang_stats {/* This context had batch pending when hang was declared */ unsigned batch_pending;
drivers/gpu/drm/i915/i915_gem_gtt.c +101 -0
158 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 28/29 drm/i915: Use the new vm [un]bind functions
In reply to Ben WidawskyBuilding on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use.
Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it.
v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot.
--- drivers/gpu/drm/i915/i915_drv.h | 9 ------ drivers/gpu/drm/i915/i915_gem.c | 31 ++++++++----------- drivers/gpu/drm/i915/i915_gem_context.c | 8 +++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 29 ++++++++---------- drivers/gpu/drm/i915/i915_gem_gtt.c | 48 ++---------------------------- 5 files changed, 34 insertions(+), 91 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +0 -9
@@ -1938,17 +1938,8 @@ int i915_gem_context_destroy_ioctl(struct drm_device *dev, void *data,/* i915_gem_gtt.c */ void i915_gem_cleanup_aliasing_ppgtt (struct drm_device *dev); -void i915_ppgtt_bind_object (struct i915_hw_ppgtt *ppgtt, -struct drm_i915_gem_object *obj, -enum i915_cache_level cache_level); -void i915_ppgtt_unbind_object (struct i915_hw_ppgtt *ppgtt, -struct drm_i915_gem_object *obj); -void i915_gem_restore_gtt_mappings (struct drm_device *dev);int __must_check i915_gem_gtt_prepare_object (struct drm_i915_gem_object *obj); -void i915_gem_gtt_bind_object (struct drm_i915_gem_object *obj, -enum i915_cache_level cache_level); -void i915_gem_gtt_unbind_object (struct drm_i915_gem_object *obj);void i915_gem_gtt_finish_object (struct drm_i915_gem_object *obj);void i915_gem_init_global_gtt (struct drm_device *dev); void i915_gem_setup_global_gtt(struct drm_device *dev, unsigned long start,
drivers/gpu/drm/i915/i915_gem.c +13 -18
@@ -2631,12 +2631,8 @@ int i915_vma_unbind(struct i915_vma *vma)trace_i915_vma_unbind (vma); -if (obj->has_global_gtt_mapping &&i915_is_ggtt (vma->vm)) -i915_gem_gtt_unbind_object (obj); -if (obj->has_aliasing_ppgtt_mapping) { -i915_ppgtt_unbind_object (dev_priv->mm.aliasing_ppgtt, obj); - obj->has_aliasing_ppgtt_mapping =0 ; - } + vma->vm->unbind_vma (vma); +i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj);
@@ -3354,7 +3350,6 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, enum i915_cache_level cache_level) {struct drm_device *dev = obj->base.dev; -drm_i915_private_t *dev_priv = dev->dev_private;struct i915_vma *vma;int ret;
@@ -3393,11 +3388,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return ret; } -if (obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (obj, cache_level); -if (obj->has_aliasing_ppgtt_mapping) -i915_ppgtt_bind_object (dev_priv->mm.aliasing_ppgtt, - obj, cache_level); +list_for_each_entry (vma, &obj->vma_list, vma_link) + vma->vm->bind_vma (vma, cache_level,0 ); }if (cache_level ==I915_CACHE_NONE ) {
@@ -3676,6 +3668,7 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj, bool map_and_fenceable, bool nonblocking) { +const u32 flags = map_and_fenceable ?GLOBAL_BIND :0 ;struct i915_vma *vma;int ret;
@@ -3704,20 +3697,22 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj, }if (!i915_gem_obj_bound (obj, vm)) { -struct drm_i915_private *dev_priv = obj->base.dev->dev_private; - ret =i915_gem_object_bind_to_vm (obj, vm, alignment, map_and_fenceable, nonblocking);if (ret)return ret; -if (!dev_priv->mm.aliasing_ppgtt) -i915_gem_gtt_bind_object (obj, obj->cache_level); - } + vma =i915_gem_obj_to_vma (obj, vm); + vm->bind_vma (vma, obj->cache_level, flags); + }else + vma =i915_gem_obj_to_vma (obj, vm); +/* Objects are created map and fenceable. If we bind an object +* the first time, and we had aliasing PPGTT (and didn't request +* GLOBAL), we'll need to do this on the second bind.*/ if (!obj->has_global_gtt_mapping && map_and_fenceable) -i915_gem_gtt_bind_object (obj, obj->cache_level); + vm->bind_vma (vma, obj->cache_level,GLOBAL_BIND ); obj->pin_count++; obj->pin_mappable |= map_and_fenceable;
drivers/gpu/drm/i915/i915_gem_context.c +6 -2
@@ -391,6 +391,7 @@ mi_set_context(struct intel_ring_buffer *ring,static int do_switch (struct i915_hw_context *to) {struct intel_ring_buffer *ring = to->ring; +struct drm_i915_private *dev_priv = ring->dev->dev_private;struct i915_hw_context *from = ring->last_context;u32 hw_flags =0 ;int ret;
@@ -415,8 +416,11 @@ static int do_switch(struct i915_hw_context *to)return ret; } -if (!to->obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (to->obj, to->obj->cache_level); +if (!to->obj->has_global_gtt_mapping) { +struct i915_vma *vma =i915_gem_obj_to_vma (to->obj, + &dev_priv->gtt.base); + vma->vm->bind_vma (vma, to->obj->cache_level,GLOBAL_BIND ); + }if (!to->is_initialized ||is_default_context (to)) hw_flags |=MI_RESTORE_INHIBIT ;
drivers/gpu/drm/i915/i915_gem_execbuffer.c +12 -17
@@ -230,8 +230,9 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,if (unlikely (IS_GEN6 (dev) && reloc->write_domain ==I915_GEM_DOMAIN_INSTRUCTION && !target_i915_obj->has_global_gtt_mapping)) { -i915_gem_gtt_bind_object (target_i915_obj, - target_i915_obj->cache_level); +struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); + vma->vm->bind_vma (vma, target_i915_obj->cache_level, +GLOBAL_BIND ); }/* Validate that the target is in a valid r/w GPU domain */
@@ -434,11 +435,12 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma,struct intel_ring_buffer *ring, bool *need_reloc) { - struct drm_i915_private *dev_priv = ring->dev->dev_private; + struct drm_i915_gem_object *obj = vma->obj;struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;bool has_fenced_gpu_access =INTEL_INFO (ring->dev)->gen <4 ;bool need_fence, need_mappable; -struct drm_i915_gem_object *obj = vma->obj; +u32 flags = (entry->flags &EXEC_OBJECT_NEEDS_GTT ) && + !vma->obj->has_global_gtt_mapping ?GLOBAL_BIND :0 ;int ret; need_fence =
@@ -467,14 +469,6 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, } } -/* Ensure ppgtt mapping exists if needed */ -if (dev_priv->mm.aliasing_ppgtt && !obj->has_aliasing_ppgtt_mapping) { -i915_ppgtt_bind_object (dev_priv->mm.aliasing_ppgtt, - obj, obj->cache_level); - - obj->has_aliasing_ppgtt_mapping =1 ; - } -if (entry->offset != vma->node.start) { entry->offset = vma->node.start; *need_reloc = true;
@@ -485,9 +479,7 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, obj->base.pending_write_domain =I915_GEM_DOMAIN_RENDER ; } -if (entry->flags &EXEC_OBJECT_NEEDS_GTT && - !obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (obj, obj->cache_level); + vma->vm->bind_vma (vma, obj->cache_level, flags);return 0 ; }
@@ -1077,8 +1069,11 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,* batch" bit. Hence we need to pin secure batches into the global gtt. * hsw should have this fixed, but let's be paranoid and do it * unconditionally for now. */ -if (flags &I915_DISPATCH_SECURE && !batch_obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (batch_obj, batch_obj->cache_level); +if (flags &I915_DISPATCH_SECURE && + !batch_obj->has_global_gtt_mapping) { +struct i915_vma *vma =i915_gem_obj_to_vma (batch_obj, vm); + vm->bind_vma (vma, batch_obj->cache_level,GLOBAL_BIND ); + } ret =i915_gem_execbuffer_move_to_gpu (ring, &eb->vmas);if (ret)
drivers/gpu/drm/i915/i915_gem_gtt.c +3 -45
73 lines not on this page · raw message
Signature · 1 line
1.8.3.4
Re: [PATCH 28/29] drm/i915: Use the new vm [un]bind functions
Quoted 4 lines · Ben Widawsky
Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use.
Quoted 8 lines
Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it.
v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot.
Signed-off-by Ben Widawsky <[email protected]>
I don't like how this is split since it's a (small but still) flip the world approach: First you create completely new code, then you rip out the old one and switch over. So this should definitely be squashed for easier review, and if its to big the split up into different refactoring steps (where each steps keeps all the code working while slowly transforming it). I'll punt for now on this here.
-Daniel
Quoted 293 lines
--- drivers/gpu/drm/i915/i915_drv.h | 9 ------ drivers/gpu/drm/i915/i915_gem.c | 31 ++++++++----------- drivers/gpu/drm/i915/i915_gem_context.c | 8 +++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 29 ++++++++---------- drivers/gpu/drm/i915/i915_gem_gtt.c | 48 ++---------------------------- 5 files changed, 34 insertions(+), 91 deletions(-)drivers/gpu/drm/i915/i915_drv.h +0
-9@@ -1938,17 +1938,8 @@ int i915_gem_context_destroy_ioctl(struct drm_device *dev, void *data,/* i915_gem_gtt.c */ void i915_gem_cleanup_aliasing_ppgtt (struct drm_device *dev); -void i915_ppgtt_bind_object (struct i915_hw_ppgtt *ppgtt, -struct drm_i915_gem_object *obj, -enum i915_cache_level cache_level); -void i915_ppgtt_unbind_object (struct i915_hw_ppgtt *ppgtt, -struct drm_i915_gem_object *obj); -void i915_gem_restore_gtt_mappings (struct drm_device *dev);int __must_check i915_gem_gtt_prepare_object (struct drm_i915_gem_object *obj); -void i915_gem_gtt_bind_object (struct drm_i915_gem_object *obj, -enum i915_cache_level cache_level); -void i915_gem_gtt_unbind_object (struct drm_i915_gem_object *obj);void i915_gem_gtt_finish_object (struct drm_i915_gem_object *obj);void i915_gem_init_global_gtt (struct drm_device *dev); void i915_gem_setup_global_gtt(struct drm_device *dev, unsigned long start,drivers/gpu/drm/i915/i915_gem.c +13
-18@@ -2631,12 +2631,8 @@ int i915_vma_unbind(struct i915_vma *vma)trace_i915_vma_unbind (vma); -if (obj->has_global_gtt_mapping &&i915_is_ggtt (vma->vm)) -i915_gem_gtt_unbind_object (obj); -if (obj->has_aliasing_ppgtt_mapping) { -i915_ppgtt_unbind_object (dev_priv->mm.aliasing_ppgtt, obj); - obj->has_aliasing_ppgtt_mapping =0 ; - } + vma->vm->unbind_vma (vma); +i915_gem_gtt_finish_object (obj);i915_gem_object_unpin_pages (obj);@@ -3354,7 +3350,6 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, enum i915_cache_level cache_level) {struct drm_device *dev = obj->base.dev; -drm_i915_private_t *dev_priv = dev->dev_private;struct i915_vma *vma;int ret;@@ -3393,11 +3388,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,return ret; } -if (obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (obj, cache_level); -if (obj->has_aliasing_ppgtt_mapping) -i915_ppgtt_bind_object (dev_priv->mm.aliasing_ppgtt, - obj, cache_level); +list_for_each_entry (vma, &obj->vma_list, vma_link) + vma->vm->bind_vma (vma, cache_level,0 ); }if (cache_level ==I915_CACHE_NONE ) {@@ -3676,6 +3668,7 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj, bool map_and_fenceable, bool nonblocking) { +const u32 flags = map_and_fenceable ?GLOBAL_BIND :0 ;struct i915_vma *vma;int ret;@@ -3704,20 +3697,22 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj, }if (!i915_gem_obj_bound (obj, vm)) { -struct drm_i915_private *dev_priv = obj->base.dev->dev_private; - ret =i915_gem_object_bind_to_vm (obj, vm, alignment, map_and_fenceable, nonblocking);if (ret)return ret; -if (!dev_priv->mm.aliasing_ppgtt) -i915_gem_gtt_bind_object (obj, obj->cache_level); - } + vma =i915_gem_obj_to_vma (obj, vm); + vm->bind_vma (vma, obj->cache_level, flags); + }else + vma =i915_gem_obj_to_vma (obj, vm); +/* Objects are created map and fenceable. If we bind an object +* the first time, and we had aliasing PPGTT (and didn't request +* GLOBAL), we'll need to do this on the second bind.*/ if (!obj->has_global_gtt_mapping && map_and_fenceable) -i915_gem_gtt_bind_object (obj, obj->cache_level); + vm->bind_vma (vma, obj->cache_level,GLOBAL_BIND ); obj->pin_count++; obj->pin_mappable |= map_and_fenceable;drivers/gpu/drm/i915/i915_gem_context.c +6
-2@@ -391,6 +391,7 @@ mi_set_context(struct intel_ring_buffer *ring,static int do_switch (struct i915_hw_context *to) {struct intel_ring_buffer *ring = to->ring; +struct drm_i915_private *dev_priv = ring->dev->dev_private;struct i915_hw_context *from = ring->last_context;u32 hw_flags =0 ;int ret;@@ -415,8 +416,11 @@ static int do_switch(struct i915_hw_context *to)return ret; } -if (!to->obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (to->obj, to->obj->cache_level); +if (!to->obj->has_global_gtt_mapping) { +struct i915_vma *vma =i915_gem_obj_to_vma (to->obj, + &dev_priv->gtt.base); + vma->vm->bind_vma (vma, to->obj->cache_level,GLOBAL_BIND ); + }if (!to->is_initialized ||is_default_context (to)) hw_flags |=MI_RESTORE_INHIBIT ;drivers/gpu/drm/i915/i915_gem_execbuffer.c +12
-17@@ -230,8 +230,9 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,if (unlikely (IS_GEN6 (dev) && reloc->write_domain ==I915_GEM_DOMAIN_INSTRUCTION && !target_i915_obj->has_global_gtt_mapping)) { -i915_gem_gtt_bind_object (target_i915_obj, - target_i915_obj->cache_level); +struct i915_vma *vma =i915_gem_obj_to_vma (obj, vm); + vma->vm->bind_vma (vma, target_i915_obj->cache_level, +GLOBAL_BIND ); }/* Validate that the target is in a valid r/w GPU domain */ @@ -434,11 +435,12 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma,struct intel_ring_buffer *ring, bool *need_reloc) { - struct drm_i915_private *dev_priv = ring->dev->dev_private; + struct drm_i915_gem_object *obj = vma->obj;struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;bool has_fenced_gpu_access =INTEL_INFO (ring->dev)->gen <4 ;bool need_fence, need_mappable; -struct drm_i915_gem_object *obj = vma->obj; +u32 flags = (entry->flags &EXEC_OBJECT_NEEDS_GTT ) && + !vma->obj->has_global_gtt_mapping ?GLOBAL_BIND :0 ;int ret; need_fence =@@ -467,14 +469,6 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, } } -/* Ensure ppgtt mapping exists if needed */ -if (dev_priv->mm.aliasing_ppgtt && !obj->has_aliasing_ppgtt_mapping) { -i915_ppgtt_bind_object (dev_priv->mm.aliasing_ppgtt, - obj, obj->cache_level); - - obj->has_aliasing_ppgtt_mapping =1 ; - } -if (entry->offset != vma->node.start) { entry->offset = vma->node.start; *need_reloc = true;@@ -485,9 +479,7 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, obj->base.pending_write_domain =I915_GEM_DOMAIN_RENDER ; } -if (entry->flags &EXEC_OBJECT_NEEDS_GTT && - !obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (obj, obj->cache_level); + vma->vm->bind_vma (vma, obj->cache_level, flags);return 0 ; }@@ -1077,8 +1069,11 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,* batch" bit. Hence we need to pin secure batches into the global gtt. * hsw should have this fixed, but let's be paranoid and do it * unconditionally for now. */ -if (flags &I915_DISPATCH_SECURE && !batch_obj->has_global_gtt_mapping) -i915_gem_gtt_bind_object (batch_obj, batch_obj->cache_level); +if (flags &I915_DISPATCH_SECURE && + !batch_obj->has_global_gtt_mapping) { +struct i915_vma *vma =i915_gem_obj_to_vma (batch_obj, vm); + vm->bind_vma (vma, batch_obj->cache_level,GLOBAL_BIND ); + } ret =i915_gem_execbuffer_move_to_gpu (ring, &eb->vmas);if (ret)drivers/gpu/drm/i915/i915_gem_gtt.c +3
-45@@ -412,15 +412,6 @@ void i915_gem_cleanup_aliasing_ppgtt(struct drm_device *dev) dev_priv->mm.aliasing_ppgtt =NULL ; } -void i915_ppgtt_bind_object (struct i915_hw_ppgtt *ppgtt, -struct drm_i915_gem_object *obj, -enum i915_cache_level cache_level) -{ - ppgtt->base.insert_entries (&ppgtt->base, obj->pages, -i915_gem_obj_ggtt_offset (obj) >>PAGE_SHIFT , - cache_level); -} - static void __always_unused gen6_ppgtt_bind_vma(struct i915_vma *vma, enum i915_cache_level cache_level,@@ -433,14 +424,6 @@ gen6_ppgtt_bind_vma(struct i915_vma *vma,gen6_ppgtt_insert_entries (vma->vm, vma->obj->pages, entry, cache_level); } -void i915_ppgtt_unbind_object (struct i915_hw_ppgtt *ppgtt, -struct drm_i915_gem_object *obj) -{ - ppgtt->base.clear_range (&ppgtt->base, -i915_gem_obj_ggtt_offset (obj) >>PAGE_SHIFT , - obj->base.size >>PAGE_SHIFT ); -} -static void __always_unused gen6_ppgtt_unbind_vma (struct i915_vma *vma) {const unsigned long entry = vma->node.start >>PAGE_SHIFT ;@@ -501,8 +484,10 @@ void i915_gem_restore_gtt_mappings(struct drm_device *dev)gen6_write_pdes (dev_priv->mm.aliasing_ppgtt);list_for_each_entry (obj, &dev_priv->mm.bound_list, global_list) { +struct i915_vma *vma =i915_gem_obj_to_vma (obj, + &dev_priv->gtt.base);i915_gem_clflush_object (obj); -i915_gem_gtt_bind_object (obj, obj->cache_level); + vma->vm->bind_vma (vma, obj->cache_level,0 ); }i915_gem_chipset_flush (dev);@@ -658,33 +643,6 @@ static void gen6_ggtt_bind_vma(struct i915_vma *vma, } } -void i915_gem_gtt_bind_object (struct drm_i915_gem_object *obj, -enum i915_cache_level cache_level) -{ -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -const unsigned long entry =i915_gem_obj_ggtt_offset (obj) >>PAGE_SHIFT ; - - dev_priv->gtt.base.insert_entries (&dev_priv->gtt.base, obj->pages, - entry, - cache_level); - - obj->has_global_gtt_mapping =1 ; -} - -void i915_gem_gtt_unbind_object (struct drm_i915_gem_object *obj) -{ -struct drm_device *dev = obj->base.dev; -struct drm_i915_private *dev_priv = dev->dev_private; -const unsigned long entry =i915_gem_obj_ggtt_offset (obj) >>PAGE_SHIFT ; - - dev_priv->gtt.base.clear_range (&dev_priv->gtt.base, - entry, - obj->base.size >>PAGE_SHIFT ); - - obj->has_global_gtt_mapping =0 ; -} -static void gen6_ggtt_unbind_vma (struct i915_vma *vma) {struct drm_device *dev = vma->vm->dev;Signature · 6 lines
1.8.3.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
PATCH 29/29 drm/i915: eliminate vm->insert_entries()
In reply to Ben WidawskyWith bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings.
--- drivers/gpu/drm/i915/i915_gem_gtt.c | 17 +---------------- 1 file changed, 1 insertion(+), 16 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +1 -16
@@ -314,8 +314,8 @@ static int gen6_ppgtt_init(struct i915_hw_ppgtt *ppgtt) ppgtt->enable = gen6_ppgtt_enable; ppgtt->base.unbind_vma =NULL ; ppgtt->base.clear_range = gen6_ppgtt_clear_range; - ppgtt->base.bind_vma =NULL ; ppgtt->base.insert_entries = gen6_ppgtt_insert_entries; + ppgtt->base.bind_vma =NULL ; ppgtt->base.cleanup = gen6_ppgtt_cleanup; ppgtt->base.scratch = dev_priv->gtt.base.scratch; ppgtt->pt_pages = kzalloc(sizeof(struct page *)*ppgtt->num_pd_entries,
@@ -569,19 +569,6 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm,readl (gtt_base); } - -static void i915_ggtt_insert_entries (struct i915_address_space *vm, -struct sg_table *st, -unsigned int pg_start, -enum i915_cache_level cache_level) -{ -unsigned int flags = (cache_level ==I915_CACHE_NONE ) ? -AGP_USER_MEMORY :AGP_USER_CACHED_MEMORY ; - -intel_gtt_insert_sg_entries (st, pg_start, flags); - -} -static void i915_ggtt_bind_vma (struct i915_vma *vma,enum i915_cache_level cache_level,u32 unused)
@@ -894,7 +881,6 @@ static int gen6_gmch_probe(struct drm_device *dev, dev_priv->gtt.base.clear_range = gen6_ggtt_clear_range; dev_priv->gtt.base.unbind_vma = gen6_ggtt_unbind_vma; - dev_priv->gtt.base.insert_entries = gen6_ggtt_insert_entries; dev_priv->gtt.base.bind_vma = gen6_ggtt_bind_vma;return ret;
@@ -928,7 +914,6 @@ static int i915_gmch_probe(struct drm_device *dev, dev_priv->gtt.do_idle_maps =needs_idle_maps (dev_priv->dev); dev_priv->gtt.base.clear_range = i915_ggtt_clear_range; dev_priv->gtt.base.unbind_vma = i915_ggtt_unbind_vma; - dev_priv->gtt.base.insert_entries = i915_ggtt_insert_entries; dev_priv->gtt.base.bind_vma = i915_ggtt_bind_vma;return 0 ;
Signature · 1 line
1.8.3.4