From: Jason Gunthorpe <[email protected]> Like habana, xe wants to check pci_p2pdma_distance(), but unlike habana, it can migrate to system memory and support non-p2p DMAs as well.
Add two exporter SGT mapping types, one that matches P2P and one that matches all of the non-p2p. The pin and map code will force migrate if the non-p2p one is matched. Signed-off-by: Jason Gunthorpe <[email protected]> [ rebased onto v7.3-rc5: CONFIG_DMABUF_MOVE_NOTIFY no longer exists (609fc8766d68 "dma-buf: Always build with DMABUF_MOVE_NOTIFY"), so the IS_ENABLED() guards are dropped, the P2P match is always offered and xe_dma_buf_pin() only walks the attachments. dma_buf_match_mapping() is passed num_match rather than ARRAY_SIZE(sgt_match) so an unfilled entry is never read. Upstream's xe_dma_buf_release() is kept. The pin walk asks the other attachments what they accept with dma_buf_attach_accepts_sgt_p2p(), since those may not have selected a type yet; the attachment being pinned has just been selected and uses dma_buf_sgt_p2p_allowed(). - apopple ] Signed-off-by: Alistair Popple <[email protected]> --- drivers/gpu/drm/xe/xe_dma_buf.c | 62 ++++++++++++++++++++++----------- 1 file changed, 41 insertions(+), 21 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_dma_buf.c b/drivers/gpu/drm/xe/xe_dma_buf.c index bf0728838ead..3ca198da6ee8 100644 --- a/drivers/gpu/drm/xe/xe_dma_buf.c +++ b/drivers/gpu/drm/xe/xe_dma_buf.c @@ -7,7 +7,7 @@ #include <kunit/test.h> #include <linux/dma-buf.h> -#include <linux/pci-p2pdma.h> +#include <linux/dma-buf-mapping.h> #include <drm/drm_device.h> #include <drm/drm_prime.h> @@ -27,13 +27,6 @@ static int xe_dma_buf_attach(struct dma_buf *dmabuf, { struct drm_gem_object *obj = attach->dmabuf->priv; - if (attach->peer2peer && - pci_p2pdma_distance(to_pci_dev(obj->dev->dev), attach->dev, false) < 0) - attach->peer2peer = false; - - if (!attach->peer2peer && !xe_bo_can_migrate(gem_to_xe_bo(obj), XE_PL_TT)) - return -EOPNOTSUPP; - xe_pm_runtime_get(to_xe_device(obj->dev)); return 0; } @@ -53,13 +46,16 @@ static int xe_dma_buf_pin(struct dma_buf_attachment *attach) struct xe_bo *bo = gem_to_xe_bo(obj); struct xe_device *xe = xe_bo_device(bo); struct drm_exec *exec = XE_VALIDATION_UNSUPPORTED; - bool allow_vram = true; + bool allow_vram = dma_buf_sgt_p2p_allowed(attach); int ret; - list_for_each_entry(attach, &dmabuf->attachments, node) { - if (!attach->peer2peer) { - allow_vram = false; - break; + if (allow_vram) { + /* Other attachments may be unmapped: ask what they accept */ + list_for_each_entry(attach, &dmabuf->attachments, node) { + if (!dma_buf_attach_accepts_sgt_p2p(attach)) { + allow_vram = false; + break; + } } } @@ -97,6 +93,8 @@ static void xe_dma_buf_unpin(struct dma_buf_attachment *attach) static struct sg_table *xe_dma_buf_map(struct dma_buf_attachment *attach, enum dma_data_direction dir) { + struct device *dma_dev = dma_buf_sgt_dma_device(attach); + bool peer2peer = dma_buf_sgt_p2p_allowed(attach); struct dma_buf *dma_buf = attach->dmabuf; struct drm_gem_object *obj = dma_buf->priv; struct xe_bo *bo = gem_to_xe_bo(obj); @@ -104,11 +102,11 @@ static struct sg_table *xe_dma_buf_map(struct dma_buf_attachment *attach, struct sg_table *sgt; int r = 0; - if (!attach->peer2peer && !xe_bo_can_migrate(bo, XE_PL_TT)) + if (!peer2peer && !xe_bo_can_migrate(bo, XE_PL_TT)) return ERR_PTR(-EOPNOTSUPP); if (!xe_bo_is_pinned(bo)) { - if (!attach->peer2peer) + if (!peer2peer) r = xe_bo_migrate(bo, XE_PL_TT, NULL, exec); else r = xe_bo_validate(bo, NULL, false, exec); @@ -124,7 +122,7 @@ static struct sg_table *xe_dma_buf_map(struct dma_buf_attachment *attach, if (IS_ERR(sgt)) return sgt; - if (dma_map_sgtable(attach->dev, sgt, dir, + if (dma_map_sgtable(dma_dev, sgt, dir, DMA_ATTR_SKIP_CPU_SYNC)) goto error_free; break; @@ -133,7 +131,7 @@ static struct sg_table *xe_dma_buf_map(struct dma_buf_attachment *attach, case XE_PL_VRAM1: r = xe_ttm_vram_mgr_alloc_sgt(xe_bo_device(bo), bo->ttm.resource, 0, - bo->ttm.base.size, attach->dev, + bo->ttm.base.size, dma_dev, dir, &sgt); if (r) return ERR_PTR(r); @@ -154,12 +152,14 @@ static void xe_dma_buf_unmap(struct dma_buf_attachment *attach, struct sg_table *sgt, enum dma_data_direction dir) { + struct device *dma_dev = dma_buf_sgt_dma_device(attach); + if (sg_page(sgt->sgl)) { - dma_unmap_sgtable(attach->dev, sgt, dir, 0); + dma_unmap_sgtable(dma_dev, sgt, dir, 0); sg_free_table(sgt); kfree(sgt); } else { - xe_ttm_vram_mgr_free_sgt(attach->dev, dir, sgt); + xe_ttm_vram_mgr_free_sgt(dma_dev, dir, sgt); } } @@ -205,18 +205,38 @@ static void xe_dma_buf_release(struct dma_buf *dmabuf) drm_gem_dmabuf_release(dmabuf); } +static const struct dma_buf_mapping_sgt_exp_ops xe_dma_buf_sgt_ops = { + .map_dma_buf = xe_dma_buf_map, + .unmap_dma_buf = xe_dma_buf_unmap, +}; + +static int xe_dma_buf_match_mapping(struct dma_buf_match_args *args) +{ + struct drm_gem_object *obj = args->dmabuf->priv; + struct dma_buf_mapping_match sgt_match[2]; + unsigned int num_match = 0; + + sgt_match[num_match++] = DMA_BUF_EMAPPING_SGT_P2P( + &xe_dma_buf_sgt_ops, to_pci_dev(obj->dev->dev)); + + if (xe_bo_can_migrate(gem_to_xe_bo(obj), XE_PL_TT)) + sgt_match[num_match++] = + DMA_BUF_EMAPPING_SGT(&xe_dma_buf_sgt_ops); + + return dma_buf_match_mapping(args, sgt_match, num_match); +} + static const struct dma_buf_ops xe_dmabuf_ops = { .attach = xe_dma_buf_attach, .detach = xe_dma_buf_detach, .pin = xe_dma_buf_pin, .unpin = xe_dma_buf_unpin, - .map_dma_buf = xe_dma_buf_map, - .unmap_dma_buf = xe_dma_buf_unmap, .release = xe_dma_buf_release, .begin_cpu_access = xe_dma_buf_begin_cpu_access, .mmap = drm_gem_dmabuf_mmap, .vmap = drm_gem_dmabuf_vmap, .vunmap = drm_gem_dmabuf_vunmap, + .match_mapping = xe_dma_buf_match_mapping, }; struct dma_buf *xe_gem_prime_export(struct drm_gem_object *obj, int flags) -- 2.54.0
