- 13 May, 2022 34 commits
-
-
SeongJae Park authored
Currently, DAMON sysfs interface doesn't provide a way for adjusting DAMON input parameters while it is turned on. Therefore, users who want to reconfigure DAMON need to stop DAMON and restart. This means all the monitoring results that accumulated so far, which could be useful, should be flushed. This would be inefficient for many cases. For an example, let's suppose a sysadmin was running a DAMON-based Operation Scheme to find memory regions not accessed for more than 5 mins and page out the regions. If it turns out the 5 mins threshold was too long and therefore the sysadmin wants to reduce it to 4 mins, the sysadmin should turn off DAMON, restart it, and wait for at least 4 more minutes so that DAMON can find the cold memory regions, even though DAMON was knowing there are regions that not accessed for 4 mins at the time of shutdown. This commit makes DAMON sysfs interface to support online DAMON input parameters updates by adding a new input keyword for the 'state' DAMON sysfs file, 'commit'. Writing the keyword to the 'state' file while the corresponding kdamond is running makes the kdamond to read the sysfs file values again and update the DAMON context. Link: https://lkml.kernel.org/r/20220429160606.127307-12-sj@kernel.orgSigned-off-by:
SeongJae Park <sj@kernel.org> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org>
-
SeongJae Park authored
Only '->kdamond' and '->kdamond_stop' are protected by 'kdamond_lock' of 'struct damon_ctx'. All other DAMON context internal data items are recommended to be accessed in DAMON callbacks, or under some additional synchronizations. But, DAMON sysfs is accessing the schemes stat under 'kdamond_lock'. It makes no big issue as the read values are not used anywhere inside kernel, but would better to be fixed. This commit moves the reads to DAMON callback context, as supposed to be used for the purpose. Link: https://lkml.kernel.org/r/20220429160606.127307-11-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
DAMON sysfs 'state' file handling code is using string literals in both 'state_show()' and 'state_store()'. This makes the code error prone and inflexible for future extensions. To improve the situation, this commit defines possible input strings and 'enum' for identifying each input keyword only once, and refactors the code to reuse those. Link: https://lkml.kernel.org/r/20220429160606.127307-10-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
'damon_set_regions()' is general enough so that it can also be used for only creating regions. This commit makes DAMON sysfs interface to reuse the function rather keeping two implementations for a same purpose. Link: https://lkml.kernel.org/r/20220429160606.127307-9-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
This commit separates DAMON sysfs interface's monitoring context targets setup code to a new function for better readability. Link: https://lkml.kernel.org/r/20220429160606.127307-8-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
Having multiple targets for physical address space monitoring makes no sense. This commit prohibits such a ridiculous DAMON context setup my making the DAMON context build function to check and return an error for the case. Link: https://lkml.kernel.org/r/20220429160606.127307-7-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
'damon_va_apply_three_regions()' is just a wrapper of its general version, 'damon_set_regions()'. This commit replaces the wrapper calls to directly call the general version. Link: https://lkml.kernel.org/r/20220429160606.127307-6-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
This commit moves 'damon_set_regions()' from vaddr to core, as it is aimed to be used by not only 'vaddr' but also other parts of DAMON. Link: https://lkml.kernel.org/r/20220429160606.127307-5-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
'damon_va_apply_three_regions()' is for adjusting address ranges to fit in three discontiguous ranges. The function can be generalized for arbitrary number of discontiguous ranges and reused for future usage, such as arbitrary online regions update. For such future usage, this commit introduces a generalized version of the function called 'damon_set_regions()'. Link: https://lkml.kernel.org/r/20220429160606.127307-4-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
When 'after_sampling()' or 'after_aggregation()' DAMON callbacks return an error, kdamond continues the remaining loop once. It makes no much sense to run the remaining part while something wrong already happened. The context might be corrupted or having invalid data. This commit therefore makes kdamond skips the remaining works and immediately finish in the cases. Link: https://lkml.kernel.org/r/20220429160606.127307-3-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
Patch series "mm/damon: Support online tuning". Effects of DAMON and DAMON-based Operation Schemes highly depends on the configurations. Wrong configurations could even result in unexpected efficiency degradations. For finding a best configuration, repeating incremental configuration changes and results measurements, in other words, online tuning, could be helpful. Nevertheless, DAMON kernel API supports only restrictive online tuning. Worse yet, the sysfs-based DAMON user interface doesn't support online tuning at all. DAMON_RECLAIM also doesn't support online tuning. This patchset makes the DAMON kernel API, DAMON sysfs interface, and DAMON_RECLAIM supports online tuning. Sequence of patches ------------------- First two patches enhance DAMON online tuning for kernel API users. Specifically, patch 1 let kernel API users to be able to do DAMON online tuning without a restriction, and patch 2 makes error handling easier. Following seven patches (patches 3-9) refactor code for better readability and easier reuse of code fragments that will be useful for online tuning support. Patch 10 introduces DAMON callback based user request handling structure for DAMON sysfs interface, and patch 11 enables DAMON online tuning via DAMON sysfs interface. Documentation patch (patch 12) for usage of it follows. Patch 13 enables online tuning of DAMON_RECLAIM and finally patch 14 documents the DAMON_RECLAIM online tuning usage. This patch (of 14): For updating input parameters for running DAMON contexts, DAMON kernel API users can use the contexts' callbacks, as it is the safe place for context internal data accesses. When the context has DAMON-based operation schemes and all schemes are deactivated due to their watermarks, however, DAMON does nothing but only watermarks checks. As a result, no callbacks will be called back, and therefore the kernel API users cannot update the input parameters including monitoring attributes, DAMON-based operation schemes, and watermarks. To let users easily update such DAMON input parameters in such a case, this commit adds a new callback, 'after_wmarks_check()'. It will be called after each watermarks check. Users can do the online input parameters update in the callback even under the schemes deactivated case. Link: https://lkml.kernel.org/r/20220429160606.127307-2-sj@kernel.orgSigned-off-by:
-
Niels Dossche authored
Add a regression test that validates that mremap fails for vma's that don't exist. Link: https://lkml.kernel.org/r/20220427224439.23828-3-dossche.niels@gmail.comSigned-off-by:
Niels Dossche <dossche.niels@gmail.com> Cc: Mina Almasry <almasrymina@google.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Signed-off-by:
-
Adrian Huang authored
Fix spelling/grammar mistakes in comments. Link: https://lkml.kernel.org/r/20220428061522.666-1-adrianhuang0701@gmail.comSigned-off-by:
Adrian Huang <ahuang12@lenovo.com> Signed-off-by:
-
Hongchen Zhang authored
A pmd migration entry should first be a swap pmd,so use is_swap_pmd(pmd) instead of !pmd_present(pmd). On the other hand, some architecture (MIPS for example) may misjudge a pmd_none entry as a pmd migration entry. Link: https://lkml.kernel.org/r/1651131333-6386-1-git-send-email-zhanghongchen@loongson.cnSigned-off-by:
Hongchen Zhang <zhanghongchen@loongson.cn> Acked-by:
Peter Xu <peterx@redhat.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by:
-
Florian Rommel authored
Commit c1e8d7c6 ("mmap locking API: convert mmap_sem comments") missed replacing some references of mmap_sem by mmap_lock due to misspelling (mm_sem instead of mmap_sem). Link: https://lkml.kernel.org/r/20220503113333.214124-1-mail@florommel.deSigned-off-by:
Florian Rommel <mail@florommel.de> Signed-off-by:
-
Peter Collingbourne authored
When CONFIG_KASAN_HW_TAGS is enabled we currently increase the minimum slab alignment to 16. This happens even if MTE is not supported in hardware or disabled via kasan=off, which creates an unnecessary memory overhead in those cases. Eliminate this overhead by making the minimum slab alignment a runtime property and only aligning to 16 if KASAN is enabled at runtime. On a DragonBoard 845c (non-MTE hardware) with a kernel built with CONFIG_KASAN_HW_TAGS, waiting for quiescence after a full Android boot I see the following Slab measurements in /proc/meminfo (median of 3 reboots): Before: 169020 kB After: 167304 kB [akpm@linux-foundation.org: make slab alignment type `unsigned int' to avoid casting] Link: https://linux-review.googlesource.com/id/I752e725179b43b144153f4b6f584ceb646473ead Link: https://lkml.kernel.org/r/20220427195820.1716975-2-pcc@google.comSigned-off-by:
Peter Collingbourne <pcc@google.com> Reviewed-by:
Andrey Konovalov <andreyknvl@gmail.com> Reviewed-by:
Hyeonggon Yoo <42.hyeyoo@gmail.com> Tested-by:
David Rientjes <rientjes@google.com> Reviewed-by:
Catalin Marinas <catalin.marinas@arm.com> Acked-by:
Vlastimil Babka <vbabka@suse.cz> Cc: Pekka Enberg <penberg@kernel.org> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Alexander Potapenko <glider@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kees Cook <keescook@chromium.org> Signed-off-by:
-
Peter Collingbourne authored
An inclusion of cache.h in printk.h was added in 2014 in commit c28aa1f0 ("printk/cache: mark printk_once test variable __read_mostly") in order to bring in the definition of __read_mostly. The usage of __read_mostly was later removed in commit 3ec25826 ("printk: Tie printk_once / printk_deferred_once into .data.once for reset") which made the inclusion of cache.h unnecessary, so remove it. We have a small amount of code that depended on the inclusion of cache.h from printk.h; fix that code to include the appropriate header. This fixes a circular inclusion on arm64 (linux/printk.h -> linux/cache.h -> asm/cache.h -> linux/kasan-enabled.h -> linux/static_key.h -> linux/jump_label.h -> linux/bug.h -> asm/bug.h -> linux/printk.h) that would otherwise be introduced by the next patch. Build tested using {allyesconfig,defconfig} x {arm64,x86_64}. Link: https://linux-review.googlesource.com/id/I8fd51f72c9ef1f2d6afd3b2cbc875aa4792c1fba Link: https://lkml.kernel.org/r/20220427195820.1716975-1-pcc@google.comSigned-off-by:
-
Baolin Wang authored
Now we will use flush_cache_page() to flush cache for anonymous hugetlb pages when unmapping or migrating a hugetlb page mapping, but the flush_cache_page() only handles a PAGE_SIZE range on some architectures (like arm32, arc and so on), which will cause potential cache issues. Thus change to use flush_cache_range() to cover the whole size of a hugetlb page. Link: https://lkml.kernel.org/r/dc903b378d1e2d26bbbe85409ab9d009631f175c.1651056365.git.baolin.wang@linux.alibaba.comSigned-off-by:
Baolin Wang <baolin.wang@linux.alibaba.com> Reviewed-by:
Mike Kravetz <mike.kravetz@oracle.com> Cc: Mina Almasry <almasrymina@google.com> Cc: Muchun Song <songmuchun@bytedance.com> Signed-off-by:
-
Baolin Wang authored
The cache level flush will always be first when changing an existing virtual–>physical mapping to a new value, since this allows us to properly handle systems whose caches are strict and require a virtual–>physical translation to exist for a virtual address. So we should move the cache flushing before huge_pmd_unshare(). As Muchun pointed out[1], now the architectures whose supporting hugetlb PMD sharing have no cache flush issues in practice. But I think we should still follow the cache/TLB flushing rules when changing a valid virtual address mapping in case of potential issues in future. [1] https://lore.kernel.org/all/YmT%2F%2FhuUbFX+KHcy@FVFYT0MHHV2J.usts.net/ Link: https://lkml.kernel.org/r/4f7ae6dfdc838ab71e1655188b657c032ff1f28f.1651056365.git.baolin.wang@linux.alibaba.comSigned-off-by:
-
Baolin Wang authored
This patchset fixes some cache flushing issues if PMD sharing is possible for hugetlb pages, which were found by code inspection. Meanwhile Mike found the flush_cache_page() can not cover the whole size of a hugetlb page on some architectures [1], so I added a new patch 3 to fix this issue, since I found only try_to_unmap_one() and try_to_migrate_one() need to fix after some investigation. [1] https://lore.kernel.org/linux-mm/064da3bb-5b4b-7332-a722-c5a541128705@oracle.com/ This patch (of 3): When moving hugetlb page tables, the cache flushing is called in move_page_tables() without considering the shared PMDs, which may be cause cache issues on some architectures. Thus we should move the hugetlb cache flushing into move_hugetlb_page_tables() with considering the shared PMDs ranges, calculated by adjust_range_if_pmd_sharing_possible(). Meanwhile also expanding the TLBs flushing range in case of shared PMDs. Note this is discovered via code inspection, and did not meet a real problem in practice so far. Link: https://lkml.kernel.org/r/cover.1651056365.git.baolin.wang@linux.alibaba.com Link: https://lkml.kernel.org/r/0443c8cf20db554d3ff4b439b30e0ff26c0181dd.1651056365.git.baolin.wang@linux.alibaba.com Fixes: 550a7d60 ("mm, hugepages: add mremap() support for hugepage backed vma") Signed-off-by:
Muchun Song <songmuchun@bytedance.com> Cc: Mina Almasry <almasrymina@google.com> Signed-off-by:
-
liusongtang authored
pgprot.pgprot is non-portable code. It should be replaced by portable macro pgprot_val. Link: https://lkml.kernel.org/r/20220426071302.220646-1-liusongtang@huawei.comSigned-off-by:
liusongtang <liusongtang@huawei.com> Reviewed-by:
Anshuman Khandual <anshuman.khandual@arm.com> Cc: Xiaoming Ni <nixiaoming@huawei.com> Signed-off-by:
-
SeongJae Park authored
This commit documents the user space support of the newly added monitoring operations set for fixed virtual address ranges monitoring, namely 'fvaddr', on the ABI and usage documents for DAMON. Link: https://lkml.kernel.org/r/20220426231750.48822-4-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
This commit makes DAMON sysfs interface to support the fixed virtual address ranges monitoring. After this commit, writing 'fvaddr' to the 'operations' DAMON sysfs file makes DAMON uses the monitoring operations set for fixed virtual address ranges, so that users can monitor accesses to only interested virtual address ranges. [sj@kernel.org: fix pid leak under fvaddr ops use case] Link: https://lkml.kernel.org/r/20220503220531.45913-1-sj@kernel.org Link: https://lkml.kernel.org/r/20220426231750.48822-3-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
Patch series "support fixed virtual address ranges monitoring". The monitoring operations set for virtual address spaces automatically updates the monitoring target regions to cover entire mappings of the virtual address spaces as much as possible. Some users could have more information about their programs than kernel and therefore have interest in not entire regions but only specific regions. For such cases, the automatic monitoring target regions updates are only unnecessary overhead or distractions. This patchset adds supports for the use case on DAMON's kernel API (DAMON_OPS_FVADDR) and sysfs interface ('fvaddr' keyword for 'operations' sysfs file). This patch (of 3): The monitoring operations set for virtual address spaces automatically updates the monitoring target regions to cover entire mappings of the virtual address spaces as much as possible. Some users could have more information about their programs than kernel and therefore have interest in not entire regions but only specific regions. For such cases, the automatic monitoring target regions updates are only unnecessary overheads or distractions. For such cases, DAMON's API users can simply set the '->init()' and '->update()' of the DAMON context's '->ops' NULL, and set the target monitoring regions when creating the context. But, that would be a dirty hack. Worse yet, the hack is unavailable for DAMON user space interface users. To support the use case in a clean way that can easily exported to the user space, this commit adds another monitoring operations set called 'fvaddr', which is same to 'vaddr' but does not automatically update the monitoring regions. Instead, it will only respect the virtual address regions which have explicitly passed at the initial context creation. Note that this commit leave sysfs interface not supporting the feature yet. The support will be made in a following commit. Link: https://lkml.kernel.org/r/20220426231750.48822-1-sj@kernel.org Link: https://lkml.kernel.org/r/20220426231750.48822-2-sj@kernel.orgSigned-off-by: -
SeongJae Park authored
This commit updates the DAMON ABI and usage documents for the new sysfs file, 'avail_operations'. Link: https://lkml.kernel.org/r/20220426203843.45238-5-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
This commit adds a selftest test case for ensuring the existence and the permission (read-only) of the 'avail_oprations' DAMON sysfs file. Link: https://lkml.kernel.org/r/20220426203843.45238-4-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
DAMON programming interface users can know if specific monitoring ops set is registered or not using 'damon_is_registered_ops()', but there is no such method for the user space. To help the case, this commit adds a new DAMON sysfs file called 'avail_operations' under each context directory for listing available monitoring ops. Reading the file will list each registered monitoring ops on each line. Link: https://lkml.kernel.org/r/20220426203843.45238-3-sj@kernel.orgSigned-off-by:
-
SeongJae Park authored
Patch series "mm/damon: allow users know which monitoring ops are available". DAMON users can configure it for vaious address spaces including virtual address spaces and the physical address space by setting its monitoring operations set with appropriate one for their purpose. However, there is no celan and simple way to know exactly which monitoring operations sets are available on the currently running kernel. This patchset adds functions for the purpose on DAMON's kernel API ('damon_is_registered_ops()') and sysfs interface ('avail_operations' file under each context directory). This patch (of 4): To know if a specific 'damon_operations' is registered, users need to check the kernel config or try 'damon_select_ops()' with the ops of the question, and then see if it successes. In the latter case, the user should also revert the change. To make the process simple and convenient, this commit adds a function for checking if a specific 'damon_operations' is registered or not. Link: https://lkml.kernel.org/r/20220426203843.45238-1-sj@kernel.org Link: https://lkml.kernel.org/r/20220426203843.45238-2-sj@kernel.orgSigned-off-by: -
Fabio M. De Francesco authored
Add VM_BUG_ON() bounds checking to make sure that, if "offset + len> PAGE_SIZE", memset() does not corrupt data in adjacent pages. Mainly to match all the similar functions in highmem.h. Link: https://lkml.kernel.org/r/20220426193020.8710-1-fmdefrancesco@gmail.comSigned-off-by:
Fabio M. De Francesco <fmdefrancesco@gmail.com> Reviewed-by:
-
huangshaobo authored
Out-of-bounds accesses that aren't caught by a guard page will result in corruption of canary memory. In pathological cases, where an object has certain alignment requirements, an out-of-bounds access might never be caught by the guard page. Such corruptions, however, are only detected on kfree() normally. If the bug causes the kernel to panic before kfree(), KFENCE has no opportunity to report the issue. Such corruptions may also indicate failing memory or other faults. To provide some more information in such cases, add the option to check canary bytes on panic. This might help narrow the search for the panic cause; but, due to only having the allocation stack trace, such reports are difficult to use to diagnose an issue alone. In most cases, such reports are inactionable, and is therefore an opt-in feature (disabled by default). [akpm@linux-foundation.org: add __read_mostly, per Marco] Link: https://lkml.kernel.org/r/20220425022456.44300-1-huangshaobo6@huawei.comSigned-off-by:
huangshaobo <huangshaobo6@huawei.com> Suggested-by:
chenzefeng <chenzefeng2@huawei.com> Reviewed-by:
Marco Elver <elver@google.com> Cc: Alexander Potapenko <glider@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Xiaoming Ni <nixiaoming@huawei.com> Cc: Wangbing <wangbing6@huawei.com> Cc: Jubin Zhong <zhongjubin@huawei.com> Signed-off-by:
-
Mina Almasry authored
After commit db71ef79 ("hugetlb: make free_huge_page irq safe"), the subpool lock should be locked with spin_lock_irq() and all call sites was modified as such, except for the ones in hugetlbfs_statfs(). Link: https://lkml.kernel.org/r/20220429202207.3045-1-almasrymina@google.com Fixes: db71ef79 ("hugetlb: make free_huge_page irq safe") Signed-off-by:
Mina Almasry <almasrymina@google.com> Reviewed-by:
-
Nadav Amit authored
Calls to change_protection_range() on THP can trigger, at least on x86, two TLB flushes for one page: one immediately, when pmdp_invalidate() is called by change_huge_pmd(), and then another one later (that can be batched) when change_protection_range() finishes. The first TLB flush is only necessary to prevent the dirty bit (and with a lesser importance the access bit) from changing while the PTE is modified. However, this is not necessary as the x86 CPUs set the dirty-bit atomically with an additional check that the PTE is (still) present. One caveat is Intel's Knights Landing that has a bug and does not do so. Leverage this behavior to eliminate the unnecessary TLB flush in change_huge_pmd(). Introduce a new arch specific pmdp_invalidate_ad() that only invalidates the access and dirty bit from further changes. Link: https://lkml.kernel.org/r/20220401180821.1986781-4-namit@vmware.comSigned-off-by:
Nadav Amit <namit@vmware.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Andrew Cooper <andrew.cooper3@citrix.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Nick Piggin <npiggin@gmail.com> Signed-off-by:
-
Nadav Amit authored
Currently, using mprotect() to unprotect a memory region or uffd to unprotect a memory region causes a TLB flush. However, in such cases the PTE is often not modified (i.e., remain RO) and therefore not TLB flush is needed. Add an arch-specific pte_needs_flush() which tells whether a TLB flush is needed based on the old PTE and the new one. Implement an x86 pte_needs_flush(). Always flush the TLB when it is architecturally needed even when skipping a TLB flush might only result in a spurious page-faults by skipping the flush. Even with such conservative manner, we can in the future further refine the checks to test whether a PTE is present by only considering the architectural _PAGE_PRESENT flag instead of {pte|pmd}_preesnt(). For not be careful and use the latter. Link: https://lkml.kernel.org/r/20220401180821.1986781-3-namit@vmware.comSigned-off-by: -
Nadav Amit authored
Patch series "mm/mprotect: avoid unnecessary TLB flushes", v6. This patchset is intended to remove unnecessary TLB flushes during mprotect() syscalls. Once this patch-set make it through, similar and further optimizations for MADV_COLD and userfaultfd would be possible. Basically, there are 3 optimizations in this patch-set: 1. Use TLB batching infrastructure to batch flushes across VMAs and do better/fewer flushes. This would also be handy for later userfaultfd enhancements. 2. Avoid unnecessary TLB flushes. This optimization is the one that provides most of the performance benefits. Unlike previous versions, we now only avoid flushes that would not result in spurious page-faults. 3. Avoiding TLB flushes on change_huge_pmd() that are only needed to prevent the A/D bits from changing. Andrew asked for some benchmark numbers. I do not have an easy determinate macrobenchmark in which it is easy to show benefit. I therefore ran a microbenchmark: a loop that does the following on anonymous memory, just as a sanity check to see that time is saved by avoiding TLB flushes. The loop goes: mprotect(p, PAGE_SIZE, PROT_READ) mprotect(p, PAGE_SIZE, PROT_READ|PROT_WRITE) *p = 0; // make the page writable The test was run in KVM guest with 1 or 2 threads (the second thread was busy-looping). I measured the time (cycles) of each operation: 1 thread 2 threads mmots +patch mmots +patch PROT_READ 3494 2725 (-22%) 8630 7788 (-10%) PROT_READ|WRITE 3952 2724 (-31%) 9075 2865 (-68%) [ mmots = v5.17-rc6-mmots-2022-03-06-20-38 ] The exact numbers are really meaningless, but the benefit is clear. There are 2 interesting results though. (1) PROT_READ is cheaper, while one can expect it not to be affected. This is presumably due to TLB miss that is saved (2) Without memory access (*p = 0), the speedup of the patch is even greater. In that scenario mprotect(PROT_READ) also avoids the TLB flush. As a result both operations on the patched kernel take roughly ~1500 cycles (with either 1 or 2 threads), whereas on mmotm their cost is as high as presented in the table. This patch (of 3): change_pXX_range() currently does not use mmu_gather, but instead implements its own deferred TLB flushes scheme. This both complicates the code, as developers need to be aware of different invalidation schemes, and prevents opportunities to avoid TLB flushes or perform them in finer granularity. The use of mmu_gather for modified PTEs has benefits in various scenarios even if pages are not released. For instance, if only a single page needs to be flushed out of a range of many pages, only that page would be flushed. If a THP page is flushed, on x86 a single TLB invlpg instruction can be used instead of 512 instructions (or a full TLB flush, which would Linux would actually use by default). mprotect() over multiple VMAs requires a single flush. Use mmu_gather in change_pXX_range(). As the pages are not released, only record the flushed range using tlb_flush_pXX_range(). Handle THP similarly and get rid of flush_cache_range() which becomes redundant since tlb_start_vma() calls it when needed. Link: https://lkml.kernel.org/r/20220401180821.1986781-1-namit@vmware.com Link: https://lkml.kernel.org/r/20220401180821.1986781-2-namit@vmware.comSigned-off-by:
Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Andrew Cooper <andrew.cooper3@citrix.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Peter Xu <peterx@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Nick Piggin <npiggin@gmail.com> Signed-off-by:
-
- 10 May, 2022 6 commits
-
-
NeilBrown authored
Currently various places test if direct IO is possible on a file by checking for the existence of the direct_IO address space operation. This is a poor choice, as the direct_IO operation may not be used - it is only used if the generic_file_*_iter functions are called for direct IO and some filesystems - particularly NFS - don't do this. Instead, introduce a new f_mode flag: FMODE_CAN_ODIRECT and change the various places to check this (avoiding pointer dereferences). do_dentry_open() will set this flag if ->direct_IO is present, so filesystems do not need to be changed. NFS *is* changed, to set the flag explicitly and discard the direct_IO entry in the address_space_operations for files. Other filesystems which currently use noop_direct_IO could usefully be changed to set this flag instead. Link: https://lkml.kernel.org/r/164859778128.29473.15189737957277399416.stgit@noble.brownReviewed-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
NeilBrown <neilb@suse.de> Tested-by:
David Howells <dhowells@redhat.com> Tested-by:
Geert Uytterhoeven <geert+renesas@glider.be> Cc: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Trond Myklebust <trond.myklebust@hammerspace.com> Cc: Miaohe Lin <linmiaohe@huawei.com> Signed-off-by:
-
NeilBrown authored
We need to use count_swpout_vm_event() for sio_write_complete() to get correct counting. Note that THP swap in (if it ever happens) is current accounted 1 for each page, whether HUGE or normal. This is different from swap-out accounting. This patch should be squashed into MM: handle THP in swap_*page_fs() Link: https://lkml.kernel.org/r/165146948934.24404.5909750610552745025@noble.neil.brown.nameSigned-off-by:Miaohe Lin <linmiaohe@huawei.com> Reviewed-by:
-
NeilBrown authored
Pages passed to swap_readpage()/swap_writepage() are not necessarily all the same size - there may be transparent-huge-pages involves. The BIO paths of swap_*page() handle this correctly, but the SWP_FS_OPS path does not. So we need to use thp_size() to find the size, not just assume PAGE_SIZE, and we need to track the total length of the request, not just assume it is "page * PAGE_SIZE". Link: https://lkml.kernel.org/r/165119301488.15698.9457662928942765453.stgit@noble.brownSigned-off-by:
-
NeilBrown authored
swap_writepage() is given one page at a time, but may be called repeatedly in succession. For block-device swapspace, the blk_plug functionality allows the multiple pages to be combined together at lower layers. That cannot be used for SWP_FS_OPS as blk_plug may not exist - it is only active when CONFIG_BLOCK=y. Consequently all swap reads over NFS are single page reads. With this patch we pass a pointer-to-pointer via the wbc. swap_writepage can store state between calls - much like the pointer passed explicitly to swap_readpage. After calling swap_writepage() some number of times, the state will be passed to swap_write_unplug() which can submit the combined request. Link: https://lkml.kernel.org/r/164859778128.29473.5191868522654408537.stgit@noble.brownSigned-off-by:
-
NeilBrown authored
swap_readpage() is given one page at a time, but may be called repeatedly in succession. For block-device swap-space, the blk_plug functionality allows the multiple pages to be combined together at lower layers. That cannot be used for SWP_FS_OPS as blk_plug may not exist - it is only active when CONFIG_BLOCK=y. Consequently all swap reads over NFS are single page reads. With this patch we pass in a pointer-to-pointer when swap_readpage can store state between calls - much like the effect of blk_plug. After calling swap_readpage() some number of times, the state will be passed to swap_read_unplug() which can submit the combined request. Link: https://lkml.kernel.org/r/164859778127.29473.14059420492644907783.stgit@noble.brownSigned-off-by:
-
NeilBrown authored
This documentation for ->swap_activate() has been out-of-date for a long time. This patch updates it to match recent changes, and adds documentation for the associated ->swap_rw() Link: https://lkml.kernel.org/r/164859778126.29473.6778751233552859461.stgit@noble.brownSigned-off-by:
-
