In order to reduce TLB misses and CPU usage this patchset enables huge- and giant page-table entries for TTM and TTM-enabled graphics drivers.
Patch 1 and 2 introduce a vma_is_special_huge() function to make the mm code take the same path as DAX when splitting huge- and giant page table entries, (which currently means zapping the page-table entry and rely on re-faulting).
Patch 3 makes the mm code split existing huge page-table entries on huge_fault fallbacks. Typically on COW or on buffer-objects that want write-notify. COW and write-notification is always done on the lowest page-table level. See the patch log message for additional considerations.
Patch 4 introduces functions to allow the graphics drivers to manipulate the caching- and encryption flags of huge page-table entries without ugly hacks.
Patch 5 implements the huge_fault handler in TTM. This enables huge page-table entries, provided that the kernel is configured to support transhuge pages, either by default or using madvise(). However, they are unlikely to be inserted unless the kernel buffer object pfns and user-space addresses align perfectly. There are various options here, but since buffer objects that reside in system pages typically start at huge page boundaries if they are backed by huge pages, we try to enforce buffer object starting pfns and user-space addresses to be huge page-size aligned if their size exceeds a huge page-size. If pud-size transhuge ("giant") pages are enabled by the arch, the same holds for those.
Patch 6 implements a specialized huge_fault handler for vmwgfx. The vmwgfx driver may perform dirty-tracking and needs some special code to handle that correctly.
Patch 7 implements a drm helper to align user-space addresses according to the above scheme, if possible.
Patch 8 implements a TTM range manager for vmwgfx that does the same for graphics IO memory. This may later be reused by other graphics drivers if necessary.
Patch 9 finally hooks up the helpers of patch 7 and 8 to the vmwgfx driver. A similar change is needed for graphics drivers that want a reasonable likelyhood of actually using huge page-table entries.
If a buffer object size is not huge-page or giant-page aligned, its size will NOT be inflated by this patchset. This means that the buffer object tail will use smaller size page-table entries and thus no memory overhead occurs. Drivers that want to pay the memory overhead price need to implement their own scheme to inflate buffer-object sizes.
PMD size huge page-table-entries have been tested with vmwgfx and found to work well both with system memory backed and IO memory backed buffer objects.
PUD size giant page-table-entries have seen limited (fault and COW) testing using a modified kernel (to support 1GB page allocations) and a fake vmwgfx TTM memory type. The vmwgfx driver does otherwise not support 1GB-size IO memory resources.
Comments and suggestions welcome. Thomas
Changes since RFC: * Check for buffer objects present in contigous IO Memory (Christian König) * Rebased on the vmwgfx emulated coherent memory functionality. That rebase adds patch 5. Changes since v1: * Make the new TTM range manager vmwgfx-specific. (Christian König) * Minor fixes for configs that don't support or only partially support transhuge pages. Changes since v2: * Minor coding style and doc fixes in patch 5/9 (Christian König) * Patch 5/9 doesn't touch mm. Remove from the patch title. Changes since v3: * Added reviews and acks * Implemented ugly but generic ttm_pgprot_is_wrprotecting() instead of arch specific code.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com
From: Thomas Hellstrom thellstrom@vmware.com
The vma argument is only dereferenced for reading.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Acked-by: Christian König christian.koenig@amd.com --- include/linux/fs.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/linux/fs.h b/include/linux/fs.h index 98e0349adb52..4f41fdbf402f 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -3367,7 +3367,7 @@ static inline bool io_is_direct(struct file *filp) return (filp->f_flags & O_DIRECT) || IS_DAX(filp->f_mapping->host); }
-static inline bool vma_is_dax(struct vm_area_struct *vma) +static inline bool vma_is_dax(const struct vm_area_struct *vma) { return vma->vm_file && IS_DAX(vma->vm_file->f_mapping->host); }
On Thu, Feb 20, 2020 at 4:27 AM Thomas Hellström (VMware) thomas_os@shipmail.org wrote:
From: Thomas Hellstrom thellstrom@vmware.com
The vma argument is only dereferenced for reading.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Acked-by: Christian König christian.koenig@amd.com
include/linux/fs.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/linux/fs.h b/include/linux/fs.h index 98e0349adb52..4f41fdbf402f 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -3367,7 +3367,7 @@ static inline bool io_is_direct(struct file *filp) return (filp->f_flags & O_DIRECT) || IS_DAX(filp->f_mapping->host); }
-static inline bool vma_is_dax(struct vm_area_struct *vma) +static inline bool vma_is_dax(const struct vm_area_struct *vma) { return vma->vm_file && IS_DAX(vma->vm_file->f_mapping->host); }
Looks good to me:
Acked-by: Dan Williams dan.j.williams@intel.com
From: Thomas Hellstrom thellstrom@vmware.com
For VM_PFNMAP and VM_MIXEDMAP vmas that want to support transhuge pages and -page table entries, introduce vma_is_special_huge() that takes the same codepaths as vma_is_dax().
The use of "special" follows the definition in memory.c, vm_normal_page(): "Special" mappings do not wish to be associated with a "struct page" (either it doesn't exist, or it exists but they don't want to touch it)
For PAGE_SIZE pages, "special" is determined per page table entry to be able to deal with COW pages. But since we don't have huge COW pages, we can classify a vma as either "special huge" or "normal huge".
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Acked-by: Christian König christian.koenig@amd.com --- include/linux/mm.h | 17 +++++++++++++++++ mm/huge_memory.c | 6 +++--- 2 files changed, 20 insertions(+), 3 deletions(-)
diff --git a/include/linux/mm.h b/include/linux/mm.h index 0157d293935f..d370ce2932a1 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -2822,6 +2822,23 @@ extern long copy_huge_page_from_user(struct page *dst_page, const void __user *usr_src, unsigned int pages_per_huge_page, bool allow_pagefault); + +/** + * vma_is_special_huge - Are transhuge page-table entries considered special? + * @vma: Pointer to the struct vm_area_struct to consider + * + * Whether transhuge page-table entries are considered "special" following + * the definition in vm_normal_page(). + * + * Return: true if transhuge page-table entries should be considered special, + * false otherwise. + */ +static inline bool vma_is_special_huge(const struct vm_area_struct *vma) +{ + return vma_is_dax(vma) || (vma->vm_file && + (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))); +} + #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
#ifdef CONFIG_DEBUG_PAGEALLOC diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 41a0fbddc96b..f8d24fc3f4df 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -1789,7 +1789,7 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma, orig_pmd = pmdp_huge_get_and_clear_full(tlb->mm, addr, pmd, tlb->fullmm); tlb_remove_pmd_tlb_entry(tlb, pmd, addr); - if (vma_is_dax(vma)) { + if (vma_is_special_huge(vma)) { if (arch_needs_pgtable_deposit()) zap_deposited_table(tlb->mm, pmd); spin_unlock(ptl); @@ -2053,7 +2053,7 @@ int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma, */ pudp_huge_get_and_clear_full(tlb->mm, addr, pud, tlb->fullmm); tlb_remove_pud_tlb_entry(tlb, pud, addr); - if (vma_is_dax(vma)) { + if (vma_is_special_huge(vma)) { spin_unlock(ptl); /* No zero page support yet */ } else { @@ -2162,7 +2162,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd, */ if (arch_needs_pgtable_deposit()) zap_deposited_table(mm, pmd); - if (vma_is_dax(vma)) + if (vma_is_special_huge(vma)) return; page = pmd_page(_pmd); if (!PageDirty(page) && pmd_dirty(_pmd))
From: Thomas Hellstrom thellstrom@vmware.com
We currently only do COW and write-notify on the PTE level, so if the huge_fault() handler returns VM_FAULT_FALLBACK on wp faults, split the huge pages and page-table entries. Also do this for huge PUDs if there is no huge_fault() handler and the vma is not anonymous, similar to how it's done for PMDs.
Note that fs/dax.c does the splitting in the huge_fault() handler, but as huge_fault() is implemented by modules we need to consider whether to export the splitting functions for use in the modules or whether to try to keep calls in the core. Opt for the latter. A follow-up patch can remove the dax.c split_huge_pmd() if needed.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Acked-by: Christian König christian.koenig@amd.com --- mm/memory.c | 27 +++++++++++++++++++-------- 1 file changed, 19 insertions(+), 8 deletions(-)
diff --git a/mm/memory.c b/mm/memory.c index 17aadc751e5c..4c49fe963e5c 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -3880,11 +3880,14 @@ static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) { if (vma_is_anonymous(vmf->vma)) return do_huge_pmd_wp_page(vmf, orig_pmd); - if (vmf->vma->vm_ops->huge_fault) - return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); + if (vmf->vma->vm_ops->huge_fault) { + vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
- /* COW handled on pte level: split pmd */ - VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); + if (!(ret & VM_FAULT_FALLBACK)) + return ret; + } + + /* COW or write-notify handled on pte level: split pmd. */ __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL);
return VM_FAULT_FALLBACK; @@ -3897,12 +3900,20 @@ static inline bool vma_is_accessible(struct vm_area_struct *vma)
static vm_fault_t create_huge_pud(struct vm_fault *vmf) { -#ifdef CONFIG_TRANSPARENT_HUGEPAGE +#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ + defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) /* No support for anonymous transparent PUD pages yet */ if (vma_is_anonymous(vmf->vma)) - return VM_FAULT_FALLBACK; - if (vmf->vma->vm_ops->huge_fault) - return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); + goto split; + if (vmf->vma->vm_ops->huge_fault) { + vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); + + if (!(ret & VM_FAULT_FALLBACK)) + return ret; + } +split: + /* COW or write-notify not handled on PUD level: split pud.*/ + __split_huge_pud(vmf->vma, vmf->pud, vmf->address); #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ return VM_FAULT_FALLBACK; }
From: Thomas Hellstrom thellstrom@vmware.com
For graphics drivers needing to modify the page-protection, add huge page-table entries counterparts to vmf_insert_pfn_prot().
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Acked-by: Christian König christian.koenig@amd.com --- include/linux/huge_mm.h | 41 +++++++++++++++++++++++++++++++++++++++-- mm/huge_memory.c | 38 ++++++++++++++++++++++++++++++++------ 2 files changed, 71 insertions(+), 8 deletions(-)
diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 0b84e13e88e2..a95d1bc8ffe8 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -47,8 +47,45 @@ extern bool move_huge_pmd(struct vm_area_struct *vma, unsigned long old_addr, extern int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd, unsigned long addr, pgprot_t newprot, int prot_numa); -vm_fault_t vmf_insert_pfn_pmd(struct vm_fault *vmf, pfn_t pfn, bool write); -vm_fault_t vmf_insert_pfn_pud(struct vm_fault *vmf, pfn_t pfn, bool write); +vm_fault_t vmf_insert_pfn_pmd_prot(struct vm_fault *vmf, pfn_t pfn, + pgprot_t pgprot, bool write); + +/** + * vmf_insert_pfn_pmd - insert a pmd size pfn + * @vmf: Structure describing the fault + * @pfn: pfn to insert + * @pgprot: page protection to use + * @write: whether it's a write fault + * + * Insert a pmd size pfn. See vmf_insert_pfn() for additional info. + * + * Return: vm_fault_t value. + */ +static inline vm_fault_t vmf_insert_pfn_pmd(struct vm_fault *vmf, pfn_t pfn, + bool write) +{ + return vmf_insert_pfn_pmd_prot(vmf, pfn, vmf->vma->vm_page_prot, write); +} +vm_fault_t vmf_insert_pfn_pud_prot(struct vm_fault *vmf, pfn_t pfn, + pgprot_t pgprot, bool write); + +/** + * vmf_insert_pfn_pud - insert a pud size pfn + * @vmf: Structure describing the fault + * @pfn: pfn to insert + * @pgprot: page protection to use + * @write: whether it's a write fault + * + * Insert a pud size pfn. See vmf_insert_pfn() for additional info. + * + * Return: vm_fault_t value. + */ +static inline vm_fault_t vmf_insert_pfn_pud(struct vm_fault *vmf, pfn_t pfn, + bool write) +{ + return vmf_insert_pfn_pud_prot(vmf, pfn, vmf->vma->vm_page_prot, write); +} + enum transparent_hugepage_flag { TRANSPARENT_HUGEPAGE_FLAG, TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, diff --git a/mm/huge_memory.c b/mm/huge_memory.c index f8d24fc3f4df..b2ec62cca3ae 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -811,11 +811,24 @@ static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr, pte_free(mm, pgtable); }
-vm_fault_t vmf_insert_pfn_pmd(struct vm_fault *vmf, pfn_t pfn, bool write) +/** + * vmf_insert_pfn_pmd_prot - insert a pmd size pfn + * @vmf: Structure describing the fault + * @pfn: pfn to insert + * @pgprot: page protection to use + * @write: whether it's a write fault + * + * Insert a pmd size pfn. See vmf_insert_pfn() for additional info and + * also consult the vmf_insert_mixed_prot() documentation when + * @pgprot != @vmf->vma->vm_page_prot. + * + * Return: vm_fault_t value. + */ +vm_fault_t vmf_insert_pfn_pmd_prot(struct vm_fault *vmf, pfn_t pfn, + pgprot_t pgprot, bool write) { unsigned long addr = vmf->address & PMD_MASK; struct vm_area_struct *vma = vmf->vma; - pgprot_t pgprot = vma->vm_page_prot; pgtable_t pgtable = NULL;
/* @@ -843,7 +856,7 @@ vm_fault_t vmf_insert_pfn_pmd(struct vm_fault *vmf, pfn_t pfn, bool write) insert_pfn_pmd(vma, addr, vmf->pmd, pfn, pgprot, write, pgtable); return VM_FAULT_NOPAGE; } -EXPORT_SYMBOL_GPL(vmf_insert_pfn_pmd); +EXPORT_SYMBOL_GPL(vmf_insert_pfn_pmd_prot);
#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD static pud_t maybe_pud_mkwrite(pud_t pud, struct vm_area_struct *vma) @@ -889,11 +902,24 @@ static void insert_pfn_pud(struct vm_area_struct *vma, unsigned long addr, spin_unlock(ptl); }
-vm_fault_t vmf_insert_pfn_pud(struct vm_fault *vmf, pfn_t pfn, bool write) +/** + * vmf_insert_pfn_pud_prot - insert a pud size pfn + * @vmf: Structure describing the fault + * @pfn: pfn to insert + * @pgprot: page protection to use + * @write: whether it's a write fault + * + * Insert a pud size pfn. See vmf_insert_pfn() for additional info and + * also consult the vmf_insert_mixed_prot() documentation when + * @pgprot != @vmf->vma->vm_page_prot. + * + * Return: vm_fault_t value. + */ +vm_fault_t vmf_insert_pfn_pud_prot(struct vm_fault *vmf, pfn_t pfn, + pgprot_t pgprot, bool write) { unsigned long addr = vmf->address & PUD_MASK; struct vm_area_struct *vma = vmf->vma; - pgprot_t pgprot = vma->vm_page_prot;
/* * If we had pud_special, we could avoid all these restrictions, @@ -914,7 +940,7 @@ vm_fault_t vmf_insert_pfn_pud(struct vm_fault *vmf, pfn_t pfn, bool write) insert_pfn_pud(vma, addr, vmf->pud, pfn, pgprot, write); return VM_FAULT_NOPAGE; } -EXPORT_SYMBOL_GPL(vmf_insert_pfn_pud); +EXPORT_SYMBOL_GPL(vmf_insert_pfn_pud_prot); #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
static void touch_pmd(struct vm_area_struct *vma, unsigned long addr,
From: Thomas Hellstrom thellstrom@vmware.com
Support huge (PMD-size and PUD-size) page-table entries by providing a huge_fault() callback. We still support private mappings and write-notify by splitting the huge page-table entries on write-access.
Note that for huge page-faults to occur, either the kernel needs to be compiled with trans-huge-pages always enabled, or the kernel needs to be compiled with trans-huge-pages enabled using madvise, and the user-space app needs to call madvise() to enable trans-huge pages on a per-mapping basis.
Furthermore huge page-faults will not succeed unless buffer objects and user-space addresses are aligned on huge page size boundaries.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Reviewed-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/ttm/ttm_bo_vm.c | 161 ++++++++++++++++++++- drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c | 2 +- include/drm/ttm/ttm_bo_api.h | 3 +- 3 files changed, 161 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index 389128b8c4dd..0af14835504c 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -156,6 +156,89 @@ vm_fault_t ttm_bo_vm_reserve(struct ttm_buffer_object *bo, } EXPORT_SYMBOL(ttm_bo_vm_reserve);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE +/** + * ttm_bo_vm_insert_huge - Insert a pfn for PUD or PMD faults + * @vmf: Fault data + * @bo: The buffer object + * @page_offset: Page offset from bo start + * @fault_page_size: The size of the fault in pages. + * @pgprot: The page protections. + * Does additional checking whether it's possible to insert a PUD or PMD + * pfn and performs the insertion. + * + * Return: VM_FAULT_NOPAGE on successful insertion, VM_FAULT_FALLBACK if + * a huge fault was not possible, or on insertion error. + */ +static vm_fault_t ttm_bo_vm_insert_huge(struct vm_fault *vmf, + struct ttm_buffer_object *bo, + pgoff_t page_offset, + pgoff_t fault_page_size, + pgprot_t pgprot) +{ + pgoff_t i; + vm_fault_t ret; + unsigned long pfn; + pfn_t pfnt; + struct ttm_tt *ttm = bo->ttm; + bool write = vmf->flags & FAULT_FLAG_WRITE; + + /* Fault should not cross bo boundary. */ + page_offset &= ~(fault_page_size - 1); + if (page_offset + fault_page_size > bo->num_pages) + goto out_fallback; + + if (bo->mem.bus.is_iomem) + pfn = ttm_bo_io_mem_pfn(bo, page_offset); + else + pfn = page_to_pfn(ttm->pages[page_offset]); + + /* pfn must be fault_page_size aligned. */ + if ((pfn & (fault_page_size - 1)) != 0) + goto out_fallback; + + /* Check that memory is contiguous. */ + if (!bo->mem.bus.is_iomem) { + for (i = 1; i < fault_page_size; ++i) { + if (page_to_pfn(ttm->pages[page_offset + i]) != pfn + i) + goto out_fallback; + } + } else if (bo->bdev->driver->io_mem_pfn) { + for (i = 1; i < fault_page_size; ++i) { + if (ttm_bo_io_mem_pfn(bo, page_offset + i) != pfn + i) + goto out_fallback; + } + } + + pfnt = __pfn_to_pfn_t(pfn, PFN_DEV); + if (fault_page_size == (HPAGE_PMD_SIZE >> PAGE_SHIFT)) + ret = vmf_insert_pfn_pmd_prot(vmf, pfnt, pgprot, write); +#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD + else if (fault_page_size == (HPAGE_PUD_SIZE >> PAGE_SHIFT)) + ret = vmf_insert_pfn_pud_prot(vmf, pfnt, pgprot, write); +#endif + else + WARN_ON_ONCE(ret = VM_FAULT_FALLBACK); + + if (ret != VM_FAULT_NOPAGE) + goto out_fallback; + + return VM_FAULT_NOPAGE; +out_fallback: + count_vm_event(THP_FAULT_FALLBACK); + return VM_FAULT_FALLBACK; +} +#else +static vm_fault_t ttm_bo_vm_insert_huge(struct vm_fault *vmf, + struct ttm_buffer_object *bo, + pgoff_t page_offset, + pgoff_t fault_page_size, + pgprot_t pgprot) +{ + return VM_FAULT_FALLBACK; +} +#endif + /** * ttm_bo_vm_fault_reserved - TTM fault helper * @vmf: The struct vm_fault given as argument to the fault callback @@ -163,6 +246,7 @@ EXPORT_SYMBOL(ttm_bo_vm_reserve); * @num_prefault: Maximum number of prefault pages. The caller may want to * specify this based on madvice settings and the size of the GPU object * backed by the memory. + * @fault_page_size: The size of the fault in pages. * * This function inserts one or more page table entries pointing to the * memory backing the buffer object, and then returns a return code @@ -176,7 +260,8 @@ EXPORT_SYMBOL(ttm_bo_vm_reserve); */ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, pgprot_t prot, - pgoff_t num_prefault) + pgoff_t num_prefault, + pgoff_t fault_page_size) { struct vm_area_struct *vma = vmf->vma; struct ttm_buffer_object *bo = vma->vm_private_data; @@ -268,6 +353,13 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, prot = pgprot_decrypted(prot); }
+ /* We don't prefault on huge faults. Yet. */ + if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && fault_page_size != 1) { + ret = ttm_bo_vm_insert_huge(vmf, bo, page_offset, + fault_page_size, prot); + goto out_io_unlock; + } + /* * Speculatively prefault a number of pages. Only error on * first page. @@ -334,7 +426,7 @@ vm_fault_t ttm_bo_vm_fault(struct vm_fault *vmf) return ret;
prot = vma->vm_page_prot; - ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT); + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT, 1); if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) return ret;
@@ -344,6 +436,66 @@ vm_fault_t ttm_bo_vm_fault(struct vm_fault *vmf) } EXPORT_SYMBOL(ttm_bo_vm_fault);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE +/** + * ttm_pgprot_is_wrprotecting - Is a page protection value write-protecting? + * @prot: The page protection value + * + * Return: true if @prot is write-protecting. false otherwise. + */ +static bool ttm_pgprot_is_wrprotecting(pgprot_t prot) +{ + /* + * This is meant to say "pgprot_wrprotect(prot) == prot" in a generic + * way. Unfortunately there is no generic pgprot_wrprotect. + */ + return pte_val(pte_wrprotect(__pte(pgprot_val(prot)))) == + pgprot_val(prot); +} + +static vm_fault_t ttm_bo_vm_huge_fault(struct vm_fault *vmf, + enum page_entry_size pe_size) +{ + struct vm_area_struct *vma = vmf->vma; + pgprot_t prot; + struct ttm_buffer_object *bo = vma->vm_private_data; + vm_fault_t ret; + pgoff_t fault_page_size = 0; + bool write = vmf->flags & FAULT_FLAG_WRITE; + + switch (pe_size) { + case PE_SIZE_PMD: + fault_page_size = HPAGE_PMD_SIZE >> PAGE_SHIFT; + break; +#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD + case PE_SIZE_PUD: + fault_page_size = HPAGE_PUD_SIZE >> PAGE_SHIFT; + break; +#endif + default: + WARN_ON_ONCE(1); + return VM_FAULT_FALLBACK; + } + + /* Fallback on write dirty-tracking or COW */ + if (write && ttm_pgprot_is_wrprotecting(vma->vm_page_prot)) + return VM_FAULT_FALLBACK; + + ret = ttm_bo_vm_reserve(bo, vmf); + if (ret) + return ret; + + prot = vm_get_page_prot(vma->vm_flags); + ret = ttm_bo_vm_fault_reserved(vmf, prot, 1, fault_page_size); + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) + return ret; + + dma_resv_unlock(bo->base.resv); + + return ret; +} +#endif + void ttm_bo_vm_open(struct vm_area_struct *vma) { struct ttm_buffer_object *bo = vma->vm_private_data; @@ -445,7 +597,10 @@ static const struct vm_operations_struct ttm_bo_vm_ops = { .fault = ttm_bo_vm_fault, .open = ttm_bo_vm_open, .close = ttm_bo_vm_close, - .access = ttm_bo_vm_access + .access = ttm_bo_vm_access, +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + .huge_fault = ttm_bo_vm_huge_fault, +#endif };
static struct ttm_buffer_object *ttm_bo_vm_lookup(struct ttm_bo_device *bdev, diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c index f07aa857587c..17a5dca7b921 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c @@ -477,7 +477,7 @@ vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf) else prot = vm_get_page_prot(vma->vm_flags);
- ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault); + ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault, 1); if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) return ret;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 66ca49db9633..4fc90d53aa15 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -732,7 +732,8 @@ vm_fault_t ttm_bo_vm_reserve(struct ttm_buffer_object *bo,
vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, pgprot_t prot, - pgoff_t num_prefault); + pgoff_t num_prefault, + pgoff_t fault_page_size);
vm_fault_t ttm_bo_vm_fault(struct vm_fault *vmf);
From: Thomas Hellstrom thellstrom@vmware.com
With vmwgfx dirty-tracking we need a specialized huge_fault callback. Implement and hook it up.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Acked-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/vmwgfx/vmwgfx_drv.h | 4 ++ drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c | 74 +++++++++++++++++++++- drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c | 5 +- 3 files changed, 81 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index a31e726d6d71..82d86f2d2569 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h @@ -1428,6 +1428,10 @@ void vmw_bo_dirty_unmap(struct vmw_buffer_object *vbo, pgoff_t start, pgoff_t end); vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf); vm_fault_t vmw_bo_vm_mkwrite(struct vm_fault *vmf); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +vm_fault_t vmw_bo_vm_huge_fault(struct vm_fault *vmf, + enum page_entry_size pe_size); +#endif
/** * VMW_DEBUG_KMS - Debug output for kernel mode-setting diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c index 17a5dca7b921..cde3e07ebaf7 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c @@ -473,7 +473,7 @@ vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf) * a lot of unnecessary write faults. */ if (vbo->dirty && vbo->dirty->method == VMW_BO_DIRTY_MKWRITE) - prot = vma->vm_page_prot; + prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED); else prot = vm_get_page_prot(vma->vm_flags);
@@ -486,3 +486,75 @@ vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf)
return ret; } + +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +vm_fault_t vmw_bo_vm_huge_fault(struct vm_fault *vmf, + enum page_entry_size pe_size) +{ + struct vm_area_struct *vma = vmf->vma; + struct ttm_buffer_object *bo = (struct ttm_buffer_object *) + vma->vm_private_data; + struct vmw_buffer_object *vbo = + container_of(bo, struct vmw_buffer_object, base); + pgprot_t prot; + vm_fault_t ret; + pgoff_t fault_page_size; + bool write = vmf->flags & FAULT_FLAG_WRITE; + bool is_cow_mapping = + (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; + + switch (pe_size) { + case PE_SIZE_PMD: + fault_page_size = HPAGE_PMD_SIZE >> PAGE_SHIFT; + break; +#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD + case PE_SIZE_PUD: + fault_page_size = HPAGE_PUD_SIZE >> PAGE_SHIFT; + break; +#endif + default: + WARN_ON_ONCE(1); + return VM_FAULT_FALLBACK; + } + + /* Always do write dirty-tracking and COW on PTE level. */ + if (write && (READ_ONCE(vbo->dirty) || is_cow_mapping)) + return VM_FAULT_FALLBACK; + + ret = ttm_bo_vm_reserve(bo, vmf); + if (ret) + return ret; + + if (vbo->dirty) { + pgoff_t allowed_prefault; + unsigned long page_offset; + + page_offset = vmf->pgoff - + drm_vma_node_start(&bo->base.vma_node); + if (page_offset >= bo->num_pages || + vmw_resources_clean(vbo, page_offset, + page_offset + PAGE_SIZE, + &allowed_prefault)) { + ret = VM_FAULT_SIGBUS; + goto out_unlock; + } + + /* + * Write protect, so we get a new fault on write, and can + * split. + */ + prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED); + } else { + prot = vm_get_page_prot(vma->vm_flags); + } + + ret = ttm_bo_vm_fault_reserved(vmf, prot, 1, fault_page_size); + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) + return ret; + +out_unlock: + dma_resv_unlock(bo->base.resv); + + return ret; +} +#endif diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c index ce288756531b..34100d1f5a9d 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c @@ -34,7 +34,10 @@ int vmw_mmap(struct file *filp, struct vm_area_struct *vma) .page_mkwrite = vmw_bo_vm_mkwrite, .fault = vmw_bo_vm_fault, .open = ttm_bo_vm_open, - .close = ttm_bo_vm_close + .close = ttm_bo_vm_close, +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + .huge_fault = vmw_bo_vm_huge_fault, +#endif }; struct drm_file *file_priv = filp->private_data; struct vmw_private *dev_priv = vmw_priv(file_priv->minor->dev);
From: Thomas Hellstrom thellstrom@vmware.com
Unaligned virtual addresses makes it unlikely that huge page-table entries can be used. So align virtual buffer object address huge page boundaries to the underlying physical address huge page boundaries taking buffer object sizes into account to determine when it might be possible to use huge page-table entries.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Acked-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/drm_file.c | 136 +++++++++++++++++++++++++++++++++++++ include/drm/drm_file.h | 5 ++ 2 files changed, 141 insertions(+)
diff --git a/drivers/gpu/drm/drm_file.c b/drivers/gpu/drm/drm_file.c index 92d16724f949..40fae356d202 100644 --- a/drivers/gpu/drm/drm_file.c +++ b/drivers/gpu/drm/drm_file.c @@ -39,10 +39,13 @@ #include <linux/poll.h> #include <linux/slab.h>
+#include <uapi/asm/mman.h> + #include <drm/drm_client.h> #include <drm/drm_drv.h> #include <drm/drm_file.h> #include <drm/drm_print.h> +#include <drm/drm_vma_manager.h>
#include "drm_crtc_internal.h" #include "drm_internal.h" @@ -796,3 +799,136 @@ struct file *mock_drm_getfile(struct drm_minor *minor, unsigned int flags) return file; } EXPORT_SYMBOL_FOR_TESTS_ONLY(mock_drm_getfile); + +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +/* + * drm_addr_inflate() attempts to construct an aligned area by inflating + * the area size and skipping the unaligned start of the area. + * adapted from shmem_get_unmapped_area() + */ +static unsigned long drm_addr_inflate(unsigned long addr, + unsigned long len, + unsigned long pgoff, + unsigned long flags, + unsigned long huge_size) +{ + unsigned long offset, inflated_len; + unsigned long inflated_addr; + unsigned long inflated_offset; + + offset = (pgoff << PAGE_SHIFT) & (huge_size - 1); + if (offset && offset + len < 2 * huge_size) + return addr; + if ((addr & (huge_size - 1)) == offset) + return addr; + + inflated_len = len + huge_size - PAGE_SIZE; + if (inflated_len > TASK_SIZE) + return addr; + if (inflated_len < len) + return addr; + + inflated_addr = current->mm->get_unmapped_area(NULL, 0, inflated_len, + 0, flags); + if (IS_ERR_VALUE(inflated_addr)) + return addr; + if (inflated_addr & ~PAGE_MASK) + return addr; + + inflated_offset = inflated_addr & (huge_size - 1); + inflated_addr += offset - inflated_offset; + if (inflated_offset > offset) + inflated_addr += huge_size; + + if (inflated_addr > TASK_SIZE - len) + return addr; + + return inflated_addr; +} + +/** + * drm_get_unmapped_area() - Get an unused user-space virtual memory area + * suitable for huge page table entries. + * @file: The struct file representing the address space being mmap()'d. + * @uaddr: Start address suggested by user-space. + * @len: Length of the area. + * @pgoff: The page offset into the address space. + * @flags: mmap flags + * @mgr: The address space manager used by the drm driver. This argument can + * probably be removed at some point when all drivers use the same + * address space manager. + * + * This function attempts to find an unused user-space virtual memory area + * that can accommodate the size we want to map, and that is properly + * aligned to facilitate huge page table entries matching actual + * huge pages or huge page aligned memory in buffer objects. Buffer objects + * are assumed to start at huge page boundary pfns (io memory) or be + * populated by huge pages aligned to the start of the buffer object + * (system- or coherent memory). Adapted from shmem_get_unmapped_area. + * + * Return: aligned user-space address. + */ +unsigned long drm_get_unmapped_area(struct file *file, + unsigned long uaddr, unsigned long len, + unsigned long pgoff, unsigned long flags, + struct drm_vma_offset_manager *mgr) +{ + unsigned long addr; + unsigned long inflated_addr; + struct drm_vma_offset_node *node; + + if (len > TASK_SIZE) + return -ENOMEM; + + /* + * @pgoff is the file page-offset the huge page boundaries of + * which typically aligns to physical address huge page boundaries. + * That's not true for DRM, however, where physical address huge + * page boundaries instead are aligned with the offset from + * buffer object start. So adjust @pgoff to be the offset from + * buffer object start. + */ + drm_vma_offset_lock_lookup(mgr); + node = drm_vma_offset_lookup_locked(mgr, pgoff, 1); + if (node) + pgoff -= node->vm_node.start; + drm_vma_offset_unlock_lookup(mgr); + + addr = current->mm->get_unmapped_area(file, uaddr, len, pgoff, flags); + if (IS_ERR_VALUE(addr)) + return addr; + if (addr & ~PAGE_MASK) + return addr; + if (addr > TASK_SIZE - len) + return addr; + + if (len < HPAGE_PMD_SIZE) + return addr; + if (flags & MAP_FIXED) + return addr; + /* + * Our priority is to support MAP_SHARED mapped hugely; + * and support MAP_PRIVATE mapped hugely too, until it is COWed. + * But if caller specified an address hint, respect that as before. + */ + if (uaddr) + return addr; + + inflated_addr = drm_addr_inflate(addr, len, pgoff, flags, + HPAGE_PMD_SIZE); + + if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) && + len >= HPAGE_PUD_SIZE) + inflated_addr = drm_addr_inflate(inflated_addr, len, pgoff, + flags, HPAGE_PUD_SIZE); + return inflated_addr; +} +#else /* CONFIG_TRANSPARENT_HUGEPAGE */ +unsigned long drm_get_unmapped_area(struct file *file, + unsigned long uaddr, unsigned long len, + unsigned long pgoff, unsigned long flags, + struct drm_vma_offset_manager *mgr) +{ + return current->mm->get_unmapped_area(file, uaddr, len, pgoff, flags); +} +#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ diff --git a/include/drm/drm_file.h b/include/drm/drm_file.h index 8b099b347817..06fe3da09b27 100644 --- a/include/drm/drm_file.h +++ b/include/drm/drm_file.h @@ -387,6 +387,11 @@ void drm_event_cancel_free(struct drm_device *dev, struct drm_pending_event *p); void drm_send_event_locked(struct drm_device *dev, struct drm_pending_event *e); void drm_send_event(struct drm_device *dev, struct drm_pending_event *e); +struct drm_vma_offset_manager; +unsigned long drm_get_unmapped_area(struct file *file, + unsigned long uaddr, unsigned long len, + unsigned long pgoff, unsigned long flags, + struct drm_vma_offset_manager *mgr);
struct file *mock_drm_getfile(struct drm_minor *minor, unsigned int flags);
From: Thomas Hellstrom thellstrom@vmware.com
Using huge page-table entries requires that the physical address of the start of a buffer object is huge page size aligned. Make a special version of the TTM range manager that accomplishes this, but falls back to a smaller page size alignment (PUD->PMD, PMD->NORMAL) to avoid eviction. If other drivers want to use it in the future, it can be made a TTM generic helper. Note that drivers can force eviction for a certain alignment by assigning the TTM GPU alignment correspondingly.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Acked-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/vmwgfx/Makefile | 1 + drivers/gpu/drm/vmwgfx/vmwgfx_drv.h | 7 ++ drivers/gpu/drm/vmwgfx/vmwgfx_thp.c | 166 ++++++++++++++++++++++++++++ 3 files changed, 174 insertions(+) create mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_thp.c
diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile index c877a21a0739..421dd2a497a5 100644 --- a/drivers/gpu/drm/vmwgfx/Makefile +++ b/drivers/gpu/drm/vmwgfx/Makefile @@ -11,4 +11,5 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \ vmwgfx_validation.o vmwgfx_page_dirty.o \ ttm_object.o ttm_lock.o
+vmwgfx-$(CONFIG_TRANSPARENT_HUGEPAGE) += vmwgfx_thp.o obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index 82d86f2d2569..06267184aa0a 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h @@ -1433,6 +1433,13 @@ vm_fault_t vmw_bo_vm_huge_fault(struct vm_fault *vmf, enum page_entry_size pe_size); #endif
+/* Transparent hugepage support - vmwgfx_thp.c */ +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +extern const struct ttm_mem_type_manager_func vmw_thp_func; +#else +#define vmw_thp_func ttm_bo_manager_func +#endif + /** * VMW_DEBUG_KMS - Debug output for kernel mode-setting * diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_thp.c b/drivers/gpu/drm/vmwgfx/vmwgfx_thp.c new file mode 100644 index 000000000000..b7c816ba7166 --- /dev/null +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_thp.c @@ -0,0 +1,166 @@ +// SPDX-License-Identifier: GPL-2.0 OR MIT +/* + * Huge page-table-entry support for IO memory. + * + * Copyright (C) 2007-2019 Vmware, Inc. All rights reservedd. + */ +#include "vmwgfx_drv.h" +#include <drm/ttm/ttm_module.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_placement.h> + +/** + * struct vmw_thp_manager - Range manager implementing huge page alignment + * + * @mm: The underlying range manager. Protected by @lock. + * @lock: Manager lock. + */ +struct vmw_thp_manager { + struct drm_mm mm; + spinlock_t lock; +}; + +static int vmw_thp_insert_aligned(struct drm_mm *mm, struct drm_mm_node *node, + unsigned long align_pages, + const struct ttm_place *place, + struct ttm_mem_reg *mem, + unsigned long lpfn, + enum drm_mm_insert_mode mode) +{ + if (align_pages >= mem->page_alignment && + (!mem->page_alignment || align_pages % mem->page_alignment == 0)) { + return drm_mm_insert_node_in_range(mm, node, + mem->num_pages, + align_pages, 0, + place->fpfn, lpfn, mode); + } + + return -ENOSPC; +} + +static int vmw_thp_get_node(struct ttm_mem_type_manager *man, + struct ttm_buffer_object *bo, + const struct ttm_place *place, + struct ttm_mem_reg *mem) +{ + struct vmw_thp_manager *rman = (struct vmw_thp_manager *) man->priv; + struct drm_mm *mm = &rman->mm; + struct drm_mm_node *node; + unsigned long align_pages; + unsigned long lpfn; + enum drm_mm_insert_mode mode = DRM_MM_INSERT_BEST; + int ret; + + node = kzalloc(sizeof(*node), GFP_KERNEL); + if (!node) + return -ENOMEM; + + lpfn = place->lpfn; + if (!lpfn) + lpfn = man->size; + + mode = DRM_MM_INSERT_BEST; + if (place->flags & TTM_PL_FLAG_TOPDOWN) + mode = DRM_MM_INSERT_HIGH; + + spin_lock(&rman->lock); + if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)) { + align_pages = (HPAGE_PUD_SIZE >> PAGE_SHIFT); + if (mem->num_pages >= align_pages) { + ret = vmw_thp_insert_aligned(mm, node, align_pages, + place, mem, lpfn, mode); + if (!ret) + goto found_unlock; + } + } + + align_pages = (HPAGE_PMD_SIZE >> PAGE_SHIFT); + if (mem->num_pages >= align_pages) { + ret = vmw_thp_insert_aligned(mm, node, align_pages, place, mem, + lpfn, mode); + if (!ret) + goto found_unlock; + } + + ret = drm_mm_insert_node_in_range(mm, node, mem->num_pages, + mem->page_alignment, 0, + place->fpfn, lpfn, mode); +found_unlock: + spin_unlock(&rman->lock); + + if (unlikely(ret)) { + kfree(node); + } else { + mem->mm_node = node; + mem->start = node->start; + } + + return 0; +} + + + +static void vmw_thp_put_node(struct ttm_mem_type_manager *man, + struct ttm_mem_reg *mem) +{ + struct vmw_thp_manager *rman = (struct vmw_thp_manager *) man->priv; + + if (mem->mm_node) { + spin_lock(&rman->lock); + drm_mm_remove_node(mem->mm_node); + spin_unlock(&rman->lock); + + kfree(mem->mm_node); + mem->mm_node = NULL; + } +} + +static int vmw_thp_init(struct ttm_mem_type_manager *man, + unsigned long p_size) +{ + struct vmw_thp_manager *rman; + + rman = kzalloc(sizeof(*rman), GFP_KERNEL); + if (!rman) + return -ENOMEM; + + drm_mm_init(&rman->mm, 0, p_size); + spin_lock_init(&rman->lock); + man->priv = rman; + return 0; +} + +static int vmw_thp_takedown(struct ttm_mem_type_manager *man) +{ + struct vmw_thp_manager *rman = (struct vmw_thp_manager *) man->priv; + struct drm_mm *mm = &rman->mm; + + spin_lock(&rman->lock); + if (drm_mm_clean(mm)) { + drm_mm_takedown(mm); + spin_unlock(&rman->lock); + kfree(rman); + man->priv = NULL; + return 0; + } + spin_unlock(&rman->lock); + return -EBUSY; +} + +static void vmw_thp_debug(struct ttm_mem_type_manager *man, + struct drm_printer *printer) +{ + struct vmw_thp_manager *rman = (struct vmw_thp_manager *) man->priv; + + spin_lock(&rman->lock); + drm_mm_print(&rman->mm, printer); + spin_unlock(&rman->lock); +} + +const struct ttm_mem_type_manager_func vmw_thp_func = { + .init = vmw_thp_init, + .takedown = vmw_thp_takedown, + .get_node = vmw_thp_get_node, + .put_node = vmw_thp_put_node, + .debug = vmw_thp_debug +};
From: Thomas Hellstrom thellstrom@vmware.com
Start using the helpers that align buffer object user-space addresses and buffer object vram addresses to huge page boundaries. This is to improve the chances of allowing huge page-table entries.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Roland Scheidegger sroland@vmware.com Acked-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/drm_file.c | 1 + drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 13 +++++++++++++ drivers/gpu/drm/vmwgfx/vmwgfx_drv.h | 1 + drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c | 2 +- 4 files changed, 16 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/drm_file.c b/drivers/gpu/drm/drm_file.c index 40fae356d202..1df2fca608c3 100644 --- a/drivers/gpu/drm/drm_file.c +++ b/drivers/gpu/drm/drm_file.c @@ -932,3 +932,4 @@ unsigned long drm_get_unmapped_area(struct file *file, return current->mm->get_unmapped_area(file, uaddr, len, pgoff, flags); } #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ +EXPORT_SYMBOL_GPL(drm_get_unmapped_area); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c index e962048f65d2..5452cabb4a2e 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c @@ -1215,6 +1215,18 @@ static void vmw_remove(struct pci_dev *pdev) drm_put_dev(dev); }
+static unsigned long +vmw_get_unmapped_area(struct file *file, unsigned long uaddr, + unsigned long len, unsigned long pgoff, + unsigned long flags) +{ + struct drm_file *file_priv = file->private_data; + struct vmw_private *dev_priv = vmw_priv(file_priv->minor->dev); + + return drm_get_unmapped_area(file, uaddr, len, pgoff, flags, + &dev_priv->vma_manager); +} + static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val, void *ptr) { @@ -1386,6 +1398,7 @@ static const struct file_operations vmwgfx_driver_fops = { .compat_ioctl = vmw_compat_ioctl, #endif .llseek = noop_llseek, + .get_unmapped_area = vmw_get_unmapped_area, };
static struct drm_driver driver = { diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index 06267184aa0a..9ea145cffa3d 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h @@ -929,6 +929,7 @@ extern int vmw_mmap(struct file *filp, struct vm_area_struct *vma);
extern void vmw_validation_mem_init_ttm(struct vmw_private *dev_priv, size_t gran); + /** * TTM buffer object driver - vmwgfx_ttm_buffer.c */ diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c index d8ea3dd10af0..34c721ab3ff3 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c @@ -754,7 +754,7 @@ static int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, break; case TTM_PL_VRAM: /* "On-card" video ram */ - man->func = &ttm_bo_manager_func; + man->func = &vmw_thp_func; man->gpu_offset = 0; man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE; man->available_caching = TTM_PL_FLAG_CACHED;
Andrew, Michal
I'm wondering what's the best way here to get the patches touching mm reviewed and accepted? While drm people and VMware internal people have looked at them, I think the huge_fault() fallback splitting and the introduction of vma_is_special_huge() needs looking at more thoroughly.
Apart from that, if possible, I think the best way to merge this series is also through a DRM tree.
Thanks, Thomas
On 2/20/20 1:27 PM, Thomas Hellström (VMware) wrote:
In order to reduce TLB misses and CPU usage this patchset enables huge- and giant page-table entries for TTM and TTM-enabled graphics drivers.
Patch 1 and 2 introduce a vma_is_special_huge() function to make the mm code take the same path as DAX when splitting huge- and giant page table entries, (which currently means zapping the page-table entry and rely on re-faulting).
Patch 3 makes the mm code split existing huge page-table entries on huge_fault fallbacks. Typically on COW or on buffer-objects that want write-notify. COW and write-notification is always done on the lowest page-table level. See the patch log message for additional considerations.
Patch 4 introduces functions to allow the graphics drivers to manipulate the caching- and encryption flags of huge page-table entries without ugly hacks.
Patch 5 implements the huge_fault handler in TTM. This enables huge page-table entries, provided that the kernel is configured to support transhuge pages, either by default or using madvise(). However, they are unlikely to be inserted unless the kernel buffer object pfns and user-space addresses align perfectly. There are various options here, but since buffer objects that reside in system pages typically start at huge page boundaries if they are backed by huge pages, we try to enforce buffer object starting pfns and user-space addresses to be huge page-size aligned if their size exceeds a huge page-size. If pud-size transhuge ("giant") pages are enabled by the arch, the same holds for those.
Patch 6 implements a specialized huge_fault handler for vmwgfx. The vmwgfx driver may perform dirty-tracking and needs some special code to handle that correctly.
Patch 7 implements a drm helper to align user-space addresses according to the above scheme, if possible.
Patch 8 implements a TTM range manager for vmwgfx that does the same for graphics IO memory. This may later be reused by other graphics drivers if necessary.
Patch 9 finally hooks up the helpers of patch 7 and 8 to the vmwgfx driver. A similar change is needed for graphics drivers that want a reasonable likelyhood of actually using huge page-table entries.
If a buffer object size is not huge-page or giant-page aligned, its size will NOT be inflated by this patchset. This means that the buffer object tail will use smaller size page-table entries and thus no memory overhead occurs. Drivers that want to pay the memory overhead price need to implement their own scheme to inflate buffer-object sizes.
PMD size huge page-table-entries have been tested with vmwgfx and found to work well both with system memory backed and IO memory backed buffer objects.
PUD size giant page-table-entries have seen limited (fault and COW) testing using a modified kernel (to support 1GB page allocations) and a fake vmwgfx TTM memory type. The vmwgfx driver does otherwise not support 1GB-size IO memory resources.
Comments and suggestions welcome. Thomas
Changes since RFC:
- Check for buffer objects present in contigous IO Memory (Christian König)
- Rebased on the vmwgfx emulated coherent memory functionality. That rebase adds patch 5.
Changes since v1:
- Make the new TTM range manager vmwgfx-specific. (Christian König)
- Minor fixes for configs that don't support or only partially support transhuge pages.
Changes since v2:
- Minor coding style and doc fixes in patch 5/9 (Christian König)
- Patch 5/9 doesn't touch mm. Remove from the patch title.
Changes since v3:
- Added reviews and acks
- Implemented ugly but generic ttm_pgprot_is_wrprotecting() instead of arch specific code.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Michal Hocko mhocko@suse.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Ralph Campbell rcampbell@nvidia.com Cc: "Jérôme Glisse" jglisse@redhat.com Cc: "Christian König" christian.koenig@amd.com Cc: Dan Williams dan.j.williams@intel.com
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Fri, 28 Feb 2020 14:08:04 +0100 Thomas Hellström (VMware) thomas_os@shipmail.org wrote:
I'm wondering what's the best way here to get the patches touching mm reviewed and accepted? While drm people and VMware internal people have looked at them, I think the huge_fault() fallback splitting and the introduction of vma_is_special_huge() needs looking at more thoroughly.
Apart from that, if possible, I think the best way to merge this series is also through a DRM tree.
Patches 1-3 look OK to me. I just had a few commenting/changelogging niggles.
On 3/1/20 5:04 AM, Andrew Morton wrote:
On Fri, 28 Feb 2020 14:08:04 +0100 Thomas Hellström (VMware) thomas_os@shipmail.org wrote:
I'm wondering what's the best way here to get the patches touching mm reviewed and accepted? While drm people and VMware internal people have looked at them, I think the huge_fault() fallback splitting and the introduction of vma_is_special_huge() needs looking at more thoroughly.
Apart from that, if possible, I think the best way to merge this series is also through a DRM tree.
Patches 1-3 look OK to me. I just had a few commenting/changelogging niggles.
Thanks for reviewing, Andrew.
I just updated the series following your comments.
Thanks,
Thomas
On Fri 28-02-20 14:08:04, Thomas Hellström (VMware) wrote:
Andrew, Michal
I'm wondering what's the best way here to get the patches touching mm reviewed and accepted?
I am sorry, but I am busy with other stuff and unlikely to find time to review this series.
On Thu, 20 Feb 2020 13:27:15 +0100 Thomas Hellstrom wrote:
+static vm_fault_t ttm_bo_vm_huge_fault(struct vm_fault *vmf,
enum page_entry_size pe_size)+{
- struct vm_area_struct *vma = vmf->vma;
- pgprot_t prot;
- struct ttm_buffer_object *bo = vma->vm_private_data;
- vm_fault_t ret;
- pgoff_t fault_page_size = 0;
- bool write = vmf->flags & FAULT_FLAG_WRITE;
- switch (pe_size) {
- case PE_SIZE_PMD:
fault_page_size = HPAGE_PMD_SIZE >> PAGE_SHIFT;break;+#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
- case PE_SIZE_PUD:
fault_page_size = HPAGE_PUD_SIZE >> PAGE_SHIFT;break;+#endif
- default:
WARN_ON_ONCE(1);return VM_FAULT_FALLBACK;- }
- /* Fallback on write dirty-tracking or COW */
- if (write && ttm_pgprot_is_wrprotecting(vma->vm_page_prot))
return VM_FAULT_FALLBACK;- ret = ttm_bo_vm_reserve(bo, vmf);
- if (ret)
return ret;- prot = vm_get_page_prot(vma->vm_flags);
- ret = ttm_bo_vm_fault_reserved(vmf, prot, 1, fault_page_size);
- if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
return ret;
Seems it does not make much sense to check VM_FAULT_RETRY and return as at least resv lock is left behind without care.
- dma_resv_unlock(bo->base.resv);
- return ret;
+}
On Sun, 2020-03-01 at 21:49 +0800, Hillf Danton wrote:
On Thu, 20 Feb 2020 13:27:15 +0100 Thomas Hellstrom wrote:
+static vm_fault_t ttm_bo_vm_huge_fault(struct vm_fault *vmf,
enum page_entry_size pe_size)+{
- struct vm_area_struct *vma = vmf->vma;
- pgprot_t prot;
- struct ttm_buffer_object *bo = vma->vm_private_data;
- vm_fault_t ret;
- pgoff_t fault_page_size = 0;
- bool write = vmf->flags & FAULT_FLAG_WRITE;
- switch (pe_size) {
- case PE_SIZE_PMD:
fault_page_size = HPAGE_PMD_SIZE >> PAGE_SHIFT;break;+#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
- case PE_SIZE_PUD:
fault_page_size = HPAGE_PUD_SIZE >> PAGE_SHIFT;break;+#endif
- default:
WARN_ON_ONCE(1);return VM_FAULT_FALLBACK;- }
- /* Fallback on write dirty-tracking or COW */
- if (write && ttm_pgprot_is_wrprotecting(vma->vm_page_prot))
return VM_FAULT_FALLBACK;- ret = ttm_bo_vm_reserve(bo, vmf);
- if (ret)
return ret;- prot = vm_get_page_prot(vma->vm_flags);
- ret = ttm_bo_vm_fault_reserved(vmf, prot, 1, fault_page_size);
- if (ret == VM_FAULT_RETRY && !(vmf->flags &
FAULT_FLAG_RETRY_NOWAIT))
return ret;Seems it does not make much sense to check VM_FAULT_RETRY and return as at least resv lock is left behind without care.
With this particular flag combination, both the mm_sem and the dma_resv lock have already been released by TTM.
It's a special case allowing for drivers to release the mmap_sem when waiting for IO.
That should probably be documented better in TTM.
/Thomas
dri-devel@lists.freedesktop.org
-
Andrew Morton -
Dan Williams -
Hillf Danton -
Michal Hocko -
Thomas Hellstrom -
Thomas Hellström (VMware)