Command parser batch buffer copy
Subsystems (3)
v7 cover Command parser batch buffer copy
From: "Michael H. Nguyen" <[email protected]>
This is v7 in response to
http://lists.freedesktop.org/archives/intel-gfx/2014-December/057033.html
Minor updates from the last rev
- Keep batch pool in LRU order (Daniel)
- Remove duplicate madv assignments (Daniel)
Brad Volkin (5):
drm/i915: Implement a framework for batch buffer pools drm/i915: Use batch pools with the command parser drm/i915: Use batch length instead of object size in command parser drm/i915: Mark shadow batch buffers as purgeable drm/i915: Tidy up execbuffer command parsing code
Documentation/DocBook/drm.tmpl | 5 ++ drivers/gpu/drm/i915/Makefile | 1 + drivers/gpu/drm/i915/i915_cmd_parser.c | 99 +++++++++++++++++---- drivers/gpu/drm/i915/i915_debugfs.c | 71 +++++++++++++-- drivers/gpu/drm/i915/i915_dma.c | 1 + drivers/gpu/drm/i915/i915_drv.h | 23 +++++ drivers/gpu/drm/i915/i915_gem.c | 3 + drivers/gpu/drm/i915/i915_gem_batch_pool.c | 137 +++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 98 +++++++++++++++++---- 9 files changed, 396 insertions(+), 42 deletions(-) create mode 100644 drivers/gpu/drm/i915/i915_gem_batch_pool.c
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH v7 1/5 drm/i915: Implement a framework for batch buffer pools
From: Brad Volkin <[email protected]>
This adds a small module for managing a pool of batch buffers. The only current use case is for the command parser, as described in the kerneldoc in the patch. The code is simple, but separating it out makes it easier to change the underlying algorithms and to extend to future use cases should they arise.
The interface is simple: init to create an empty pool, fini to clean it up, get to obtain a new buffer. Note that all buffers are expected to be inactive before cleaning up the pool.
Locking is currently based on the caller holding the struct_mutex. We already do that in the places where we will use the batch pool for the command parser.
v2:
- s/BUG_ON/WARN_ON/ for locking assertions
- Remove the cap on pool size
- Switch from alloc/free to init/fini
v3:
- Idiomatic looping structure in _fini
- Correct handling of purged objects
- Don't return a buffer that's too much larger than needed
v4:
- Rebased to latest -nightly
v5:
- Remove _put() function and clean up comments to match
v6:
- Move purged check inside the loop (danvet, from v4 1/7 feedback)
v7:
- Use single list instead of two. (Chris W)
- s/active_list/cache_list
- Squashed in debug patches (Chris W)
drm/i915: Add a batch pool debugfs file
It provides some useful information about the buffers in the global command parser batch pool.
v2: rebase on global pool instead of per-ring pools
v3: rebase
drm/i915: Add batch pool details to i915_gem_objects debugfs
To better account for the potentially large memory consumption of the batch pool.
v8:
- Keep cache in LRU order (danvet, from v6 1/5 feedback)
Issue: VIZ-4719
--- Documentation/DocBook/drm.tmpl | 5 ++ drivers/gpu/drm/i915/Makefile | 1 + drivers/gpu/drm/i915/i915_debugfs.c | 71 +++++++++++++-- drivers/gpu/drm/i915/i915_drv.h | 21 +++++ drivers/gpu/drm/i915/i915_gem.c | 1 + drivers/gpu/drm/i915/i915_gem_batch_pool.c | 135 +++++++++++++++++++++++++++++ 6 files changed, 225 insertions(+), 9 deletions(-) create mode 100644 drivers/gpu/drm/i915/i915_gem_batch_pool.c
Documentation/DocBook/drm.tmpl +5 -0
@@ -4059,6 +4059,11 @@ int num_ioctls;</synopsis> !Idrivers/gpu/drm/i915/i915_cmd_parser.c </sect2> <sect2> + <title>Batchbuffer Pools</title> +!Pdrivers/gpu/drm/i915/i915_gem_batch_pool.c batch pool +!Idrivers/gpu/drm/i915/i915_gem_batch_pool.c + </sect2> + <sect2> <title>Logical Rings, Logical Ring Contexts and Execlists</title> !Pdrivers/gpu/drm/i915/intel_lrc.c Logical Rings, Logical Ring Contexts and Execlists !Idrivers/gpu/drm/i915/intel_lrc.c
drivers/gpu/drm/i915/Makefile +1 -0
@@ -19,6 +19,7 @@ i915-$(CONFIG_DEBUG_FS) += i915_debugfs.o # GEM code i915-y += i915_cmd_parser.o \ + i915_gem_batch_pool.o \ i915_gem_context.o \ i915_gem_render_state.o \ i915_gem_debug.o \
drivers/gpu/drm/i915/i915_debugfs.c +62 -9
101 lines not on this page · raw message
drivers/gpu/drm/i915/i915_drv.h +21 -0
@@ -1142,6 +1142,11 @@ struct intel_l3_parity {int which_slice; }; +struct i915_gem_batch_pool { +struct drm_device *dev; +struct list_head cache_list; +}; +struct i915_gem_mm {/** Memory allocator for GTT stolen memory */ struct drm_mm stolen;
@@ -1155,6 +1160,13 @@ struct i915_gem_mm {*/ struct list_head unbound_list; +/* +* A pool of objects to use as shadow copies of client batch buffers +* when the command parser is enabled. Prevents the client from +* modifying the batch contents after software parsing. +*/ +struct i915_gem_batch_pool batch_pool; +/** Usable portion of the GTT for GEM */ unsigned long stolen_base;/* limited to low memory (32-bit) */
@@ -1885,6 +1897,8 @@ struct drm_i915_gem_object {/** Used in execbuf to temporarily hold a ref */ struct list_head obj_exec_link; +struct list_head batch_pool_list; +/** * This is set if the object is on the active lists (has pending * rendering and so a non-zero seqno), and is not set if it i s on
@@ -2889,6 +2903,13 @@ void i915_destroy_error_state(struct drm_device *dev);void i915_get_extra_instdone (struct drm_device *dev,uint32_t *instdone);const char *i915_cache_level_str (struct drm_i915_private *i915,int type); +/* i915_gem_batch_pool.c */ +void i915_gem_batch_pool_init (struct drm_device *dev, +struct i915_gem_batch_pool *pool); +void i915_gem_batch_pool_fini (struct i915_gem_batch_pool *pool); +struct drm_i915_gem_object * +i915_gem_batch_pool_get (struct i915_gem_batch_pool *pool,size_t size); +/* i915_cmd_parser.c */ int i915_cmd_parser_get_version (void );int i915_cmd_parser_init_ring (struct intel_engine_cs *ring);
drivers/gpu/drm/i915/i915_gem.c +1 -0
@@ -4384,6 +4384,7 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj,INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->obj_exec_link);INIT_LIST_HEAD (&obj->vma_list); +INIT_LIST_HEAD (&obj->batch_pool_list); obj->ops = ops;
drivers/gpu/drm/i915/i915_gem_batch_pool.c new file +135 -0
135 lines not on this page · raw message
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 1/5] drm/i915: Implement a framework for batch buffer pools
Quoted 436 lines
-----Original Message-----
From: Nguyen, Michael H
Sent: Thursday, December 11, 2014 8:13 PM
To: [email protected]
Cc: Bloomfield, Jon; Brad Volkin
Subject: [PATCH v7 1/5] drm/i915: Implement a framework for batch buffer poolsFrom: Brad Volkin <[email protected]>
This adds a small module for managing a pool of batch buffers. The only current use case is for the command parser, as described in the kerneldoc in the patch. The code is simple, but separating it out makes it easier to change the underlying algorithms and to extend to future use cases should they arise.
The interface is simple: init to create an empty pool, fini to clean it up, get to obtain a new buffer. Note that all buffers are expected to be inactive before cleaning up the pool.
Locking is currently based on the caller holding the
struct_mutex. We already do that in the places where we will use the batch pool for the command parser.v2:
- s/BUG_ON/WARN_ON/ for locking assertions
- Remove the cap on pool size
- Switch from alloc/free to init/finiv3:
- Idiomatic looping structure in_fini
- Correct handling of purged objects
- Don't return a buffer that's too much larger than neededv4:
- Rebased to latest -nightlyv5:
- Remove_put()function and clean up comments to matchv6:
- Move purged check inside the loop (danvet, from v4 1/7 feedback)v7:
- Use single list instead of two. (Chris W)
- s/active_list/cache_list
- Squashed in debug patches (Chris W)
drm/i915: Add a batch pool debugfs fileIt provides some useful information about the buffers in the global command parser batch pool.
v2: rebase on global pool instead of per-ring pools
v3: rebasedrm/i915: Add batch pool details to
i915_gem_objectsdebugfsTo better account for the potentially large memory consumption of the batch pool.
v8:
- Keep cache in LRU order (danvet, from v6 1/5 feedback)Issue: VIZ-4719
Signed-off-by Brad Volkin <[email protected]>--- Documentation/DocBook/drm.tmpl | 5 ++ drivers/gpu/drm/i915/Makefile | 1 + drivers/gpu/drm/i915/i915_debugfs.c | 71 +++++++++++++-- drivers/gpu/drm/i915/i915_drv.h | 21 +++++ drivers/gpu/drm/i915/i915_gem.c | 1 + drivers/gpu/drm/i915/i915_gem_batch_pool.c | 135 +++++++++++++++++++++++++++++ 6 files changed, 225 insertions(+), 9 deletions(-) create mode 100644
drivers/gpu/drm/i915/i915_gem_batch_pool.c(quoted diff) +0
-0b/Documentation/DocBook/drm.tmpl index 25c23ca..21cbcdb 100644
Documentation/DocBook/drm.tmpl +0
-0@@ -4059,6 +4059,11 @@ int num_ioctls;</synopsis>!Idrivers/gpu/drm/i915/i915_cmd_parser.c
</sect2> <sect2>Documentation/DocBook/drm.tmpl +5
-0+ <title>Batchbuffer Pools</title> +!Pdrivers/gpu/drm/i915/i915_gem_batch_pool.c batch pool +!Idrivers/gpu/drm/i915/i915_gem_batch_pool.c + </sect2> + <sect2> <title>Logical Rings, Logical Ring Contexts and Execlists</title>!Pdrivers/gpu/drm/i915/intel_lrc.c Logical Rings, Logical Ring Contexts and Execlists !Idrivers/gpu/drm/i915/intel_lrc.c
drivers/gpu/drm/i915/Makefile +1
-0@@ -19,6 +19,7 @@ i915-$(CONFIG_DEBUG_FS) += i915_debugfs.o # GEM code i915-y += i915_cmd_parser.o \ + i915_gem_batch_pool.o \ i915_gem_context.o \ i915_gem_render_state.o \ i915_gem_debug.o \drivers/gpu/drm/i915/i915_debugfs.c +30
-9@@ -359,6 +359,33 @@ static int per_file_stats(int id, void *ptr, void *data)return 0 ; } +#define print_file_stats (m, name, stats) \ + seq_printf(m, "%s: %u objects, %zu bytes (%zu active, %zu inactive, %zu global, %zu shared, %zu unbound)\n", \ +name, \ +stats.count, \ +stats.total, \ +stats.active, \ +stats.inactive, \ +stats.global, \ +stats.shared, \ + stats.unbound) + +static void print_batch_pool_stats(struct seq_file *m, + struct drm_i915_private *dev_priv) { +struct drm_i915_gem_object *obj; +struct file_stats stats; + +memset (&stats,0 ,sizeof (stats)); + +list_for_each_entry (obj, + &dev_priv->mm.batch_pool.cache_list, + batch_pool_list) +per_file_stats (0 , obj, &stats); + +print_file_stats (m,"batch pool" , stats); } +#define count_vmas (list, member) do { \ list_for_each_entry(vma, list, member) { \ size += i915_gem_obj_ggtt_size(vma->obj); \ @@ -441,6 +468,9 @@ static int i915_gem_object_info(struct seq_file *m, void* data) dev_priv->gtt.mappable_end - dev_priv->gtt.base.start);seq_putc (m,'\n' ); +print_batch_pool_stats (m, dev_priv); + +seq_putc (m,'\n' );list_for_each_entry_reverse (file, &dev->filelist, lhead) {struct file_stats stats;struct task_struct *task;@@ -458,15 +488,7 @@ static int i915_gem_object_info(struct seq_file *m,void* data) */ rcu_read_lock (); task =pid_task (file->pid,PIDTYPE_PID ); - seq_printf(m, "%s: %u objects, %zu bytes (%zu active, %zu inactive, %zu global, %zu shared, %zu unbound)\n", - task ? task->comm :"<unknown>" , - stats.count, - stats.total, - stats.active, - stats.inactive, - stats.global, - stats.shared, - stats.unbound); + print_file_stats(m, task ? task->comm :"<unknown>" , stats);rcu_read_unlock ();drivers/gpu/drm/i915/i915_debugfs.c +29
-0}@@ -583,6 +605,36 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)return 0 ; } +static int i915_gem_batch_pool_info (struct seq_file *m,void *data) { +struct drm_info_node *node = m->private; +struct drm_device *dev = node->minor->dev; +struct drm_i915_private *dev_priv = dev->dev_private; +struct drm_i915_gem_object *obj; +int count =0 ; +int ret; + + ret =mutex_lock_interruptible (&dev->struct_mutex); +if (ret) +return ret; + +seq_puts (m,"cache:\n" ); +list_for_each_entry (obj, + &dev_priv->mm.batch_pool.cache_list, + batch_pool_list) { +seq_puts (m," " ); +describe_obj (m, obj); +seq_putc (m,'\n' ); + count++; + } + +seq_printf (m,"total: %d\n" , count); + +mutex_unlock (&dev->struct_mutex); + +return 0 ; +} +static int i915_gem_request_info (struct seq_file *m,void *data) { struct drm_info_node *node = m->private; @@ -4356,6 +4408,7 @@static const struct
drm_info_listi915_debugfs_list[] = {drivers/gpu/drm/i915/i915_debugfs.c +1
-0{"i915_gem_hws_blt", i915_hws_info, 0, (void *)BCS}, {"i915_gem_hws_bsd", i915_hws_info, 0, (void *)VCS}, {"i915_gem_hws_vebox", i915_hws_info, 0, (void *)VECS}, + {"i915_gem_batch_pool", i915_gem_batch_pool_info, 0}, {"i915_frequency_info", i915_frequency_info, 0}, {"i915_drpc_info", i915_drpc_info, 0}, {"i915_emon_status", i915_emon_status, 0}, diff --git
a/drivers/gpu/drm/i915/i915_drv.hb/drivers/gpu/drm/i915/i915_drv.hindex d68c75f..e1e503a 100644drivers/gpu/drm/i915/i915_drv.h +14
-0@@ -1142,6 +1142,11 @@ struct intel_l3_parity {int which_slice; }; +struct i915_gem_batch_pool { +struct drm_device *dev; +struct list_head cache_list; +}; +struct i915_gem_mm {/** Memory allocator for GTT stolen memory */ struct drm_mm stolen;@@ -1155,6 +1160,13 @@ struct i915_gem_mm {*/ struct list_head unbound_list; +/* +* A pool of objects to use as shadow copies of client batch buffers +* when the command parser is enabled. Prevents the client from +* modifying the batch contents after software parsing. +*/ +struct i915_gem_batch_pool batch_pool; +/** Usable portion of the GTT for GEM */ unsigned long stolen_base;/* limited to low memory (32-bit) */ @@ -1885,6 +1897,8 @@ struct drm_i915_gem_object {/** Used in execbuf to temporarily hold a ref */ struct list_head obj_exec_link; +struct list_head batch_pool_list; +/** * This is set if the object is on the active lists (has pending * rendering and so a non-zero seqno), and is not set if it i s on @@ - 2889,6 +2903,13 @@ void
i915_destroy_error_state(struct drm_device *dev); voidi915_get_extra_instdone(struct drm_device *dev, uint32_t *instdone); const char *i915_cache_level_str(structdrm_i915_private*i915, int type);drivers/gpu/drm/i915/i915_drv.h +5
-0+/* i915_gem_batch_pool.c */ +void i915_gem_batch_pool_init (struct drm_device *dev, +struct i915_gem_batch_pool *pool);void +i915_gem_batch_pool_fini (struct i915_gem_batch_pool *pool);struct +drm_i915_gem_object* i915_gem_batch_pool_get(struct
i915_gem_batch_pooldrivers/gpu/drm/i915/i915_drv.h +2
-0+*pool, size_t size); +/* i915_cmd_parser.c */ int i915_cmd_parser_get_version (void ); int i915_cmd_parser_init_ring(struct intel_engine_cs *ring); diff --git
a/drivers/gpu/drm/i915/i915_gem.cb/drivers/gpu/drm/i915/i915_gem.cindex de241eb..2f14ae1 100644drivers/gpu/drm/i915/i915_gem.c +1
-0@@ -4384,6 +4384,7 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj,INIT_LIST_HEAD (&obj->ring_list);INIT_LIST_HEAD (&obj->obj_exec_link);INIT_LIST_HEAD (&obj->vma_list); +INIT_LIST_HEAD (&obj->batch_pool_list); obj->ops = ops;drivers/gpu/drm/i915/i915_gem_batch_pool.c new file +26
-0@@ -0,0 +1,135 @@ +/* +* Copyright © 2014 Intel Corporation +* +* Permission is hereby granted, free of charge, to any person +obtaining a +* copy of this software and associated documentation files (the +"Software"), +* to deal in the Software without restriction, including without +limitation +* the rights to use, copy, modify, merge, publish, distribute, +sublicense, +* and/or sell copies of the Software, and to permit persons to whom +the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the +next +* paragraph) shall be included in all copies or substantial portions +of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT
+SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES ORdrivers/gpu/drm/i915/i915_gem_batch_pool.c new file +4
-0+OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, +ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OROTHER
drivers/gpu/drm/i915/i915_gem_batch_pool.c new file +109
-0+DEALINGS +* IN THE SOFTWARE. +* +*/ + +#include "i915_drv.h" + +/** +* DOC: batch pool +* +* In order to submit batch buffers as 'secure', the software command +parser +* must ensure that a batch buffer cannot be modified after parsing. It +does +* this by copying the user provided batch buffer contents to a kernel +owned +* buffer from which the hardware will actually execute, and by +carefully +* managing the address space bindings for such buffers. +* +* The batch pool framework provides a mechanism for the driver to +manage a +* set of scratch buffers to use for this purpose. The framework can be +* extended to support other uses cases should they arise. +*/ + +/** +* i915_gem_batch_pool_init() - initialize a batch buffer pool +* @dev: the drm device +* @pool: the batch buffer pool +*/ +void i915_gem_batch_pool_init (struct drm_device *dev, +struct i915_gem_batch_pool *pool) { + pool->dev = dev; +INIT_LIST_HEAD (&pool->cache_list); +} + +/** +* i915_gem_batch_pool_fini() - clean up a batch buffer pool +* @pool: the pool to clean up +* +* Note: Callers must hold the struct_mutex. +*/ +void i915_gem_batch_pool_fini (struct i915_gem_batch_pool *pool) { +WARN_ON (!mutex_is_locked (&pool->dev->struct_mutex)); + +while (!list_empty (&pool->cache_list)) { +struct drm_i915_gem_object *obj = +list_first_entry (&pool->cache_list, + struct drm_i915_gem_object, + batch_pool_list); + +WARN_ON (obj->active); + +list_del_init (&obj->batch_pool_list); +drm_gem_object_unreference (&obj->base); + } +} + +/** +* i915_gem_batch_pool_get() - select a buffer from the pool +* @pool: the batch buffer pool +* @size: the minimum desired size of the returned buffer +* +* Finds or allocates a batch buffer in the pool with at least the +requested +* size. The caller is responsible for any domain, active/inactive, or +* purgeability management for the returned buffer. +* +* Note: Callers must hold the struct_mutex +* +* Return: the selected batch buffer object */ struct +drm_i915_gem_object *i915_gem_batch_pool_get (struct +i915_gem_batch_pool *pool, +size_t size) +{ +struct drm_i915_gem_object *obj =NULL ; +struct drm_i915_gem_object *tmp, *next; + +WARN_ON (!mutex_is_locked (&pool->dev->struct_mutex)); + +list_for_each_entry_safe (tmp, next, + &pool->cache_list, batch_pool_list) { + +if (tmp->active) +continue ; + +/* While we're looping, do some clean up */ +if (tmp->madv == __I915_MADV_PURGED) { +list_del (&tmp->batch_pool_list); +drm_gem_object_unreference (&tmp->base); +continue ; + } + +/* +* Select a buffer that is at least as big as needed +* but not 'too much' bigger. A better way to do this +* might be to bucket the pool objects based on size. +*/ +if (tmp->base.size >= size && + tmp->base.size <= (2 * size)) { + obj = tmp; +break ; + } + } + +if (!obj) { + obj =i915_gem_alloc_object (pool->dev, size); +if (!obj)
drivers/gpu/drm/i915/i915_gem_batch_pool.c line 135
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_batch_pool.c new file +10
-0+return ERR_PTR (-ENOMEM ); + +list_add_tail (&obj->batch_pool_list, &pool->cache_list); + } +else +/* Keep list in LRU order */ +list_move_tail (&obj->batch_pool_list, &pool->cache_list); + +return obj; +}
Quoted 2 lines
--
1.9.1
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH v7 2/5 drm/i915: Use batch pools with the command parser
In reply to [email protected]From: Brad Volkin <[email protected]>
This patch sets up all of the tracking and copying necessary to use batch pools with the command parser and dispatches the copied (shadow) batch to the hardware.
After this patch, the parser is in 'enabling' mode.
Note that performance takes a hit from the copy in some cases and will likely need some work. At a rough pass, the memcpy appears to be the bottleneck. Without having done a deeper analysis, two ideas that come to mind are:
1) Copy sections of the batch at a time, as they are reached by parsing. Might improve cache locality.
2) Copy only up to the userspace-supplied batch length and memset the rest of the buffer. Reduces the number of reads.
v2:
- Remove setting the capacity of the pool
- One global pool instead of per-ring pools
- Replace batch_obj with shadow_batch_obj and hook into eb->vmas
- Memset any space in the shadow batch beyond what gets copied
- Rebased on execlist prep refactoring
v3:
- Rebase on chained batch handling
- Squash in setting the secure dispatch flag
- Add a note about the interaction w/secure dispatch pinning
- Check for request->batch_obj == NULL in i915_gem_free_request
v4:
- Fix read domains for shadow_batch_obj
- Remove the set_to_gtt_domain call from i915_parse_cmds
- ggtt_pin/unpin in the parser block to simplify error handling
- Check USES_FULL_PPGTT before setting DISPATCH_SECURE flag
- Remove i915_gem_batch_pool_put calls
v5:
- Move 'pending_read_domains |= I915_GEM_DOMAIN_COMMAND' after the parser (danvet, from v4 0/7 feedback)
Issue: VIZ-4719
--- drivers/gpu/drm/i915/i915_cmd_parser.c | 79 +++++++++++++++++++++++------- drivers/gpu/drm/i915/i915_dma.c | 1 + drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/i915_gem.c | 2 + drivers/gpu/drm/i915/i915_gem_execbuffer.c | 52 +++++++++++++++++--- 5 files changed, 112 insertions(+), 23 deletions(-)
drivers/gpu/drm/i915/i915_cmd_parser.c +62 -17
@@ -848,6 +848,56 @@ finish:return (u32 *)addr; } +/* Returns a vmap'd pointer to dest_obj, which the caller must unmap */ +static u32 *copy_batch (struct drm_i915_gem_object *dest_obj, +struct drm_i915_gem_object *src_obj) +{ +int ret =0 ; +int needs_clflush =0 ; +u32 *src_addr, *dest_addr =NULL ; + + ret =i915_gem_obj_prepare_shmem_read (src_obj, &needs_clflush); +if (ret) { +DRM_DEBUG_DRIVER ("CMD: failed to prep read\n" ); +return ERR_PTR (ret); + } + + src_addr =vmap_batch (src_obj); +if (!src_addr) { +DRM_DEBUG_DRIVER ("CMD: Failed to vmap batch\n" ); + ret = -ENOMEM ; + goto unpin_src; + } + +if (needs_clflush) +drm_clflush_virt_range ((char *)src_addr, src_obj->base.size); + + ret =i915_gem_object_set_to_cpu_domain (dest_obj, true); +if (ret) { +DRM_DEBUG_DRIVER ("CMD: Failed to set batch CPU domain\n" ); + goto unmap_src; + } + + dest_addr =vmap_batch (dest_obj); +if (!dest_addr) { +DRM_DEBUG_DRIVER ("CMD: Failed to vmap shadow batch\n" ); + ret = -ENOMEM ; + goto unmap_src; + } + +memcpy (dest_addr, src_addr, src_obj->base.size); +if (dest_obj->base.size > src_obj->base.size) +memset ((u8 *)dest_addr + src_obj->base.size,0 , + dest_obj->base.size - src_obj->base.size); + +unmap_src: +vunmap (src_addr); +unpin_src: +i915_gem_object_unpin_pages (src_obj); + +return ret ?ERR_PTR (ret) : dest_addr; +} +/** * i915_needs_cmd_parser() - should a given ring use software command parsing? * @ring: the ring in question
@@ -964,6 +1014,7 @@ static bool check_cmd(const struct intel_engine_cs *ring,* i915_parse_cmds() - parse a submitted batch buffer for privilege violations * @ring: the ring on which the batch is to execute * @batch_obj: the batch buffer in question +* @shadow_batch_obj: copy of the batch buffer in question * @batch_start_offset: byte offset in the batch at which execution starts * @is_master: is the submitting process the drm master? *
@@ -975,32 +1026,28 @@ static bool check_cmd(const struct intel_engine_cs *ring,*/ int i915_parse_cmds (struct intel_engine_cs *ring,struct drm_i915_gem_object *batch_obj, +struct drm_i915_gem_object *shadow_batch_obj,u32 batch_start_offset,bool is_master) {int ret =0 ;u32 *cmd, *batch_base, *batch_end;struct drm_i915_cmd_descriptor default_desc = {0 }; -int needs_clflush =0 ;bool oacontrol_set = false;/* OACONTROL tracking. See check_cmd() */ - ret =i915_gem_obj_prepare_shmem_read (batch_obj, &needs_clflush); -if (ret) { -DRM_DEBUG_DRIVER ("CMD: failed to prep read\n" ); -return ret; + batch_base =copy_batch (shadow_batch_obj, batch_obj); +if (IS_ERR (batch_base)) { +DRM_DEBUG_DRIVER ("CMD: Failed to copy batch\n" ); +return PTR_ERR (batch_base); } - batch_base =vmap_batch (batch_obj); -if (!batch_base) { -DRM_DEBUG_DRIVER ("CMD: Failed to vmap batch\n" ); -i915_gem_object_unpin_pages (batch_obj); -return -ENOMEM ; - } - -if (needs_clflush) -drm_clflush_virt_range ((char *)batch_base, batch_obj->base.size); - cmd = batch_base + (batch_start_offset /sizeof (*cmd)); + +/* +* We use the source object's size because the shadow object is as +* large or larger and copy_batch() will write MI_NOPs to the extra +* space. Parsing should be faster in some cases this way. +*/ batch_end = cmd + (batch_obj->base.size /sizeof (*batch_end));while (cmd < batch_end) {
@@ -1062,8 +1109,6 @@ int i915_parse_cmds(struct intel_engine_cs *ring,vunmap (batch_base); -i915_gem_object_unpin_pages (batch_obj); -return ret; }
drivers/gpu/drm/i915/i915_dma.c +1 -0
@@ -928,6 +928,7 @@ int i915_driver_unload(struct drm_device *dev)mutex_lock (&dev->struct_mutex);i915_gem_cleanup_ringbuffer (dev); +i915_gem_batch_pool_fini (&dev_priv->mm.batch_pool);i915_gem_context_fini (dev);mutex_unlock (&dev->struct_mutex);i915_gem_cleanup_stolen (dev);
drivers/gpu/drm/i915/i915_drv.h +1 -0
@@ -2917,6 +2917,7 @@ void i915_cmd_parser_fini_ring(struct intel_engine_cs *ring);bool i915_needs_cmd_parser (struct intel_engine_cs *ring);int i915_parse_cmds (struct intel_engine_cs *ring,struct drm_i915_gem_object *batch_obj, +struct drm_i915_gem_object *shadow_batch_obj,u32 batch_start_offset,bool is_master);
drivers/gpu/drm/i915/i915_gem.c +2 -0
@@ -4997,6 +4997,8 @@ i915_gem_load(struct drm_device *dev) dev_priv->mm.oom_notifier.notifier_call = i915_gem_shrinker_oom;register_oom_notifier (&dev_priv->mm.oom_notifier); +i915_gem_batch_pool_init (dev, &dev_priv->mm.batch_pool); +mutex_init (&dev_priv->fb_tracking.lock); }
drivers/gpu/drm/i915/i915_gem_execbuffer.c +46 -6
80 lines not on this page · raw message
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 2/5] drm/i915: Use batch pools with the command parser
Quoted 44 lines
-----Original Message-----
From: Nguyen, Michael H
Sent: Thursday, December 11, 2014 8:13 PM
To: [email protected]
Cc: Bloomfield, Jon; Brad Volkin
Subject: [PATCH v7 2/5] drm/i915: Use batch pools with the command parserFrom: Brad Volkin <[email protected]>
This patch sets up all of the tracking and copying necessary to use batch pools with the command parser and dispatches the copied
(shadow) batch to the hardware.After this patch, the parser is in 'enabling' mode.
Note that performance takes a hit from the copy in some cases and will likely need some work. At a rough pass, the memcpy appears to be the bottleneck. Without having done a deeper analysis, two ideas that come to mind are:
1) Copy sections of the batch at a time, as they are reached
by parsing. Might improve cache locality.
2) Copy only up to the userspace-supplied batch length and
memset the rest of the buffer. Reduces the number of reads.v2:
- Remove setting the capacity of the pool
- One global pool instead of per-ring pools
- Replacebatch_objwithshadow_batch_objand hook intoeb->vmas
- Memset any space in the shadow batch beyond what gets copied
- Rebased on execlist prep refactoringv3:
- Rebase on chained batch handling
- Squash in setting the secure dispatch flag
- Add a note about the interaction w/secure dispatch pinning
- Check forrequest->batch_obj== NULL ini915_gem_free_requestv4:
- Fix read domains forshadow_batch_obj
- Remove theset_to_gtt_domaincall fromi915_parse_cmds
- ggtt_pin/unpin in the parser block to simplify error handling
- CheckUSES_FULL_PPGTTbefore settingDISPATCH_SECUREflag
- Removei915_gem_batch_pool_putcalls
Quoted 7 lines
v5:
- Move 'pending_read_domains|=I915_GEM_DOMAIN_COMMAND' after the parser (danvet, from v4 0/7 feedback)Issue: VIZ-4719
Signed-off-by Brad Volkin <[email protected]>---
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH v7 3/5 drm/i915: Use batch length instead of object size in command parser
In reply to [email protected]From: Brad Volkin <[email protected]>
Previously we couldn't trust the user-supplied batch length because it came directly from userspace (i.e. untrusted code). It would have affected what commands software parsed without regard to what hardware would actually execute, leaving a potential hole.
With the parser now copying the user supplied batch buffer and writing MI_NOP commands to any space after the copied region, we can safely use the batch length input. This should be a performance win as the actual batch length is frequently much smaller than the allocated object size.
v2: Fix handling of non-zero batch_start_offset
Issue: VIZ-4719
--- drivers/gpu/drm/i915/i915_cmd_parser.c | 48 ++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/i915_gem_execbuffer.c | 1 + 3 files changed, 34 insertions(+), 16 deletions(-)
drivers/gpu/drm/i915/i915_cmd_parser.c +32 -16
@@ -850,11 +850,19 @@ finish:/* Returns a vmap'd pointer to dest_obj, which the caller must unmap */ static u32 *copy_batch (struct drm_i915_gem_object *dest_obj, -struct drm_i915_gem_object *src_obj) +struct drm_i915_gem_object *src_obj, +u32 batch_start_offset, +u32 batch_len) {int ret =0 ;int needs_clflush =0 ; -u32 *src_addr, *dest_addr =NULL ; +u32 *src_base, *dest_base =NULL ; +u32 *src_addr, *dest_addr; +u32 offset = batch_start_offset /sizeof (*dest_addr); +u32 end = batch_start_offset + batch_len; + +if (end > dest_obj->base.size || end > src_obj->base.size) +return ERR_PTR (-E2BIG ); ret =i915_gem_obj_prepare_shmem_read (src_obj, &needs_clflush);if (ret) {
@@ -862,15 +870,17 @@ static u32 *copy_batch(struct drm_i915_gem_object *dest_obj,return ERR_PTR (ret); } - src_addr =vmap_batch (src_obj); -if (!src_addr) { + src_base =vmap_batch (src_obj); +if (!src_base) {DRM_DEBUG_DRIVER ("CMD: Failed to vmap batch\n" ); ret = -ENOMEM ; goto unpin_src; } + src_addr = src_base + offset; +if (needs_clflush) -drm_clflush_virt_range ((char *)src_addr, src_obj->base.size); +drm_clflush_virt_range ((char *)src_addr, batch_len); ret =i915_gem_object_set_to_cpu_domain (dest_obj, true);if (ret) {
@@ -878,24 +888,27 @@ static u32 *copy_batch(struct drm_i915_gem_object *dest_obj, goto unmap_src; } - dest_addr =vmap_batch (dest_obj); -if (!dest_addr) { + dest_base =vmap_batch (dest_obj); +if (!dest_base) {DRM_DEBUG_DRIVER ("CMD: Failed to vmap shadow batch\n" ); ret = -ENOMEM ; goto unmap_src; } -memcpy (dest_addr, src_addr, src_obj->base.size); -if (dest_obj->base.size > src_obj->base.size) -memset ((u8 *)dest_addr + src_obj->base.size,0 , - dest_obj->base.size - src_obj->base.size); + dest_addr = dest_base + offset; + +if (batch_start_offset !=0 ) +memset ((u8 *)dest_base,0 , batch_start_offset); + +memcpy (dest_addr, src_addr, batch_len); +memset ((u8 *)dest_addr + batch_len,0 , dest_obj->base.size - end); unmap_src: -vunmap (src_addr); +vunmap (src_base); unpin_src:i915_gem_object_unpin_pages (src_obj); -return ret ?ERR_PTR (ret) : dest_addr; +return ret ?ERR_PTR (ret) : dest_base; }/**
@@ -1016,6 +1029,7 @@ static bool check_cmd(const struct intel_engine_cs *ring,* @batch_obj: the batch buffer in question * @shadow_batch_obj: copy of the batch buffer in question * @batch_start_offset: byte offset in the batch at which execution starts +* @batch_len: length of the commands in batch_obj * @is_master: is the submitting process the drm master? * * Parses the specified batch buffer looking for privilege violations as
@@ -1028,6 +1042,7 @@ int i915_parse_cmds(struct intel_engine_cs *ring,struct drm_i915_gem_object *batch_obj, struct drm_i915_gem_object *shadow_batch_obj, u32 batch_start_offset, + u32 batch_len, bool is_master) { int ret = 0;
@@ -1035,7 +1050,8 @@ int i915_parse_cmds(struct intel_engine_cs *ring,struct drm_i915_cmd_descriptor default_desc = {0 };bool oacontrol_set = false;/* OACONTROL tracking. See check_cmd() */ - batch_base =copy_batch (shadow_batch_obj, batch_obj); + batch_base =copy_batch (shadow_batch_obj, batch_obj, + batch_start_offset, batch_len);if (IS_ERR (batch_base)) {DRM_DEBUG_DRIVER ("CMD: Failed to copy batch\n" );return PTR_ERR (batch_base);
@@ -1044,11 +1060,11 @@ int i915_parse_cmds(struct intel_engine_cs *ring, cmd = batch_base + (batch_start_offset /sizeof (*cmd));/* -* We use the source object's size because the shadow object is as +* We use the batch length as size because the shadow object is as * large or larger and copy_batch() will write MI_NOPs to the extra * space. Parsing should be faster in some cases this way. */ - batch_end = cmd + (batch_obj->base.size /sizeof (*batch_end)); + batch_end = cmd + (batch_len /sizeof (*batch_end));while (cmd < batch_end) {const struct drm_i915_cmd_descriptor *desc;
drivers/gpu/drm/i915/i915_drv.h +1 -0
@@ -2919,6 +2919,7 @@ int i915_parse_cmds(struct intel_engine_cs *ring,struct drm_i915_gem_object *batch_obj, struct drm_i915_gem_object *shadow_batch_obj, u32 batch_start_offset, + u32 batch_len, bool is_master);/* i915_suspend.c */
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -0
@@ -1418,6 +1418,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, batch_obj, shadow_batch_obj, args->batch_start_offset, + args->batch_len, file->is_master);i915_gem_object_ggtt_unpin (shadow_batch_obj);
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 3/5] drm/i915: Use batch length instead of object size in command parser
Quoted 20 lines
-----Original Message-----
From: Nguyen, Michael H
Sent: Thursday, December 11, 2014 8:13 PM
To: [email protected]
Cc: Bloomfield, Jon; Brad Volkin
Subject: [PATCH v7 3/5] drm/i915: Use batch length instead of object size in command parserFrom: Brad Volkin <[email protected]>
Previously we couldn't trust the user-supplied batch length because it came directly from userspace (i.e. untrusted code). It would have affected what commands software parsed without regard to what hardware would actually execute, leaving a potential hole.
With the parser now copying the user supplied batch buffer and writing
MI_NOPcommands to any space after the copied region, we can safely use the batch length input. This should be a performance win as the actual batch length is frequently much smaller than the allocated object size.
Quoted 4 lines
v2: Fix handling of non-zero
batch_start_offsetIssue: VIZ-4719
Signed-off-by Brad Volkin <[email protected]>
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH v7 4/5 drm/i915: Mark shadow batch buffers as purgeable
In reply to [email protected]From: Brad Volkin <[email protected]>
By adding a new exec_entry flag, we cleanly mark the shadow objects as purgeable after they are on the active list.
v2:
- Move 'shadow_batch_obj->madv = I915_MADV_WILLNEED' inside _get fnc (danvet, from v4 6/7 feedback)
v3:
- Remove duplicate 'madv = I915_MADV_WILLNEED' (danvet, from v6 4/5)
Issue: VIZ-4719
--- drivers/gpu/drm/i915/i915_gem_batch_pool.c | 2 ++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 9 ++++++++- 2 files changed, 10 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_gem_batch_pool.c +2 -0
@@ -131,5 +131,7 @@ i915_gem_batch_pool_get(struct i915_gem_batch_pool *pool,/* Keep list in LRU order */ list_move_tail (&obj->batch_pool_list, &pool->cache_list); + obj->madv =I915_MADV_WILLNEED ; +return obj; }
drivers/gpu/drm/i915/i915_gem_execbuffer.c +8 -1
@@ -37,6 +37,7 @@#define __EXEC_OBJECT_HAS_FENCE (1<<30)#define __EXEC_OBJECT_NEEDS_MAP (1<<29)#define __EXEC_OBJECT_NEEDS_BIAS (1<<28) +#define __EXEC_OBJECT_PURGEABLE (1<<27)#define BATCH_OFFSET_BIAS (256*1024)
@@ -226,7 +227,12 @@ i915_gem_execbuffer_unreserve_vma(struct i915_vma *vma)if (entry->flags & __EXEC_OBJECT_HAS_PIN) vma->pin_count--; - entry->flags &= ~(__EXEC_OBJECT_HAS_FENCE | __EXEC_OBJECT_HAS_PIN); +if (entry->flags & __EXEC_OBJECT_PURGEABLE) + obj->madv =I915_MADV_DONTNEED ; + + entry->flags &= ~(__EXEC_OBJECT_HAS_FENCE | + __EXEC_OBJECT_HAS_PIN | + __EXEC_OBJECT_PURGEABLE); }static void eb_destroy (struct eb_vmas *eb)
@@ -1433,6 +1439,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data, vma =i915_gem_obj_to_ggtt (shadow_batch_obj); vma->exec_entry = &shadow_exec_entry; + vma->exec_entry->flags = __EXEC_OBJECT_PURGEABLE;drm_gem_object_reference (&shadow_batch_obj->base);list_add_tail (&vma->exec_list, &eb->vmas);
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 4/5] drm/i915: Mark shadow batch buffers as purgeable
Quoted 16 lines
-----Original Message-----
From: Nguyen, Michael H
Sent: Thursday, December 11, 2014 8:13 PM
To: [email protected]
Cc: Bloomfield, Jon; Brad Volkin
Subject: [PATCH v7 4/5] drm/i915: Mark shadow batch buffers as purgeableFrom: Brad Volkin <[email protected]>
By adding a new
exec_entryflag, we cleanly mark the shadow objects as purgeable after they are on the active list.v2:
- Move 'shadow_batch_obj->madv=I915_MADV_WILLNEED' inside_getfnc (danvet, from v4 6/7 feedback)
Quoted 5 lines
v3:
- Remove duplicate 'madv =I915_MADV_WILLNEED' (danvet, from v6 4/5)Issue: VIZ-4719
Signed-off-by Brad Volkin <[email protected]>
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH v7 5/5 drm/i915: Tidy up execbuffer command parsing code
In reply to [email protected]From: Brad Volkin <[email protected]>
Move it to a separate function since the main do_execbuffer function already has so much going on.
v2:
- Move pin/unpin calls inside i915_parse_cmds() (Chris W, v4 7/7 feedback)
Issue: VIZ-4719
--- drivers/gpu/drm/i915/i915_cmd_parser.c | 8 ++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 124 ++++++++++++++++------------- 2 files changed, 77 insertions(+), 55 deletions(-)
drivers/gpu/drm/i915/i915_cmd_parser.c +8 -0
@@ -1050,10 +1050,17 @@ int i915_parse_cmds(struct intel_engine_cs *ring,struct drm_i915_cmd_descriptor default_desc = {0 };bool oacontrol_set = false;/* OACONTROL tracking. See check_cmd() */ + ret =i915_gem_obj_ggtt_pin (shadow_batch_obj,4096 ,0 ); +if (ret) { +DRM_DEBUG_DRIVER ("CMD: Failed to pin shadow batch\n" ); +return -1 ; + } + batch_base =copy_batch (shadow_batch_obj, batch_obj, batch_start_offset, batch_len);if (IS_ERR (batch_base)) {DRM_DEBUG_DRIVER ("CMD: Failed to copy batch\n" ); +i915_gem_object_ggtt_unpin (shadow_batch_obj);return PTR_ERR (batch_base); }
@@ -1124,6 +1131,7 @@ int i915_parse_cmds(struct intel_engine_cs *ring, }vunmap (batch_base); +i915_gem_object_ggtt_unpin (shadow_batch_obj);return ret; }
drivers/gpu/drm/i915/i915_gem_execbuffer.c +69 -55
144 lines not on this page · raw message
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 5/5] drm/i915: Tidy up execbuffer command parsing code
Quoted 12 lines
-----Original Message-----
From: Nguyen, Michael H
Sent: Thursday, December 11, 2014 8:13 PM
To: [email protected]
Cc: Bloomfield, Jon; Brad Volkin
Subject: [PATCH v7 5/5] drm/i915: Tidy up execbuffer command parsing codeFrom: Brad Volkin <[email protected]>
Move it to a separate function since the main
do_execbufferfunction already has so much going on.
Quoted 6 lines
v2:
- Move pin/unpin calls insidei915_parse_cmds()(Chris W, v4 7/7 feedback)Issue: VIZ-4719
Signed-off-by Brad Volkin <[email protected]>
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 5/5] drm/i915: Tidy up execbuffer command parsing code
Quoted 12 lines · Jon
Quoted 12 lines
-----Original Message-----
From: Nguyen, Michael H
Sent: Thursday, December 11, 2014 8:13 PM
To: [email protected]
Cc: Bloomfield, Jon; Brad Volkin
Subject: [PATCH v7 5/5] drm/i915: Tidy up execbuffer command parsing codeFrom: Brad Volkin <[email protected]>
Move it to a separate function since the main
do_execbufferfunction already has so much going on.
Quoted 8 lines
Quoted 6 lines
v2:
- Move pin/unpin calls insidei915_parse_cmds()(Chris W, v4 7/7 feedback)Issue: VIZ-4719
Signed-off-by Brad Volkin <[email protected]>Reviewed-by Jon Bloomfield <[email protected]>?
Vacuumed up entire series, thanks for patches&review. -Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH v7 5/5] drm/i915: Tidy up execbuffer command parsing code
In reply to [email protected]-------------------------------------Summary-------------------------------------
Platform Delta drm-intel-nightly Series Applied
PNV 364/364 364/364
ILK +1-3 364/366 362/366
SNB 448/450 448/450 IVB 497/498 497/498 BYT 289/289 289/289 HSW 563/564 563/564 BDW 417/417 417/417 -------------------------------------Detailed------------------------------------- Platform Test drm-intel-nightly Series Applied ILK igt_kms_flip_flip-vs-panning DMESG_WARN(1, M26)PASS(8, M26) DMESG_WARN(1, M26) ILK igt_kms_flip_plain-flip-fb-recreate-interruptible DMESG_WARN(1, M26)PASS(5, M26) DMESG_WARN(1, M26) *ILK igt_kms_flip_rcs-flip-vs-panning PASS(5, M26) DMESG_WARN(1, M26) ILK igt_kms_flip_wf_vblank-ts-check DMESG_WARN(8, M26)PASS(26, M26M37) PASS(1, M26)
Note: You need to pay more attention to line start with '*'
_______________________________________________
Intel-gfx mailing list [email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx