From: jglisse@redhat.com To: dri-devel@lists.freedesktop.org, nouveau@lists.freedesktop.org Cc: "Jérôme Glisse" <jglisse@redhat.com>, linux-mm@kvack.org, "John Hubbard" <jhubbard@nvidia.com>, "Evgeny Baskakov" <ebaskakov@nvidia.com>, "Ralph Campbell" <rcampbell@nvidia.com>, "Ben Skeggs" <bskeggs@redhat.com> Subject: [RFC PATCH 01/13] drm/nouveau/vmm: enable page table iterator over non populated range Date: Fri, 9 Mar 2018 22:21:29 -0500 [thread overview] Message-ID: <20180310032141.6096-2-jglisse@redhat.com> (raw) In-Reply-To: <20180310032141.6096-1-jglisse@redhat.com> From: JA(C)rA'me Glisse <jglisse@redhat.com> This patch modify the page table iterator to support empty range when unmaping a range (ie when it is not trying to populate the range). Signed-off-by: JA(C)rA'me Glisse <jglisse@redhat.com> Cc: Ben Skeggs <bskeggs@redhat.com> --- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c | 75 ++++++++++++++++++--------- 1 file changed, 51 insertions(+), 24 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c index 93946dcee319..20d31526ba8f 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c @@ -75,6 +75,7 @@ struct nvkm_vmm_iter { struct nvkm_vmm *vmm; u64 cnt; u16 max, lvl; + u64 start, addr; u32 pte[NVKM_VMM_LEVELS_MAX]; struct nvkm_vmm_pt *pt[NVKM_VMM_LEVELS_MAX]; int flush; @@ -485,6 +486,23 @@ nvkm_vmm_ref_swpt(struct nvkm_vmm_iter *it, struct nvkm_vmm_pt *pgd, u32 pdei) return true; } +static inline u64 +nvkm_vmm_iter_addr(const struct nvkm_vmm_iter *it, + const struct nvkm_vmm_desc *desc) +{ + int max = it->max; + u64 addr; + + /* Reconstruct address */ + addr = it->pte[max--]; + do { + addr = addr << desc[max].bits; + addr |= it->pte[max]; + } while (max--); + + return addr; +} + static inline u64 nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, u64 addr, u64 size, const char *name, bool ref, @@ -494,21 +512,23 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, { const struct nvkm_vmm_desc *desc = page->desc; struct nvkm_vmm_iter it; - u64 bits = addr >> page->shift; + u64 addr_bits = addr >> page->shift; it.page = page; it.desc = desc; it.vmm = vmm; it.cnt = size >> page->shift; it.flush = NVKM_VMM_LEVELS_MAX; + it.start = it.addr = addr; /* Deconstruct address into PTE indices for each mapping level. */ for (it.lvl = 0; desc[it.lvl].bits; it.lvl++) { - it.pte[it.lvl] = bits & ((1 << desc[it.lvl].bits) - 1); - bits >>= desc[it.lvl].bits; + it.pte[it.lvl] = addr_bits & ((1 << desc[it.lvl].bits) - 1); + addr_bits >>= desc[it.lvl].bits; } it.max = --it.lvl; it.pt[it.max] = vmm->pd; + addr_bits = addr >> page->shift; it.lvl = 0; TRA(&it, "%s: %016llx %016llx %d %lld PTEs", name, @@ -521,7 +541,8 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, const int type = desc->type == SPT; const u32 pten = 1 << desc->bits; const u32 ptei = it.pte[0]; - const u32 ptes = min_t(u64, it.cnt, pten - ptei); + u32 ptes = min_t(u64, it.cnt, pten - ptei); + u64 tmp; /* Walk down the tree, finding page tables for each level. */ for (; it.lvl; it.lvl--) { @@ -529,9 +550,14 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, struct nvkm_vmm_pt *pgd = pgt; /* Software PT. */ - if (ref && NVKM_VMM_PDE_INVALID(pgd->pde[pdei])) { - if (!nvkm_vmm_ref_swpt(&it, pgd, pdei)) - goto fail; + if (NVKM_VMM_PDE_INVALID(pgd->pde[pdei])) { + if (ref) { + if (!nvkm_vmm_ref_swpt(&it, pgd, pdei)) + goto fail; + } else { + it.pte[it.lvl] += 1; + goto next; + } } it.pt[it.lvl - 1] = pgt = pgd->pde[pdei]; @@ -545,9 +571,16 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, if (!nvkm_vmm_ref_hwpt(&it, pgd, pdei)) goto fail; } + + /* With HMM we might walk down un-populated range */ + if (!pgt) { + it.pte[it.lvl] += 1; + goto next; + } } /* Handle PTE updates. */ + it.addr = nvkm_vmm_iter_addr(&it, desc) << PAGE_SHIFT; if (!REF_PTES || REF_PTES(&it, ptei, ptes)) { struct nvkm_mmu_pt *pt = pgt->pt[type]; if (MAP_PTES || CLR_PTES) { @@ -558,32 +591,26 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, nvkm_vmm_flush_mark(&it); } } + it.pte[it.lvl] += ptes; +next: /* Walk back up the tree to the next position. */ - it.pte[it.lvl] += ptes; - it.cnt -= ptes; - if (it.cnt) { - while (it.pte[it.lvl] == (1 << desc[it.lvl].bits)) { - it.pte[it.lvl++] = 0; - it.pte[it.lvl]++; - } + while (it.pte[it.lvl] == (1 << desc[it.lvl].bits)) { + it.pte[it.lvl++] = 0; + if (it.lvl == it.max) + break; + it.pte[it.lvl]++; } + tmp = nvkm_vmm_iter_addr(&it, desc); + it.cnt -= min_t(u64, it.cnt, tmp - addr_bits); + addr_bits = tmp; }; nvkm_vmm_flush(&it); return ~0ULL; fail: - /* Reconstruct the failure address so the caller is able to - * reverse any partially completed operations. - */ - addr = it.pte[it.max--]; - do { - addr = addr << desc[it.max].bits; - addr |= it.pte[it.max]; - } while (it.max--); - - return addr << page->shift; + return addr_bits << page->shift; } static void -- 2.14.3
WARNING: multiple messages have this Message-ID (diff)
From: jglisse@redhat.com To: dri-devel@lists.freedesktop.org, nouveau@lists.freedesktop.org Cc: "Ralph Campbell" <rcampbell@nvidia.com>, "Evgeny Baskakov" <ebaskakov@nvidia.com>, linux-mm@kvack.org, "Jérôme Glisse" <jglisse@redhat.com>, "Ben Skeggs" <bskeggs@redhat.com>, "John Hubbard" <jhubbard@nvidia.com> Subject: [RFC PATCH 01/13] drm/nouveau/vmm: enable page table iterator over non populated range Date: Fri, 9 Mar 2018 22:21:29 -0500 [thread overview] Message-ID: <20180310032141.6096-2-jglisse@redhat.com> (raw) In-Reply-To: <20180310032141.6096-1-jglisse@redhat.com> From: Jérôme Glisse <jglisse@redhat.com> This patch modify the page table iterator to support empty range when unmaping a range (ie when it is not trying to populate the range). Signed-off-by: Jérôme Glisse <jglisse@redhat.com> Cc: Ben Skeggs <bskeggs@redhat.com> --- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c | 75 ++++++++++++++++++--------- 1 file changed, 51 insertions(+), 24 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c index 93946dcee319..20d31526ba8f 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c @@ -75,6 +75,7 @@ struct nvkm_vmm_iter { struct nvkm_vmm *vmm; u64 cnt; u16 max, lvl; + u64 start, addr; u32 pte[NVKM_VMM_LEVELS_MAX]; struct nvkm_vmm_pt *pt[NVKM_VMM_LEVELS_MAX]; int flush; @@ -485,6 +486,23 @@ nvkm_vmm_ref_swpt(struct nvkm_vmm_iter *it, struct nvkm_vmm_pt *pgd, u32 pdei) return true; } +static inline u64 +nvkm_vmm_iter_addr(const struct nvkm_vmm_iter *it, + const struct nvkm_vmm_desc *desc) +{ + int max = it->max; + u64 addr; + + /* Reconstruct address */ + addr = it->pte[max--]; + do { + addr = addr << desc[max].bits; + addr |= it->pte[max]; + } while (max--); + + return addr; +} + static inline u64 nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, u64 addr, u64 size, const char *name, bool ref, @@ -494,21 +512,23 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, { const struct nvkm_vmm_desc *desc = page->desc; struct nvkm_vmm_iter it; - u64 bits = addr >> page->shift; + u64 addr_bits = addr >> page->shift; it.page = page; it.desc = desc; it.vmm = vmm; it.cnt = size >> page->shift; it.flush = NVKM_VMM_LEVELS_MAX; + it.start = it.addr = addr; /* Deconstruct address into PTE indices for each mapping level. */ for (it.lvl = 0; desc[it.lvl].bits; it.lvl++) { - it.pte[it.lvl] = bits & ((1 << desc[it.lvl].bits) - 1); - bits >>= desc[it.lvl].bits; + it.pte[it.lvl] = addr_bits & ((1 << desc[it.lvl].bits) - 1); + addr_bits >>= desc[it.lvl].bits; } it.max = --it.lvl; it.pt[it.max] = vmm->pd; + addr_bits = addr >> page->shift; it.lvl = 0; TRA(&it, "%s: %016llx %016llx %d %lld PTEs", name, @@ -521,7 +541,8 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, const int type = desc->type == SPT; const u32 pten = 1 << desc->bits; const u32 ptei = it.pte[0]; - const u32 ptes = min_t(u64, it.cnt, pten - ptei); + u32 ptes = min_t(u64, it.cnt, pten - ptei); + u64 tmp; /* Walk down the tree, finding page tables for each level. */ for (; it.lvl; it.lvl--) { @@ -529,9 +550,14 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, struct nvkm_vmm_pt *pgd = pgt; /* Software PT. */ - if (ref && NVKM_VMM_PDE_INVALID(pgd->pde[pdei])) { - if (!nvkm_vmm_ref_swpt(&it, pgd, pdei)) - goto fail; + if (NVKM_VMM_PDE_INVALID(pgd->pde[pdei])) { + if (ref) { + if (!nvkm_vmm_ref_swpt(&it, pgd, pdei)) + goto fail; + } else { + it.pte[it.lvl] += 1; + goto next; + } } it.pt[it.lvl - 1] = pgt = pgd->pde[pdei]; @@ -545,9 +571,16 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, if (!nvkm_vmm_ref_hwpt(&it, pgd, pdei)) goto fail; } + + /* With HMM we might walk down un-populated range */ + if (!pgt) { + it.pte[it.lvl] += 1; + goto next; + } } /* Handle PTE updates. */ + it.addr = nvkm_vmm_iter_addr(&it, desc) << PAGE_SHIFT; if (!REF_PTES || REF_PTES(&it, ptei, ptes)) { struct nvkm_mmu_pt *pt = pgt->pt[type]; if (MAP_PTES || CLR_PTES) { @@ -558,32 +591,26 @@ nvkm_vmm_iter(struct nvkm_vmm *vmm, const struct nvkm_vmm_page *page, nvkm_vmm_flush_mark(&it); } } + it.pte[it.lvl] += ptes; +next: /* Walk back up the tree to the next position. */ - it.pte[it.lvl] += ptes; - it.cnt -= ptes; - if (it.cnt) { - while (it.pte[it.lvl] == (1 << desc[it.lvl].bits)) { - it.pte[it.lvl++] = 0; - it.pte[it.lvl]++; - } + while (it.pte[it.lvl] == (1 << desc[it.lvl].bits)) { + it.pte[it.lvl++] = 0; + if (it.lvl == it.max) + break; + it.pte[it.lvl]++; } + tmp = nvkm_vmm_iter_addr(&it, desc); + it.cnt -= min_t(u64, it.cnt, tmp - addr_bits); + addr_bits = tmp; }; nvkm_vmm_flush(&it); return ~0ULL; fail: - /* Reconstruct the failure address so the caller is able to - * reverse any partially completed operations. - */ - addr = it.pte[it.max--]; - do { - addr = addr << desc[it.max].bits; - addr |= it.pte[it.max]; - } while (it.max--); - - return addr << page->shift; + return addr_bits << page->shift; } static void -- 2.14.3 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
next prev parent reply other threads:[~2018-03-10 3:21 UTC|newest] Thread overview: 52+ messages / expand[flat|nested] mbox.gz Atom feed top 2018-03-10 3:21 [RFC PATCH 00/13] SVM (share virtual memory) with HMM in nouveau jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` jglisse [this message] 2018-03-10 3:21 ` [RFC PATCH 01/13] drm/nouveau/vmm: enable page table iterator over non populated range jglisse 2018-03-10 3:21 ` [RFC PATCH 02/13] drm/nouveau/core/memory: add some useful accessor macros jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 03/13] drm/nouveau/core: define engine for handling replayable faults jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 04/13] drm/nouveau/mmu/gp100: allow gcc/tex to generate " jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 05/13] drm/nouveau/mc/gp100-: handle replayable fault interrupt jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 06/13] drm/nouveau/fault/gp100: initial implementation of MaxwellFaultBufferA jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 07/13] drm/nouveau: special mapping method for HMM jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 08/13] drm/nouveau: special mapping method for HMM (user interface) jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 09/13] drm/nouveau: add SVM through HMM support to nouveau client jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 10/13] drm/nouveau: add HMM area creation jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 11/13] drm/nouveau: add HMM area creation user interface jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 12/13] drm/nouveau: HMM area creation helpers for nouveau client jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 3:21 ` [RFC PATCH 13/13] drm/nouveau: HACK FOR HMM AREA jglisse 2018-03-10 3:21 ` jglisse 2018-03-10 15:01 ` [RFC PATCH 00/13] SVM (share virtual memory) with HMM in nouveau Christian König 2018-03-10 15:01 ` Christian König 2018-03-10 17:55 ` Jerome Glisse 2018-03-10 17:55 ` Jerome Glisse 2018-03-12 17:30 ` Daniel Vetter 2018-03-12 17:30 ` Daniel Vetter 2018-03-12 17:50 ` Jerome Glisse 2018-03-12 17:50 ` Jerome Glisse 2018-03-13 6:14 ` John Hubbard 2018-03-13 6:14 ` John Hubbard 2018-03-13 13:29 ` Matthew Wilcox 2018-03-13 13:29 ` Matthew Wilcox 2018-03-13 14:31 ` Jerome Glisse 2018-03-13 14:31 ` Jerome Glisse 2018-03-13 15:56 ` Jerome Glisse 2018-03-13 15:56 ` Jerome Glisse 2018-03-13 10:46 ` Daniel Vetter 2018-03-13 10:46 ` Daniel Vetter 2018-03-12 18:28 ` Felix Kuehling 2018-03-12 18:28 ` Felix Kuehling 2018-03-13 14:28 ` Jerome Glisse 2018-03-13 14:28 ` Jerome Glisse 2018-03-13 15:32 ` Felix Kuehling 2018-03-13 15:32 ` Felix Kuehling
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20180310032141.6096-2-jglisse@redhat.com \ --to=jglisse@redhat.com \ --cc=bskeggs@redhat.com \ --cc=dri-devel@lists.freedesktop.org \ --cc=ebaskakov@nvidia.com \ --cc=jhubbard@nvidia.com \ --cc=linux-mm@kvack.org \ --cc=nouveau@lists.freedesktop.org \ --cc=rcampbell@nvidia.com \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.