[PATCH 1/2] drm/ttm: Move swapped objects off the manager's LRU list
Thomas Hellström
thomas.hellstrom at linux.intel.com
Tue Sep 3 08:55:11 UTC 2024
Hi, Christian,
Thanks for reviewing.
On Tue, 2024-09-03 at 09:49 +0200, Christian König wrote:
> Am 02.09.24 um 17:41 schrieb Thomas Hellström:
> > Resources of swapped objects remains on the TTM_PL_SYSTEM manager's
> > LRU list, which is bad for the LRU walk efficiency.
> >
> > Rename the device-wide "pinned" list to "unevictable" and move
> > also resources of swapped-out objects to that list.
> >
> > An alternative would be to create an "UNEVICTABLE" priority to
> > be able to keep the pinned- and swapped objects on their
> > respective manager's LRU without affecting the LRU walk efficiency.
> >
> > Cc: Christian König <christian.koenig at amd.com>
> > Cc: Matthew Brost <matthew.brost at intel.com>
> > Cc: <dri-devel at lists.freedesktop.org>
> > Signed-off-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
> > ---
> > drivers/gpu/drm/i915/gem/i915_gem_ttm.c | 2 +-
> > drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 2 +-
> > drivers/gpu/drm/i915/gem/i915_gem_ttm_pm.c | 4 +-
> > drivers/gpu/drm/ttm/ttm_bo.c | 55
> > +++++++++++++++++++-
> > drivers/gpu/drm/ttm/ttm_bo_util.c | 6 +--
> > drivers/gpu/drm/ttm/ttm_bo_vm.c | 2 +-
> > drivers/gpu/drm/ttm/ttm_device.c | 4 +-
> > drivers/gpu/drm/ttm/ttm_resource.c | 9 ++--
> > drivers/gpu/drm/ttm/ttm_tt.c | 1 -
> > drivers/gpu/drm/xe/xe_bo.c | 4 +-
> > include/drm/ttm/ttm_bo.h | 2 +
> > include/drm/ttm/ttm_device.h | 5 +-
> > include/drm/ttm/ttm_tt.h | 5 ++
> > 13 files changed, 81 insertions(+), 20 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> > b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> > index 5c72462d1f57..7de284766f82 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> > @@ -808,7 +808,7 @@ static int __i915_ttm_get_pages(struct
> > drm_i915_gem_object *obj,
> > }
> >
> > if (bo->ttm && !ttm_tt_is_populated(bo->ttm)) {
> > - ret = ttm_tt_populate(bo->bdev, bo->ttm, &ctx);
> > + ret = ttm_bo_populate(bo, &ctx);
> > if (ret)
> > return ret;
> >
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c
> > b/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c
> > index 03b00a03a634..041dab543b78 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c
> > @@ -624,7 +624,7 @@ int i915_ttm_move(struct ttm_buffer_object *bo,
> > bool evict,
> >
> > /* Populate ttm with pages if needed. Typically system
> > memory. */
> > if (ttm && (dst_man->use_tt || (ttm->page_flags &
> > TTM_TT_FLAG_SWAPPED))) {
> > - ret = ttm_tt_populate(bo->bdev, ttm, ctx);
> > + ret = ttm_bo_populate(bo, ctx);
> > if (ret)
> > return ret;
> > }
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm_pm.c
> > b/drivers/gpu/drm/i915/gem/i915_gem_ttm_pm.c
> > index ad649523d5e0..61596cecce4d 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm_pm.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm_pm.c
> > @@ -90,7 +90,7 @@ static int i915_ttm_backup(struct
> > i915_gem_apply_to_region *apply,
> > goto out_no_lock;
> >
> > backup_bo = i915_gem_to_ttm(backup);
> > - err = ttm_tt_populate(backup_bo->bdev, backup_bo->ttm,
> > &ctx);
> > + err = ttm_bo_populate(backup_bo, &ctx);
> > if (err)
> > goto out_no_populate;
> >
> > @@ -189,7 +189,7 @@ static int i915_ttm_restore(struct
> > i915_gem_apply_to_region *apply,
> > if (!backup_bo->resource)
> > err = ttm_bo_validate(backup_bo,
> > i915_ttm_sys_placement(), &ctx);
> > if (!err)
> > - err = ttm_tt_populate(backup_bo->bdev, backup_bo-
> > >ttm, &ctx);
> > + err = ttm_bo_populate(backup_bo, &ctx);
> > if (!err) {
> > err = i915_gem_obj_copy_ttm(obj, backup, pm_apply-
> > >allow_gpu,
> > false);
> > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
> > b/drivers/gpu/drm/ttm/ttm_bo.c
> > index 320592435252..d244566a7e48 100644
> > --- a/drivers/gpu/drm/ttm/ttm_bo.c
> > +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> > @@ -139,7 +139,7 @@ static int ttm_bo_handle_move_mem(struct
> > ttm_buffer_object *bo,
> > goto out_err;
> >
> > if (mem->mem_type != TTM_PL_SYSTEM) {
> > - ret = ttm_tt_populate(bo->bdev, bo->ttm,
> > ctx);
> > + ret = ttm_bo_populate(bo, ctx);
> > if (ret)
> > goto out_err;
> > }
> > @@ -1131,6 +1131,13 @@ ttm_bo_swapout_cb(struct ttm_lru_walk *walk,
> > struct ttm_buffer_object *bo)
> > if (ttm_tt_is_populated(bo->ttm))
> > ret = ttm_tt_swapout(bo->bdev, bo->ttm,
> > swapout_walk->gfp_flags);
> >
> > + if (ttm_tt_is_swapped(bo->ttm)) {
> > + spin_lock(&bo->bdev->lru_lock);
> > + ttm_resource_del_bulk_move(bo->resource, bo);
> > + ttm_resource_move_to_lru_tail(bo->resource);
> > + spin_unlock(&bo->bdev->lru_lock);
> > + }
> > +
> > out:
> > /* Consider -ENOMEM and -ENOSPC non-fatal. */
> > if (ret == -ENOMEM || ret == -ENOSPC)
> > @@ -1180,3 +1187,49 @@ void ttm_bo_tt_destroy(struct
> > ttm_buffer_object *bo)
> > ttm_tt_destroy(bo->bdev, bo->ttm);
> > bo->ttm = NULL;
> > }
> > +
> > +/**
> > + * ttm_bo_populate() - Ensure that a buffer object has backing
> > pages
> > + * @bo: The buffer object
> > + * @ctx: The ttm_operation_ctx governing the operation.
> > + *
> > + * For buffer objects in a memory type whose manager uses
> > + * struct ttm_tt for backing pages, ensure those backing pages
> > + * are present and with valid content. The bo's resource is also
> > + * placed on the correct LRU list if it was previously swapped
> > + * out.
> > + *
> > + * Return: 0 if successful, negative error code on failure.
> > + * Note: May return -EINTR or -ERESTARTSYS if @ctx::interruptible
> > + * is set to true.
> > + */
> > +int ttm_bo_populate(struct ttm_buffer_object *bo,
> > + struct ttm_operation_ctx *ctx)
> > +{
> > + struct ttm_tt *tt = bo->ttm;
> > + bool swapped;
> > + int ret;
> > +
> > + dma_resv_assert_held(bo->base.resv);
> > +
> > + if (!tt)
> > + return 0;
> > +
> > + swapped = ttm_tt_is_swapped(tt);
> > + ret = ttm_tt_populate(bo->bdev, tt, ctx);
> > + if (ret)
> > + return ret;
> > +
> > + if (swapped && !ttm_tt_is_swapped(tt) && !bo->pin_count) {
> > + if (WARN_ON_ONCE(bo->pin_count))
> > + return 0;
>
> You have both "&& !bo->pin_count" and a WARN_ON(bo->pin_count), that
> doesn't make much sense.
Yeah, I guess that won't catch much more than buggy processors. I'll
remove.
>
> > +
> > + spin_lock(&bo->bdev->lru_lock);
> > + ttm_resource_add_bulk_move(bo->resource, bo);
> > + ttm_resource_move_to_lru_tail(bo->resource);
> > + spin_unlock(&bo->bdev->lru_lock);
> > + }
> > +
> > + return 0;
> > +}
> > +EXPORT_SYMBOL(ttm_bo_populate);
> > diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c
> > b/drivers/gpu/drm/ttm/ttm_bo_util.c
> > index 3c07f4712d5c..d939925efa81 100644
> > --- a/drivers/gpu/drm/ttm/ttm_bo_util.c
> > +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
> > @@ -163,7 +163,7 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object
> > *bo,
> > src_man = ttm_manager_type(bdev, src_mem->mem_type);
> > if (ttm && ((ttm->page_flags & TTM_TT_FLAG_SWAPPED) ||
> > dst_man->use_tt)) {
> > - ret = ttm_tt_populate(bdev, ttm, ctx);
> > + ret = ttm_bo_populate(bo, ctx);
> > if (ret)
> > return ret;
> > }
> > @@ -350,7 +350,7 @@ static int ttm_bo_kmap_ttm(struct
> > ttm_buffer_object *bo,
> >
> > BUG_ON(!ttm);
> >
> > - ret = ttm_tt_populate(bo->bdev, ttm, &ctx);
> > + ret = ttm_bo_populate(bo, &ctx);
> > if (ret)
> > return ret;
> >
> > @@ -507,7 +507,7 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo,
> > struct iosys_map *map)
> > pgprot_t prot;
> > void *vaddr;
> >
> > - ret = ttm_tt_populate(bo->bdev, ttm, &ctx);
> > + ret = ttm_bo_populate(bo, &ctx);
> > if (ret)
> > return ret;
> >
> > diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c
> > b/drivers/gpu/drm/ttm/ttm_bo_vm.c
> > index 4212b8c91dd4..2c699ed1963a 100644
> > --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
> > +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
> > @@ -224,7 +224,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct
> > vm_fault *vmf,
> > };
> >
> > ttm = bo->ttm;
> > - err = ttm_tt_populate(bdev, bo->ttm, &ctx);
> > + err = ttm_bo_populate(bo, &ctx);
> > if (err) {
> > if (err == -EINTR || err == -ERESTARTSYS
> > ||
> > err == -EAGAIN)
> > diff --git a/drivers/gpu/drm/ttm/ttm_device.c
> > b/drivers/gpu/drm/ttm/ttm_device.c
> > index e7cc4954c1bc..02e797fd1891 100644
> > --- a/drivers/gpu/drm/ttm/ttm_device.c
> > +++ b/drivers/gpu/drm/ttm/ttm_device.c
> > @@ -216,7 +216,7 @@ int ttm_device_init(struct ttm_device *bdev,
> > const struct ttm_device_funcs *func
> >
> > bdev->vma_manager = vma_manager;
> > spin_lock_init(&bdev->lru_lock);
> > - INIT_LIST_HEAD(&bdev->pinned);
> > + INIT_LIST_HEAD(&bdev->unevictable);
> > bdev->dev_mapping = mapping;
> > mutex_lock(&ttm_global_mutex);
> > list_add_tail(&bdev->device_list, &glob->device_list);
> > @@ -283,7 +283,7 @@ void ttm_device_clear_dma_mappings(struct
> > ttm_device *bdev)
> > struct ttm_resource_manager *man;
> > unsigned int i, j;
> >
> > - ttm_device_clear_lru_dma_mappings(bdev, &bdev->pinned);
> > + ttm_device_clear_lru_dma_mappings(bdev, &bdev-
> > >unevictable);
> >
> > for (i = TTM_PL_SYSTEM; i < TTM_NUM_MEM_TYPES; ++i) {
> > man = ttm_manager_type(bdev, i);
> > diff --git a/drivers/gpu/drm/ttm/ttm_resource.c
> > b/drivers/gpu/drm/ttm/ttm_resource.c
> > index 6d764ba88aab..9d54c0e3e43d 100644
> > --- a/drivers/gpu/drm/ttm/ttm_resource.c
> > +++ b/drivers/gpu/drm/ttm/ttm_resource.c
> > @@ -30,6 +30,7 @@
> > #include <drm/ttm/ttm_bo.h>
> > #include <drm/ttm/ttm_placement.h>
> > #include <drm/ttm/ttm_resource.h>
> > +#include <drm/ttm/ttm_tt.h>
> >
> > #include <drm/drm_util.h>
> >
> > @@ -259,8 +260,8 @@ void ttm_resource_move_to_lru_tail(struct
> > ttm_resource *res)
> >
> > lockdep_assert_held(&bo->bdev->lru_lock);
> >
> > - if (bo->pin_count) {
> > - list_move_tail(&res->lru.link, &bdev->pinned);
> > + if (bo->pin_count || (bo->ttm && ttm_tt_is_swapped(bo-
> > >ttm))) {
> > + list_move_tail(&res->lru.link, &bdev-
> > >unevictable);
>
> You need to change ttm_resource_add_bulk_move() and
> ttm_resource_del_bulk_move() as well.
Ugh. HMM, that will result in some slightly ugly code since we can't
remove from a bulk move twice. Those pos->first and pos->last updates
seem to have caused some grief in the past when it comes to fragility.
Perhaps moving forward aim for using special struct ttm_lru_items for
those, which would make them self-adjusting.
Anyway, I'll update.
>
> >
> > } else if (bo->bulk_move) {
> > struct ttm_lru_bulk_move_pos *pos =
> > @@ -301,8 +302,8 @@ void ttm_resource_init(struct ttm_buffer_object
> > *bo,
> >
> > man = ttm_manager_type(bo->bdev, place->mem_type);
> > spin_lock(&bo->bdev->lru_lock);
> > - if (bo->pin_count)
> > - list_add_tail(&res->lru.link, &bo->bdev->pinned);
> > + if (bo->pin_count || (bo->ttm && ttm_tt_is_swapped(bo-
> > >ttm)))
> > + list_add_tail(&res->lru.link, &bo->bdev-
> > >unevictable);
> > else
> > list_add_tail(&res->lru.link, &man->lru[bo-
> > >priority]);
> > man->usage += res->size;
> > diff --git a/drivers/gpu/drm/ttm/ttm_tt.c
> > b/drivers/gpu/drm/ttm/ttm_tt.c
> > index 4b51b9023126..d1325cf37b18 100644
> > --- a/drivers/gpu/drm/ttm/ttm_tt.c
> > +++ b/drivers/gpu/drm/ttm/ttm_tt.c
> > @@ -367,7 +367,6 @@ int ttm_tt_populate(struct ttm_device *bdev,
> > }
> > return ret;
> > }
> > -EXPORT_SYMBOL(ttm_tt_populate);
> >
> > void ttm_tt_unpopulate(struct ttm_device *bdev, struct ttm_tt
> > *ttm)
> > {
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > b/drivers/gpu/drm/xe/xe_bo.c
> > index 9df5a16662cf..d7d0add20b77 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -903,7 +903,7 @@ int xe_bo_evict_pinned(struct xe_bo *bo)
> > }
> > }
> >
> > - ret = ttm_tt_populate(bo->ttm.bdev, bo->ttm.ttm, &ctx);
> > + ret = ttm_bo_populate(&bo->ttm, &ctx);
> > if (ret)
> > goto err_res_free;
> >
> > @@ -956,7 +956,7 @@ int xe_bo_restore_pinned(struct xe_bo *bo)
> > if (ret)
> > return ret;
> >
> > - ret = ttm_tt_populate(bo->ttm.bdev, bo->ttm.ttm, &ctx);
> > + ret = ttm_bo_populate(&bo->ttm, &ctx);
> > if (ret)
> > goto err_res_free;
> >
> > diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h
> > index 7b56d1ca36d7..5804408815be 100644
> > --- a/include/drm/ttm/ttm_bo.h
> > +++ b/include/drm/ttm/ttm_bo.h
> > @@ -462,5 +462,7 @@ int ttm_bo_pipeline_gutting(struct
> > ttm_buffer_object *bo);
> > pgprot_t ttm_io_prot(struct ttm_buffer_object *bo, struct
> > ttm_resource *res,
> > pgprot_t tmp);
> > void ttm_bo_tt_destroy(struct ttm_buffer_object *bo);
> > +int ttm_bo_populate(struct ttm_buffer_object *bo,
> > + struct ttm_operation_ctx *ctx);
> >
> > #endif
> > diff --git a/include/drm/ttm/ttm_device.h
> > b/include/drm/ttm/ttm_device.h
> > index c22f30535c84..438358f72716 100644
> > --- a/include/drm/ttm/ttm_device.h
> > +++ b/include/drm/ttm/ttm_device.h
> > @@ -252,9 +252,10 @@ struct ttm_device {
> > spinlock_t lru_lock;
> >
> > /**
> > - * @pinned: Buffer objects which are pinned and so not on
> > any LRU list.
> > + * @unevictable Buffer objects which are pinned or swapped
> > and as such
> > + * not on an LRU list.
>
> Either "a LRU list" or "any LRU list".
It's actually "an" since the L in LRU is pronounced with a leading
Vowel sound.
>
> Apart from that this change makes a lot of sense.
Thanks. Will also update the broken KUNIT tests.
/Thomas
>
> Regards,
> Christian.
>
> > */
> > - struct list_head pinned;
> > + struct list_head unevictable;
> >
> > /**
> > * @dev_mapping: A pointer to the struct address_space for
> > invalidating
> > diff --git a/include/drm/ttm/ttm_tt.h b/include/drm/ttm/ttm_tt.h
> > index 2b9d856ff388..991edafdb2dd 100644
> > --- a/include/drm/ttm/ttm_tt.h
> > +++ b/include/drm/ttm/ttm_tt.h
> > @@ -129,6 +129,11 @@ static inline bool ttm_tt_is_populated(struct
> > ttm_tt *tt)
> > return tt->page_flags & TTM_TT_FLAG_PRIV_POPULATED;
> > }
> >
> > +static inline bool ttm_tt_is_swapped(const struct ttm_tt *tt)
> > +{
> > + return tt->page_flags & TTM_TT_FLAG_SWAPPED;
> > +}
> > +
> > /**
> > * ttm_tt_create
> > *
>
More information about the dri-devel
mailing list