Message ID | 20240429114104.182890-1-ryan.roberts@arm.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [v1] fs/proc/task_mmu: Fix uffd-wp confusion in pagemap_scan_pmd_entry() | expand |
On 29.04.24 13:41, Ryan Roberts wrote: > pagemap_scan_pmd_entry() checks if uffd-wp is set on each pte to avoid > unnecessary if set. However it was previously checking with > `pte_uffd_wp(ptep_get(pte))` without first confirming that the pte was > present. It is only valid to call pte_uffd_wp() for present ptes. For > swap ptes, pte_swp_uffd_wp() must be called because the uffd-wp bit may > be kept in a different position, depending on the arch. > > This was leading to test failures in the pagemap_ioctl mm selftest, when > bringing up uffd-wp support on arm64 due to incorrectly interpretting > the uffd-wp status of migration entries. > > Let's fix this by using the correct check based on pte_present(). While > we are at it, let's pass the pte to make_uffd_wp_pte() to avoid the > pointless extra ptep_get() which can't be optimized out due to > READ_ONCE() on many arches. > > Closes: https://lore.kernel.org/linux-arm-kernel/ZiuyGXt0XWwRgFh9@x1n/ > Fixes: 12f6b01a0bcb ("fs/proc/task_mmu: add fast paths to get/clear PAGE_IS_WRITTEN flag") > Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> > --- Acked-by: David Hildenbrand <david@redhat.com>
Thanks for finding and fixing Ryan! On 4/29/24 4:41 PM, Ryan Roberts wrote: > pagemap_scan_pmd_entry() checks if uffd-wp is set on each pte to avoid > unnecessary if set. However it was previously checking with > `pte_uffd_wp(ptep_get(pte))` without first confirming that the pte was > present. It is only valid to call pte_uffd_wp() for present ptes. For > swap ptes, pte_swp_uffd_wp() must be called because the uffd-wp bit may > be kept in a different position, depending on the arch. > > This was leading to test failures in the pagemap_ioctl mm selftest, when > bringing up uffd-wp support on arm64 due to incorrectly interpretting > the uffd-wp status of migration entries. > > Let's fix this by using the correct check based on pte_present(). While > we are at it, let's pass the pte to make_uffd_wp_pte() to avoid the > pointless extra ptep_get() which can't be optimized out due to > READ_ONCE() on many arches. > > Closes: https://lore.kernel.org/linux-arm-kernel/ZiuyGXt0XWwRgFh9@x1n/ > Fixes: 12f6b01a0bcb ("fs/proc/task_mmu: add fast paths to get/clear PAGE_IS_WRITTEN flag") > Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> Reviewed-by: Muhammad Usama Anjum <usama.anjum@collabora.com> Tested-by: Muhammad Usama Anjum <usama.anjum@collabora.com> > --- > fs/proc/task_mmu.c | 22 +++++++++++++--------- > 1 file changed, 13 insertions(+), 9 deletions(-) > > diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c > index af4bc1da0c01..102f48668c35 100644 > --- a/fs/proc/task_mmu.c > +++ b/fs/proc/task_mmu.c > @@ -1817,10 +1817,8 @@ static unsigned long pagemap_page_category(struct pagemap_scan_private *p, > } > > static void make_uffd_wp_pte(struct vm_area_struct *vma, > - unsigned long addr, pte_t *pte) > + unsigned long addr, pte_t *pte, pte_t ptent) > { > - pte_t ptent = ptep_get(pte); > - > if (pte_present(ptent)) { > pte_t old_pte; > > @@ -2175,9 +2173,12 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) { > /* Fast path for performing exclusive WP */ > for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { > - if (pte_uffd_wp(ptep_get(pte))) > + pte_t ptent = ptep_get(pte); > + > + if ((pte_present(ptent) && pte_uffd_wp(ptent)) || > + pte_swp_uffd_wp_any(ptent)) > continue; > - make_uffd_wp_pte(vma, addr, pte); > + make_uffd_wp_pte(vma, addr, pte, ptent); > if (!flush_end) > start = addr; > flush_end = addr + PAGE_SIZE; > @@ -2190,8 +2191,10 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > p->arg.return_mask == PAGE_IS_WRITTEN) { > for (addr = start; addr < end; pte++, addr += PAGE_SIZE) { > unsigned long next = addr + PAGE_SIZE; > + pte_t ptent = ptep_get(pte); > > - if (pte_uffd_wp(ptep_get(pte))) > + if ((pte_present(ptent) && pte_uffd_wp(ptent)) || > + pte_swp_uffd_wp_any(ptent)) > continue; > ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN, > p, addr, &next); > @@ -2199,7 +2202,7 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > break; > if (~p->arg.flags & PM_SCAN_WP_MATCHING) > continue; > - make_uffd_wp_pte(vma, addr, pte); > + make_uffd_wp_pte(vma, addr, pte, ptent); > if (!flush_end) > start = addr; > flush_end = next; > @@ -2208,8 +2211,9 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > } > > for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { > + pte_t ptent = ptep_get(pte); > unsigned long categories = p->cur_vma_category | > - pagemap_page_category(p, vma, addr, ptep_get(pte)); > + pagemap_page_category(p, vma, addr, ptent); > unsigned long next = addr + PAGE_SIZE; > > if (!pagemap_scan_is_interesting_page(categories, p)) > @@ -2224,7 +2228,7 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > if (~categories & PAGE_IS_WRITTEN) > continue; > > - make_uffd_wp_pte(vma, addr, pte); > + make_uffd_wp_pte(vma, addr, pte, ptent); > if (!flush_end) > start = addr; > flush_end = next; > -- > 2.25.1 >
On 29/04/2024 12:41, Ryan Roberts wrote: > pagemap_scan_pmd_entry() checks if uffd-wp is set on each pte to avoid > unnecessary if set. However it was previously checking with > `pte_uffd_wp(ptep_get(pte))` without first confirming that the pte was > present. It is only valid to call pte_uffd_wp() for present ptes. For > swap ptes, pte_swp_uffd_wp() must be called because the uffd-wp bit may > be kept in a different position, depending on the arch. > > This was leading to test failures in the pagemap_ioctl mm selftest, when > bringing up uffd-wp support on arm64 due to incorrectly interpretting > the uffd-wp status of migration entries. > > Let's fix this by using the correct check based on pte_present(). While > we are at it, let's pass the pte to make_uffd_wp_pte() to avoid the > pointless extra ptep_get() which can't be optimized out due to > READ_ONCE() on many arches. > > Closes: https://lore.kernel.org/linux-arm-kernel/ZiuyGXt0XWwRgFh9@x1n/ > Fixes: 12f6b01a0bcb ("fs/proc/task_mmu: add fast paths to get/clear PAGE_IS_WRITTEN flag") > Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> I guess this should have cc'ed stable but I forgot to add it. Are you able to fix this up when you take it, Andrew, or do I need to repost? > --- > fs/proc/task_mmu.c | 22 +++++++++++++--------- > 1 file changed, 13 insertions(+), 9 deletions(-) > > diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c > index af4bc1da0c01..102f48668c35 100644 > --- a/fs/proc/task_mmu.c > +++ b/fs/proc/task_mmu.c > @@ -1817,10 +1817,8 @@ static unsigned long pagemap_page_category(struct pagemap_scan_private *p, > } > > static void make_uffd_wp_pte(struct vm_area_struct *vma, > - unsigned long addr, pte_t *pte) > + unsigned long addr, pte_t *pte, pte_t ptent) > { > - pte_t ptent = ptep_get(pte); > - > if (pte_present(ptent)) { > pte_t old_pte; > > @@ -2175,9 +2173,12 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) { > /* Fast path for performing exclusive WP */ > for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { > - if (pte_uffd_wp(ptep_get(pte))) > + pte_t ptent = ptep_get(pte); > + > + if ((pte_present(ptent) && pte_uffd_wp(ptent)) || > + pte_swp_uffd_wp_any(ptent)) > continue; > - make_uffd_wp_pte(vma, addr, pte); > + make_uffd_wp_pte(vma, addr, pte, ptent); > if (!flush_end) > start = addr; > flush_end = addr + PAGE_SIZE; > @@ -2190,8 +2191,10 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > p->arg.return_mask == PAGE_IS_WRITTEN) { > for (addr = start; addr < end; pte++, addr += PAGE_SIZE) { > unsigned long next = addr + PAGE_SIZE; > + pte_t ptent = ptep_get(pte); > > - if (pte_uffd_wp(ptep_get(pte))) > + if ((pte_present(ptent) && pte_uffd_wp(ptent)) || > + pte_swp_uffd_wp_any(ptent)) > continue; > ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN, > p, addr, &next); > @@ -2199,7 +2202,7 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > break; > if (~p->arg.flags & PM_SCAN_WP_MATCHING) > continue; > - make_uffd_wp_pte(vma, addr, pte); > + make_uffd_wp_pte(vma, addr, pte, ptent); > if (!flush_end) > start = addr; > flush_end = next; > @@ -2208,8 +2211,9 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > } > > for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { > + pte_t ptent = ptep_get(pte); > unsigned long categories = p->cur_vma_category | > - pagemap_page_category(p, vma, addr, ptep_get(pte)); > + pagemap_page_category(p, vma, addr, ptent); > unsigned long next = addr + PAGE_SIZE; > > if (!pagemap_scan_is_interesting_page(categories, p)) > @@ -2224,7 +2228,7 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, > if (~categories & PAGE_IS_WRITTEN) > continue; > > - make_uffd_wp_pte(vma, addr, pte); > + make_uffd_wp_pte(vma, addr, pte, ptent); > if (!flush_end) > start = addr; > flush_end = next; > -- > 2.25.1 >
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c index af4bc1da0c01..102f48668c35 100644 --- a/fs/proc/task_mmu.c +++ b/fs/proc/task_mmu.c @@ -1817,10 +1817,8 @@ static unsigned long pagemap_page_category(struct pagemap_scan_private *p, } static void make_uffd_wp_pte(struct vm_area_struct *vma, - unsigned long addr, pte_t *pte) + unsigned long addr, pte_t *pte, pte_t ptent) { - pte_t ptent = ptep_get(pte); - if (pte_present(ptent)) { pte_t old_pte; @@ -2175,9 +2173,12 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, if ((p->arg.flags & PM_SCAN_WP_MATCHING) && !p->vec_out) { /* Fast path for performing exclusive WP */ for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { - if (pte_uffd_wp(ptep_get(pte))) + pte_t ptent = ptep_get(pte); + + if ((pte_present(ptent) && pte_uffd_wp(ptent)) || + pte_swp_uffd_wp_any(ptent)) continue; - make_uffd_wp_pte(vma, addr, pte); + make_uffd_wp_pte(vma, addr, pte, ptent); if (!flush_end) start = addr; flush_end = addr + PAGE_SIZE; @@ -2190,8 +2191,10 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, p->arg.return_mask == PAGE_IS_WRITTEN) { for (addr = start; addr < end; pte++, addr += PAGE_SIZE) { unsigned long next = addr + PAGE_SIZE; + pte_t ptent = ptep_get(pte); - if (pte_uffd_wp(ptep_get(pte))) + if ((pte_present(ptent) && pte_uffd_wp(ptent)) || + pte_swp_uffd_wp_any(ptent)) continue; ret = pagemap_scan_output(p->cur_vma_category | PAGE_IS_WRITTEN, p, addr, &next); @@ -2199,7 +2202,7 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, break; if (~p->arg.flags & PM_SCAN_WP_MATCHING) continue; - make_uffd_wp_pte(vma, addr, pte); + make_uffd_wp_pte(vma, addr, pte, ptent); if (!flush_end) start = addr; flush_end = next; @@ -2208,8 +2211,9 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, } for (addr = start; addr != end; pte++, addr += PAGE_SIZE) { + pte_t ptent = ptep_get(pte); unsigned long categories = p->cur_vma_category | - pagemap_page_category(p, vma, addr, ptep_get(pte)); + pagemap_page_category(p, vma, addr, ptent); unsigned long next = addr + PAGE_SIZE; if (!pagemap_scan_is_interesting_page(categories, p)) @@ -2224,7 +2228,7 @@ static int pagemap_scan_pmd_entry(pmd_t *pmd, unsigned long start, if (~categories & PAGE_IS_WRITTEN) continue; - make_uffd_wp_pte(vma, addr, pte); + make_uffd_wp_pte(vma, addr, pte, ptent); if (!flush_end) start = addr; flush_end = next;
pagemap_scan_pmd_entry() checks if uffd-wp is set on each pte to avoid unnecessary if set. However it was previously checking with `pte_uffd_wp(ptep_get(pte))` without first confirming that the pte was present. It is only valid to call pte_uffd_wp() for present ptes. For swap ptes, pte_swp_uffd_wp() must be called because the uffd-wp bit may be kept in a different position, depending on the arch. This was leading to test failures in the pagemap_ioctl mm selftest, when bringing up uffd-wp support on arm64 due to incorrectly interpretting the uffd-wp status of migration entries. Let's fix this by using the correct check based on pte_present(). While we are at it, let's pass the pte to make_uffd_wp_pte() to avoid the pointless extra ptep_get() which can't be optimized out due to READ_ONCE() on many arches. Closes: https://lore.kernel.org/linux-arm-kernel/ZiuyGXt0XWwRgFh9@x1n/ Fixes: 12f6b01a0bcb ("fs/proc/task_mmu: add fast paths to get/clear PAGE_IS_WRITTEN flag") Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> --- fs/proc/task_mmu.c | 22 +++++++++++++--------- 1 file changed, 13 insertions(+), 9 deletions(-) -- 2.25.1