The patch series mainly contains minor optimizations and documentation, except for patch 5/6 which is really a bugfix, but the synchronization bug shouldn't affect x86 processors.
Remove an obsolete comment about mm nodes. Document the new bo range manager interface.
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com --- drivers/gpu/drm/ttm/ttm_bo.c | 8 ---- include/drm/ttm/ttm_bo_driver.h | 79 ++++++++++++++++++++++++++++++++++++-- 2 files changed, 74 insertions(+), 13 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index a1cb783..cf47978 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -27,14 +27,6 @@ /* * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com> */ -/* Notes: - * - * We store bo pointer in drm_mm_node struct so we know which bo own a - * specific node. There is no protection on the pointer, thus to make - * sure things don't go berserk you have to access this pointer while - * holding the global lru lock and make sure anytime you free a node you - * reset the pointer to NULL. - */
#include "ttm/ttm_module.h" #include "ttm/ttm_bo_driver.h" diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index d01b4dd..8e0c848 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -206,14 +206,84 @@ struct ttm_tt { struct ttm_mem_type_manager;
struct ttm_mem_type_manager_func { + /** + * struct ttm_mem_type_manager member init + * + * @man: Pointer to a memory type manager. + * @p_size: Implementation dependent, but typically the size of the + * range to be managed in pages. + * + * Called to initialize a private range manager. The function is + * expected to initialize the man::priv member. + * Returns 0 on success, negative error code on failure. + */ int (*init)(struct ttm_mem_type_manager *man, unsigned long p_size); + + /** + * struct ttm_mem_type_manager member takedown + * + * @man: Pointer to a memory type manager. + * + * Called to undo the setup done in init. All allocated resources + * should be freed. + */ int (*takedown)(struct ttm_mem_type_manager *man); + + /** + * struct ttm_mem_type_manager member get_node + * + * @man: Pointer to a memory type manager. + * @bo: Pointer to the buffer object we're allocating space for. + * @placement: Placement details. + * @mem: Pointer to a struct ttm_mem_reg to be filled in. + * + * This function should allocate space in the memory type managed + * by @man. Placement details if + * applicable are given by @placement. If successful, + * @mem::mm_node should be set to a non-null value, and + * @mem::start should be set to a value identifying the beginning + * of the range allocated, and the function should return zero. + * If the memory region accomodate the buffer object, @mem::mm_node + * should be set to NULL, and the function should return 0. + * If a system error occured, preventing the request to be fulfilled, + * the function should return a negative error code. + * + * Note that @mem::mm_node will only be dereferenced by + * struct ttm_mem_type_manager functions and optionally by the driver, + * which has knowledge of the underlying type. + * + * This function may not be called from within atomic context, so + * an implementation can and must use either a mutex or a spinlock to + * protect any data structures managing the space. + */ int (*get_node)(struct ttm_mem_type_manager *man, struct ttm_buffer_object *bo, struct ttm_placement *placement, struct ttm_mem_reg *mem); + + /** + * struct ttm_mem_type_manager member put_node + * + * @man: Pointer to a memory type manager. + * @mem: Pointer to a struct ttm_mem_reg to be filled in. + * + * This function frees memory type resources previously allocated + * and that are identified by @mem::mm_node and @mem::start. May not + * be called from within atomic context. + */ void (*put_node)(struct ttm_mem_type_manager *man, struct ttm_mem_reg *mem); + + /** + * struct ttm_mem_type_manager member debug + * + * @man: Pointer to a memory type manager. + * @prefix: Prefix to be used in printout to identify the caller. + * + * This function is called to print out the state of the memory + * type manager to aid debugging of out-of-memory conditions. + * It may not be called from within atomic context. + */ void (*debug)(struct ttm_mem_type_manager *man, const char *prefix); };
@@ -231,14 +301,13 @@ struct ttm_mem_type_manager { uint64_t size; uint32_t available_caching; uint32_t default_caching; + const struct ttm_mem_type_manager_func *func; + void *priv;
/* - * Protected by the bdev->lru_lock. - * TODO: Consider one lru_lock per ttm_mem_type_manager. - * Plays ill with list removal, though. + * Protected by the global->lru_lock. */ - const struct ttm_mem_type_manager_func *func; - void *priv; + struct list_head lru; };
Searching for a free block in the range manager may in some situations be a lenghty operation, and we want to avoid holding the global lru lock during that time. Instead use a per-manager spinlock.
This leaves the global lru lock for quick lru list and swap list manipulation only, including list manipulation associated with reserving buffer objects.
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com --- drivers/gpu/drm/ttm/ttm_bo_manager.c | 81 +++++++++++++++++++--------------- 1 files changed, 45 insertions(+), 36 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c index 7410c19..038e947 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_manager.c +++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright (c) 2007-2009 VMware, Inc., Palo Alto, CA., USA + * Copyright (c) 2007-2010 VMware, Inc., Palo Alto, CA., USA * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -31,20 +31,29 @@ #include "ttm/ttm_module.h" #include "ttm/ttm_bo_driver.h" #include "ttm/ttm_placement.h" -#include <linux/jiffies.h> +#include "drm_mm.h" #include <linux/slab.h> -#include <linux/sched.h> -#include <linux/mm.h> -#include <linux/file.h> +#include <linux/spinlock.h> #include <linux/module.h>
+/** + * Currently we use a spinlock for the lock, but a mutex *may* be + * more appropriate to reduce scheduling latency if the range manager + * ends up with very fragmented allocation patterns. + */ + +struct ttm_range_manager { + struct drm_mm mm; + spinlock_t lock; +}; + static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, struct ttm_buffer_object *bo, struct ttm_placement *placement, struct ttm_mem_reg *mem) { - struct ttm_bo_global *glob = man->bdev->glob; - struct drm_mm *mm = man->priv; + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; + struct drm_mm *mm = &rman->mm; struct drm_mm_node *node = NULL; unsigned long lpfn; int ret; @@ -57,19 +66,19 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, if (unlikely(ret)) return ret;
- spin_lock(&glob->lru_lock); + spin_lock(&rman->lock); node = drm_mm_search_free_in_range(mm, mem->num_pages, mem->page_alignment, placement->fpfn, lpfn, 1); if (unlikely(node == NULL)) { - spin_unlock(&glob->lru_lock); + spin_unlock(&rman->lock); return 0; } node = drm_mm_get_block_atomic_range(node, mem->num_pages, - mem->page_alignment, - placement->fpfn, - lpfn); - spin_unlock(&glob->lru_lock); + mem->page_alignment, + placement->fpfn, + lpfn); + spin_unlock(&rman->lock); } while (node == NULL);
mem->mm_node = node; @@ -80,12 +89,12 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, static void ttm_bo_man_put_node(struct ttm_mem_type_manager *man, struct ttm_mem_reg *mem) { - struct ttm_bo_global *glob = man->bdev->glob; + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
if (mem->mm_node) { - spin_lock(&glob->lru_lock); + spin_lock(&rman->lock); drm_mm_put_block(mem->mm_node); - spin_unlock(&glob->lru_lock); + spin_unlock(&rman->lock); mem->mm_node = NULL; } } @@ -93,49 +102,49 @@ static void ttm_bo_man_put_node(struct ttm_mem_type_manager *man, static int ttm_bo_man_init(struct ttm_mem_type_manager *man, unsigned long p_size) { - struct drm_mm *mm; + struct ttm_range_manager *rman; int ret;
- mm = kzalloc(sizeof(*mm), GFP_KERNEL); - if (!mm) + rman = kzalloc(sizeof(*rman), GFP_KERNEL); + if (!rman) return -ENOMEM;
- ret = drm_mm_init(mm, 0, p_size); + ret = drm_mm_init(&rman->mm, 0, p_size); if (ret) { - kfree(mm); + kfree(rman); return ret; }
- man->priv = mm; + spin_lock_init(&rman->lock); + man->priv = rman; return 0; }
static int ttm_bo_man_takedown(struct ttm_mem_type_manager *man) { - struct ttm_bo_global *glob = man->bdev->glob; - struct drm_mm *mm = man->priv; - int ret = 0; + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; + struct drm_mm *mm = &rman->mm;
- spin_lock(&glob->lru_lock); + spin_lock(&rman->lock); if (drm_mm_clean(mm)) { drm_mm_takedown(mm); - kfree(mm); + spin_unlock(&rman->lock); + kfree(rman); man->priv = NULL; - } else - ret = -EBUSY; - spin_unlock(&glob->lru_lock); - return ret; + return 0; + } + spin_unlock(&rman->lock); + return -EBUSY; }
static void ttm_bo_man_debug(struct ttm_mem_type_manager *man, const char *prefix) { - struct ttm_bo_global *glob = man->bdev->glob; - struct drm_mm *mm = man->priv; + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
- spin_lock(&glob->lru_lock); - drm_mm_debug_table(mm, prefix); - spin_unlock(&glob->lru_lock); + spin_lock(&rman->lock); + drm_mm_debug_table(&rman->mm, prefix); + spin_unlock(&rman->lock); }
const struct ttm_mem_type_manager_func ttm_bo_manager_func = {
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com --- drivers/gpu/drm/ttm/ttm_bo.c | 7 ------- 1 files changed, 0 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index cf47978..e6cedf4 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -814,7 +814,6 @@ static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo, bool no_wait_gpu) { struct ttm_bo_device *bdev = bo->bdev; - struct ttm_bo_global *glob = bdev->glob; struct ttm_mem_type_manager *man = &bdev->man[mem_type]; int ret;
@@ -824,12 +823,6 @@ static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo, return ret; if (mem->mm_node) break; - spin_lock(&glob->lru_lock); - if (list_empty(&man->lru)) { - spin_unlock(&glob->lru_lock); - break; - } - spin_unlock(&glob->lru_lock); ret = ttm_mem_evict_first(bdev, mem_type, interruptible, no_wait_reserve, no_wait_gpu); if (unlikely(ret != 0))
Replace with BUG_ON(). These error messages remained from the time when TTM was initialized from user-space. Nowadays hitting one of those is really a kernel bug.
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com --- drivers/gpu/drm/ttm/ttm_bo.c | 20 ++------------------ 1 files changed, 2 insertions(+), 18 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index e6cedf4..f561eea 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -1354,18 +1354,9 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type, int ret = -EINVAL; struct ttm_mem_type_manager *man;
- if (type >= TTM_NUM_MEM_TYPES) { - printk(KERN_ERR TTM_PFX "Illegal memory type %d\n", type); - return ret; - } - + BUG_ON(type >= TTM_NUM_MEM_TYPES); man = &bdev->man[type]; - if (man->has_type) { - printk(KERN_ERR TTM_PFX - "Memory manager already initialized for type %d\n", - type); - return ret; - } + BUG_ON(man->has_type);
ret = bdev->driver->init_mem_type(bdev, type, man); if (ret) @@ -1374,13 +1365,6 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
ret = 0; if (type != TTM_PL_SYSTEM) { - if (!p_size) { - printk(KERN_ERR TTM_PFX - "Zero size memory manager type %d\n", - type); - return ret; - } - ret = (*man->func->init)(man, p_size); if (ret) return ret;
Since we're doing this outside of a spinlock to provide the necessary barriers, add an explicit barrier.
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com --- drivers/gpu/drm/ttm/ttm_bo.c | 6 ++++++ 1 files changed, 6 insertions(+), 0 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index f561eea..a32fe41 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -37,6 +37,7 @@ #include <linux/mm.h> #include <linux/file.h> #include <linux/module.h> +#include <asm/atomic.h>
#define TTM_ASSERT_LOCKED(param) #define TTM_DEBUG(fmt, arg...) @@ -444,6 +445,11 @@ static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo) ttm_bo_mem_put(bo, &bo->mem);
atomic_set(&bo->reserved, 0); + + /* + * Make processes trying to reserve really pick it up. + */ + smp_mb__after_atomic_dec(); wake_up_all(&bo->event_queue); }
The driver (for example vmwgfx) may want to silently deal with the error itself.
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com --- drivers/gpu/drm/ttm/ttm_tt.c | 4 +--- 1 files changed, 1 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index a7bab87..af789dc 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -440,10 +440,8 @@ int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem) return ret;
ret = be->func->bind(be, bo_mem); - if (ret) { - printk(KERN_ERR TTM_PFX "Couldn't bind backend.\n"); + if (unlikely(ret != 0)) return ret; - }
ttm->state = tt_bound;
dri-devel@lists.freedesktop.org
-
Thomas Hellstrom