* TTM/Nouveau cleanups
@ 2020-01-24 14:30 Christian König
2020-01-24 14:30 ` [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2 Christian König
2020-01-24 14:30 ` [PATCH 2/2] drm/ttm: remove io_reserve_lru handling Christian König
0 siblings, 2 replies; 5+ messages in thread
From: Christian König @ 2020-01-24 14:30 UTC (permalink / raw)
To: bskeggs, airlied, imirkin, dri-devel, nouveau
Hi guys,
I've already send this out in September last year, but only got a response from Daniel.
Could you guys please test this and tell me what you think about it?
Basically I'm trying to remove all driver specific features from TTM which don't need to be inside the framework.
Thanks,
Christian.
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2
2020-01-24 14:30 TTM/Nouveau cleanups Christian König
@ 2020-01-24 14:30 ` Christian König
2020-01-28 5:49 ` [Nouveau] " Ben Skeggs
2020-01-24 14:30 ` [PATCH 2/2] drm/ttm: remove io_reserve_lru handling Christian König
1 sibling, 1 reply; 5+ messages in thread
From: Christian König @ 2020-01-24 14:30 UTC (permalink / raw)
To: bskeggs, airlied, imirkin, dri-devel, nouveau
From: Christian König <ckoenig.leichtzumerken@gmail.com>
While working on TTM cleanups I've found that the io_reserve_lru used by
Nouveau is actually not working at all.
In general we should remove driver specific handling from the memory
management, so this patch moves the io_reserve_lru handling into Nouveau
instead.
The patch should be functional correct, but is only compile tested!
v2: don't call ttm_bo_unmap_virtual in nouveau_ttm_io_mem_reserve
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
drivers/gpu/drm/nouveau/nouveau_bo.c | 107 ++++++++++++++++++++------
drivers/gpu/drm/nouveau/nouveau_bo.h | 3 +
drivers/gpu/drm/nouveau/nouveau_drv.h | 2 +
drivers/gpu/drm/nouveau/nouveau_ttm.c | 43 ++++++++++-
4 files changed, 131 insertions(+), 24 deletions(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 81668104595f..acee054f77ed 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -137,6 +137,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
struct nouveau_bo *nvbo = nouveau_bo(bo);
WARN_ON(nvbo->pin_refcnt > 0);
+ nouveau_bo_del_io_reserve_lru(bo);
nv10_bo_put_tile_region(dev, nvbo->tile, NULL);
/*
@@ -304,6 +305,7 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 flags,
nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
nouveau_bo_placement_set(nvbo, flags, 0);
+ INIT_LIST_HEAD(&nvbo->io_reserve_lru);
ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type,
&nvbo->placement, align >> PAGE_SHIFT, false,
@@ -574,6 +576,26 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo)
PAGE_SIZE, DMA_FROM_DEVICE);
}
+void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo)
+{
+ struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
+ struct nouveau_bo *nvbo = nouveau_bo(bo);
+
+ mutex_lock(&drm->ttm.io_reserve_mutex);
+ list_move_tail(&nvbo->io_reserve_lru, &drm->ttm.io_reserve_lru);
+ mutex_unlock(&drm->ttm.io_reserve_mutex);
+}
+
+void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo)
+{
+ struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
+ struct nouveau_bo *nvbo = nouveau_bo(bo);
+
+ mutex_lock(&drm->ttm.io_reserve_mutex);
+ list_del_init(&nvbo->io_reserve_lru);
+ mutex_unlock(&drm->ttm.io_reserve_mutex);
+}
+
int
nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
bool no_wait_gpu)
@@ -675,8 +697,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
}
man->func = &nouveau_vram_manager;
- man->io_reserve_fastpath = false;
- man->use_io_reserve_lru = true;
} else {
man->func = &ttm_bo_manager_func;
}
@@ -1305,6 +1325,8 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
if (bo->destroy != nouveau_bo_del_ttm)
return;
+ nouveau_bo_del_io_reserve_lru(bo);
+
if (mem && new_reg->mem_type != TTM_PL_SYSTEM &&
mem->mem.page == nvbo->page) {
list_for_each_entry(vma, &nvbo->vma_list, head) {
@@ -1427,6 +1449,30 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp)
filp->private_data);
}
+static void
+nouveau_ttm_io_mem_free_locked(struct nouveau_drm *drm, struct ttm_mem_reg *reg)
+{
+ struct nouveau_mem *mem = nouveau_mem(reg);
+
+ if (!reg->bus.base)
+ return; /* already freed */
+
+ if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) {
+ switch (reg->mem_type) {
+ case TTM_PL_TT:
+ if (mem->kind)
+ nvif_object_unmap_handle(&mem->mem.object);
+ break;
+ case TTM_PL_VRAM:
+ nvif_object_unmap_handle(&mem->mem.object);
+ break;
+ default:
+ break;
+ }
+ }
+ reg->bus.base = 0;
+}
+
static int
nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
{
@@ -1434,18 +1480,26 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
struct nouveau_drm *drm = nouveau_bdev(bdev);
struct nvkm_device *device = nvxx_device(&drm->client.device);
struct nouveau_mem *mem = nouveau_mem(reg);
+ struct nouveau_bo *nvbo;
+ int ret;
+
+ if (reg->bus.base)
+ return 0; /* already mapped */
reg->bus.addr = NULL;
reg->bus.offset = 0;
reg->bus.size = reg->num_pages << PAGE_SHIFT;
- reg->bus.base = 0;
reg->bus.is_iomem = false;
if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
return -EINVAL;
+
+ mutex_lock(&drm->ttm.io_reserve_mutex);
+retry:
switch (reg->mem_type) {
case TTM_PL_SYSTEM:
/* System memory */
- return 0;
+ ret = 0;
+ goto out;
case TTM_PL_TT:
#if IS_ENABLED(CONFIG_AGP)
if (drm->agp.bridge) {
@@ -1469,7 +1523,6 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
} args;
u64 handle, length;
u32 argc = 0;
- int ret;
switch (mem->mem.object.oclass) {
case NVIF_CLASS_MEM_NV50:
@@ -1493,38 +1546,46 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
ret = nvif_object_map_handle(&mem->mem.object,
&args, argc,
&handle, &length);
- if (ret != 1)
- return ret ? ret : -EINVAL;
+ if (ret != 1) {
+ ret = ret ? ret : -EINVAL;
+ goto out;
+ }
+ ret = 0;
reg->bus.base = 0;
reg->bus.offset = handle;
}
break;
default:
- return -EINVAL;
+ ret = -EINVAL;
}
- return 0;
+
+out:
+ if (ret == -EAGAIN) {
+ nvbo = list_first_entry_or_null(&drm->ttm.io_reserve_lru,
+ typeof(*nvbo),
+ io_reserve_lru);
+ if (nvbo) {
+ list_del_init(&nvbo->io_reserve_lru);
+ drm_vma_node_unmap(&nvbo->bo.base.vma_node,
+ bdev->dev_mapping);
+ nouveau_ttm_io_mem_free_locked(drm, &nvbo->bo.mem);
+ goto retry;
+ }
+
+ }
+ mutex_unlock(&drm->ttm.io_reserve_mutex);
+ return ret;
}
static void
nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
{
struct nouveau_drm *drm = nouveau_bdev(bdev);
- struct nouveau_mem *mem = nouveau_mem(reg);
- if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) {
- switch (reg->mem_type) {
- case TTM_PL_TT:
- if (mem->kind)
- nvif_object_unmap_handle(&mem->mem.object);
- break;
- case TTM_PL_VRAM:
- nvif_object_unmap_handle(&mem->mem.object);
- break;
- default:
- break;
- }
- }
+ mutex_lock(&drm->ttm.io_reserve_mutex);
+ nouveau_ttm_io_mem_free_locked(drm, reg);
+ mutex_unlock(&drm->ttm.io_reserve_mutex);
}
static int
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index 38f9d8350963..c47fcdf80ade 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -17,6 +17,7 @@ struct nouveau_bo {
bool force_coherent;
struct ttm_bo_kmap_obj kmap;
struct list_head head;
+ struct list_head io_reserve_lru;
/* protected by ttm_bo_reserve() */
struct drm_file *reserved_by;
@@ -92,6 +93,8 @@ int nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
bool no_wait_gpu);
void nouveau_bo_sync_for_device(struct nouveau_bo *nvbo);
void nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo);
+void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo);
+void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo);
/* TODO: submit equivalent to TTM generic API upstream? */
static inline void __iomem *
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index da8c46e09943..cd19c8ce5939 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -158,6 +158,8 @@ struct nouveau_drm {
int type_vram;
int type_host[2];
int type_ncoh[2];
+ struct mutex io_reserve_mutex;
+ struct list_head io_reserve_lru;
} ttm;
/* GEM interface support */
diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c
index 77a0c6ad3cef..50518b48e9b4 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ttm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c
@@ -162,13 +162,51 @@ const struct ttm_mem_type_manager_func nv04_gart_manager = {
.debug = nouveau_manager_debug
};
+static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf)
+{
+ struct vm_area_struct *vma = vmf->vma;
+ struct ttm_buffer_object *bo = vma->vm_private_data;
+ pgprot_t prot;
+ vm_fault_t ret;
+
+ ret = ttm_bo_vm_reserve(bo, vmf);
+ if (ret)
+ return ret;
+
+ nouveau_bo_del_io_reserve_lru(bo);
+
+ prot = vm_get_page_prot(vma->vm_flags);
+ ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT);
+ if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
+ return ret;
+
+ nouveau_bo_add_io_reserve_lru(bo);
+
+ dma_resv_unlock(bo->base.resv);
+
+ return ret;
+}
+
+static struct vm_operations_struct nouveau_ttm_vm_ops = {
+ .fault = nouveau_ttm_fault,
+ .open = ttm_bo_vm_open,
+ .close = ttm_bo_vm_close,
+ .access = ttm_bo_vm_access
+};
+
int
nouveau_ttm_mmap(struct file *filp, struct vm_area_struct *vma)
{
struct drm_file *file_priv = filp->private_data;
struct nouveau_drm *drm = nouveau_drm(file_priv->minor->dev);
+ int ret;
- return ttm_bo_mmap(filp, vma, &drm->ttm.bdev);
+ ret = ttm_bo_mmap(filp, vma, &drm->ttm.bdev);
+ if (ret)
+ return ret;
+
+ vma->vm_ops = &nouveau_ttm_vm_ops;
+ return 0;
}
static int
@@ -273,6 +311,9 @@ nouveau_ttm_init(struct nouveau_drm *drm)
return ret;
}
+ mutex_init(&drm->ttm.io_reserve_mutex);
+ INIT_LIST_HEAD(&drm->ttm.io_reserve_lru);
+
NV_INFO(drm, "VRAM: %d MiB\n", (u32)(drm->gem.vram_available >> 20));
NV_INFO(drm, "GART: %d MiB\n", (u32)(drm->gem.gart_available >> 20));
return 0;
--
2.17.1
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH 2/2] drm/ttm: remove io_reserve_lru handling
2020-01-24 14:30 TTM/Nouveau cleanups Christian König
2020-01-24 14:30 ` [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2 Christian König
@ 2020-01-24 14:30 ` Christian König
1 sibling, 0 replies; 5+ messages in thread
From: Christian König @ 2020-01-24 14:30 UTC (permalink / raw)
To: bskeggs, airlied, imirkin, dri-devel, nouveau
From: Christian König <ckoenig.leichtzumerken@gmail.com>
That is not used any more.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
drivers/gpu/drm/ttm/ttm_bo.c | 37 +++-------
drivers/gpu/drm/ttm/ttm_bo_util.c | 114 +-----------------------------
drivers/gpu/drm/ttm/ttm_bo_vm.c | 33 +++------
include/drm/ttm/ttm_bo_api.h | 5 --
include/drm/ttm/ttm_bo_driver.h | 5 --
5 files changed, 20 insertions(+), 174 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 1fbc36f05d89..568329e517d2 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -311,13 +311,8 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
int ret = 0;
if (old_is_pci || new_is_pci ||
- ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) {
- ret = ttm_mem_io_lock(old_man, true);
- if (unlikely(ret != 0))
- goto out_err;
- ttm_bo_unmap_virtual_locked(bo);
- ttm_mem_io_unlock(old_man);
- }
+ ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0))
+ ttm_bo_unmap_virtual(bo);
/*
* Create and bind a ttm if required.
@@ -646,15 +641,12 @@ static void ttm_bo_release(struct kref *kref)
struct ttm_buffer_object *bo =
container_of(kref, struct ttm_buffer_object, kref);
struct ttm_bo_device *bdev = bo->bdev;
- struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
if (bo->bdev->driver->release_notify)
bo->bdev->driver->release_notify(bo);
drm_vma_offset_remove(bdev->vma_manager, &bo->base.vma_node);
- ttm_mem_io_lock(man, false);
- ttm_mem_io_free_vm(bo);
- ttm_mem_io_unlock(man);
+ ttm_mem_io_free(bdev, &bo->mem);
ttm_bo_cleanup_refs_or_queue(bo);
kref_put(&bo->list_kref, ttm_bo_release_list);
}
@@ -703,8 +695,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo,
evict_mem = bo->mem;
evict_mem.mm_node = NULL;
- evict_mem.bus.io_reserved_vm = false;
- evict_mem.bus.io_reserved_count = 0;
+ evict_mem.bus.base = 0;
ret = ttm_bo_mem_space(bo, &placement, &evict_mem, ctx);
if (ret) {
@@ -1155,8 +1146,7 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
mem.num_pages = bo->num_pages;
mem.size = mem.num_pages << PAGE_SHIFT;
mem.page_alignment = bo->mem.page_alignment;
- mem.bus.io_reserved_vm = false;
- mem.bus.io_reserved_count = 0;
+ mem.bus.base = 0;
/*
* Determine where to move the buffer.
*/
@@ -1301,8 +1291,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev,
bo->mem.num_pages = bo->num_pages;
bo->mem.mm_node = NULL;
bo->mem.page_alignment = page_alignment;
- bo->mem.bus.io_reserved_vm = false;
- bo->mem.bus.io_reserved_count = 0;
+ bo->mem.bus.base = 0;
bo->moving = NULL;
bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED);
bo->acc_size = acc_size;
@@ -1748,22 +1737,12 @@ bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
return true;
}
-void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo)
-{
- struct ttm_bo_device *bdev = bo->bdev;
-
- drm_vma_node_unmap(&bo->base.vma_node, bdev->dev_mapping);
- ttm_mem_io_free_vm(bo);
-}
-
void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
{
struct ttm_bo_device *bdev = bo->bdev;
- struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
- ttm_mem_io_lock(man, false);
- ttm_bo_unmap_virtual_locked(bo);
- ttm_mem_io_unlock(man);
+ drm_vma_node_unmap(&bo->base.vma_node, bdev->dev_mapping);
+ ttm_mem_io_free(bdev, &bo->mem);
}
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 2b0e5a088da0..ace31ad65997 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -91,124 +91,30 @@ int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
}
EXPORT_SYMBOL(ttm_bo_move_ttm);
-int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool interruptible)
-{
- if (likely(man->io_reserve_fastpath))
- return 0;
-
- if (interruptible)
- return mutex_lock_interruptible(&man->io_reserve_mutex);
-
- mutex_lock(&man->io_reserve_mutex);
- return 0;
-}
-
-void ttm_mem_io_unlock(struct ttm_mem_type_manager *man)
-{
- if (likely(man->io_reserve_fastpath))
- return;
-
- mutex_unlock(&man->io_reserve_mutex);
-}
-
-static int ttm_mem_io_evict(struct ttm_mem_type_manager *man)
-{
- struct ttm_buffer_object *bo;
-
- if (!man->use_io_reserve_lru || list_empty(&man->io_reserve_lru))
- return -EAGAIN;
-
- bo = list_first_entry(&man->io_reserve_lru,
- struct ttm_buffer_object,
- io_reserve_lru);
- list_del_init(&bo->io_reserve_lru);
- ttm_bo_unmap_virtual_locked(bo);
-
- return 0;
-}
-
-
int ttm_mem_io_reserve(struct ttm_bo_device *bdev,
struct ttm_mem_reg *mem)
{
- struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
- int ret = 0;
-
if (!bdev->driver->io_mem_reserve)
return 0;
- if (likely(man->io_reserve_fastpath))
- return bdev->driver->io_mem_reserve(bdev, mem);
-
- if (bdev->driver->io_mem_reserve &&
- mem->bus.io_reserved_count++ == 0) {
-retry:
- ret = bdev->driver->io_mem_reserve(bdev, mem);
- if (ret == -EAGAIN) {
- ret = ttm_mem_io_evict(man);
- if (ret == 0)
- goto retry;
- }
- }
- return ret;
+
+ return bdev->driver->io_mem_reserve(bdev, mem);
}
void ttm_mem_io_free(struct ttm_bo_device *bdev,
struct ttm_mem_reg *mem)
{
- struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
-
- if (likely(man->io_reserve_fastpath))
- return;
-
- if (bdev->driver->io_mem_reserve &&
- --mem->bus.io_reserved_count == 0 &&
- bdev->driver->io_mem_free)
+ if (bdev->driver->io_mem_free)
bdev->driver->io_mem_free(bdev, mem);
-
-}
-
-int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo)
-{
- struct ttm_mem_reg *mem = &bo->mem;
- int ret;
-
- if (!mem->bus.io_reserved_vm) {
- struct ttm_mem_type_manager *man =
- &bo->bdev->man[mem->mem_type];
-
- ret = ttm_mem_io_reserve(bo->bdev, mem);
- if (unlikely(ret != 0))
- return ret;
- mem->bus.io_reserved_vm = true;
- if (man->use_io_reserve_lru)
- list_add_tail(&bo->io_reserve_lru,
- &man->io_reserve_lru);
- }
- return 0;
-}
-
-void ttm_mem_io_free_vm(struct ttm_buffer_object *bo)
-{
- struct ttm_mem_reg *mem = &bo->mem;
-
- if (mem->bus.io_reserved_vm) {
- mem->bus.io_reserved_vm = false;
- list_del_init(&bo->io_reserve_lru);
- ttm_mem_io_free(bo->bdev, mem);
- }
}
static int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
void **virtual)
{
- struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
int ret;
void *addr;
*virtual = NULL;
- (void) ttm_mem_io_lock(man, false);
ret = ttm_mem_io_reserve(bdev, mem);
- ttm_mem_io_unlock(man);
if (ret || !mem->bus.is_iomem)
return ret;
@@ -220,9 +126,7 @@ static int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *m
else
addr = ioremap_nocache(mem->bus.base + mem->bus.offset, mem->bus.size);
if (!addr) {
- (void) ttm_mem_io_lock(man, false);
ttm_mem_io_free(bdev, mem);
- ttm_mem_io_unlock(man);
return -ENOMEM;
}
}
@@ -239,9 +143,7 @@ static void ttm_mem_reg_iounmap(struct ttm_bo_device *bdev, struct ttm_mem_reg *
if (virtual && mem->bus.addr == NULL)
iounmap(virtual);
- (void) ttm_mem_io_lock(man, false);
ttm_mem_io_free(bdev, mem);
- ttm_mem_io_unlock(man);
}
static int ttm_copy_io_page(void *dst, void *src, unsigned long page)
@@ -616,8 +518,6 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
unsigned long start_page, unsigned long num_pages,
struct ttm_bo_kmap_obj *map)
{
- struct ttm_mem_type_manager *man =
- &bo->bdev->man[bo->mem.mem_type];
unsigned long offset, size;
int ret;
@@ -628,9 +528,7 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
if (start_page > bo->num_pages)
return -EINVAL;
- (void) ttm_mem_io_lock(man, false);
ret = ttm_mem_io_reserve(bo->bdev, &bo->mem);
- ttm_mem_io_unlock(man);
if (ret)
return ret;
if (!bo->mem.bus.is_iomem) {
@@ -645,10 +543,6 @@ EXPORT_SYMBOL(ttm_bo_kmap);
void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
{
- struct ttm_buffer_object *bo = map->bo;
- struct ttm_mem_type_manager *man =
- &bo->bdev->man[bo->mem.mem_type];
-
if (!map->virtual)
return;
switch (map->bo_kmap_type) {
@@ -666,9 +560,7 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
default:
BUG();
}
- (void) ttm_mem_io_lock(man, false);
ttm_mem_io_free(map->bo->bdev, &map->bo->mem);
- ttm_mem_io_unlock(man);
map->virtual = NULL;
map->page = NULL;
}
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index eebb4c06c04d..11d49fa90bdc 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -191,8 +191,6 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf,
pgoff_t i;
vm_fault_t ret = VM_FAULT_NOPAGE;
unsigned long address = vmf->address;
- struct ttm_mem_type_manager *man =
- &bdev->man[bo->mem.mem_type];
/*
* Refuse to fault imported pages. This should be handled
@@ -231,24 +229,17 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf,
if (unlikely(ret != 0))
return ret;
- err = ttm_mem_io_lock(man, true);
+ err = ttm_mem_io_reserve(bdev, &bo->mem);
if (unlikely(err != 0))
- return VM_FAULT_NOPAGE;
- err = ttm_mem_io_reserve_vm(bo);
- if (unlikely(err != 0)) {
- ret = VM_FAULT_SIGBUS;
- goto out_io_unlock;
- }
+ return VM_FAULT_SIGBUS;
page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) +
vma->vm_pgoff - drm_vma_node_start(&bo->base.vma_node);
page_last = vma_pages(vma) + vma->vm_pgoff -
drm_vma_node_start(&bo->base.vma_node);
- if (unlikely(page_offset >= bo->num_pages)) {
- ret = VM_FAULT_SIGBUS;
- goto out_io_unlock;
- }
+ if (unlikely(page_offset >= bo->num_pages))
+ return VM_FAULT_SIGBUS;
cvma.vm_page_prot = ttm_io_prot(bo->mem.placement, prot);
if (!bo->mem.bus.is_iomem) {
@@ -260,10 +251,8 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf,
};
ttm = bo->ttm;
- if (ttm_tt_populate(bo->ttm, &ctx)) {
- ret = VM_FAULT_OOM;
- goto out_io_unlock;
- }
+ if (ttm_tt_populate(bo->ttm, &ctx))
+ return VM_FAULT_OOM;
} else {
/* Iomem should not be marked encrypted */
cvma.vm_page_prot = pgprot_decrypted(cvma.vm_page_prot);
@@ -279,8 +268,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf,
} else {
page = ttm->pages[page_offset];
if (unlikely(!page && i == 0)) {
- ret = VM_FAULT_OOM;
- goto out_io_unlock;
+ return VM_FAULT_OOM;
} else if (unlikely(!page)) {
break;
}
@@ -298,7 +286,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf,
/* Never error on prefaulted PTEs */
if (unlikely((ret & VM_FAULT_ERROR))) {
if (i == 0)
- goto out_io_unlock;
+ return VM_FAULT_NOPAGE;
else
break;
}
@@ -307,10 +295,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf,
if (unlikely(++page_offset >= page_last))
break;
}
- ret = VM_FAULT_NOPAGE;
-out_io_unlock:
- ttm_mem_io_unlock(man);
- return ret;
+ return VM_FAULT_NOPAGE;
}
EXPORT_SYMBOL(ttm_bo_vm_fault_reserved);
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 66ca49db9633..3c082d3d6959 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -62,8 +62,6 @@ struct ttm_lru_bulk_move;
* @is_iomem: is this io memory ?
* @size: size in byte
* @offset: offset from the base address
- * @io_reserved_vm: The VM system has a refcount in @io_reserved_count
- * @io_reserved_count: Refcounting the numbers of callers to ttm_mem_io_reserve
*
* Structure indicating the bus placement of an object.
*/
@@ -73,11 +71,8 @@ struct ttm_bus_placement {
unsigned long size;
unsigned long offset;
bool is_iomem;
- bool io_reserved_vm;
- uint64_t io_reserved_count;
};
-
/**
* struct ttm_mem_reg
*
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..5cb6b28f954a 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -609,11 +609,6 @@ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo);
*/
void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo);
-int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo);
-void ttm_mem_io_free_vm(struct ttm_buffer_object *bo);
-int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool interruptible);
-void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);
-
/**
* __ttm_bo_reserve:
*
--
2.17.1
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [Nouveau] [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2
2020-01-24 14:30 ` [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2 Christian König
@ 2020-01-28 5:49 ` Ben Skeggs
2020-02-13 12:58 ` Christian König
0 siblings, 1 reply; 5+ messages in thread
From: Ben Skeggs @ 2020-01-28 5:49 UTC (permalink / raw)
To: Christian König; +Cc: Dave Airlie, ML nouveau, ML dri-devel, Ben Skeggs
On Sat, 25 Jan 2020 at 00:30, Christian König
<ckoenig.leichtzumerken@gmail.com> wrote:
>
> From: Christian König <ckoenig.leichtzumerken@gmail.com>
>
> While working on TTM cleanups I've found that the io_reserve_lru used by
> Nouveau is actually not working at all.
>
> In general we should remove driver specific handling from the memory
> management, so this patch moves the io_reserve_lru handling into Nouveau
> instead.
>
> The patch should be functional correct, but is only compile tested!
NACK on this as it currently stands. It not only causes invalid io
accesses somehow on module load, but while attempting to track down
why, I realised there's a more severe issue. This removes the
distinction between kmap() and mapping into userspace, the former of
which should not be placed onto the LRU as an eviction candidate.
We *do* require the LRU, so it's not something that can just be
dropped completely. There's a user report where they're getting a
SIGBUS due to the bug you noticed causing it to not work right now.
Ben.
>
> v2: don't call ttm_bo_unmap_virtual in nouveau_ttm_io_mem_reserve
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
> Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> ---
> drivers/gpu/drm/nouveau/nouveau_bo.c | 107 ++++++++++++++++++++------
> drivers/gpu/drm/nouveau/nouveau_bo.h | 3 +
> drivers/gpu/drm/nouveau/nouveau_drv.h | 2 +
> drivers/gpu/drm/nouveau/nouveau_ttm.c | 43 ++++++++++-
> 4 files changed, 131 insertions(+), 24 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index 81668104595f..acee054f77ed 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -137,6 +137,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
> struct nouveau_bo *nvbo = nouveau_bo(bo);
>
> WARN_ON(nvbo->pin_refcnt > 0);
> + nouveau_bo_del_io_reserve_lru(bo);
> nv10_bo_put_tile_region(dev, nvbo->tile, NULL);
>
> /*
> @@ -304,6 +305,7 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 flags,
>
> nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
> nouveau_bo_placement_set(nvbo, flags, 0);
> + INIT_LIST_HEAD(&nvbo->io_reserve_lru);
>
> ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type,
> &nvbo->placement, align >> PAGE_SHIFT, false,
> @@ -574,6 +576,26 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo)
> PAGE_SIZE, DMA_FROM_DEVICE);
> }
>
> +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo)
> +{
> + struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
> + struct nouveau_bo *nvbo = nouveau_bo(bo);
> +
> + mutex_lock(&drm->ttm.io_reserve_mutex);
> + list_move_tail(&nvbo->io_reserve_lru, &drm->ttm.io_reserve_lru);
> + mutex_unlock(&drm->ttm.io_reserve_mutex);
> +}
> +
> +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo)
> +{
> + struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
> + struct nouveau_bo *nvbo = nouveau_bo(bo);
> +
> + mutex_lock(&drm->ttm.io_reserve_mutex);
> + list_del_init(&nvbo->io_reserve_lru);
> + mutex_unlock(&drm->ttm.io_reserve_mutex);
> +}
> +
> int
> nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
> bool no_wait_gpu)
> @@ -675,8 +697,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
> }
>
> man->func = &nouveau_vram_manager;
> - man->io_reserve_fastpath = false;
> - man->use_io_reserve_lru = true;
> } else {
> man->func = &ttm_bo_manager_func;
> }
> @@ -1305,6 +1325,8 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
> if (bo->destroy != nouveau_bo_del_ttm)
> return;
>
> + nouveau_bo_del_io_reserve_lru(bo);
> +
> if (mem && new_reg->mem_type != TTM_PL_SYSTEM &&
> mem->mem.page == nvbo->page) {
> list_for_each_entry(vma, &nvbo->vma_list, head) {
> @@ -1427,6 +1449,30 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp)
> filp->private_data);
> }
>
> +static void
> +nouveau_ttm_io_mem_free_locked(struct nouveau_drm *drm, struct ttm_mem_reg *reg)
> +{
> + struct nouveau_mem *mem = nouveau_mem(reg);
> +
> + if (!reg->bus.base)
> + return; /* already freed */
> +
> + if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) {
> + switch (reg->mem_type) {
> + case TTM_PL_TT:
> + if (mem->kind)
> + nvif_object_unmap_handle(&mem->mem.object);
> + break;
> + case TTM_PL_VRAM:
> + nvif_object_unmap_handle(&mem->mem.object);
> + break;
> + default:
> + break;
> + }
> + }
> + reg->bus.base = 0;
> +}
> +
> static int
> nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
> {
> @@ -1434,18 +1480,26 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
> struct nouveau_drm *drm = nouveau_bdev(bdev);
> struct nvkm_device *device = nvxx_device(&drm->client.device);
> struct nouveau_mem *mem = nouveau_mem(reg);
> + struct nouveau_bo *nvbo;
> + int ret;
> +
> + if (reg->bus.base)
> + return 0; /* already mapped */
>
> reg->bus.addr = NULL;
> reg->bus.offset = 0;
> reg->bus.size = reg->num_pages << PAGE_SHIFT;
> - reg->bus.base = 0;
> reg->bus.is_iomem = false;
> if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
> return -EINVAL;
> +
> + mutex_lock(&drm->ttm.io_reserve_mutex);
> +retry:
> switch (reg->mem_type) {
> case TTM_PL_SYSTEM:
> /* System memory */
> - return 0;
> + ret = 0;
> + goto out;
> case TTM_PL_TT:
> #if IS_ENABLED(CONFIG_AGP)
> if (drm->agp.bridge) {
> @@ -1469,7 +1523,6 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
> } args;
> u64 handle, length;
> u32 argc = 0;
> - int ret;
>
> switch (mem->mem.object.oclass) {
> case NVIF_CLASS_MEM_NV50:
> @@ -1493,38 +1546,46 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
> ret = nvif_object_map_handle(&mem->mem.object,
> &args, argc,
> &handle, &length);
> - if (ret != 1)
> - return ret ? ret : -EINVAL;
> + if (ret != 1) {
> + ret = ret ? ret : -EINVAL;
> + goto out;
> + }
> + ret = 0;
>
> reg->bus.base = 0;
> reg->bus.offset = handle;
> }
> break;
> default:
> - return -EINVAL;
> + ret = -EINVAL;
> }
> - return 0;
> +
> +out:
> + if (ret == -EAGAIN) {
> + nvbo = list_first_entry_or_null(&drm->ttm.io_reserve_lru,
> + typeof(*nvbo),
> + io_reserve_lru);
> + if (nvbo) {
> + list_del_init(&nvbo->io_reserve_lru);
> + drm_vma_node_unmap(&nvbo->bo.base.vma_node,
> + bdev->dev_mapping);
> + nouveau_ttm_io_mem_free_locked(drm, &nvbo->bo.mem);
> + goto retry;
> + }
> +
> + }
> + mutex_unlock(&drm->ttm.io_reserve_mutex);
> + return ret;
> }
>
> static void
> nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
> {
> struct nouveau_drm *drm = nouveau_bdev(bdev);
> - struct nouveau_mem *mem = nouveau_mem(reg);
>
> - if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) {
> - switch (reg->mem_type) {
> - case TTM_PL_TT:
> - if (mem->kind)
> - nvif_object_unmap_handle(&mem->mem.object);
> - break;
> - case TTM_PL_VRAM:
> - nvif_object_unmap_handle(&mem->mem.object);
> - break;
> - default:
> - break;
> - }
> - }
> + mutex_lock(&drm->ttm.io_reserve_mutex);
> + nouveau_ttm_io_mem_free_locked(drm, reg);
> + mutex_unlock(&drm->ttm.io_reserve_mutex);
> }
>
> static int
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
> index 38f9d8350963..c47fcdf80ade 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.h
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
> @@ -17,6 +17,7 @@ struct nouveau_bo {
> bool force_coherent;
> struct ttm_bo_kmap_obj kmap;
> struct list_head head;
> + struct list_head io_reserve_lru;
>
> /* protected by ttm_bo_reserve() */
> struct drm_file *reserved_by;
> @@ -92,6 +93,8 @@ int nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
> bool no_wait_gpu);
> void nouveau_bo_sync_for_device(struct nouveau_bo *nvbo);
> void nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo);
> +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo);
> +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo);
>
> /* TODO: submit equivalent to TTM generic API upstream? */
> static inline void __iomem *
> diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
> index da8c46e09943..cd19c8ce5939 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_drv.h
> +++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
> @@ -158,6 +158,8 @@ struct nouveau_drm {
> int type_vram;
> int type_host[2];
> int type_ncoh[2];
> + struct mutex io_reserve_mutex;
> + struct list_head io_reserve_lru;
> } ttm;
>
> /* GEM interface support */
> diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c
> index 77a0c6ad3cef..50518b48e9b4 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_ttm.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c
> @@ -162,13 +162,51 @@ const struct ttm_mem_type_manager_func nv04_gart_manager = {
> .debug = nouveau_manager_debug
> };
>
> +static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf)
> +{
> + struct vm_area_struct *vma = vmf->vma;
> + struct ttm_buffer_object *bo = vma->vm_private_data;
> + pgprot_t prot;
> + vm_fault_t ret;
> +
> + ret = ttm_bo_vm_reserve(bo, vmf);
> + if (ret)
> + return ret;
> +
> + nouveau_bo_del_io_reserve_lru(bo);
> +
> + prot = vm_get_page_prot(vma->vm_flags);
> + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT);
> + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
> + return ret;
> +
> + nouveau_bo_add_io_reserve_lru(bo);
> +
> + dma_resv_unlock(bo->base.resv);
> +
> + return ret;
> +}
> +
> +static struct vm_operations_struct nouveau_ttm_vm_ops = {
> + .fault = nouveau_ttm_fault,
> + .open = ttm_bo_vm_open,
> + .close = ttm_bo_vm_close,
> + .access = ttm_bo_vm_access
> +};
> +
> int
> nouveau_ttm_mmap(struct file *filp, struct vm_area_struct *vma)
> {
> struct drm_file *file_priv = filp->private_data;
> struct nouveau_drm *drm = nouveau_drm(file_priv->minor->dev);
> + int ret;
>
> - return ttm_bo_mmap(filp, vma, &drm->ttm.bdev);
> + ret = ttm_bo_mmap(filp, vma, &drm->ttm.bdev);
> + if (ret)
> + return ret;
> +
> + vma->vm_ops = &nouveau_ttm_vm_ops;
> + return 0;
> }
>
> static int
> @@ -273,6 +311,9 @@ nouveau_ttm_init(struct nouveau_drm *drm)
> return ret;
> }
>
> + mutex_init(&drm->ttm.io_reserve_mutex);
> + INIT_LIST_HEAD(&drm->ttm.io_reserve_lru);
> +
> NV_INFO(drm, "VRAM: %d MiB\n", (u32)(drm->gem.vram_available >> 20));
> NV_INFO(drm, "GART: %d MiB\n", (u32)(drm->gem.gart_available >> 20));
> return 0;
> --
> 2.17.1
>
> _______________________________________________
> Nouveau mailing list
> Nouveau@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/nouveau
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Nouveau] [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2
2020-01-28 5:49 ` [Nouveau] " Ben Skeggs
@ 2020-02-13 12:58 ` Christian König
0 siblings, 0 replies; 5+ messages in thread
From: Christian König @ 2020-02-13 12:58 UTC (permalink / raw)
To: Ben Skeggs; +Cc: Dave Airlie, ML nouveau, ML dri-devel, Ben Skeggs
Hi Ben,
sorry for the delayed response. Haven been rather busy recently.
Am 28.01.20 um 06:49 schrieb Ben Skeggs:
> On Sat, 25 Jan 2020 at 00:30, Christian König
> <ckoenig.leichtzumerken@gmail.com> wrote:
>> From: Christian König <ckoenig.leichtzumerken@gmail.com>
>>
>> While working on TTM cleanups I've found that the io_reserve_lru used by
>> Nouveau is actually not working at all.
>>
>> In general we should remove driver specific handling from the memory
>> management, so this patch moves the io_reserve_lru handling into Nouveau
>> instead.
>>
>> The patch should be functional correct, but is only compile tested!
> NACK on this as it currently stands. It not only causes invalid io
> accesses somehow on module load, but while attempting to track down
> why, I realised there's a more severe issue. This removes the
> distinction between kmap() and mapping into userspace, the former of
> which should not be placed onto the LRU as an eviction candidate.
Yeah, I already feared that this won't work of hand. It's just a bit
hard to write stuff without being able to test it.
> We *do* require the LRU, so it's not something that can just be
> dropped completely. There's a user report where they're getting a
> SIGBUS due to the bug you noticed causing it to not work right now.
Well that you require an LRU is obvious. But since this is completely
Nouveau specificthat handling doesn't needs to be and actually should
not be in TTM.
And as Daniel pointed out the current handling in the combination of TTM
and Nouveau is fundamentally broken.
Userspace can either always force other applications into a SIGBUS or
the page faulting into a live lock. Both of that is rather bad.
To outline how this could work correctly:
1. You split up your PCIe BAR into pages/segments of fixed size. I'm not
sure what the NVidia hardware can handle, but I would use something like
2MB segments for that to match the x86 huge page size.
2. Whenever you get a page fault or a kmap request for a buffer you walk
your LRU and grab a free segment.
3. kmap requests obviously are not put back on the LRU (or are skipped
while grabbed). Segments for page fault requests are put on the LRU
immediately again.
The idea here is that you have at least enough segments (in this example
256 MB BAR / 2 MB segment size = 128 segments) so that you always have
more segments than page faults can happen at the same time.
I'm completely fine with keeping the code around for now, but as
outlined above the whole handling is rather broken.
Regards,
Christian.
>
> Ben.
>
>> v2: don't call ttm_bo_unmap_virtual in nouveau_ttm_io_mem_reserve
>>
>> Signed-off-by: Christian König <christian.koenig@amd.com>
>> Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>> ---
>> drivers/gpu/drm/nouveau/nouveau_bo.c | 107 ++++++++++++++++++++------
>> drivers/gpu/drm/nouveau/nouveau_bo.h | 3 +
>> drivers/gpu/drm/nouveau/nouveau_drv.h | 2 +
>> drivers/gpu/drm/nouveau/nouveau_ttm.c | 43 ++++++++++-
>> 4 files changed, 131 insertions(+), 24 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index 81668104595f..acee054f77ed 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -137,6 +137,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
>> struct nouveau_bo *nvbo = nouveau_bo(bo);
>>
>> WARN_ON(nvbo->pin_refcnt > 0);
>> + nouveau_bo_del_io_reserve_lru(bo);
>> nv10_bo_put_tile_region(dev, nvbo->tile, NULL);
>>
>> /*
>> @@ -304,6 +305,7 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 flags,
>>
>> nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
>> nouveau_bo_placement_set(nvbo, flags, 0);
>> + INIT_LIST_HEAD(&nvbo->io_reserve_lru);
>>
>> ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type,
>> &nvbo->placement, align >> PAGE_SHIFT, false,
>> @@ -574,6 +576,26 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo)
>> PAGE_SIZE, DMA_FROM_DEVICE);
>> }
>>
>> +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo)
>> +{
>> + struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
>> + struct nouveau_bo *nvbo = nouveau_bo(bo);
>> +
>> + mutex_lock(&drm->ttm.io_reserve_mutex);
>> + list_move_tail(&nvbo->io_reserve_lru, &drm->ttm.io_reserve_lru);
>> + mutex_unlock(&drm->ttm.io_reserve_mutex);
>> +}
>> +
>> +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo)
>> +{
>> + struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
>> + struct nouveau_bo *nvbo = nouveau_bo(bo);
>> +
>> + mutex_lock(&drm->ttm.io_reserve_mutex);
>> + list_del_init(&nvbo->io_reserve_lru);
>> + mutex_unlock(&drm->ttm.io_reserve_mutex);
>> +}
>> +
>> int
>> nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
>> bool no_wait_gpu)
>> @@ -675,8 +697,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
>> }
>>
>> man->func = &nouveau_vram_manager;
>> - man->io_reserve_fastpath = false;
>> - man->use_io_reserve_lru = true;
>> } else {
>> man->func = &ttm_bo_manager_func;
>> }
>> @@ -1305,6 +1325,8 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
>> if (bo->destroy != nouveau_bo_del_ttm)
>> return;
>>
>> + nouveau_bo_del_io_reserve_lru(bo);
>> +
>> if (mem && new_reg->mem_type != TTM_PL_SYSTEM &&
>> mem->mem.page == nvbo->page) {
>> list_for_each_entry(vma, &nvbo->vma_list, head) {
>> @@ -1427,6 +1449,30 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp)
>> filp->private_data);
>> }
>>
>> +static void
>> +nouveau_ttm_io_mem_free_locked(struct nouveau_drm *drm, struct ttm_mem_reg *reg)
>> +{
>> + struct nouveau_mem *mem = nouveau_mem(reg);
>> +
>> + if (!reg->bus.base)
>> + return; /* already freed */
>> +
>> + if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) {
>> + switch (reg->mem_type) {
>> + case TTM_PL_TT:
>> + if (mem->kind)
>> + nvif_object_unmap_handle(&mem->mem.object);
>> + break;
>> + case TTM_PL_VRAM:
>> + nvif_object_unmap_handle(&mem->mem.object);
>> + break;
>> + default:
>> + break;
>> + }
>> + }
>> + reg->bus.base = 0;
>> +}
>> +
>> static int
>> nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
>> {
>> @@ -1434,18 +1480,26 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
>> struct nouveau_drm *drm = nouveau_bdev(bdev);
>> struct nvkm_device *device = nvxx_device(&drm->client.device);
>> struct nouveau_mem *mem = nouveau_mem(reg);
>> + struct nouveau_bo *nvbo;
>> + int ret;
>> +
>> + if (reg->bus.base)
>> + return 0; /* already mapped */
>>
>> reg->bus.addr = NULL;
>> reg->bus.offset = 0;
>> reg->bus.size = reg->num_pages << PAGE_SHIFT;
>> - reg->bus.base = 0;
>> reg->bus.is_iomem = false;
>> if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
>> return -EINVAL;
>> +
>> + mutex_lock(&drm->ttm.io_reserve_mutex);
>> +retry:
>> switch (reg->mem_type) {
>> case TTM_PL_SYSTEM:
>> /* System memory */
>> - return 0;
>> + ret = 0;
>> + goto out;
>> case TTM_PL_TT:
>> #if IS_ENABLED(CONFIG_AGP)
>> if (drm->agp.bridge) {
>> @@ -1469,7 +1523,6 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
>> } args;
>> u64 handle, length;
>> u32 argc = 0;
>> - int ret;
>>
>> switch (mem->mem.object.oclass) {
>> case NVIF_CLASS_MEM_NV50:
>> @@ -1493,38 +1546,46 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
>> ret = nvif_object_map_handle(&mem->mem.object,
>> &args, argc,
>> &handle, &length);
>> - if (ret != 1)
>> - return ret ? ret : -EINVAL;
>> + if (ret != 1) {
>> + ret = ret ? ret : -EINVAL;
>> + goto out;
>> + }
>> + ret = 0;
>>
>> reg->bus.base = 0;
>> reg->bus.offset = handle;
>> }
>> break;
>> default:
>> - return -EINVAL;
>> + ret = -EINVAL;
>> }
>> - return 0;
>> +
>> +out:
>> + if (ret == -EAGAIN) {
>> + nvbo = list_first_entry_or_null(&drm->ttm.io_reserve_lru,
>> + typeof(*nvbo),
>> + io_reserve_lru);
>> + if (nvbo) {
>> + list_del_init(&nvbo->io_reserve_lru);
>> + drm_vma_node_unmap(&nvbo->bo.base.vma_node,
>> + bdev->dev_mapping);
>> + nouveau_ttm_io_mem_free_locked(drm, &nvbo->bo.mem);
>> + goto retry;
>> + }
>> +
>> + }
>> + mutex_unlock(&drm->ttm.io_reserve_mutex);
>> + return ret;
>> }
>>
>> static void
>> nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
>> {
>> struct nouveau_drm *drm = nouveau_bdev(bdev);
>> - struct nouveau_mem *mem = nouveau_mem(reg);
>>
>> - if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) {
>> - switch (reg->mem_type) {
>> - case TTM_PL_TT:
>> - if (mem->kind)
>> - nvif_object_unmap_handle(&mem->mem.object);
>> - break;
>> - case TTM_PL_VRAM:
>> - nvif_object_unmap_handle(&mem->mem.object);
>> - break;
>> - default:
>> - break;
>> - }
>> - }
>> + mutex_lock(&drm->ttm.io_reserve_mutex);
>> + nouveau_ttm_io_mem_free_locked(drm, reg);
>> + mutex_unlock(&drm->ttm.io_reserve_mutex);
>> }
>>
>> static int
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
>> index 38f9d8350963..c47fcdf80ade 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.h
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
>> @@ -17,6 +17,7 @@ struct nouveau_bo {
>> bool force_coherent;
>> struct ttm_bo_kmap_obj kmap;
>> struct list_head head;
>> + struct list_head io_reserve_lru;
>>
>> /* protected by ttm_bo_reserve() */
>> struct drm_file *reserved_by;
>> @@ -92,6 +93,8 @@ int nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
>> bool no_wait_gpu);
>> void nouveau_bo_sync_for_device(struct nouveau_bo *nvbo);
>> void nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo);
>> +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo);
>> +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo);
>>
>> /* TODO: submit equivalent to TTM generic API upstream? */
>> static inline void __iomem *
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
>> index da8c46e09943..cd19c8ce5939 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_drv.h
>> +++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
>> @@ -158,6 +158,8 @@ struct nouveau_drm {
>> int type_vram;
>> int type_host[2];
>> int type_ncoh[2];
>> + struct mutex io_reserve_mutex;
>> + struct list_head io_reserve_lru;
>> } ttm;
>>
>> /* GEM interface support */
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c
>> index 77a0c6ad3cef..50518b48e9b4 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_ttm.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c
>> @@ -162,13 +162,51 @@ const struct ttm_mem_type_manager_func nv04_gart_manager = {
>> .debug = nouveau_manager_debug
>> };
>>
>> +static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf)
>> +{
>> + struct vm_area_struct *vma = vmf->vma;
>> + struct ttm_buffer_object *bo = vma->vm_private_data;
>> + pgprot_t prot;
>> + vm_fault_t ret;
>> +
>> + ret = ttm_bo_vm_reserve(bo, vmf);
>> + if (ret)
>> + return ret;
>> +
>> + nouveau_bo_del_io_reserve_lru(bo);
>> +
>> + prot = vm_get_page_prot(vma->vm_flags);
>> + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT);
>> + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
>> + return ret;
>> +
>> + nouveau_bo_add_io_reserve_lru(bo);
>> +
>> + dma_resv_unlock(bo->base.resv);
>> +
>> + return ret;
>> +}
>> +
>> +static struct vm_operations_struct nouveau_ttm_vm_ops = {
>> + .fault = nouveau_ttm_fault,
>> + .open = ttm_bo_vm_open,
>> + .close = ttm_bo_vm_close,
>> + .access = ttm_bo_vm_access
>> +};
>> +
>> int
>> nouveau_ttm_mmap(struct file *filp, struct vm_area_struct *vma)
>> {
>> struct drm_file *file_priv = filp->private_data;
>> struct nouveau_drm *drm = nouveau_drm(file_priv->minor->dev);
>> + int ret;
>>
>> - return ttm_bo_mmap(filp, vma, &drm->ttm.bdev);
>> + ret = ttm_bo_mmap(filp, vma, &drm->ttm.bdev);
>> + if (ret)
>> + return ret;
>> +
>> + vma->vm_ops = &nouveau_ttm_vm_ops;
>> + return 0;
>> }
>>
>> static int
>> @@ -273,6 +311,9 @@ nouveau_ttm_init(struct nouveau_drm *drm)
>> return ret;
>> }
>>
>> + mutex_init(&drm->ttm.io_reserve_mutex);
>> + INIT_LIST_HEAD(&drm->ttm.io_reserve_lru);
>> +
>> NV_INFO(drm, "VRAM: %d MiB\n", (u32)(drm->gem.vram_available >> 20));
>> NV_INFO(drm, "GART: %d MiB\n", (u32)(drm->gem.gart_available >> 20));
>> return 0;
>> --
>> 2.17.1
>>
>> _______________________________________________
>> Nouveau mailing list
>> Nouveau@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/nouveau
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2020-02-13 12:58 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-01-24 14:30 TTM/Nouveau cleanups Christian König
2020-01-24 14:30 ` [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver v2 Christian König
2020-01-28 5:49 ` [Nouveau] " Ben Skeggs
2020-02-13 12:58 ` Christian König
2020-01-24 14:30 ` [PATCH 2/2] drm/ttm: remove io_reserve_lru handling Christian König
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.