diff options
author | Thomas Hellstrom <thellstrom@vmware.com> | 2009-08-17 16:28:39 +0200 |
---|---|---|
committer | Dave Airlie <airlied@linux.ie> | 2009-08-19 16:09:53 +1000 |
commit | 5fd9cbad3a4ae82c83c55b9c621d156c326724ef (patch) | |
tree | 1a0868a3bd2751fa861c083aeb3ac27f3f695694 /drivers | |
parent | e9840be8c23601285a70520b4898818f28ce8c2b (diff) |
drm/ttm: Memory accounting rework.
Use inclusive zones to simplify accounting and its sysfs representation.
Use DMA32 accounting where applicable.
Add a sysfs interface to make the heuristically determined limits
readable and configurable.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Diffstat (limited to 'drivers')
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo.c | 6 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_global.c | 4 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_memory.c | 488 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_tt.c | 29 |
4 files changed, 427 insertions, 100 deletions
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index c1c407f7cca3..f16909ceec93 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -70,7 +70,7 @@ static void ttm_bo_release_list(struct kref *list_kref) if (bo->destroy) bo->destroy(bo); else { - ttm_mem_global_free(bdev->mem_glob, bo->acc_size, false); + ttm_mem_global_free(bdev->mem_glob, bo->acc_size); kfree(bo); } } @@ -1065,14 +1065,14 @@ int ttm_buffer_object_create(struct ttm_bo_device *bdev, size_t acc_size = ttm_bo_size(bdev, (size + PAGE_SIZE - 1) >> PAGE_SHIFT); - ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false, false); + ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false); if (unlikely(ret != 0)) return ret; bo = kzalloc(sizeof(*bo), GFP_KERNEL); if (unlikely(bo == NULL)) { - ttm_mem_global_free(mem_glob, acc_size, false); + ttm_mem_global_free(mem_glob, acc_size); return -ENOMEM; } diff --git a/drivers/gpu/drm/ttm/ttm_global.c b/drivers/gpu/drm/ttm/ttm_global.c index 0b14eb1972b8..541744d00d3e 100644 --- a/drivers/gpu/drm/ttm/ttm_global.c +++ b/drivers/gpu/drm/ttm/ttm_global.c @@ -71,7 +71,7 @@ int ttm_global_item_ref(struct ttm_global_reference *ref) mutex_lock(&item->mutex); if (item->refcount == 0) { - item->object = kmalloc(ref->size, GFP_KERNEL); + item->object = kzalloc(ref->size, GFP_KERNEL); if (unlikely(item->object == NULL)) { ret = -ENOMEM; goto out_err; @@ -89,7 +89,6 @@ int ttm_global_item_ref(struct ttm_global_reference *ref) mutex_unlock(&item->mutex); return 0; out_err: - kfree(item->object); mutex_unlock(&item->mutex); item->object = NULL; return ret; @@ -105,7 +104,6 @@ void ttm_global_item_unref(struct ttm_global_reference *ref) BUG_ON(ref->object != item->object); if (--item->refcount == 0) { ref->release(ref); - kfree(item->object); item->object = NULL; } mutex_unlock(&item->mutex); diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c index 87323d4ff68d..62fb5cf0899e 100644 --- a/drivers/gpu/drm/ttm/ttm_memory.c +++ b/drivers/gpu/drm/ttm/ttm_memory.c @@ -26,15 +26,180 @@ **************************************************************************/ #include "ttm/ttm_memory.h" +#include "ttm/ttm_module.h" #include <linux/spinlock.h> #include <linux/sched.h> #include <linux/wait.h> #include <linux/mm.h> #include <linux/module.h> -#define TTM_PFX "[TTM] " #define TTM_MEMORY_ALLOC_RETRIES 4 +struct ttm_mem_zone { + struct kobject kobj; + struct ttm_mem_global *glob; + const char *name; + uint64_t zone_mem; + uint64_t emer_mem; + uint64_t max_mem; + uint64_t swap_limit; + uint64_t used_mem; +}; + +static struct attribute ttm_mem_sys = { + .name = "zone_memory", + .mode = S_IRUGO +}; +static struct attribute ttm_mem_emer = { + .name = "emergency_memory", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_mem_max = { + .name = "available_memory", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_mem_swap = { + .name = "swap_limit", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_mem_used = { + .name = "used_memory", + .mode = S_IRUGO +}; + +static void ttm_mem_zone_kobj_release(struct kobject *kobj) +{ + struct ttm_mem_zone *zone = + container_of(kobj, struct ttm_mem_zone, kobj); + + printk(KERN_INFO TTM_PFX + "Zone %7s: Used memory at exit: %llu kiB.\n", + zone->name, (unsigned long long) zone->used_mem >> 10); + kfree(zone); +} + +static ssize_t ttm_mem_zone_show(struct kobject *kobj, + struct attribute *attr, + char *buffer) +{ + struct ttm_mem_zone *zone = + container_of(kobj, struct ttm_mem_zone, kobj); + uint64_t val = 0; + + spin_lock(&zone->glob->lock); + if (attr == &ttm_mem_sys) + val = zone->zone_mem; + else if (attr == &ttm_mem_emer) + val = zone->emer_mem; + else if (attr == &ttm_mem_max) + val = zone->max_mem; + else if (attr == &ttm_mem_swap) + val = zone->swap_limit; + else if (attr == &ttm_mem_used) + val = zone->used_mem; + spin_unlock(&zone->glob->lock); + + return snprintf(buffer, PAGE_SIZE, "%llu\n", + (unsigned long long) val >> 10); +} + +static void ttm_check_swapping(struct ttm_mem_global *glob); + +static ssize_t ttm_mem_zone_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, + size_t size) +{ + struct ttm_mem_zone *zone = + container_of(kobj, struct ttm_mem_zone, kobj); + int chars; + unsigned long val; + uint64_t val64; + + chars = sscanf(buffer, "%lu", &val); + if (chars == 0) + return size; + + val64 = val; + val64 <<= 10; + + spin_lock(&zone->glob->lock); + if (val64 > zone->zone_mem) + val64 = zone->zone_mem; + if (attr == &ttm_mem_emer) { + zone->emer_mem = val64; + if (zone->max_mem > val64) + zone->max_mem = val64; + } else if (attr == &ttm_mem_max) { + zone->max_mem = val64; + if (zone->emer_mem < val64) + zone->emer_mem = val64; + } else if (attr == &ttm_mem_swap) + zone->swap_limit = val64; + spin_unlock(&zone->glob->lock); + + ttm_check_swapping(zone->glob); + + return size; +} + +static struct attribute *ttm_mem_zone_attrs[] = { + &ttm_mem_sys, + &ttm_mem_emer, + &ttm_mem_max, + &ttm_mem_swap, + &ttm_mem_used, + NULL +}; + +static struct sysfs_ops ttm_mem_zone_ops = { + .show = &ttm_mem_zone_show, + .store = &ttm_mem_zone_store +}; + +static struct kobj_type ttm_mem_zone_kobj_type = { + .release = &ttm_mem_zone_kobj_release, + .sysfs_ops = &ttm_mem_zone_ops, + .default_attrs = ttm_mem_zone_attrs, +}; + +static void ttm_mem_global_kobj_release(struct kobject *kobj) +{ + struct ttm_mem_global *glob = + container_of(kobj, struct ttm_mem_global, kobj); + + kfree(glob); +} + +static struct kobj_type ttm_mem_glob_kobj_type = { + .release = &ttm_mem_global_kobj_release, +}; + +static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob, + bool from_wq, uint64_t extra) +{ + unsigned int i; + struct ttm_mem_zone *zone; + uint64_t target; + + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + + if (from_wq) + target = zone->swap_limit; + else if (capable(CAP_SYS_ADMIN)) + target = zone->emer_mem; + else + target = zone->max_mem; + + target = (extra > target) ? 0ULL : target; + + if (zone->used_mem > target) + return true; + } + return false; +} + /** * At this point we only support a single shrink callback. * Extend this if needed, perhaps using a linked list of callbacks. @@ -42,34 +207,17 @@ * many threads may try to swap out at any given time. */ -static void ttm_shrink(struct ttm_mem_global *glob, bool from_workqueue, +static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq, uint64_t extra) { int ret; struct ttm_mem_shrink *shrink; - uint64_t target; - uint64_t total_target; spin_lock(&glob->lock); if (glob->shrink == NULL) goto out; - if (from_workqueue) { - target = glob->swap_limit; - total_target = glob->total_memory_swap_limit; - } else if (capable(CAP_SYS_ADMIN)) { - total_target = glob->emer_total_memory; - target = glob->emer_memory; - } else { - total_target = glob->max_total_memory; - target = glob->max_memory; - } - - total_target = (extra >= total_target) ? 0 : total_target - extra; - target = (extra >= target) ? 0 : target - extra; - - while (glob->used_memory > target || - glob->used_total_memory > total_target) { + while (ttm_zones_above_swap_target(glob, from_wq, extra)) { shrink = glob->shrink; spin_unlock(&glob->lock); ret = shrink->do_shrink(shrink); @@ -81,6 +229,8 @@ out: spin_unlock(&glob->lock); } + + static void ttm_shrink_work(struct work_struct *work) { struct ttm_mem_global *glob = @@ -89,63 +239,178 @@ static void ttm_shrink_work(struct work_struct *work) ttm_shrink(glob, true, 0ULL); } +static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob, + const struct sysinfo *si) +{ + struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL); + uint64_t mem; + + if (unlikely(!zone)) + return -ENOMEM; + + mem = si->totalram - si->totalhigh; + mem *= si->mem_unit; + + zone->name = "kernel"; + zone->zone_mem = mem; + zone->max_mem = mem >> 1; + zone->emer_mem = (mem >> 1) + (mem >> 2); + zone->swap_limit = zone->max_mem - (mem >> 3); + zone->used_mem = 0; + zone->glob = glob; + glob->zone_kernel = zone; + glob->zones[glob->num_zones++] = zone; + kobject_init(&zone->kobj, &ttm_mem_zone_kobj_type); + return kobject_add(&zone->kobj, &glob->kobj, zone->name); +} + +#ifdef CONFIG_HIGHMEM +static int ttm_mem_init_highmem_zone(struct ttm_mem_global *glob, + const struct sysinfo *si) +{ + struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL); + uint64_t mem; + + if (unlikely(!zone)) + return -ENOMEM; + + if (si->totalhigh == 0) + return 0; + + mem = si->totalram; + mem *= si->mem_unit; + + zone->name = "highmem"; + zone->zone_mem = mem; + zone->max_mem = mem >> 1; + zone->emer_mem = (mem >> 1) + (mem >> 2); + zone->swap_limit = zone->max_mem - (mem >> 3); + zone->used_mem = 0; + zone->glob = glob; + glob->zone_highmem = zone; + glob->zones[glob->num_zones++] = zone; + kobject_init(&zone->kobj, &ttm_mem_zone_kobj_type); + return kobject_add(&zone->kobj, &glob->kobj, zone->name); +} +#else +static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob, + const struct sysinfo *si) +{ + struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL); + uint64_t mem; + + if (unlikely(!zone)) + return -ENOMEM; + + mem = si->totalram; + mem *= si->mem_unit; + + /** + * No special dma32 zone needed. + */ + + if (mem <= ((uint64_t) 1ULL << 32)) + return 0; + + /* + * Limit max dma32 memory to 4GB for now + * until we can figure out how big this + * zone really is. + */ + + mem = ((uint64_t) 1ULL << 32); + zone->name = "dma32"; + zone->zone_mem = mem; + zone->max_mem = mem >> 1; + zone->emer_mem = (mem >> 1) + (mem >> 2); + zone->swap_limit = zone->max_mem - (mem >> 3); + zone->used_mem = 0; + zone->glob = glob; + glob->zone_dma32 = zone; + glob->zones[glob->num_zones++] = zone; + kobject_init(&zone->kobj, &ttm_mem_zone_kobj_type); + return kobject_add(&zone->kobj, &glob->kobj, zone->name); +} +#endif + int ttm_mem_global_init(struct ttm_mem_global *glob) { struct sysinfo si; - uint64_t mem; + int ret; + int i; + struct ttm_mem_zone *zone; spin_lock_init(&glob->lock); glob->swap_queue = create_singlethread_workqueue("ttm_swap"); INIT_WORK(&glob->work, ttm_shrink_work); init_waitqueue_head(&glob->queue); + kobject_init(&glob->kobj, &ttm_mem_glob_kobj_type); + ret = kobject_add(&glob->kobj, + ttm_get_kobj(), + "memory_accounting"); + if (unlikely(ret != 0)) + goto out_no_zone; si_meminfo(&si); - mem = si.totalram - si.totalhigh; - mem *= si.mem_unit; - - glob->max_memory = mem >> 1; - glob->emer_memory = (mem >> 1) + (mem >> 2); - glob->swap_limit = glob->max_memory - (mem >> 3); - glob->used_memory = 0; - glob->used_total_memory = 0; - glob->shrink = NULL; - - mem = si.totalram; - mem *= si.mem_unit; - - glob->max_total_memory = mem >> 1; - glob->emer_total_memory = (mem >> 1) + (mem >> 2); - - glob->total_memory_swap_limit = glob->max_total_memory - (mem >> 3); - - printk(KERN_INFO TTM_PFX "TTM available graphics memory: %llu MiB\n", - glob->max_total_memory >> 20); - printk(KERN_INFO TTM_PFX "TTM available object memory: %llu MiB\n", - glob->max_memory >> 20); - + ret = ttm_mem_init_kernel_zone(glob, &si); + if (unlikely(ret != 0)) + goto out_no_zone; +#ifdef CONFIG_HIGHMEM + ret = ttm_mem_init_highmem_zone(glob, &si); + if (unlikely(ret != 0)) + goto out_no_zone; +#else + ret = ttm_mem_init_dma32_zone(glob, &si); + if (unlikely(ret != 0)) + goto out_no_zone; +#endif + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + printk(KERN_INFO TTM_PFX + "Zone %7s: Available graphics memory: %llu kiB.\n", + zone->name, (unsigned long long) zone->max_mem >> 10); + } return 0; +out_no_zone: + ttm_mem_global_release(glob); + return ret; } EXPORT_SYMBOL(ttm_mem_global_init); void ttm_mem_global_release(struct ttm_mem_global *glob) { - printk(KERN_INFO TTM_PFX "Used total memory is %llu bytes.\n", - (unsigned long long)glob->used_total_memory); + unsigned int i; + struct ttm_mem_zone *zone; + flush_workqueue(glob->swap_queue); destroy_workqueue(glob->swap_queue); glob->swap_queue = NULL; + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + kobject_del(&zone->kobj); + kobject_put(&zone->kobj); + } + kobject_del(&glob->kobj); + kobject_put(&glob->kobj); } EXPORT_SYMBOL(ttm_mem_global_release); -static inline void ttm_check_swapping(struct ttm_mem_global *glob) +static void ttm_check_swapping(struct ttm_mem_global *glob) { - bool needs_swapping; + bool needs_swapping = false; + unsigned int i; + struct ttm_mem_zone *zone; spin_lock(&glob->lock); - needs_swapping = (glob->used_memory > glob->swap_limit || - glob->used_total_memory > - glob->total_memory_swap_limit); + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (zone->used_mem > zone->swap_limit) { + needs_swapping = true; + break; + } + } + spin_unlock(&glob->lock); if (unlikely(needs_swapping)) @@ -153,44 +418,60 @@ static inline void ttm_check_swapping(struct ttm_mem_global *glob) } -void ttm_mem_global_free(struct ttm_mem_global *glob, - uint64_t amount, bool himem) +static void ttm_mem_global_free_zone(struct ttm_mem_global *glob, + struct ttm_mem_zone *single_zone, + uint64_t amount) { + unsigned int i; + struct ttm_mem_zone *zone; + spin_lock(&glob->lock); - glob->used_total_memory -= amount; - if (!himem) - glob->used_memory -= amount; - wake_up_all(&glob->queue); + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (single_zone && zone != single_zone) + continue; + zone->used_mem -= amount; + } spin_unlock(&glob->lock); } +void ttm_mem_global_free(struct ttm_mem_global *glob, + uint64_t amount) +{ + return ttm_mem_global_free_zone(glob, NULL, amount); +} + static int ttm_mem_global_reserve(struct ttm_mem_global *glob, - uint64_t amount, bool himem, bool reserve) + struct ttm_mem_zone *single_zone, + uint64_t amount, bool reserve) { uint64_t limit; - uint64_t lomem_limit; int ret = -ENOMEM; + unsigned int i; + struct ttm_mem_zone *zone; spin_lock(&glob->lock); + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (single_zone && zone != single_zone) + continue; - if (capable(CAP_SYS_ADMIN)) { - limit = glob->emer_total_memory; - lomem_limit = glob->emer_memory; - } else { - limit = glob->max_total_memory; - lomem_limit = glob->max_memory; - } + limit = (capable(CAP_SYS_ADMIN)) ? + zone->emer_mem : zone->max_mem; - if (unlikely(glob->used_total_memory + amount > limit)) - goto out_unlock; - if (unlikely(!himem && glob->used_memory + amount > lomem_limit)) - goto out_unlock; + if (zone->used_mem > limit) + goto out_unlock; + } if (reserve) { - glob->used_total_memory += amount; - if (!himem) - glob->used_memory += amount; + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (single_zone && zone != single_zone) + continue; + zone->used_mem += amount; + } } + ret = 0; out_unlock: spin_unlock(&glob->lock); @@ -199,12 +480,17 @@ out_unlock: return ret; } -int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory, - bool no_wait, bool interruptible, bool himem) + +static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob, + struct ttm_mem_zone *single_zone, + uint64_t memory, + bool no_wait, bool interruptible) { int count = TTM_MEMORY_ALLOC_RETRIES; - while (unlikely(ttm_mem_global_reserve(glob, memory, himem, true) + while (unlikely(ttm_mem_global_reserve(glob, + single_zone, + memory, true) != 0)) { if (no_wait) return -ENOMEM; @@ -216,6 +502,56 @@ int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory, return 0; } +int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory, + bool no_wait, bool interruptible) +{ + /** + * Normal allocations of kernel memory are registered in + * all zones. + */ + + return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait, + interruptible); +} + +int ttm_mem_global_alloc_page(struct ttm_mem_global *glob, + struct page *page, + bool no_wait, bool interruptible) +{ + + struct ttm_mem_zone *zone = NULL; + + /** + * Page allocations may be registed in a single zone + * only if highmem or !dma32. + */ + +#ifdef CONFIG_HIGHMEM + if (PageHighMem(page) && glob->zone_highmem != NULL) + zone = glob->zone_highmem; +#else + if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL) + zone = glob->zone_kernel; +#endif + return ttm_mem_global_alloc_zone(glob, zone, PAGE_SIZE, no_wait, + interruptible); +} + +void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page) +{ + struct ttm_mem_zone *zone = NULL; + +#ifdef CONFIG_HIGHMEM + if (PageHighMem(page) && glob->zone_highmem != NULL) + zone = glob->zone_highmem; +#else + if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL) + zone = glob->zone_kernel; +#endif + ttm_mem_global_free_zone(glob, zone, PAGE_SIZE); +} + + size_t ttm_round_pot(size_t size) { if ((size & (size - 1)) == 0) diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index 75dc8bd24592..4e1e2566d519 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -166,7 +166,7 @@ static void ttm_tt_free_user_pages(struct ttm_tt *ttm) set_page_dirty_lock(page); ttm->pages[i] = NULL; - ttm_mem_global_free(ttm->bdev->mem_glob, PAGE_SIZE, false); + ttm_mem_global_free(ttm->bdev->mem_glob, PAGE_SIZE); put_page(page); } ttm->state = tt_unpopulated; @@ -187,21 +187,14 @@ static struct page *__ttm_tt_get_page(struct ttm_tt *ttm, int index) if (!p) return NULL; - if (PageHighMem(p)) { - ret = - ttm_mem_global_alloc(mem_glob, PAGE_SIZE, - false, false, true); - if (unlikely(ret != 0)) - goto out_err; + ret = ttm_mem_global_alloc_page(mem_glob, p, false, false); + if (unlikely(ret != 0)) + goto out_err; + + if (PageHighMem(p)) ttm->pages[--ttm->first_himem_page] = p; - } else { - ret = - ttm_mem_global_alloc(mem_glob, PAGE_SIZE, - false, false, false); - if (unlikely(ret != 0)) - goto out_err; + else ttm->pages[++ttm->last_lomem_page] = p; - } } return p; out_err: @@ -355,8 +348,8 @@ static void ttm_tt_free_alloced_pages(struct ttm_tt *ttm) printk(KERN_ERR TTM_PFX "Erroneous page count. " "Leaking pages.\n"); - ttm_mem_global_free(ttm->bdev->mem_glob, PAGE_SIZE, - PageHighMem(cur_page)); + ttm_mem_global_free_page(ttm->bdev->mem_glob, + cur_page); __free_page(cur_page); } } @@ -411,7 +404,7 @@ int ttm_tt_set_user(struct ttm_tt *ttm, */ ret = ttm_mem_global_alloc(mem_glob, num_pages * PAGE_SIZE, - false, false, false); + false, false); if (unlikely(ret != 0)) return ret; @@ -422,7 +415,7 @@ int ttm_tt_set_user(struct ttm_tt *ttm, if (ret != num_pages && write) { ttm_tt_free_user_pages(ttm); - ttm_mem_global_free(mem_glob, num_pages * PAGE_SIZE, false); + ttm_mem_global_free(mem_glob, num_pages * PAGE_SIZE); return -ENOMEM; } |