Replace xe_hmm with gpusvm
Subsystems (4)
v2 cover Replace xe_hmm with gpusvm
As a first step to moving userptr handling over to drm, replace the hmm usage in xe over to gpusvm, which already offers similar functionality. As some prep steps we also align on some of the missing pieces that were already handled in xe_hmm.
v2:
- Rework the gpusvm API based on feedback.
- Unify SVM and userptr vm struct so we use the same notifier lock.
- Drop the mark pages as dirty patch.
- Various other improvements.
Signature · 1 line
2.48.1
PATCH v2 1/7 drm/gpusvm: fix hmm_pfn_to_map_order() usage
Handle the case where the hmm range partially covers a huge page (like 2M), otherwise we can potentially end up doing something nasty like mapping memory which is outside the range, and maybe not even mapped by the mm. Fix is based on the xe userptr code, which in a future patch will directly use gpusvm, so needs alignment here.
v2:
- Add kernel-doc (Matt B)
- s/fls/ilog2/ (Thomas)
--- drivers/gpu/drm/drm_gpusvm.c | 33 +++++++++++++++++++++++++++++++-- 1 file changed, 31 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gpusvm.c +31 -2
@@ -817,6 +817,35 @@ drm_gpusvm_range_alloc(struct drm_gpusvm *gpusvm,return range; } +/** +* drm_gpusvm_hmm_pfn_to_order() - Get the largest CPU mapping order. +* @hmm_pfn: The current hmm_pfn. +* @hmm_pfn_index: Index of the @hmm_pfn within the pfn array. +* @npages: Number of pages within the pfn array i.e the hmm range size. +* +* To allow skipping PFNs with the same flags (like when they belong to +* the same huge PTE) when looping over the pfn array, take a given a hmm_pfn, +* and return the largest order that will fit inside the CPU PTE, but also +* crucially accounting for the original hmm range boundaries. +* +* Return: The largest order that will safely fit within the size of the hmm_pfn +* CPU PTE. +*/ +static unsigned int drm_gpusvm_hmm_pfn_to_order (unsigned long hmm_pfn, +unsigned long hmm_pfn_index, +unsigned long npages) +{ +unsigned long size; + + size =1UL <<hmm_pfn_to_map_order (hmm_pfn); + size -= (hmm_pfn & ~HMM_PFN_FLAGS ) & (size -1 ); + hmm_pfn_index += size; +if (hmm_pfn_index > npages) + size -= (hmm_pfn_index - npages); + +return ilog2 (size); +} +/** * drm_gpusvm_check_pages() - Check pages * @gpusvm: Pointer to the GPU SVM structure
@@ -875,7 +904,7 @@ static bool drm_gpusvm_check_pages(struct drm_gpusvm *gpusvm, err = -EFAULT ; goto err_free; } - i +=0x1 <<hmm_pfn_to_map_order (pfns[i]); + i +=0x1 <<drm_gpusvm_hmm_pfn_to_order (pfns[i], i, npages); } err_free:
@@ -1408,7 +1437,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,for (i =0 , j =0 ; i < npages; ++j) {struct page *page =hmm_pfn_to_page (pfns[i]); - order =hmm_pfn_to_map_order (pfns[i]); + order =drm_gpusvm_hmm_pfn_to_order (pfns[i], i, npages);if (is_device_private_page (page) ||is_device_coherent_page (page)) {if (zdd != page->zone_device_data && i >0 ) {
Signature · 1 line
2.48.1
PATCH v2 2/7 drm/gpusvm: use more selective dma dir in get_pages()
In reply to Matthew AuldIf we are only reading the memory then from the device pov the direction can be DMA_TO_DEVICE. This aligns with the xe-userptr code. Using the most restrictive data direction to represent the access is normally a good idea.
--- drivers/gpu/drm/drm_gpusvm.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_gpusvm.c +5 -3
@@ -1363,6 +1363,8 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,int err =0 ;struct dev_pagemap *pagemap;struct drm_pagemap *dpagemap; +enum dma_data_direction dma_dir = ctx->read_only ?DMA_TO_DEVICE : +DMA_BIDIRECTIONAL ; retry: hmm_range.notifier_seq =mmu_interval_read_begin (notifier);
@@ -1467,7 +1469,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, dpagemap->ops->device_map (dpagemap, gpusvm->drm->dev, page, order, -DMA_BIDIRECTIONAL ); + dma_dir);if (dma_mapping_error (gpusvm->drm->dev, range->dma_addr[j].addr)) { err = -EFAULT ;
@@ -1486,7 +1488,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, addr =dma_map_page (gpusvm->drm->dev, page,0 ,PAGE_SIZE << order, -DMA_BIDIRECTIONAL ); + dma_dir);if (dma_mapping_error (gpusvm->drm->dev, addr)) { err = -EFAULT ; goto err_unmap;
@@ -1494,7 +1496,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, range->dma_addr[j] =drm_pagemap_device_addr_encode (addr,DRM_INTERCONNECT_SYSTEM , order, -DMA_BIDIRECTIONAL ); + dma_dir); } i +=1 << order; num_dma_mapped = i;
Signature · 1 line
2.48.1
PATCH v2 3/7 drm/gpusvm: pull out drm_gpusvm_pages substructure
In reply to Matthew AuldPull the pages stuff from the svm range into its own substructure, with the idea of having the main pages related routines, like get_pages(), unmap_pages() and free_pages() all operating on some lower level structures, which can then be re-used for stuff like userptr.
v2:
- Move seq into pages struct (Matt B)
--- drivers/gpu/drm/drm_gpusvm.c | 66 ++++++++++++++++++++---------------- drivers/gpu/drm/xe/xe_pt.c | 2 +- drivers/gpu/drm/xe/xe_svm.c | 8 ++--- drivers/gpu/drm/xe/xe_svm.h | 6 ++-- include/drm/drm_gpusvm.h | 53 +++++++++++++++++------------ 5 files changed, 76 insertions(+), 59 deletions(-)
drivers/gpu/drm/drm_gpusvm.c +36 -30
@@ -811,8 +811,8 @@ drm_gpusvm_range_alloc(struct drm_gpusvm *gpusvm, range->itree.start =ALIGN_DOWN (fault_addr, chunk_size); range->itree.last =ALIGN (fault_addr +1 , chunk_size) -1 ;INIT_LIST_HEAD (&range->entry); - range->notifier_seq =LONG_MAX ; - range->flags.migrate_devmem = migrate_devmem ?1 :0 ; + range->pages.notifier_seq =LONG_MAX ; + range->pages.flags.migrate_devmem = migrate_devmem ?1 :0 ;return range; }
@@ -1140,15 +1140,16 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm, struct drm_gpusvm_range *range, unsigned long npages) { -unsigned long i, j; -struct drm_pagemap *dpagemap = range->dpagemap; +struct drm_gpusvm_pages *svm_pages = &range->pages; +struct drm_pagemap *dpagemap = svm_pages->dpagemap;struct device *dev = gpusvm->drm->dev; +unsigned long i, j;lockdep_assert_held (&gpusvm->notifier_lock); -if (range->flags.has_dma_mapping) { +if (svm_pages->flags.has_dma_mapping) {for (i =0 , j =0 ; i < npages; j++) { -struct drm_pagemap_device_addr *addr = &range->dma_addr[j]; +struct drm_pagemap_device_addr *addr = &svm_pages->dma_addr[j];if (addr->proto ==DRM_INTERCONNECT_SYSTEM ) dma_unmap_page(dev,
@@ -1160,9 +1161,9 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm, dev, *addr); i +=1 << addr->order; } - range->flags.has_devmem_pages = false; - range->flags.has_dma_mapping = false; - range->dpagemap =NULL ; + svm_pages->flags.has_devmem_pages = false; + svm_pages->flags.has_dma_mapping = false; + svm_pages->dpagemap =NULL ; } }
@@ -1176,11 +1177,13 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,static void drm_gpusvm_range_free_pages (struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range) { +struct drm_gpusvm_pages *svm_pages = &range->pages; +lockdep_assert_held (&gpusvm->notifier_lock); -if (range->dma_addr) { -kvfree (range->dma_addr); - range->dma_addr =NULL ; +if (svm_pages->dma_addr) { +kvfree (svm_pages->dma_addr); + svm_pages->dma_addr =NULL ; } }
@@ -1291,9 +1294,11 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_put);bool drm_gpusvm_range_pages_valid (struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range) { +struct drm_gpusvm_pages *svm_pages = &range->pages; +lockdep_assert_held (&gpusvm->notifier_lock); -return range->flags.has_devmem_pages || range->flags.has_dma_mapping; +return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping; }EXPORT_SYMBOL_GPL (drm_gpusvm_range_pages_valid);
@@ -1311,9 +1316,10 @@ static bool drm_gpusvm_range_pages_valid_unlocked(struct drm_gpusvm *gpusvm, struct drm_gpusvm_range *range) { + struct drm_gpusvm_pages *svm_pages = &range->pages;bool pages_valid; -if (!range->dma_addr) +if (!svm_pages->dma_addr)return false;drm_gpusvm_notifier_lock (gpusvm);
@@ -1340,6 +1346,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range, const struct drm_gpusvm_ctx *ctx) { + struct drm_gpusvm_pages *svm_pages = &range->pages;struct mmu_interval_notifier *notifier = &range->notifier->notifier;struct hmm_range hmm_range = { .default_flags = HMM_PFN_REQ_FAULT | (ctx->read_only ? 0 :
@@ -1409,7 +1416,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,*/ drm_gpusvm_notifier_lock (gpusvm); -if (range->flags.unmapped) { +if (svm_pages->flags.unmapped) {drm_gpusvm_notifier_unlock (gpusvm); err = -EFAULT ; goto err_free;
@@ -1421,13 +1428,12 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto retry; } -if (!range->dma_addr) { +if (!svm_pages->dma_addr) {/* Unlock and restart mapping to allocate memory. */ drm_gpusvm_notifier_unlock (gpusvm); - range->dma_addr =kvmalloc_array (npages, -sizeof (*range->dma_addr), -GFP_KERNEL ); -if (!range->dma_addr) { + svm_pages->dma_addr = +kvmalloc_array (npages,sizeof (*svm_pages->dma_addr),GFP_KERNEL ); +if (!svm_pages->dma_addr) { err = -ENOMEM ; goto err_free; }
@@ -1465,13 +1471,13 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto err_unmap; } } - range->dma_addr[j] = + svm_pages->dma_addr[j] = dpagemap->ops->device_map (dpagemap, gpusvm->drm->dev, page, order, dma_dir);if (dma_mapping_error (gpusvm->drm->dev, - range->dma_addr[j].addr)) { + svm_pages->dma_addr[j].addr)) { err = -EFAULT ; goto err_unmap; }
@@ -1494,7 +1500,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto err_unmap; } - range->dma_addr[j] =drm_pagemap_device_addr_encode + svm_pages->dma_addr[j] =drm_pagemap_device_addr_encode (addr,DRM_INTERCONNECT_SYSTEM , order, dma_dir); }
@@ -1502,16 +1508,16 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, num_dma_mapped = i; } - range->flags.has_dma_mapping = true; + svm_pages->flags.has_dma_mapping = true;if (zdd) { - range->flags.has_devmem_pages = true; - range->dpagemap = dpagemap; + svm_pages->flags.has_devmem_pages = true; + svm_pages->dpagemap = dpagemap; }drm_gpusvm_notifier_unlock (gpusvm);kvfree (pfns); set_seqno: - range->notifier_seq = hmm_range.notifier_seq; + svm_pages->notifier_seq = hmm_range.notifier_seq;return 0 ;
@@ -1718,7 +1724,7 @@ int drm_gpusvm_migrate_to_devmem(struct drm_gpusvm *gpusvm,mmap_assert_locked (gpusvm->mm); -if (!range->flags.migrate_devmem) +if (!range->pages.flags.migrate_devmem)return -EINVAL ; if (!ops->populate_devmem_pfn || !ops->copy_to_devmem ||
@@ -2247,10 +2253,10 @@ void drm_gpusvm_range_set_unmapped(struct drm_gpusvm_range *range, {lockdep_assert_held_write (&range->gpusvm->notifier_lock); - range->flags.unmapped = true; + range->pages.flags.unmapped = true;if (drm_gpusvm_range_start (range) < mmu_range->start ||drm_gpusvm_range_end (range) > mmu_range->end) - range->flags.partial_unmap = true; + range->pages.flags.partial_unmap = true; }EXPORT_SYMBOL_GPL (drm_gpusvm_range_set_unmapped);
drivers/gpu/drm/xe/xe_pt.c +1 -1
@@ -701,7 +701,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,return -EAGAIN ; }if (xe_svm_range_has_dma_mapping (range)) { -xe_res_first_dma (range->base.dma_addr,0 , +xe_res_first_dma (range->base.pages.dma_addr,0 , range->base.itree.last +1 - range->base.itree.start, &curs);xe_svm_range_debug (range,"BIND PREPARE - MIXED" );
drivers/gpu/drm/xe/xe_svm.c +4 -4
@@ -17,7 +17,7 @@static bool xe_svm_range_in_vram (struct xe_svm_range *range) {/* Not reliable without notifier lock */ -return range->base.flags.has_devmem_pages; +return range->base.pages.flags.has_devmem_pages; }static bool xe_svm_range_has_vram_binding (struct xe_svm_range *range)
@@ -59,7 +59,7 @@ static unsigned long xe_svm_range_size(struct xe_svm_range *range) (r__)->base.gpusvm, \xe_svm_range_in_vram ((r__)) ?1 :0 , \xe_svm_range_has_vram_binding ((r__)) ?1 :0 , \ - (r__)->base.notifier_seq, \ + (r__)->base.pages.notifier_seq, \xe_svm_range_start ((r__)),xe_svm_range_end ((r__)), \ xe_svm_range_size((r__)))
@@ -135,7 +135,7 @@ xe_svm_range_notifier_event_begin(struct xe_vm *vm, struct drm_gpusvm_range *r,range_debug (range,"NOTIFIER" );/* Skip if already unmapped or if no binding exist */ -if (range->base.flags.unmapped || !range->tile_present) +if (range->base.pages.flags.unmapped || !range->tile_present)return 0 ;range_debug (range,"NOTIFIER - EXECUTE" );
@@ -783,7 +783,7 @@ int xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma,range_debug (range,"PAGE FAULT" );/* XXX: Add migration policy, for now migrate range once */ -if (!range->skip_migrate && range->base.flags.migrate_devmem && +if (!range->skip_migrate && range->base.pages.flags.migrate_devmem &&xe_svm_range_size (range) >=SZ_64K ) { range->skip_migrate = true;
drivers/gpu/drm/xe/xe_svm.h +4 -2
@@ -84,7 +84,7 @@ void xe_svm_range_debug(struct xe_svm_range *range, const char *operation);static inline bool xe_svm_range_has_dma_mapping (struct xe_svm_range *range) {lockdep_assert_held (&range->base.gpusvm->notifier_lock); -return range->base.flags.has_dma_mapping; +return range->base.pages.flags.has_dma_mapping; }#define xe_svm_assert_in_notifier (vm__) \
@@ -112,7 +112,9 @@ struct xe_vram_region;struct xe_svm_range {struct {struct interval_tree_node itree; -const struct drm_pagemap_device_addr *dma_addr; +struct { +struct drm_pagemap_device_addr *dma_addr; + } pages; } base;u32 tile_present;u32 tile_invalidated;
include/drm/drm_gpusvm.h +31 -22
71 lines not on this page · raw message
Signature · 1 line
2.48.1
Re: [PATCH v2 3/7] drm/gpusvm: pull out drm_gpusvm_pages substructure
Quoted 9 lines · Matthew Auld
Pull the pages stuff from the svm range into its own substructure, with the idea of having the main pages related routines, like
get_pages(),unmap_pages()andfree_pages()all operating on some lower level structures, which can then be re-used for stuff like userptr.v2:
- Move seq into pages struct (Matt B)Suggested-by Matthew Brost <[email protected]>
Quoted 362 lines
Signed-off-by Matthew Auld <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/drm_gpusvm.c | 66 ++++++++++++++++++++---------------- drivers/gpu/drm/xe/xe_pt.c | 2 +- drivers/gpu/drm/xe/xe_svm.c | 8 ++--- drivers/gpu/drm/xe/xe_svm.h | 6 ++-- include/drm/drm_gpusvm.h | 53 +++++++++++++++++------------ 5 files changed, 76 insertions(+), 59 deletions(-)drivers/gpu/drm/drm_gpusvm.c +36
-30@@ -811,8 +811,8 @@ drm_gpusvm_range_alloc(struct drm_gpusvm *gpusvm, range->itree.start =ALIGN_DOWN (fault_addr, chunk_size); range->itree.last =ALIGN (fault_addr +1 , chunk_size) -1 ;INIT_LIST_HEAD (&range->entry); - range->notifier_seq =LONG_MAX ; - range->flags.migrate_devmem = migrate_devmem ?1 :0 ; + range->pages.notifier_seq =LONG_MAX ; + range->pages.flags.migrate_devmem = migrate_devmem ?1 :0 ;return range; }@@ -1140,15 +1140,16 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm, struct drm_gpusvm_range *range, unsigned long npages) { -unsigned long i, j; -struct drm_pagemap *dpagemap = range->dpagemap; +struct drm_gpusvm_pages *svm_pages = &range->pages; +struct drm_pagemap *dpagemap = svm_pages->dpagemap;struct device *dev = gpusvm->drm->dev; +unsigned long i, j;lockdep_assert_held (&gpusvm->notifier_lock); -if (range->flags.has_dma_mapping) { +if (svm_pages->flags.has_dma_mapping) {for (i =0 , j =0 ; i < npages; j++) { -struct drm_pagemap_device_addr *addr = &range->dma_addr[j]; +struct drm_pagemap_device_addr *addr = &svm_pages->dma_addr[j];if (addr->proto ==DRM_INTERCONNECT_SYSTEM ) dma_unmap_page(dev,@@ -1160,9 +1161,9 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm, dev, *addr); i +=1 << addr->order; } - range->flags.has_devmem_pages = false; - range->flags.has_dma_mapping = false; - range->dpagemap =NULL ; + svm_pages->flags.has_devmem_pages = false; + svm_pages->flags.has_dma_mapping = false; + svm_pages->dpagemap =NULL ; } }@@ -1176,11 +1177,13 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,static void drm_gpusvm_range_free_pages (struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range) { +struct drm_gpusvm_pages *svm_pages = &range->pages; +lockdep_assert_held (&gpusvm->notifier_lock); -if (range->dma_addr) { -kvfree (range->dma_addr); - range->dma_addr =NULL ; +if (svm_pages->dma_addr) { +kvfree (svm_pages->dma_addr); + svm_pages->dma_addr =NULL ; } }@@ -1291,9 +1294,11 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_put);bool drm_gpusvm_range_pages_valid (struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range) { +struct drm_gpusvm_pages *svm_pages = &range->pages; +lockdep_assert_held (&gpusvm->notifier_lock); -return range->flags.has_devmem_pages || range->flags.has_dma_mapping; +return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping; }EXPORT_SYMBOL_GPL (drm_gpusvm_range_pages_valid);@@ -1311,9 +1316,10 @@ static bool drm_gpusvm_range_pages_valid_unlocked(struct drm_gpusvm *gpusvm, struct drm_gpusvm_range *range) { + struct drm_gpusvm_pages *svm_pages = &range->pages;bool pages_valid; -if (!range->dma_addr) +if (!svm_pages->dma_addr)return false;drm_gpusvm_notifier_lock (gpusvm);@@ -1340,6 +1346,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range, const struct drm_gpusvm_ctx *ctx) { + struct drm_gpusvm_pages *svm_pages = &range->pages;struct mmu_interval_notifier *notifier = &range->notifier->notifier;struct hmm_range hmm_range = { .default_flags = HMM_PFN_REQ_FAULT | (ctx->read_only ? 0 :@@ -1409,7 +1416,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,*/ drm_gpusvm_notifier_lock (gpusvm); -if (range->flags.unmapped) { +if (svm_pages->flags.unmapped) {drm_gpusvm_notifier_unlock (gpusvm); err = -EFAULT ; goto err_free;@@ -1421,13 +1428,12 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto retry; } -if (!range->dma_addr) { +if (!svm_pages->dma_addr) {/* Unlock and restart mapping to allocate memory. */ drm_gpusvm_notifier_unlock (gpusvm); - range->dma_addr =kvmalloc_array (npages, -sizeof (*range->dma_addr), -GFP_KERNEL ); -if (!range->dma_addr) { + svm_pages->dma_addr = +kvmalloc_array (npages,sizeof (*svm_pages->dma_addr),GFP_KERNEL ); +if (!svm_pages->dma_addr) { err = -ENOMEM ; goto err_free; }@@ -1465,13 +1471,13 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto err_unmap; } } - range->dma_addr[j] = + svm_pages->dma_addr[j] = dpagemap->ops->device_map (dpagemap, gpusvm->drm->dev, page, order, dma_dir);if (dma_mapping_error (gpusvm->drm->dev, - range->dma_addr[j].addr)) { + svm_pages->dma_addr[j].addr)) { err = -EFAULT ; goto err_unmap; }@@ -1494,7 +1500,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto err_unmap; } - range->dma_addr[j] =drm_pagemap_device_addr_encode + svm_pages->dma_addr[j] =drm_pagemap_device_addr_encode (addr,DRM_INTERCONNECT_SYSTEM , order, dma_dir); }@@ -1502,16 +1508,16 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, num_dma_mapped = i; } - range->flags.has_dma_mapping = true; + svm_pages->flags.has_dma_mapping = true;if (zdd) { - range->flags.has_devmem_pages = true; - range->dpagemap = dpagemap; + svm_pages->flags.has_devmem_pages = true; + svm_pages->dpagemap = dpagemap; }drm_gpusvm_notifier_unlock (gpusvm);kvfree (pfns); set_seqno: - range->notifier_seq = hmm_range.notifier_seq; + svm_pages->notifier_seq = hmm_range.notifier_seq;return 0 ;@@ -1718,7 +1724,7 @@ int drm_gpusvm_migrate_to_devmem(struct drm_gpusvm *gpusvm,mmap_assert_locked (gpusvm->mm); -if (!range->flags.migrate_devmem) +if (!range->pages.flags.migrate_devmem)return -EINVAL ; if (!ops->populate_devmem_pfn || !ops->copy_to_devmem ||@@ -2247,10 +2253,10 @@ void drm_gpusvm_range_set_unmapped(struct drm_gpusvm_range *range, {lockdep_assert_held_write (&range->gpusvm->notifier_lock); - range->flags.unmapped = true; + range->pages.flags.unmapped = true;if (drm_gpusvm_range_start (range) < mmu_range->start ||drm_gpusvm_range_end (range) > mmu_range->end) - range->flags.partial_unmap = true; + range->pages.flags.partial_unmap = true; }EXPORT_SYMBOL_GPL (drm_gpusvm_range_set_unmapped);drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -701,7 +701,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,return -EAGAIN ; }if (xe_svm_range_has_dma_mapping (range)) { -xe_res_first_dma (range->base.dma_addr,0 , +xe_res_first_dma (range->base.pages.dma_addr,0 , range->base.itree.last +1 - range->base.itree.start, &curs);xe_svm_range_debug (range,"BIND PREPARE - MIXED" );drivers/gpu/drm/xe/xe_svm.c +4
-4@@ -17,7 +17,7 @@static bool xe_svm_range_in_vram (struct xe_svm_range *range) {/* Not reliable without notifier lock */ -return range->base.flags.has_devmem_pages; +return range->base.pages.flags.has_devmem_pages; }static bool xe_svm_range_has_vram_binding (struct xe_svm_range *range)@@ -59,7 +59,7 @@ static unsigned long xe_svm_range_size(struct xe_svm_range *range) (r__)->base.gpusvm, \xe_svm_range_in_vram ((r__)) ?1 :0 , \xe_svm_range_has_vram_binding ((r__)) ?1 :0 , \ - (r__)->base.notifier_seq, \ + (r__)->base.pages.notifier_seq, \xe_svm_range_start ((r__)),xe_svm_range_end ((r__)), \ xe_svm_range_size((r__)))@@ -135,7 +135,7 @@ xe_svm_range_notifier_event_begin(struct xe_vm *vm, struct drm_gpusvm_range *r,range_debug (range,"NOTIFIER" );/* Skip if already unmapped or if no binding exist */ -if (range->base.flags.unmapped || !range->tile_present) +if (range->base.pages.flags.unmapped || !range->tile_present)return 0 ;range_debug (range,"NOTIFIER - EXECUTE" );@@ -783,7 +783,7 @@ int xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma,range_debug (range,"PAGE FAULT" );/* XXX: Add migration policy, for now migrate range once */ -if (!range->skip_migrate && range->base.flags.migrate_devmem && +if (!range->skip_migrate && range->base.pages.flags.migrate_devmem &&xe_svm_range_size (range) >=SZ_64K ) { range->skip_migrate = true;drivers/gpu/drm/xe/xe_svm.h +4
-2@@ -84,7 +84,7 @@ void xe_svm_range_debug(struct xe_svm_range *range, const char *operation);static inline bool xe_svm_range_has_dma_mapping (struct xe_svm_range *range) {lockdep_assert_held (&range->base.gpusvm->notifier_lock); -return range->base.flags.has_dma_mapping; +return range->base.pages.flags.has_dma_mapping; }#define xe_svm_assert_in_notifier (vm__) \@@ -112,7 +112,9 @@ struct xe_vram_region;struct xe_svm_range {struct {struct interval_tree_node itree; -const struct drm_pagemap_device_addr *dma_addr; +struct { +struct drm_pagemap_device_addr *dma_addr; + } pages; } base;u32 tile_present;u32 tile_invalidated;include/drm/drm_gpusvm.h +31
-22@@ -185,6 +185,35 @@ struct drm_gpusvm_notifier { } flags; }; +/** +* struct drm_gpusvm_pages - Structure representing a GPU SVM mapped pages +* +* @dma_addr: Device address array +* @dpagemap: The struct drm_pagemap of the device pages we're dma-mapping. +* Note this is assuming only one drm_pagemap per range is allowed. +* @notifier_seq: Notifier sequence number of the range's pages +* @flags: Flags for range +* @flags.migrate_devmem: Flag indicating whether the range can be migrated to device memory +* @flags.unmapped: Flag indicating if the range has been unmapped +* @flags.partial_unmap: Flag indicating if the range has been partially unmapped +* @flags.has_devmem_pages: Flag indicating if the range has devmem pages +* @flags.has_dma_mapping: Flag indicating if the range has a DMA mapping +*/ +struct drm_gpusvm_pages { +struct drm_pagemap_device_addr *dma_addr; +struct drm_pagemap *dpagemap; +unsigned long notifier_seq; +struct { +/* All flags below must be set upon creation */ +u16 migrate_devmem :1 ; +/* All flags below must be set / cleared under notifier lock */ +u16 unmapped :1 ; +u16 partial_unmap :1 ; +u16 has_devmem_pages :1 ; +u16 has_dma_mapping :1 ; + } flags; +}; +/** * struct drm_gpusvm_range - Structure representing a GPU SVM range * @@ -193,16 +222,7 @@ struct drm_gpusvm_notifier {* @refcount: Reference count for the range * @itree: Interval tree node for the range (inserted in GPU SVM notifier) * @entry: List entry to fast interval tree traversal -* @notifier_seq: Notifier sequence number of the range's pages -* @dma_addr: Device address array -* @dpagemap: The struct drm_pagemap of the device pages we're dma-mapping. -* Note this is assuming only one drm_pagemap per range is allowed. -* @flags: Flags for range -* @flags.migrate_devmem: Flag indicating whether the range can be migrated to device memory -* @flags.unmapped: Flag indicating if the range has been unmapped -* @flags.partial_unmap: Flag indicating if the range has been partially unmapped -* @flags.has_devmem_pages: Flag indicating if the range has devmem pages -* @flags.has_dma_mapping: Flag indicating if the range has a DMA mapping +* @pages: The pages for this range. * * This structure represents a GPU SVM range used for tracking memory ranges * mapped in a DRM device. @@ -213,18 +233,7 @@ struct drm_gpusvm_range {struct kref refcount;struct interval_tree_node itree;struct list_head entry; -unsigned long notifier_seq; -struct drm_pagemap_device_addr *dma_addr; -struct drm_pagemap *dpagemap; -struct { -/* All flags below must be set upon creation */ -u16 migrate_devmem :1 ; -/* All flags below must be set / cleared under notifier lock */ -u16 unmapped :1 ; -u16 partial_unmap :1 ; -u16 has_devmem_pages :1 ; -u16 has_dma_mapping :1 ; - } flags; +struct drm_gpusvm_pages pages; };/** Signature · 1 line
2.48.1
PATCH v2 4/7 drm/gpusvm: refactor core API to use pages struct
In reply to Matthew AuldRefactor the core API of get/unmap/free pages to all operate on drm_gpusvm_pages. In the next patch we want to export a simplified core API without needing fully blown svm range etc.
--- drivers/gpu/drm/drm_gpusvm.c | 161 ++++++++++++++++++++++++----------- 1 file changed, 110 insertions(+), 51 deletions(-)
drivers/gpu/drm/drm_gpusvm.c +110 -51
@@ -1128,19 +1128,19 @@ drm_gpusvm_range_find_or_insert(struct drm_gpusvm *gpusvm,EXPORT_SYMBOL_GPL (drm_gpusvm_range_find_or_insert);/** -* __drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range (internal) +* __drm_gpusvm_unmap_pages() - Unmap pages associated with GPU SVM pages +* (internal) * @gpusvm: Pointer to the GPU SVM structure -* @range: Pointer to the GPU SVM range structure +* @pages: Pointer to the GPU SVM pages structure * @npages: Number of pages to unmap * -* This function unmap pages associated with a GPU SVM range. Assumes and +* This function unmap pages associated with a GPU SVM pages struct. Assumes and * asserts correct locking is in place when called. */ -static void __drm_gpusvm_range_unmap_pages (struct drm_gpusvm *gpusvm, -struct drm_gpusvm_range *range, -unsigned long npages) +static void __drm_gpusvm_unmap_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +unsigned long npages) { -struct drm_gpusvm_pages *svm_pages = &range->pages;struct drm_pagemap *dpagemap = svm_pages->dpagemap;struct device *dev = gpusvm->drm->dev;unsigned long i, j;
@@ -1168,17 +1168,15 @@ static void __drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm, }/** -* drm_gpusvm_range_free_pages() - Free pages associated with a GPU SVM range +* __drm_gpusvm_free_pages() - Free dma array associated with GPU SVM pages * @gpusvm: Pointer to the GPU SVM structure -* @range: Pointer to the GPU SVM range structure +* @svm_pages: Pointer to the GPU SVM pages structure * * This function frees the dma address array associated with a GPU SVM range. */ -static void drm_gpusvm_range_free_pages (struct drm_gpusvm *gpusvm, -struct drm_gpusvm_range *range) +static void __drm_gpusvm_free_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages) { -struct drm_gpusvm_pages *svm_pages = &range->pages; -lockdep_assert_held (&gpusvm->notifier_lock);if (svm_pages->dma_addr) {
@@ -1211,8 +1209,8 @@ void drm_gpusvm_range_remove(struct drm_gpusvm *gpusvm,return ;drm_gpusvm_notifier_lock (gpusvm); -__drm_gpusvm_range_unmap_pages (gpusvm, range, npages); -drm_gpusvm_range_free_pages (gpusvm, range); +__drm_gpusvm_unmap_pages (gpusvm, &range->pages, npages); +__drm_gpusvm_free_pages (gpusvm, &range->pages);__drm_gpusvm_range_remove (notifier, range);drm_gpusvm_notifier_unlock (gpusvm);
@@ -1277,6 +1275,28 @@ void drm_gpusvm_range_put(struct drm_gpusvm_range *range) }EXPORT_SYMBOL_GPL (drm_gpusvm_range_put); +/** +* drm_gpusvm_pages_valid() - GPU SVM range pages valid +* @gpusvm: Pointer to the GPU SVM structure +* @svm_pages: Pointer to the GPU SVM pages structure +* +* This function determines if a GPU SVM range pages are valid. Expected be +* called holding gpusvm->notifier_lock and as the last step before committing a +* GPU binding. This is akin to a notifier seqno check in the HMM documentation +* but due to wider notifiers (i.e., notifiers which span multiple ranges) this +* function is required for finer grained checking (i.e., per range) if pages +* are valid. +* +* Return: True if GPU SVM range has valid pages, False otherwise +*/ +static bool drm_gpusvm_pages_valid (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages) +{ +lockdep_assert_held (&gpusvm->notifier_lock); + +return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping; +} +/** * drm_gpusvm_range_pages_valid() - GPU SVM range pages valid * @gpusvm: Pointer to the GPU SVM structure
@@ -1294,11 +1314,7 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_put);bool drm_gpusvm_range_pages_valid (struct drm_gpusvm *gpusvm,struct drm_gpusvm_range *range) { -struct drm_gpusvm_pages *svm_pages = &range->pages; - -lockdep_assert_held (&gpusvm->notifier_lock); - -return svm_pages->flags.has_devmem_pages || svm_pages->flags.has_dma_mapping; +return drm_gpusvm_pages_valid (gpusvm, &range->pages); }EXPORT_SYMBOL_GPL (drm_gpusvm_range_pages_valid);
@@ -1312,57 +1328,59 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_pages_valid);* * Return: True if GPU SVM range has valid pages, False otherwise */ -static bool -drm_gpusvm_range_pages_valid_unlocked (struct drm_gpusvm *gpusvm, -struct drm_gpusvm_range *range) +static bool drm_gpusvm_pages_valid_unlocked (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages) { -struct drm_gpusvm_pages *svm_pages = &range->pages;bool pages_valid;if (!svm_pages->dma_addr)return false;drm_gpusvm_notifier_lock (gpusvm); - pages_valid =drm_gpusvm_range_pages_valid (gpusvm, range); + pages_valid =drm_gpusvm_pages_valid (gpusvm, svm_pages);if (!pages_valid) -drm_gpusvm_range_free_pages (gpusvm, range); +__drm_gpusvm_free_pages (gpusvm, svm_pages);drm_gpusvm_notifier_unlock (gpusvm);return pages_valid; }/** -* drm_gpusvm_range_get_pages() - Get pages for a GPU SVM range +* drm_gpusvm_get_pages() - Get pages and populate GPU SVM pages struct * @gpusvm: Pointer to the GPU SVM structure -* @range: Pointer to the GPU SVM range structure +* @svm_pages: The SVM pages to populate. This will contain the dma-addresses +* @mm: The mm corresponding to the CPU range +* @notifier: The corresponding notifier for the given CPU range +* @pages_start: Start CPU address for the pages +* @pages_end: End CPU address for the pages (exclusive) * @ctx: GPU SVM context * -* This function gets pages for a GPU SVM range and ensures they are mapped for -* DMA access. +* This function gets and maps pages for CPU range and ensures they are +* mapped for DMA access. * * Return: 0 on success, negative error code on failure. */ -int drm_gpusvm_range_get_pages (struct drm_gpusvm *gpusvm, -struct drm_gpusvm_range *range, -const struct drm_gpusvm_ctx *ctx) +static int drm_gpusvm_get_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +struct mm_struct *mm, +struct mmu_interval_notifier *notifier, +unsigned long pages_start, +unsigned long pages_end, +const struct drm_gpusvm_ctx *ctx) { -struct drm_gpusvm_pages *svm_pages = &range->pages; -struct mmu_interval_notifier *notifier = &range->notifier->notifier;struct hmm_range hmm_range = { .default_flags =HMM_PFN_REQ_FAULT | (ctx->read_only ?0 :HMM_PFN_REQ_WRITE ), .notifier = notifier, - .start =drm_gpusvm_range_start (range), - .end =drm_gpusvm_range_end (range), + .start = pages_start, + .end = pages_end, .dev_private_owner = gpusvm->device_private_page_owner, }; -struct mm_struct *mm = gpusvm->mm;struct drm_gpusvm_zdd *zdd;unsigned long timeout = jiffies +msecs_to_jiffies (HMM_RANGE_DEFAULT_TIMEOUT );unsigned long i, j; -unsigned long npages =npages_in_range (drm_gpusvm_range_start (range), -drm_gpusvm_range_end (range)); +unsigned long npages =npages_in_range (pages_start, pages_end);unsigned long num_dma_mapped;unsigned int order =0 ;unsigned long *pfns;
@@ -1375,7 +1393,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, retry: hmm_range.notifier_seq =mmu_interval_read_begin (notifier); -if (drm_gpusvm_range_pages_valid_unlocked (gpusvm, range)) +if (drm_gpusvm_pages_valid_unlocked (gpusvm, svm_pages)) goto set_seqno; pfns =kvmalloc_array (npages,sizeof (*pfns),GFP_KERNEL );
@@ -1522,7 +1540,7 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm,return 0 ; err_unmap: -__drm_gpusvm_range_unmap_pages (gpusvm, range, num_dma_mapped); +__drm_gpusvm_unmap_pages (gpusvm, svm_pages, num_dma_mapped);drm_gpusvm_notifier_unlock (gpusvm); err_free:kvfree (pfns);
@@ -1530,8 +1548,57 @@ int drm_gpusvm_range_get_pages(struct drm_gpusvm *gpusvm, goto retry;return err; } + +/** +* drm_gpusvm_range_get_pages() - Get pages for a GPU SVM range +* @gpusvm: Pointer to the GPU SVM structure +* @range: Pointer to the GPU SVM range structure +* @ctx: GPU SVM context +* +* This function gets pages for a GPU SVM range and ensures they are mapped for +* DMA access. +* +* Return: 0 on success, negative error code on failure. +*/ +int drm_gpusvm_range_get_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_range *range, +const struct drm_gpusvm_ctx *ctx) +{ +return drm_gpusvm_get_pages (gpusvm, &range->pages, gpusvm->mm, + &range->notifier->notifier, +drm_gpusvm_range_start (range), +drm_gpusvm_range_end (range), ctx); +}EXPORT_SYMBOL_GPL (drm_gpusvm_range_get_pages); +/** +* drm_gpusvm_unmap_pages() - Unmap GPU svm pages +* @gpusvm: Pointer to the GPU SVM structure +* @pages: Pointer to the GPU SVM pages structure +* @ctx: GPU SVM context +* +* This function unmaps pages associated with a GPU SVM pages struct. If +* @in_notifier is set, it is assumed that gpusvm->notifier_lock is held in +* write mode; if it is clear, it acquires gpusvm->notifier_lock in read mode. +* Must be called in the invalidate() callback of the corresponding notifier for +* IOMMU security model. +*/ +static void drm_gpusvm_unmap_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +unsigned long npages, +const struct drm_gpusvm_ctx *ctx) +{ +if (ctx->in_notifier) +lockdep_assert_held_write (&gpusvm->notifier_lock); +else +drm_gpusvm_notifier_lock (gpusvm); + +__drm_gpusvm_unmap_pages (gpusvm, svm_pages, npages); + +if (!ctx->in_notifier) +drm_gpusvm_notifier_unlock (gpusvm); +} +/** * drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range * @gpusvm: Pointer to the GPU SVM structure
@@ -1551,15 +1618,7 @@ void drm_gpusvm_range_unmap_pages(struct drm_gpusvm *gpusvm,unsigned long npages =npages_in_range (drm_gpusvm_range_start (range),drm_gpusvm_range_end (range)); -if (ctx->in_notifier) -lockdep_assert_held_write (&gpusvm->notifier_lock); -else -drm_gpusvm_notifier_lock (gpusvm); - -__drm_gpusvm_range_unmap_pages (gpusvm, range, npages); - -if (!ctx->in_notifier) -drm_gpusvm_notifier_unlock (gpusvm); +return drm_gpusvm_unmap_pages (gpusvm, &range->pages, npages, ctx); }EXPORT_SYMBOL_GPL (drm_gpusvm_range_unmap_pages);
Signature · 1 line
2.48.1
PATCH v2 5/7 drm/gpusvm: export drm_gpusvm_pages API
In reply to Matthew AuldExport get/unmap/free pages API. We also need to tweak the SVM init to allow skipping much of the unneeded parts.
--- drivers/gpu/drm/drm_gpusvm.c | 66 ++++++++++++++++++++++++++++-------- include/drm/drm_gpusvm.h | 16 +++++++++ 2 files changed, 67 insertions(+), 15 deletions(-)
drivers/gpu/drm/drm_gpusvm.c +51 -15
@@ -539,6 +539,12 @@ static const struct mmu_interval_notifier_ops drm_gpusvm_notifier_ops = {* * This function initializes the GPU SVM. * +* Note: If only using the simple drm_gpusvm_pages API (get/unmap/free), +* then only @gpusvm, @name, and @drm are expected. However, the same base +* @gpusvm can also be used with both modes together in which case the full +* setup is needed, where the core drm_gpusvm_pages API will simply never use +* the other fields. +* * Return: 0 on success, a negative error code on failure. */ int drm_gpusvm_init(struct drm_gpusvm *gpusvm,
@@ -549,8 +555,16 @@ int drm_gpusvm_init(struct drm_gpusvm *gpusvm,const struct drm_gpusvm_ops *ops, const unsigned long *chunk_sizes, int num_chunks) { - if (!ops->invalidate || !num_chunks) - return -EINVAL; + if (mm) { + if (!ops->invalidate || !num_chunks) + return -EINVAL; +mmgrab (mm ); + } else { +/* No full SVM mode, only core drm_gpusvm_pages API. */ +if (ops || num_chunks || mm_range || notifier_size || + device_private_page_owner) +return -EINVAL ; + } gpusvm->name = name; gpusvm->drm = drm;
@@ -563,7 +577,6 @@ int drm_gpusvm_init(struct drm_gpusvm *gpusvm, gpusvm->chunk_sizes = chunk_sizes; gpusvm->num_chunks = num_chunks; -mmgrab (mm); gpusvm->root =RB_ROOT_CACHED ;INIT_LIST_HEAD (&gpusvm->notifier_list);
@@ -671,7 +684,8 @@ void drm_gpusvm_fini(struct drm_gpusvm *gpusvm)drm_gpusvm_range_remove (gpusvm, range); } -mmdrop (gpusvm->mm); +if (gpusvm->mm) +mmdrop (gpusvm->mm);WARN_ON (!RB_EMPTY_ROOT (&gpusvm->root.rb_root)); }EXPORT_SYMBOL_GPL (drm_gpusvm_fini);
@@ -1185,6 +1199,27 @@ static void __drm_gpusvm_free_pages(struct drm_gpusvm *gpusvm, } } +/** +* drm_gpusvm_free_pages() - Free dma-mapping associated with GPU SVM pages +* struct +* @gpusvm: Pointer to the GPU SVM structure +* @svm_pages: Pointer to the GPU SVM pages structure +* @npages: Number of mapped pages +* +* This function unmaps and frees the dma address array associated with a GPU +* SVM pages struct. +*/ +void drm_gpusvm_free_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +unsigned long npages) +{ +drm_gpusvm_notifier_lock (gpusvm); +__drm_gpusvm_unmap_pages (gpusvm, svm_pages, npages); +__drm_gpusvm_free_pages (gpusvm, svm_pages); +drm_gpusvm_notifier_unlock (gpusvm); +} +EXPORT_SYMBOL_GPL (drm_gpusvm_free_pages); +/** * drm_gpusvm_range_remove() - Remove GPU SVM range * @gpusvm: Pointer to the GPU SVM structure
@@ -1360,13 +1395,12 @@ static bool drm_gpusvm_pages_valid_unlocked(struct drm_gpusvm *gpusvm,* * Return: 0 on success, negative error code on failure. */ -static int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm, - struct drm_gpusvm_pages *svm_pages, - struct mm_struct *mm, - struct mmu_interval_notifier *notifier, - unsigned long pages_start, - unsigned long pages_end, - const struct drm_gpusvm_ctx *ctx) +int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm, + struct drm_gpusvm_pages *svm_pages, + struct mm_struct *mm, + struct mmu_interval_notifier *notifier, + unsigned long pages_start, unsigned long pages_end, + const struct drm_gpusvm_ctx *ctx) { struct hmm_range hmm_range = { .default_flags = HMM_PFN_REQ_FAULT | (ctx->read_only ? 0 :
@@ -1548,6 +1582,7 @@ static int drm_gpusvm_get_pages(struct drm_gpusvm *gpusvm, goto retry;return err; } +EXPORT_SYMBOL_GPL (drm_gpusvm_get_pages);/** * drm_gpusvm_range_get_pages() - Get pages for a GPU SVM range
@@ -1583,10 +1618,10 @@ EXPORT_SYMBOL_GPL(drm_gpusvm_range_get_pages);* Must be called in the invalidate() callback of the corresponding notifier for * IOMMU security model. */ -static void drm_gpusvm_unmap_pages (struct drm_gpusvm *gpusvm, -struct drm_gpusvm_pages *svm_pages, -unsigned long npages, -const struct drm_gpusvm_ctx *ctx) +void drm_gpusvm_unmap_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +unsigned long npages, +const struct drm_gpusvm_ctx *ctx) {if (ctx->in_notifier)lockdep_assert_held_write (&gpusvm->notifier_lock);
@@ -1598,6 +1633,7 @@ static void drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm,if (!ctx->in_notifier)drm_gpusvm_notifier_unlock (gpusvm); } +EXPORT_SYMBOL_GPL (drm_gpusvm_unmap_pages);/** * drm_gpusvm_range_unmap_pages() - Unmap pages associated with a GPU SVM range
include/drm/drm_gpusvm.h +16 -0
@@ -370,6 +370,22 @@ void drm_gpusvm_devmem_init(struct drm_gpusvm_devmem *devmem_allocation,const struct drm_gpusvm_devmem_ops *ops, struct drm_pagemap *dpagemap, size_t size); +int drm_gpusvm_get_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +struct mm_struct *mm, +struct mmu_interval_notifier *notifier, +unsigned long pages_start,unsigned long pages_end, +const struct drm_gpusvm_ctx *ctx); + +void drm_gpusvm_unmap_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +unsigned long npages, +const struct drm_gpusvm_ctx *ctx); + +void drm_gpusvm_free_pages (struct drm_gpusvm *gpusvm, +struct drm_gpusvm_pages *svm_pages, +unsigned long npages); +#ifdef CONFIG_LOCKDEP /** * drm_gpusvm_driver_set_lock() - Set the lock protecting accesses to GPU SVM
Signature · 1 line
2.48.1
PATCH v2 6/7 drm/xe/userptr: replace xe_hmm with gpusvm
In reply to Matthew AuldGoal here is cut over to gpusvm and remove xe_hmm, relying instead on common code. The core facilities we need are get_pages(), unmap_pages() and free_pages() for a given useptr range, plus a vm level notifier lock, which is now provided by gpusvm.
v2:
- Reuse the same SVM vm struct we use for full SVM, that way we can use the same lock (Matt B & Himal)
--- drivers/gpu/drm/xe/Kconfig | 3 +- drivers/gpu/drm/xe/Makefile | 1 - drivers/gpu/drm/xe/xe_exec.c | 4 +- drivers/gpu/drm/xe/xe_hmm.c | 349 ------------------------------- drivers/gpu/drm/xe/xe_hmm.h | 18 -- drivers/gpu/drm/xe/xe_pt.c | 22 +- drivers/gpu/drm/xe/xe_vm.c | 85 +++++--- drivers/gpu/drm/xe/xe_vm_types.h | 26 +-- 8 files changed, 73 insertions(+), 435 deletions(-) delete mode 100644 drivers/gpu/drm/xe/xe_hmm.c delete mode 100644 drivers/gpu/drm/xe/xe_hmm.h
drivers/gpu/drm/xe/Kconfig +1 -2
@@ -43,7 +43,7 @@ config DRM_XE select MMU_NOTIFIER select WANT_DEV_COREDUMP select AUXILIARY_BUS - select HMM_MIRROR + select DRM_GPUSVM help Experimental driver for Intel Xe series GPUs
@@ -79,7 +79,6 @@ config DRM_XE_GPUSVM depends on !UML depends on DEVICE_PRIVATE default y - select DRM_GPUSVM help Enable this option if you want support for CPU to GPU address mirroring.
drivers/gpu/drm/xe/Makefile +0 -1
@@ -124,7 +124,6 @@ xe-y += xe_bb.o \ xe_wait_user_fence.o \ xe_wopcm.o -xe-$(CONFIG_HMM_MIRROR) += xe_hmm.o xe-$(CONFIG_DRM_XE_GPUSVM) += xe_svm.o # graphics hardware monitoring (HWMON) support
drivers/gpu/drm/xe/xe_exec.c +2 -2
@@ -294,7 +294,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (err) goto err_put_job; - err =down_read_interruptible (&vm->userptr.notifier_lock); + err =down_read_interruptible (&vm->svm.gpusvm.notifier_lock);if (err) goto err_put_job;
@@ -336,7 +336,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) err_repin:if (!xe_vm_in_lr_mode (vm)) -up_read (&vm->userptr.notifier_lock); +up_read (&vm->svm.gpusvm.notifier_lock); err_put_job:if (err)xe_sched_job_put (job);
drivers/gpu/drm/xe/xe_hmm.c deleted +0 -349
349 lines not on this page · raw message
drivers/gpu/drm/xe/xe_hmm.h deleted +0 -18
@@ -1,18 +0,0 @@ -/* SPDX-License-Identifier: MIT -* -* Copyright © 2024 Intel Corporation -*/ - -#ifndef _XE_HMM_H_ -#define _XE_HMM_H_ - -#include <linux/types.h> - -struct xe_userptr_vma ; - -int xe_hmm_userptr_populate_range (struct xe_userptr_vma *uvma,bool is_mm_mmap_locked); - -void xe_hmm_userptr_free_sg (struct xe_userptr_vma *uvma); - -void xe_hmm_userptr_unmap (struct xe_userptr_vma *uvma); -#endif
drivers/gpu/drm/xe/xe_pt.c +11 -11
@@ -729,8 +729,8 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,if (!xe_vma_is_null (vma) && !range) {if (xe_vma_is_userptr (vma)) -xe_res_first_sg (to_userptr_vma (vma)->userptr.sg,0 , -xe_vma_size (vma), &curs); +xe_res_first_dma (to_userptr_vma (vma)->userptr.pages.dma_addr,0 , +xe_vma_size (vma), &curs);else if (xe_bo_is_vram (bo) ||xe_bo_is_stolen (bo))xe_res_first (bo->ttm.resource,xe_vma_bo_offset (vma),xe_vma_size (vma), &curs);
@@ -1000,7 +1000,7 @@ static void xe_pt_commit_locks_assert(struct xe_vma *vma)xe_pt_commit_prepare_locks_assert (vma);if (xe_vma_is_userptr (vma)) -lockdep_assert_held_read (&vm->userptr.notifier_lock); +lockdep_assert_held_read (&vm->svm.gpusvm.notifier_lock); } static void xe_pt_commit(struct xe_vma *vma,
@@ -1338,7 +1338,7 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,struct xe_userptr_vma *uvma;unsigned long notifier_seq; -lockdep_assert_held_read (&vm->userptr.notifier_lock); +lockdep_assert_held_read (&vm->svm.gpusvm.notifier_lock);if (!xe_vma_is_userptr (vma))return 0 ;
@@ -1347,7 +1347,7 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,if (xe_pt_userptr_inject_eagain (uvma))xe_vma_userptr_force_invalidate (uvma); - notifier_seq = uvma->userptr.notifier_seq; + notifier_seq = uvma->userptr.pages.notifier_seq;if (!mmu_interval_read_retry (&uvma->userptr.notifier, notifier_seq))
@@ -1368,7 +1368,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op, {int err =0 ; -lockdep_assert_held_read (&vm->userptr.notifier_lock); +lockdep_assert_held_read (&vm->svm.gpusvm.notifier_lock);switch (op->base.op) {case DRM_GPUVA_OP_MAP :
@@ -1409,12 +1409,12 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->userptr.notifier_lock); +down_read (&vm->svm.gpusvm.notifier_lock);list_for_each_entry (op, &vops->list, link) { err =op_check_userptr (vm, op, pt_update_ops);if (err) { -up_read (&vm->userptr.notifier_lock); +up_read (&vm->svm.gpusvm.notifier_lock);break ; } }
@@ -2137,7 +2137,7 @@ static void bind_op_commit(struct xe_vm *vm, struct xe_tile *tile, vma->tile_present |=BIT (tile->id); vma->tile_staged &= ~BIT (tile->id);if (xe_vma_is_userptr (vma)) { -lockdep_assert_held_read (&vm->userptr.notifier_lock); +lockdep_assert_held_read (&vm->svm.gpusvm.notifier_lock);to_userptr_vma (vma)->userptr.initial_bind = true; }
@@ -2173,7 +2173,7 @@ static void unbind_op_commit(struct xe_vm *vm, struct xe_tile *tile,if (!vma->tile_present) {list_del_init (&vma->combined_links.rebind);if (xe_vma_is_userptr (vma)) { -lockdep_assert_held_read (&vm->userptr.notifier_lock); +lockdep_assert_held_read (&vm->svm.gpusvm.notifier_lock);spin_lock (&vm->userptr.invalidated_lock);list_del_init (&to_userptr_vma (vma)->userptr.invalidate_link);
@@ -2422,7 +2422,7 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)if (pt_update_ops->needs_svm_lock)xe_svm_notifier_unlock (vm);if (pt_update_ops->needs_userptr_lock) -up_read (&vm->userptr.notifier_lock); +up_read (&vm->svm.gpusvm.notifier_lock);return fence;
drivers/gpu/drm/xe/xe_vm.c +51 -34
266 lines not on this page · raw message
drivers/gpu/drm/xe/xe_vm_types.h +8 -18
@@ -52,26 +52,21 @@ struct xe_userptr {struct list_head invalidate_link;/** @userptr: link into VM repin list if userptr. */ struct list_head repin_link; +/** +* @pages: gpusvm pages for this user pointer. +*/ +struct drm_gpusvm_pages pages;/** * @notifier: MMU notifier for user pointer (invalidation call back) */ struct mmu_interval_notifier notifier; -/** @sgt: storage for a scatter gather table */ -struct sg_table sgt; -/** @sg: allocated scatter gather table */ -struct sg_table *sg; -/** @notifier_seq: notifier sequence number */ -unsigned long notifier_seq; -/** @unmap_mutex: Mutex protecting dma-unmapping */ -struct mutex unmap_mutex; +/** * @initial_bind: user pointer has been bound at least once. -* write: vm->userptr.notifier_lock in read mode and vm->resv held. -* read: vm->userptr.notifier_lock in write mode or vm->resv held. +* write: vm->svm.gpusvm.notifier_lock in read mode and vm->resv held. +* read: vm->svm.gpusvm.notifier_lock in write mode or vm->resv held. */ bool initial_bind; -/** @mapped: Whether the @sgt sg-table is dma-mapped. Protected by @unmap_mutex. */ -bool mapped;#if IS_ENABLED (CONFIG_DRM_XE_USERPTR_INVAL_INJECT )u32 divisor;#endif
@@ -109,7 +104,7 @@ struct xe_vma {/** * @tile_present: GT mask of binding are present for this VMA. * protected by vm->lock, vm->resv and for userptrs, -* vm->userptr.notifier_lock for writing. Needs either for reading, +* vm->svm.gpusvm.notifier_lock for writing. Needs either for reading, * but if reading is done under the vm->lock only, it needs to be held * in write mode. */
@@ -243,11 +238,6 @@ struct xe_vm {* and needs repinning. Protected by @lock. */ struct list_head repin_list; -/** -* @notifier_lock: protects notifier in write mode and -* submission in read mode. -*/ -struct rw_semaphore notifier_lock;/** * @userptr.invalidated_lock: Protects the * @userptr.invalidated list.
Signature · 1 line
2.48.1
PATCH v2 7/7 drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
In reply to Matthew AuldWe now use the same notifier lock for SVM and userptr, with that we can combine xe_pt_userptr_pre_commit and xe_pt_svm_pre_commit.
--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)
drivers/gpu/drm/xe/xe_pt.c +33 -62
@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;struct xe_vma_ops *vops = pt_update->vops;
@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops); +if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;
@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations
@@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,
@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;
Signature · 1 line
2.48.1
Matthew Brost
maintainerRe: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Quoted 47 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +13
-6@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;struct xe_vma_ops *vops = pt_update->vops;@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma));
drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1435
Quoted 10 lines
drivers/gpu/drm/xe/xe_pt.c +10
-0+xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops);
I think it would be better to combine needs_svm_lock into a single variable, rename op_check_userptr to op_check_svm_userptr, and then add the code in the above if statement to op_check_svm_userptr. Yes, the way have it coded now works but when Himal's code for SVM prefetch lands a VM operation list could have both userptr binds and binds SVM within single list. Changing it per my suggestion should make that work too.
Matt
Quoted 103 lines
drivers/gpu/drm/xe/xe_pt.c +10
-56+if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Re: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Quoted 57 lines · Matthew Brost
Quoted 56 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +22
-6@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;struct xe_vma_ops *vops = pt_update->vops;@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else {
Quoted 8 lines
drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1435
Quoted 1 line
drivers/gpu/drm/xe/xe_pt.c +1
-0+ err =op_check_userptr (vm, op, pt_update_ops);I think it would be better to combine
needs_svm_lockinto a single variable, renameop_check_userptrtoop_check_svm_userptr, and then add the code in the above if statement toop_check_svm_userptr. Yes, the way have it coded now works but when Himal's code for SVM prefetch lands a VM operation list could have both userptr binds and binds SVM within single list. Changing it per my suggestion should make that work too.
Will try this. Thanks for the comments.
Quoted 106 lines
Matt
Quoted 103 lines
drivers/gpu/drm/xe/xe_pt.c +10
-56+if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Matthew Brost
maintainerRe: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
In reply to Matthew AuldQuoted 25 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;struct xe_vma_ops *vops = pt_update->vops;
drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1412
Quoted 5 lines
drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);
Also any reason not use xe_svm_notifier_lock / xe_svm_notifier_unlock wrappers?
Matt
Quoted 130 lines
drivers/gpu/drm/xe/xe_pt.c +31
-60list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops); +if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Re: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Quoted 29 lines · Matthew Brost
Quoted 28 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;struct xe_vma_ops *vops = pt_update->vops;@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err;
Quoted 4 lines
drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1412
Quoted 2 lines
drivers/gpu/drm/xe/xe_pt.c +1
-1-down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);Also any reason not use
xe_svm_notifier_lock/xe_svm_notifier_unlockwrappers?
Just that those are hidden behind CONFIG_DRM_XE_GPUSVM, so looks possible to disable svm, but then I think we should still have working userptr. Should I pull xe_svm_notifier_lock out of CONFIG_DRM_XE_GPUSVM?
Quoted 133 lines
Matt
Quoted 130 lines
drivers/gpu/drm/xe/xe_pt.c +31
-60list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops); +if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Matthew Brost
maintainerRe: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Quoted 33 lines · Matthew Auld
Quoted 23 lines · Matthew Brost
Quoted 22 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;Quoted 8 lines
drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1411
Quoted 6 lines
drivers/gpu/drm/xe/xe_pt.c +1
-1struct xe_vma_ops *vops = pt_update->vops;@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);Also any reason not use
xe_svm_notifier_lock/xe_svm_notifier_unlockwrappers?
Quoted 4 lines
Just that those are hidden behind
CONFIG_DRM_XE_GPUSVM, so looks possible to disable svm, but then I think we should still have working userptr. Should I pullxe_svm_notifier_lockout ofCONFIG_DRM_XE_GPUSVM?
How would userptr work without GPU SVM if we building it on top of it?
Matt
Quoted 126 lines
Quoted 125 lines
Matt
Quoted 122 lines
drivers/gpu/drm/xe/xe_pt.c +31
-60list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops); +if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Re: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Quoted 34 lines · Matthew Brost
Quoted 33 lines · Matthew Auld
Quoted 31 lines · Matthew Brost
Quoted 24 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;struct xe_vma_ops *vops = pt_update->vops;drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1411
Quoted 4 lines
drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);Also any reason not use
xe_svm_notifier_lock/xe_svm_notifier_unlockwrappers?
Quoted 6 lines
Quoted 4 lines
Just that those are hidden behind
CONFIG_DRM_XE_GPUSVM, so looks possible to disable svm, but then I think we should still have working userptr. Should I pullxe_svm_notifier_lockout ofCONFIG_DRM_XE_GPUSVM?How would userptr work without GPU SVM if we building it on top of it?
Here I mean CONFIG_DRM_*XE*_GPUSVM, which looks to be specific to everything around xe_svm.c AFAICT, which useptr doesn't touch. We do now ofc require DRM_GPUSVM by default, but that's a different config.
Quoted 129 lines
Matt
Quoted 126 lines
Quoted 125 lines
Matt
Quoted 122 lines
drivers/gpu/drm/xe/xe_pt.c +31
-60list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops); +if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Matthew Brost
maintainerRe: [PATCH v2 7/7] drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Quoted 40 lines · Matthew Auld
Quoted 39 lines · Matthew Brost
Quoted 37 lines · Matthew Auld
Quoted 23 lines · Matthew Brost
Quoted 22 lines · Matthew Auld
We now use the same notifier lock for SVM and userptr, with that we can combine
xe_pt_userptr_pre_commitandxe_pt_svm_pre_commit.Suggested-by Matthew Brost <[email protected]>Signed-off-by Matthew Auld <[email protected]>Cc Himal Prasad Ghimiray <[email protected]>Cc Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_pt.c | 95 +++++++++++++------------------------- 1 file changed, 33 insertions(+), 62 deletions(-)drivers/gpu/drm/xe/xe_pt.c +1
-1@@ -1396,7 +1396,7 @@ static int op_check_userptr(struct xe_vm *vm, struct xe_vma_op *op,return err; } -static int xe_pt_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) +static int xe_pt_svm_userptr_pre_commit (struct xe_migrate_pt_update *pt_update) {struct xe_vm *vm = pt_update->vops->vm;Quoted 8 lines
drivers/gpu/drm/xe/xe_pt.c static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) · line 1411
Quoted 6 lines
drivers/gpu/drm/xe/xe_pt.c +1
-1struct xe_vma_ops *vops = pt_update->vops;@@ -1409,55 +1409,40 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)if (err)return err; -down_read (&vm->svm.gpusvm.notifier_lock); +drm_gpusvm_notifier_lock (&vm->svm.gpusvm);Also any reason not use
xe_svm_notifier_lock/xe_svm_notifier_unlockwrappers?Just that those are hidden behind
CONFIG_DRM_XE_GPUSVM, so looks possible to disable svm, but then I think we should still have working userptr. Should I pullxe_svm_notifier_lockout ofCONFIG_DRM_XE_GPUSVM?
Quoted 6 lines
Quoted 1 line
How would userptr work without GPU SVM if we building it on top of it?
Here I mean CONFIG_DRM_*XE*_GPUSVM, which looks to be specific to everything around
xe_svm.cAFAICT, which useptr doesn't touch. We do now ofc requireDRM_GPUSVMby default, but that's a different config.
Ah ok, that makes sense I guess. We could pull xe_svm_notifier_lock / xe_svm_notifier_unlock out these functions to not be hidden behind CONFIG_DRM_*XE*_GPUSVM though.
Matt
Quoted 130 lines
Quoted 129 lines
Matt
Quoted 126 lines
Quoted 125 lines
Matt
Quoted 122 lines
drivers/gpu/drm/xe/xe_pt.c +31
-60list_for_each_entry (op, &vops->list, link) { - err =op_check_userptr (vm, op, pt_update_ops); -if (err) { -up_read (&vm->svm.gpusvm.notifier_lock); -break ; +if (pt_update_ops->needs_svm_lock) { +#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) +struct xe_svm_range *range = op->map_range.range; + +if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) +continue ; + +xe_svm_range_debug (range,"PRE-COMMIT" ); + +xe_assert (vm->xe, +xe_vma_is_cpu_addr_mirror (op->map_range.vma)); +xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); + +if (!xe_svm_range_pages_valid (range)) { +xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +return -EAGAIN ; + } +#endif + }else { + err =op_check_userptr (vm, op, pt_update_ops); +if (err) { +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm); +break ; + } } }return err; } -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static int xe_pt_svm_pre_commit (struct xe_migrate_pt_update *pt_update) -{ -struct xe_vm *vm = pt_update->vops->vm; -struct xe_vma_ops *vops = pt_update->vops; -struct xe_vma_op *op; -int err; - - err =xe_pt_pre_commit (pt_update); -if (err) -return err; - -xe_svm_notifier_lock (vm); - -list_for_each_entry (op, &vops->list, link) { -struct xe_svm_range *range = op->map_range.range; - -if (op->subop ==XE_VMA_SUBOP_UNMAP_RANGE ) -continue ; - -xe_svm_range_debug (range,"PRE-COMMIT" ); - -xe_assert (vm->xe,xe_vma_is_cpu_addr_mirror (op->map_range.vma)); -xe_assert (vm->xe, op->subop ==XE_VMA_SUBOP_MAP_RANGE ); - -if (!xe_svm_range_pages_valid (range)) { -xe_svm_range_debug (range,"PRE-COMMIT - RETRY" ); -xe_svm_notifier_unlock (vm); -return -EAGAIN ; - } - } - -return 0 ; -} -#endif -struct invalidation_fence {struct xe_gt_tlb_invalidation_fence base;struct xe_gt *gt;@@ -2255,22 +2240,12 @@ static const struct xe_migrate_pt_update_ops migrate_ops = { .pre_commit = xe_pt_pre_commit, }; -static const struct xe_migrate_pt_update_ops userptr_migrate_ops = { +static const struct xe_migrate_pt_update_ops svm_userptr_migrate_ops = { .populate = xe_vm_populate_pgtable, .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_userptr_pre_commit, + .pre_commit = xe_pt_svm_userptr_pre_commit, }; -#if IS_ENABLED (CONFIG_DRM_XE_GPUSVM ) -static const struct xe_migrate_pt_update_ops svm_migrate_ops = { - .populate = xe_vm_populate_pgtable, - .clear = xe_migrate_clear_pgtable_callback, - .pre_commit = xe_pt_svm_pre_commit, -}; -#else -static const struct xe_migrate_pt_update_ops svm_migrate_ops; -#endif -/** * xe_pt_update_ops_run() - Run PT update operations * @tile: Tile of PT update operations @@ -2296,10 +2271,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops)struct xe_vma_op *op;int err =0 , i;struct xe_migrate_pt_update update = { - .ops = pt_update_ops->needs_svm_lock ? - &svm_migrate_ops : - pt_update_ops->needs_userptr_lock ? - &userptr_migrate_ops : + .ops = pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock ? + &svm_userptr_migrate_ops : &migrate_ops, .vops = vops, .tile_id = tile->id,@@ -2419,10 +2392,8 @@ xe_pt_update_ops_run(struct xe_tile *tile, struct xe_vma_ops *vops) &ifence->base.base, &mfence->base.base); } -if (pt_update_ops->needs_svm_lock) -xe_svm_notifier_unlock (vm); -if (pt_update_ops->needs_userptr_lock) -up_read (&vm->svm.gpusvm.notifier_lock); +if (pt_update_ops->needs_svm_lock || pt_update_ops->needs_userptr_lock) +drm_gpusvm_notifier_unlock (&vm->svm.gpusvm);return fence;Signature · 1 line
2.48.1
Patchwork
CI✓ CI.Patch_applied: success for Replace xe_hmm with gpusvm (rev2)
In reply to Matthew Auld== Series Details ==
== Series Details ==
Series: Replace xe_hmm with gpusvm (rev2)
URL : https://patchwork.freedesktop.org/series/146553/
State : success
== Summary ==
=== Applying kernel patches on branch 'drm-tip' with base: ===
Base commit: b8a000ee8dbf drm-tip: 2025y-03m-28d-19h-59m-23s UTC integration manifest === git am output follows ===
Applying: drm/gpusvm: fix hmm_pfn_to_map_order() usage
Applying: drm/gpusvm: use more selective dma dir in get_pages()
Applying: drm/gpusvm: pull out drm_gpusvm_pages substructure
Applying: drm/gpusvm: refactor core API to use pages struct
Applying: drm/gpusvm: export drm_gpusvm_pages API
Applying: drm/xe/userptr: replace xe_hmm with gpusvm
Applying: drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Patchwork
CI✗ CI.checkpatch: warning for Replace xe_hmm with gpusvm (rev2)
In reply to Matthew Auld== Series Details ==
== Series Details ==
Series: Replace xe_hmm with gpusvm (rev2)
URL : https://patchwork.freedesktop.org/series/146553/
State : warning
== Summary ==
+ KERNEL=/kernel
+ git clone https://gitlab.freedesktop.org/drm/maintainer-tools mt Cloning into 'mt'...
warning: redirecting to https://gitlab.freedesktop.org/drm/maintainer-tools.git/
+ git -C mt rev-list -n1 origin/master 99e5a866b5e13f134e606a3e29d9508d97826fb3
+ cd /kernel
+ git config --global --add safe.directory /kernel
+ git log -n1
commit 3f609c821a5c6f5a929e9c2956db2af12ca961c8
Author: Matthew Auld <[email protected]> Date: Fri Mar 28 18:10:36 2025 +0000
drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
We now use the same notifier lock for SVM and userptr, with that we can combine xe_pt_userptr_pre_commit and xe_pt_svm_pre_commit.
+ /mt/dim checkpatch b8a000ee8dbf41db533796cefbfb275cd306eb5b drm-intel929bce4fd08d drm/gpusvm: fix hmm_pfn_to_map_order() usage
-:19: WARNING:BAD_REPORTED_BY_LINK: Reported-by: should be immediately followed by Closes: with a URL to the report #19:
total: 0 errors, 1 warnings, 0 checks, 51 lines checked616e715847a0 drm/gpusvm: use more selective dma dir in get_pages()cfa7c02bfb7d drm/gpusvm: pull out drm_gpusvm_pages substructure6d8e98d48cd9 drm/gpusvm: refactor core API to use pages struct9f6fcb323d0a drm/gpusvm: export drm_gpusvm_pages API2a3bd6491d32 drm/xe/userptr: replace xe_hmm with gpusvm
-:79: WARNING:FILE_PATH_CHANGES: added, moved or deleted file(s), does MAINTAINERS need updating? #79:
deleted file mode 100644
total: 0 errors, 1 warnings, 0 checks, 436 lines checked 3f609c821a5c drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
Patchwork
CI✗ CI.KUnit: failure for Replace xe_hmm with gpusvm (rev2)
In reply to Matthew Auld== Series Details ==
== Series Details ==
Series: Replace xe_hmm with gpusvm (rev2)
URL : https://patchwork.freedesktop.org/series/146553/
State : failure
== Summary ==
+ trap cleanup EXIT
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/xe/.kunitconfig ERROR:root:../lib/iomap.c:156:5: warning: no previous prototype for ‘ioread64_lo_hi’ [-Wmissing-prototypes] 156 | u64 ioread64_lo_hi(const void __iomem *addr)
| ^~~~~~~~~~~~~~
../lib/iomap.c:163:5: warning: no previous prototype for ‘ioread64_hi_lo’ [-Wmissing-prototypes]
163 | u64 ioread64_hi_lo(const void __iomem *addr)
| ^~~~~~~~~~~~~~
../lib/iomap.c:170:5: warning: no previous prototype for ‘ioread64be_lo_hi’ [-Wmissing-prototypes]
170 | u64 ioread64be_lo_hi(const void __iomem *addr)
| ^~~~~~~~~~~~~~~~
../lib/iomap.c:178:5: warning: no previous prototype for ‘ioread64be_hi_lo’ [-Wmissing-prototypes]
178 | u64 ioread64be_hi_lo(const void __iomem *addr)
| ^~~~~~~~~~~~~~~~
../lib/iomap.c:264:6: warning: no previous prototype for ‘iowrite64_lo_hi’ [-Wmissing-prototypes]
264 | void iowrite64_lo_hi(u64 val, void __iomem *addr)
| ^~~~~~~~~~~~~~~
../lib/iomap.c:272:6: warning: no previous prototype for ‘iowrite64_hi_lo’ [-Wmissing-prototypes]
272 | void iowrite64_hi_lo(u64 val, void __iomem *addr)
| ^~~~~~~~~~~~~~~
../lib/iomap.c:280:6: warning: no previous prototype for ‘iowrite64be_lo_hi’ [-Wmissing-prototypes]
280 | void iowrite64be_lo_hi(u64 val, void __iomem *addr)
| ^~~~~~~~~~~~~~~~~
../lib/iomap.c:288:6: warning: no previous prototype for ‘iowrite64be_hi_lo’ [-Wmissing-prototypes]
288 | void iowrite64be_hi_lo(u64 val, void __iomem *addr)
| ^~~~~~~~~~~~~~~~~
../drivers/gpu/drm/drm_gpusvm.c: In function ‘drm_gpusvm_get_devmem_page’:
../drivers/gpu/drm/drm_gpusvm.c:1709:9: error: implicit declaration of function ‘zone_device_page_init’ [-Werror=implicit-function-declaration]
1709 | zone_device_page_init(page);
| ^~~~~~~~~~~~~~~~~~~~~cc1: some warnings being treated as errors
make[6]: *** [../scripts/Makefile.build:207: drivers/gpu/drm/drm_gpusvm.o] Error 1 make[6]: *** Waiting for unfinished jobs....
make[5]: *** [../scripts/Makefile.build:465: drivers/gpu/drm] Error 2 make[4]: *** [../scripts/Makefile.build:465: drivers/gpu] Error 2 make[3]: *** [../scripts/Makefile.build:465: drivers] Error 2 make[2]: *** [/kernel/Makefile:1994: .] Error 2
make[1]: *** [/kernel/Makefile:251: __sub-make] Error 2
make: *** [Makefile:251: __sub-make] Error 2
[20:47:08] Configuring KUnit Kernel ...
Generating .config ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig [20:47:12] Building KUnit Kernel ... Populating config with: $ make ARCH=um O=.kunit olddefconfig Building with: $ make all compile_commands.json ARCH=um O=.kunit --jobs=48 + cleanup ++ stat -c %u:%g /kernel + chown -R 1003:1003 /kernel