From: Yang Erkun yangerkun@huawei.com

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9DN5Z CVE: NA

--------------------------------

iomap_file_buffered_write() will always call iomap_iter to find the iomap for buffer io, and then call iomap_write_iter() to get and fill the folio, which is too complicated compare to generic_perform_write(). Unixbench test for ext4 show this with a performance degradation since we switch buffer io for ext4 to iomap.

Fix it by support .write_begin() and .write_end(), and for the case write range has already been mark as dirty, no need to get the iomap.

Signed-off-by: Yang Erkun yangerkun@huawei.com --- fs/ext4/file.c | 3 +- fs/ext4/inode.c | 124 ++++++++++++++++++++++++++++++++++++ include/trace/events/ext4.h | 15 +++++ 3 files changed, 141 insertions(+), 1 deletion(-)

diff --git a/fs/ext4/file.c b/fs/ext4/file.c index 67b0e2212ca0..27d4eff79941 100644 --- a/fs/ext4/file.c +++ b/fs/ext4/file.c @@ -314,7 +314,8 @@ static ssize_t ext4_buffered_write_iter(struct kiocb *iocb, if (ret <= 0) goto out;

- if (ext4_test_inode_state(inode, EXT4_STATE_BUFFERED_IOMAP)) + if (ext4_test_inode_state(inode, EXT4_STATE_BUFFERED_IOMAP) && + iov_iter_count(from) > PAGE_SIZE) ret = ext4_iomap_buffered_write(iocb, from); else ret = generic_perform_write(iocb, from); diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c index 93a9dd03cb5c..2c0e61f531f1 100644 --- a/fs/ext4/inode.c +++ b/fs/ext4/inode.c @@ -3955,6 +3955,128 @@ static int ext4_iomap_writepages(struct address_space *mapping, return ret; }

+static int ext4_iomap_write_begin(struct file *file, + struct address_space *mapping, loff_t pos, + unsigned len, struct page **pagep, + void **fsdata) +{ + struct inode *inode = mapping->host; + struct iomap_iter iter = { + .inode = inode, + .flags = IOMAP_WRITE, + }; + int ret = 0, retries = 0; + struct folio *folio; + bool delalloc; + + if (unlikely(ext4_forced_shutdown(inode->i_sb))) + return -EIO; + + trace_ext4_iomap_write_begin(inode, pos, len); + + delalloc = test_opt(inode->i_sb, DELALLOC) && + !ext4_nonda_switch(inode->i_sb); + *fsdata = delalloc ? (void *)0 : (void *)FALL_BACK_TO_NONDELALLOC; + +retry: + iter.pos = pos; + iter.len = len; + + folio = iomap_get_folio(&iter, pos, len); + if (IS_ERR(folio)) + return PTR_ERR(folio); + + WARN_ON_ONCE(pos + len > folio_pos(folio) + folio_size(folio)); + + if (folio_test_dirty(folio) && (i_blocks_per_folio(inode, folio) == 1)) + goto out; + + do { + int length; + + ret = __ext4_iomap_buffered_io_begin(inode, iter.pos, iter.len, + iter.flags, &iter.iomap, NULL, delalloc); + if (ret) + goto out; + + WARN_ON_ONCE(iter.iomap.offset > iter.pos); + WARN_ON_ONCE(iter.iomap.length == 0); + WARN_ON_ONCE(iter.iomap.offset + iter.iomap.length <= iter.pos); + + length = iomap_length(&iter); + ret = __iomap_write_begin(&iter, iter.pos, length, folio); + if (ret) + goto out; + + iter.pos += length; + iter.len -= length; + } while (iter.len); + +out: + if (ret < 0) { + folio_unlock(folio); + folio_put(folio); + + /* + * __ext4_iomap_buffered_io_begin() may have instantiated + * a few blocks outside i_size. Trim these off again. Don't + * need i_size_read because we hold inode lock. + */ + if (pos + len > inode->i_size) + ext4_truncate_failed_write(inode); + + if (ret == -ENOSPC && + ext4_should_retry_alloc(inode->i_sb, &retries)) + goto retry; + } + + *pagep = folio_file_page(folio, pos >> PAGE_SHIFT); + return ret; +} + +static int ext4_iomap_write_end(struct file *file, + struct address_space *mapping, + loff_t pos, unsigned len, unsigned copied, + struct page *page, void *fsdata) +{ + struct inode *inode = mapping->host; + int write_mode = (int)(unsigned long)fsdata; + struct folio *folio = page_folio(page); + loff_t old_size = inode->i_size; + size_t written; + + trace_ext4_iomap_write_end(inode, pos, len, copied); + + written = __iomap_write_end(inode, pos, len, copied, folio) ? + copied : 0; + + /* + * Update the in-memory inode size after copying the data into + * the page cache. It's important to update i_size while still + * holding folio lock, because folio writeout could otherwise + * come in and zero beyond i_size. + */ + if (pos + written > old_size) + i_size_write(inode, pos + written); + + folio_unlock(folio); + folio_put(folio); + + if (old_size < pos) + pagecache_isize_extended(inode, old_size, pos); + + /* + * For delalloc, if we have pre-allocated more blocks and copied + * less, we will have delalloc extents allocated outside i_size, + * drop pre-allocated blocks that were not used, prevent the + * write back path from allocating blocks for them. + */ + if (unlikely(!written) && write_mode != FALL_BACK_TO_NONDELALLOC) + ext4_truncate_failed_write(inode); + + return written; +} + /* * For data=journal mode, folio should be marked dirty only when it was * writeably mapped. When that happens, it was already attached to the @@ -4048,6 +4170,8 @@ static const struct address_space_operations ext4_iomap_aops = { .read_folio = ext4_iomap_read_folio, .readahead = ext4_iomap_readahead, .writepages = ext4_iomap_writepages, + .write_begin = ext4_iomap_write_begin, + .write_end = ext4_iomap_write_end, .dirty_folio = iomap_dirty_folio, .bmap = ext4_bmap, .invalidate_folio = iomap_invalidate_folio, diff --git a/include/trace/events/ext4.h b/include/trace/events/ext4.h index 588991b57c12..d500568daeb1 100644 --- a/include/trace/events/ext4.h +++ b/include/trace/events/ext4.h @@ -389,6 +389,13 @@ DEFINE_EVENT(ext4__write_begin, ext4_da_write_begin, TP_ARGS(inode, pos, len) );

+DEFINE_EVENT(ext4__write_begin, ext4_iomap_write_begin, + + TP_PROTO(struct inode *inode, loff_t pos, unsigned int len), + + TP_ARGS(inode, pos, len) +); + DECLARE_EVENT_CLASS(ext4__write_end, TP_PROTO(struct inode *inode, loff_t pos, unsigned int len, unsigned int copied), @@ -441,6 +448,14 @@ DEFINE_EVENT(ext4__write_end, ext4_da_write_end, TP_ARGS(inode, pos, len, copied) );

+DEFINE_EVENT(ext4__write_end, ext4_iomap_write_end, + + TP_PROTO(struct inode *inode, loff_t pos, unsigned int len, + unsigned int copied), + + TP_ARGS(inode, pos, len, copied) +); + TRACE_EVENT(ext4_writepages, TP_PROTO(struct inode *inode, struct writeback_control *wbc),

From: Ryan Roberts ryan.roberts@arm.com

maillist inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I98AW9 CVE: NA

Reference: https://lore.kernel.org/linux-mm/Zc6zD4zfsrkRlHe6@casper.infradead.org/ -------------------------------------------------

Change the readahead config so that if it is being requested for an executable mapping, do a synchronous read of an arch-specified size in a naturally aligned manner.

On arm64 if memory is physically contiguous and naturally aligned to the "contpte" size, we can use contpte mappings, which improves utilization of the TLB. When paired with the "multi-size THP" changes, this works well to reduce dTLB pressure. However iTLB pressure is still high due to executable mappings having a low liklihood of being in the required folio size and mapping alignment, even when the filesystem supports readahead into large folios (e.g. XFS).

The reason for the low liklihood is that the current readahead algorithm starts with an order-2 folio and increases the folio order by 2 every time the readahead mark is hit. But most executable memory is faulted in fairly randomly and so the readahead mark is rarely hit and most executable folios remain order-2. This is observed impirically and confirmed from discussion with a gnu linker expert; in general, the linker does nothing to group temporally accessed text together spacially. Additionally, with the current read-around approach there are no alignment guarrantees between the file and folio. This is insufficient for arm64's contpte mapping requirement (order-4 for 4K base pages).

So it seems reasonable to special-case the read(ahead) logic for executable mappings. The trade-off is performance improvement (due to more efficient storage of the translations in iTLB) vs potential read amplification (due to reading too much data around the fault which won't be used), and the latter is independent of base page size. I've chosen 64K folio size for arm64 which benefits both the 4K and 16K base page size configs and shouldn't lead to any further read-amplification since the old read-around path was (usually) reading blocks of 128K (with the last 32K being async).

Performance Benchmarking ------------------------

The below shows kernel compilation and speedometer javascript benchmarks on Ampere Altra arm64 system. (The contpte patch series is applied in the baseline).

First, confirmation that this patch causes more memory to be contained in 64K folios (this is for all file-backed memory so includes non-executable too):

| File-backed folios | Speedometer | Kernel Compile | | by size as percentage |-----------------|-----------------| | of all mapped file mem | before | after | before | after | |=========================|========|========|========|========| |file-thp-aligned-16kB | 45% | 9% | 46% | 7% | |file-thp-aligned-32kB | 2% | 0% | 3% | 1% | |file-thp-aligned-64kB | 3% | 63% | 5% | 80% | |file-thp-aligned-128kB | 11% | 11% | 0% | 0% | |file-thp-unaligned-16kB | 1% | 0% | 3% | 1% | |file-thp-unaligned-128kB | 1% | 0% | 0% | 0% | |file-thp-partial | 0% | 0% | 0% | 0% | |-------------------------|--------|--------|--------|--------| |file-cont-aligned-64kB | 16% | 75% | 5% | 80% |

The above shows that for both use cases, the amount of file memory backed by 16K folios reduces and the amount backed by 64K folios increases significantly. And the amount of memory that is contpte-mapped significantly increases (last line).

And this is reflected in performance improvement:

Kernel Compilation (smaller is faster): | kernel | real-time | kern-time | user-time | peak memory | |----------|-------------|-------------|-------------|---------------| | before | 0.0% | 0.0% | 0.0% | 0.0% | | after | -1.6% | -2.1% | -1.7% | 0.0% |

Speedometer (bigger is faster): | kernel | runs_per_min | peak memory | |----------|----------------|---------------| | before | 0.0% | 0.0% | | after | 1.3% | 1.0% |

Both benchmarks show a ~1.5% improvement once the patch is applied.

Alternatives ------------

I considered (and rejected for now - but I anticipate this patch will stimulate discussion around what the best approach is) alternative approaches:

- Expose a global user-controlled knob to set the preferred folio size; this would move policy to user space and allow (e.g.) setting it to PMD-size for even better iTLB utilizaiton. But this would add ABI, and I prefer to start with the simplest approach first. It also has the downside that a change wouldn't apply to memory already in the page cache that is in active use (e.g. libc) so we don't get the same level of utilization as for something that is fixed from boot.

- Add a per-vma attribute to allow user space to specify preferred folio size for memory faulted from the range. (we've talked about such a control in the context of mTHP). The dynamic loader would then be responsible for adding the annotations. Again this feels like something that could be added later if value was demonstrated.

- Enhance MADV_COLLAPSE to collapse to THP sizes less than PMD-size. This would still require dynamic linker involvement, but would additionally neccessitate a copy and all memory in the range would be synchronously faulted in, adding to application load time. It would work for filesystems that don't support large folios though.

Signed-off-by: Ryan Roberts ryan.roberts@arm.com --- arch/arm64/include/asm/pgtable.h | 12 ++++++++++++ include/linux/pgtable.h | 12 ++++++++++++ mm/filemap.c | 19 +++++++++++++++++++ 3 files changed, 43 insertions(+)

diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h index 07948fe59b9d..8d68d00de0a4 100644 --- a/arch/arm64/include/asm/pgtable.h +++ b/arch/arm64/include/asm/pgtable.h @@ -1147,6 +1147,18 @@ static inline void update_mmu_cache_range(struct vm_fault *vmf, */ #define arch_wants_old_prefaulted_pte cpu_has_hw_af

+/* + * Request exec memory is read into pagecache in at least 64K folios. The + * trade-off here is performance improvement due to storing translations more + * effciently in the iTLB vs the potential for read amplification due to reading + * data from disk that won't be used. The latter is independent of base page + * size, so we set a page-size independent block size of 64K. This size can be + * contpte-mapped when 4K base pages are in use (16 pages into 1 iTLB entry), + * and HPA can coalesce it (4 pages into 1 TLB entry) when 16K base pages are in + * use. + */ +#define arch_wants_exec_folio_order() ilog2(SZ_64K >> PAGE_SHIFT) + static inline bool pud_sect_supported(void) { return PAGE_SIZE == SZ_4K; diff --git a/include/linux/pgtable.h b/include/linux/pgtable.h index ecc561d49d5b..a0fafb8e7005 100644 --- a/include/linux/pgtable.h +++ b/include/linux/pgtable.h @@ -435,6 +435,18 @@ static inline bool arch_has_hw_pte_young(void) } #endif

+#ifndef arch_wants_exec_folio_order +/* + * Returns preferred minimum folio order for executable file-backed memory. Must + * be in range [0, PMD_ORDER]. Negative value implies that the HW has no + * preference and mm will not special-case executable memory in the pagecache. + */ +static inline int arch_wants_exec_folio_order(void) +{ + return -1; +} +#endif + #ifndef arch_check_zapped_pte static inline void arch_check_zapped_pte(struct vm_area_struct *vma, pte_t pte) diff --git a/mm/filemap.c b/mm/filemap.c index a274d2c5e232..b7881f1db472 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -3197,6 +3197,25 @@ static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) } #endif

+ /* + * Allow arch to request a preferred minimum folio order for executable + * memory. This can often be beneficial to performance if (e.g.) arm64 + * can contpte-map the folio. Executable memory rarely benefits from + * read-ahead anyway, due to its random access nature. + */ + if (vm_flags & VM_EXEC) { + int order = arch_wants_exec_folio_order(); + + if (order >= 0) { + fpin = maybe_unlock_mmap_for_io(vmf, fpin); + ra->size = 1UL << order; + ra->async_size = 0; + ractl._index &= ~((unsigned long)ra->size - 1); + page_cache_ra_order(&ractl, ra, order); + return fpin; + } + } + /* If we don't want any read-ahead, don't bother */ if (vm_flags & VM_RAND_READ) return fpin;

From: Kefeng Wang wangkefeng.wang@huawei.com

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I98AW9 CVE: NA

-------------------------------------------------

Let's document the new thp_exec_enabled sysfs interfaces for transhuge.

Signed-off-by: Kefeng Wang wangkefeng.wang@huawei.com --- Documentation/admin-guide/mm/transhuge.rst | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst index 936da10c5260..22f0e0009371 100644 --- a/Documentation/admin-guide/mm/transhuge.rst +++ b/Documentation/admin-guide/mm/transhuge.rst @@ -203,11 +203,13 @@ PMD-mappable transparent hugepage:: cat /sys/kernel/mm/transparent_hugepage/hpage_pmd_size

The kernel tries to use huge, PMD-mappable page on read page fault for -file exec mapping if CONFIG_READ_ONLY_THP_FOR_FS enabled. It's possible -to enabled the feature by writing 1 or disablt by writing 0:: +if CONFIG_READ_ONLY_THP_FOR_FS enabled, or try non-PMD size page(eg, +64K arm64) for file exec mapping, BIT0 for PMD THP, BIT1 for mTHP. It's +possible to enable/disable it by configurate the corresponding bit::

- echo 0x0 >/sys/kernel/mm/transparent_hugepage/thp_exec_enabled echo 0x1 >/sys/kernel/mm/transparent_hugepage/thp_exec_enabled + echo 0x2 >/sys/kernel/mm/transparent_hugepage/thp_exec_enabled + echo 0x3 >/sys/kernel/mm/transparent_hugepage/thp_exec_enabled

khugepaged will be automatically started when one or more hugepage sizes are enabled (either by directly setting "always" or "madvise",

From: Kefeng Wang wangkefeng.wang@huawei.com

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I98AW9 CVE: NA

-------------------------------------------------

Since folio_get_unmapped_area() is experimental, add a sysfs interface to enable/disable it, it is disabled by default, and could be enabled by /sys/kernel/mm/transparent_hugepage/thp_mapping_align.

Signed-off-by: Kefeng Wang wangkefeng.wang@huawei.com --- include/linux/huge_mm.h | 2 ++ mm/huge_memory.c | 67 +++++++++++++++++++++++++++++++++++++---- 2 files changed, 63 insertions(+), 6 deletions(-)

diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 896d7870ecd9..8fdf17e80359 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -52,6 +52,8 @@ enum transparent_hugepage_flag { TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG, TRANSPARENT_HUGEPAGE_FILE_EXEC_THP_FLAG, TRANSPARENT_HUGEPAGE_FILE_EXEC_MTHP_FLAG, + TRANSPARENT_HUGEPAGE_FILE_MAPPING_ALIGN_FLAG, + TRANSPARENT_HUGEPAGE_ANON_MAPPING_ALIGN_FLAG, };

struct kobject; diff --git a/mm/huge_memory.c b/mm/huge_memory.c index c1e27a1b3d74..8cb3e014a881 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -434,8 +434,7 @@ static struct kobj_attribute hpage_pmd_size_attr = #define FILE_EXEC_MTHP_ENABLE BIT(1) #define FILE_EXEC_THP_ALL (FILE_EXEC_THP_ENABLE | FILE_EXEC_MTHP_ENABLE)

-static void thp_exec_enabled_set(enum transparent_hugepage_flag flag, - bool enable) +static void thp_flag_set(enum transparent_hugepage_flag flag, bool enable) { if (enable) set_bit(flag, &transparent_hugepage_flags); @@ -473,20 +472,61 @@ static ssize_t thp_exec_enabled_store(struct kobject *kobj, return -EINVAL;

#ifdef CONFIG_READ_ONLY_THP_FOR_FS - thp_exec_enabled_set(TRANSPARENT_HUGEPAGE_FILE_EXEC_THP_FLAG, - val & FILE_EXEC_THP_ENABLE); + thp_flag_set(TRANSPARENT_HUGEPAGE_FILE_EXEC_THP_FLAG, + val & FILE_EXEC_THP_ENABLE); ret = start_stop_khugepaged(); if (ret) return ret; #endif - thp_exec_enabled_set(TRANSPARENT_HUGEPAGE_FILE_EXEC_MTHP_FLAG, - val & FILE_EXEC_MTHP_ENABLE); + thp_flag_set(TRANSPARENT_HUGEPAGE_FILE_EXEC_MTHP_FLAG, + val & FILE_EXEC_MTHP_ENABLE);

return count; } static struct kobj_attribute thp_exec_enabled_attr = __ATTR_RW(thp_exec_enabled);

+#define FILE_MAPPING_ALIGN BIT(0) +#define ANON_MAPPING_ALIGN BIT(1) +#define THP_MAPPING_ALIGN_ALL (FILE_MAPPING_ALIGN | ANON_MAPPING_ALIGN) + +static ssize_t thp_mapping_align_show(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + unsigned long val = 0; + + if (test_bit(TRANSPARENT_HUGEPAGE_FILE_MAPPING_ALIGN_FLAG, + &transparent_hugepage_flags)) + val |= FILE_MAPPING_ALIGN; + + if (test_bit(TRANSPARENT_HUGEPAGE_ANON_MAPPING_ALIGN_FLAG, + &transparent_hugepage_flags)) + val |= ANON_MAPPING_ALIGN; + + return sysfs_emit(buf, "0x%lx\n", val); +} +static ssize_t thp_mapping_align_store(struct kobject *kobj, + struct kobj_attribute *attr, const char *buf, size_t count) +{ + unsigned long val; + int ret; + + ret = kstrtoul(buf, 16, &val); + if (ret < 0) + return ret; + if (val & ~THP_MAPPING_ALIGN_ALL) + return -EINVAL; + + thp_flag_set(TRANSPARENT_HUGEPAGE_FILE_MAPPING_ALIGN_FLAG, + val & FILE_MAPPING_ALIGN); + thp_flag_set(TRANSPARENT_HUGEPAGE_ANON_MAPPING_ALIGN_FLAG, + val & ANON_MAPPING_ALIGN); + + return count; +} +static struct kobj_attribute thp_mapping_align_attr = + __ATTR_RW(thp_mapping_align); + static struct attribute *hugepage_attr[] = { &enabled_attr.attr, &defrag_attr.attr, @@ -496,6 +536,7 @@ static struct attribute *hugepage_attr[] = { &shmem_enabled_attr.attr, #endif &thp_exec_enabled_attr.attr, + &thp_mapping_align_attr.attr, NULL, };

@@ -887,10 +928,21 @@ static unsigned long __thp_get_unmapped_area(struct file *filp, return ret; }

+#define transparent_hugepage_file_mapping_align_enabled() \ + (transparent_hugepage_flags & \ + (1<<TRANSPARENT_HUGEPAGE_FILE_MAPPING_ALIGN_FLAG)) + +#define transparent_hugepage_anon_mapping_align_enabled() \ + (transparent_hugepage_flags & \ + (1<<TRANSPARENT_HUGEPAGE_ANON_MAPPING_ALIGN_FLAG)) + static bool file_mapping_align_enabled(struct file *filp) { struct address_space *mapping;

+ if (!transparent_hugepage_file_mapping_align_enabled()) + return false; + if (!filp) return false;

@@ -905,6 +957,9 @@ static bool anon_mapping_align_enabled(int order) { unsigned long mask;

+ if (!transparent_hugepage_anon_mapping_align_enabled()) + return 0; + mask = READ_ONCE(huge_anon_orders_always) | READ_ONCE(huge_anon_orders_madvise);