LUF(Lazy Unmap Flush) reducing tlb numbers over 90%
Subsystems (13)
Byungchul Park
reviewerv10 cover LUF(Lazy Unmap Flush) reducing tlb numbers over 90%
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit 4d4b6d66db ("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:
https://lore.kernel.org/lkml/[email protected]/
However, it's only for migration through hinting fault. I thought it'd be much better if we have a general mechanism to reduce all the tlb numbers that we can apply to any unmap code, that we normally believe tlb flush should be followed.
I'm suggesting a new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
tlb flush can be defered when folios get unmapped as long as it guarantees to perform tlb flush needed, before the folios actually become used, of course, only if all the corresponding ptes don't have write permission. Otherwise, the system will get messed up.
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through mmap(), give up luf mechanism and perform tlb flush required right away.
No matter what type of workload is used for performance evaluation, the result would be positive thanks to the unconditional reduction of tlb flushes, tlb misses and interrupts. For the test, I picked up one of the most popular and heavy workload, llama.cpp that is a LLM(Large Language Model) inference engine.
The result would depend on memory latency and how often reclaim runs, which implies tlb miss overhead and how many times unmapping happens. In my system, the result shows:
1. tlb flushes are reduced about 95%.
2. tlb misses(itlb) are reduced about 80%.
3. tlb misses(dtlb store) are reduced about 57%.
4. tlb misses(dtlb load) are reduced about 24%.
5. tlb shootdown interrupts are reduced about 95%.
6. The test program runtime is reduced about 5%.
The test environment and the result is like:
Machine: bare metal, x86_64, Intel(R) Xeon(R) Gold 6430
CPU: 1 socket 64 core with hyper thread on Numa: 2 nodes (64 CPUs DRAM 42GB, no CPUs CXL expander 98GB) Config: swap off, numa balancing tiering on, demotion enabled
The test set:
llama.cpp/main -m $(70G_model1) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model2) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model3) -p "who are you?" -s 1 -t 15 -n 20 & wait
where -t: nr of threads, -s: seed used to make the runtime stable, -n: nr of tokens that determines the runtime, -p: prompt to ask, -m: LLM model to use.
Run the test set 10 times successively with caches dropped every run via 'echo 3 > /proc/sys/vm/drop_caches'. Each inference prints its runtime at the end of each.
1. Runtime from the output of llama.cpp:
BEFORE
------
llama_print_timings: total time = 1002461.95 ms / 24 tokens llama_print_timings: total time = 1044978.38 ms / 24 tokens llama_print_timings: total time = 1000653.09 ms / 24 tokens llama_print_timings: total time = 1047104.80 ms / 24 tokens llama_print_timings: total time = 1069430.36 ms / 24 tokens llama_print_timings: total time = 1068201.16 ms / 24 tokens llama_print_timings: total time = 1078092.59 ms / 24 tokens llama_print_timings: total time = 1073200.45 ms / 24 tokens llama_print_timings: total time = 1067136.00 ms / 24 tokens llama_print_timings: total time = 1076442.56 ms / 24 tokens llama_print_timings: total time = 1004142.64 ms / 24 tokens llama_print_timings: total time = 1042942.65 ms / 24 tokens llama_print_timings: total time = 999933.76 ms / 24 tokens llama_print_timings: total time = 1046548.83 ms / 24 tokens llama_print_timings: total time = 1068671.48 ms / 24 tokens llama_print_timings: total time = 1068285.76 ms / 24 tokens llama_print_timings: total time = 1077789.63 ms / 24 tokens llama_print_timings: total time = 1071558.93 ms / 24 tokens llama_print_timings: total time = 1066181.55 ms / 24 tokens llama_print_timings: total time = 1076767.53 ms / 24 tokens llama_print_timings: total time = 1004065.63 ms / 24 tokens llama_print_timings: total time = 1044522.13 ms / 24 tokens llama_print_timings: total time = 999725.33 ms / 24 tokens llama_print_timings: total time = 1047510.77 ms / 24 tokens llama_print_timings: total time = 1068010.27 ms / 24 tokens llama_print_timings: total time = 1068999.31 ms / 24 tokens llama_print_timings: total time = 1077648.05 ms / 24 tokens llama_print_timings: total time = 1071378.96 ms / 24 tokens llama_print_timings: total time = 1066326.32 ms / 24 tokens llama_print_timings: total time = 1077088.92 ms / 24 tokens
AFTER
-----
llama_print_timings: total time = 988522.03 ms / 24 tokens llama_print_timings: total time = 997204.52 ms / 24 tokens llama_print_timings: total time = 996605.86 ms / 24 tokens llama_print_timings: total time = 991985.50 ms / 24 tokens llama_print_timings: total time = 1035143.31 ms / 24 tokens llama_print_timings: total time = 993660.18 ms / 24 tokens llama_print_timings: total time = 983082.14 ms / 24 tokens llama_print_timings: total time = 990431.36 ms / 24 tokens llama_print_timings: total time = 992707.09 ms / 24 tokens llama_print_timings: total time = 992673.27 ms / 24 tokens llama_print_timings: total time = 989285.43 ms / 24 tokens llama_print_timings: total time = 996710.06 ms / 24 tokens llama_print_timings: total time = 996534.64 ms / 24 tokens llama_print_timings: total time = 991344.17 ms / 24 tokens llama_print_timings: total time = 1035210.84 ms / 24 tokens llama_print_timings: total time = 994714.13 ms / 24 tokens llama_print_timings: total time = 984184.15 ms / 24 tokens llama_print_timings: total time = 990909.45 ms / 24 tokens llama_print_timings: total time = 991881.48 ms / 24 tokens llama_print_timings: total time = 993918.03 ms / 24 tokens llama_print_timings: total time = 990061.34 ms / 24 tokens llama_print_timings: total time = 998076.69 ms / 24 tokens llama_print_timings: total time = 997082.59 ms / 24 tokens llama_print_timings: total time = 990677.58 ms / 24 tokens llama_print_timings: total time = 1036054.94 ms / 24 tokens llama_print_timings: total time = 994125.93 ms / 24 tokens llama_print_timings: total time = 982467.01 ms / 24 tokens llama_print_timings: total time = 990191.60 ms / 24 tokens llama_print_timings: total time = 993319.24 ms / 24 tokens llama_print_timings: total time = 992540.57 ms / 24 tokens
2. tlb shootdowns from 'cat /proc/interrupts':
BEFORE
------
TLB:
125553646 141418810 161932620 176853972 186655697 190399283 192143823 196414038 192872439 193313658 193395617 192521416 190788161 195067598 198016061 193607347 194293972 190786732 191545637 194856822 191801931 189634535 190399803 196365922 195268398 190115840 188050050 193194908 195317617 190820190 190164820 185556071 226797214 229592631 216112464 209909495 205575979 205950252 204948111 197999795 198892232 205287952 199344631 195015158 195869844 198858745 195692876 200961904 203463252 205921722 199850838 206145986 199613202 199961345 200129577 203020521 207873649 203697671 197093386 204243803 205993323 200934664 204193128 194435376 TLB shootdowns
AFTER
-----
TLB:
5648092 6610142 7032849 7882308 8088518 8352310
8656536 8705136 8647426 8905583 8985408 8704522
8884344 9026261 8929974 8869066 8877575 8810096
8770984 8754503 8801694 8865925 8787524 8656432
8755912 8682034 8773935 8832925 8797997 8515777
8481240 8891258 10595243 10285973 9756935 9573681
9398968 9069244 9242984 8899009 9310690 9029095
9069758 9105825 9092703 9270202 9460287 9258546
9180415 9232723 9270611 9175020 9490420 9360316
9420818 9057663 9525631 9310152 9152242 8654483
9181804 9050847 8919916 8883856 TLB shootdowns3. tlb numbers from 'perf stat' per test set:
BEFORE
------
3163679332 dTLB-load-misses
2017751856 dTLB-store-misses
327092903 iTLB-load-misses
1357543886 tlb:tlb_flush
AFTER
-----
2394694609 dTLB-load-misses
861144167 dTLB-store-misses
64055579 iTLB-load-misses
69175002 tlb:tlb_flush
---
Changes from v9:
1. Expand the candidate to apply this mechanism:
BEFORE - The souce folios at any type of migration. AFTER - Any folios that have been unmapped and freed.
2. Change the workload for test:
BEFORE - XSBench
AFTER - llama.cpp (one of the most popluar real workload)
3. Change the test environment:
BEFORE - qemu machine, too small DRAM(1GB), large remote mem AFTER - bare metal, real CXL memory, practical memory size
4. Rename the mechanism from MIGRC(Migration Read Copy) to LUF(Lazy Unmap Flush) to reflect the current version of the mechanism can be applied not only to unmap during migration but any unmap code e.g. unmap in shrink_folio_list().
5. Fix build error for riscv. (feedbacked by kernel test bot)
6. Supplement commit messages to describe what this mechanism is for, especially in the patches for arch code. (feedbacked by Thomas Gleixner)
7. Clean up some trivial things.
Changes from v8:
1. Rebase on akpm/mm.git mm-unstable as of April 18, 2024.
2. Supplement comments and commit message.
3. Change the candidate to apply migrc mechanism:
BEFORE - The source folios at demotion and promotion. AFTER - The souce folios at any type of migration.
4. Change how migrc mechanism works:
BEFORE - Reduce tlb flushes by deferring folio_free() for
source folios during demotion and promotion.
AFTER - Reduce tlb flushes by deferring tlb flush until they
actually become used, out of pcp or buddy. The current version of migrc does *not* defer calling folio_free() but let it go as it is as the same as vanilla kernel, with the folios marked kind of 'need to tlb flush'. And then handle the flush when the page exits from pcp or buddy so as to prevent changing vm stats e.g. free pages.
Changes from v7:
1. Rewrite cover letter to explain what 'migrc' mechasism is. (feedbacked by Andrew Morton)
2. Supplement the commit message of a patch 'mm: Add APIs to free a folio directly to the buddy bypassing pcp'. (feedbacked by Andrew Morton)
Changes from v6:
1. Fix build errors in case of CONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH disabled by moving migrc_flush_{start,end}() calls from arch code to try_to_unmap_flush() in mm/rmap.c.
Changes from v5:
1. Fix build errors in case of CONFIG_MIGRATION disabled or CONFIG_HWPOISON_INJECT moduled. (feedbacked by kernel test bot and Raymond Jay Golo)
2. Organize migrc code with two kconfigs, CONFIG_MIGRATION and CONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH.
Changes from v4:
1. Rebase on v6.7.
2. Fix build errors in arm64 that is doing nothing for tlb flush but has CONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH. (reported by kernel test robot)
3. Don't use any page flag. So the system would give up migrc mechanism more often but it's okay. The final improvement is good enough.
4. Instead, optimize full tlb flush(arch_tlbbatch_flush()) by avoiding redundant CPUs from tlb flush.
Changes from v3:
1. Don't use the kconfig, CONFIG_MIGRC, and remove sysctl knob, migrc_enable. (feedbacked by Nadav)
2. Remove the optimization skipping CPUs that have already performed tlb flushes needed by any reason when performing tlb flushes by migrc because I can't tell the performance difference between w/ the optimization and w/o that. (feedbacked by Nadav)
3. Minimize arch-specific code. While at it, move all the migrc declarations and inline functions from include/linux/mm.h to mm/internal.h (feedbacked by Dave Hansen, Nadav)
4. Separate a part making migrc paused when the system is in high memory pressure to another patch. (feedbacked by Nadav)
5. Rename:
a. arch_tlbbatch_clean() to arch_tlbbatch_clear(), b. tlb_ubc_nowr to tlb_ubc_ro,
c. migrc_try_flush_free_folios() to migrc_flush_free_folios(), d. migrc_stop to migrc_pause.
(feedbacked by Nadav)
6. Use ->lru list_head instead of introducing a new llist_head. (feedbacked by Nadav)
7. Use non-atomic operations of page-flag when it's safe. (feedbacked by Nadav)
8. Use stack instead of keeping a pointer of 'struct migrc_req' in struct task, which is for manipulating it locally. (feedbacked by Nadav)
9. Replace a lot of simple functions to inline functions placed in a header, mm/internal.h. (feedbacked by Nadav)
10. Add additional sufficient comments. (feedbacked by Nadav)
11. Remove a lot of wrapper functions. (feedbacked by Nadav)
Changes from RFC v2:
1. Remove additional occupation in struct page. To do that, unioned with lru field for migrc's list and added a page flag. I know page flag is a thing that we don't like to add but no choice because migrc should distinguish folios under migrc's control from others. Instead, I force migrc to be used only on 64 bit system to mitigate you guys from getting angry.
2. Remove meaningless internal object allocator that I introduced to minimize impact onto the system. However, a ton of tests showed there was no difference.
3. Stop migrc from working when the system is in high memory pressure like about to perform direct reclaim. At the condition where the swap mechanism is heavily used, I found the system suffered from regression without this control.
4. Exclude folios that pte_dirty() == true from migrc's interest so that migrc can work simpler.
5. Combine several patches that work tightly coupled to one.
6. Add sufficient comments for better review.
7. Manage migrc's request in per-node manner (from globally).
8. Add tlb miss improvement in commit message.
9. Test with more CPUs(4 -> 16) to see bigger improvement.
Changes from RFC:
1. Fix a bug triggered when a destination folio at the previous migration becomes a source folio at the next migration, before the folio gets handled properly so that the folio can play with another migration. There was inconsistency in the folio's state. Fixed it.
2. Split the patch set into more pieces so that the folks can review better. (Feedbacked by Nadav Amit)
3. Fix a wrong usage of barrier e.g. smp_mb__after_atomic(). (Feedbacked by Nadav Amit)
4. Tried to add sufficient comments to explain the patch set better. (Feedbacked by Nadav Amit)
Byungchul Park (12):
x86/tlb: add APIs manipulating tlb batch's arch data
arm64: tlbflush: add APIs manipulating tlb batch's arch data riscv, tlb: add APIs manipulating tlb batch's arch data x86/tlb, riscv/tlb, mm/rmap: separate arch_tlbbatch_clear() out of
arch_tlbbatch_flush()
mm: buddy: make room for a new variable, ugen, in struct page
mm: add folio_put_ugen() to deliver unmap generation number to pcp or
buddy
mm: add a parameter, unmap generation number, to free_unref_folios() mm/rmap: recognize read-only tlb entries during batched tlb flush
mm: implement LUF(Lazy Unmap Flush) defering tlb flush when folios get
unmapped
mm: separate move/undo parts from migrate_pages_batch() mm, migrate: apply luf mechanism to unmapping during migration mm, vmscan: apply luf mechanism to unmapping during folio reclaim
arch/arm64/include/asm/tlbflush.h | 18 ++ arch/riscv/include/asm/tlbflush.h | 21 ++ arch/riscv/mm/tlbflush.c | 1 - arch/x86/include/asm/tlbflush.h | 18 ++ arch/x86/mm/tlb.c | 2 - include/linux/mm.h | 22 ++ include/linux/mm_types.h | 40 +++- include/linux/rmap.h | 7 +- include/linux/sched.h | 11 + mm/compaction.c | 10 + mm/internal.h | 115 +++++++++- mm/memory.c | 8 + mm/migrate.c | 184 ++++++++++------ mm/mmap.c | 8 + mm/page_alloc.c | 157 +++++++++++--- mm/page_isolation.c | 6 + mm/page_reporting.c | 10 + mm/rmap.c | 345 +++++++++++++++++++++++++++++- mm/swap.c | 18 +- mm/vmscan.c | 29 ++- 20 files changed, 904 insertions(+), 126 deletions(-)
base-commit: f52bcd4a9f6058704a6f6b6b50418f579defd4fe
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 1/12 x86/tlb: add APIs manipulating tlb batch's arch data
A new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios wouldn't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
This is a preparation for the mechanism that needs to recognize read-only tlb entries by separating tlb batch arch data into two, one is for read-only entries and the other is for writable ones, and merging those two when needed.
It also optimizes tlb shootdown by skipping CPUs that have already performed tlb flush needed since. To support it, added APIs manipulating arch data for x86.
--- arch/x86/include/asm/tlbflush.h | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
arch/x86/include/asm/tlbflush.h +18 -0
@@ -5,6 +5,7 @@#include <linux/mm_types.h> #include <linux/mmu_notifier.h> #include <linux/sched.h> +#include <linux/cpumask.h> #include <asm/processor.h> #include <asm/cpufeature.h>
@@ -293,6 +294,23 @@ static inline void arch_flush_tlb_batched_pending(struct mm_struct *mm)extern void arch_tlbbatch_flush (struct arch_tlbflush_unmap_batch *batch); +static inline void arch_tlbbatch_clear (struct arch_tlbflush_unmap_batch *batch) +{ +cpumask_clear (&batch->cpumask); +} + +static inline void arch_tlbbatch_fold (struct arch_tlbflush_unmap_batch *bdst, +struct arch_tlbflush_unmap_batch *bsrc) +{ +cpumask_or (&bdst->cpumask, &bdst->cpumask, &bsrc->cpumask); +} + +static inline bool arch_tlbbatch_done (struct arch_tlbflush_unmap_batch *bdst, +struct arch_tlbflush_unmap_batch *bsrc) +{ +return !cpumask_andnot (&bdst->cpumask, &bdst->cpumask, &bsrc->cpumask); +} +static inline bool pte_flags_need_flush (unsigned long oldflags,unsigned long newflags,bool ignore_access)
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 2/12 arm64: tlbflush: add APIs manipulating tlb batch's arch data
In reply to Byungchul ParkA new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
This is a preparation for the mechanism that requires to manipulate tlb batch's arch data. Even though arm64 does nothing for tlb things, arch with CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH should provide the APIs.
--- arch/arm64/include/asm/tlbflush.h | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
arch/arm64/include/asm/tlbflush.h +18 -0
@@ -347,6 +347,24 @@ static inline void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch)dsb (ish); } +static inline void arch_tlbbatch_clear (struct arch_tlbflush_unmap_batch *batch) +{ +/* nothing to do */ +} + +static inline void arch_tlbbatch_fold (struct arch_tlbflush_unmap_batch *bdst, +struct arch_tlbflush_unmap_batch *bsrc) +{ +/* nothing to do */ +} + +static inline bool arch_tlbbatch_done (struct arch_tlbflush_unmap_batch *bdst, +struct arch_tlbflush_unmap_batch *bsrc) +{ +/* Kernel can consider tlb batch always has been done. */ +return true; +} +/* * This is meant to avoid soft lock-ups on large TLB flushing ranges and not * necessarily a performance improvement.
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 3/12 riscv, tlb: add APIs manipulating tlb batch's arch data
In reply to Byungchul ParkA new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
This is a preparation for the mechanism that needs to recognize read-only tlb entries by separating tlb batch arch data into two, one is for read-only entries and the other is for writable ones, and merging those two when needed.
It also optimizes tlb shootdown by skipping CPUs that have already performed tlb flush needed since. To support it, added APIs manipulating arch data for riscv.
--- arch/riscv/include/asm/tlbflush.h | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)
arch/riscv/include/asm/tlbflush.h +21 -0
@@ -8,6 +8,7 @@#define _ASM_RISCV_TLBFLUSH_H#include <linux/mm_types.h> +#include <linux/cpumask.h> #include <asm/smp.h> #include <asm/errata_list.h>
@@ -55,6 +56,26 @@ void arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch *batch,void arch_flush_tlb_batched_pending (struct mm_struct *mm);void arch_tlbbatch_flush (struct arch_tlbflush_unmap_batch *batch); +static inline void arch_tlbbatch_clear (struct arch_tlbflush_unmap_batch *batch) +{ +cpumask_clear (&batch->cpumask); + +} + +static inline void arch_tlbbatch_fold (struct arch_tlbflush_unmap_batch *bdst, +struct arch_tlbflush_unmap_batch *bsrc) +{ +cpumask_or (&bdst->cpumask, &bdst->cpumask, &bsrc->cpumask); + +} + +static inline bool arch_tlbbatch_done (struct arch_tlbflush_unmap_batch *bdst, +struct arch_tlbflush_unmap_batch *bsrc) +{ +return !cpumask_andnot (&bdst->cpumask, &bdst->cpumask, &bsrc->cpumask); + +} +#else /* CONFIG_SMP && CONFIG_MMU */ #define flush_tlb_all () local_flush_tlb_all()
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 4/12 x86/tlb, riscv/tlb, mm/rmap: separate arch_tlbbatch_clear() out of arch_tlbbatch_flush()
In reply to Byungchul ParkA new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
This is a preparation for the mechanism that requires to avoid redundant tlb flush by manipulating tlb batch's arch data. To achieve that, we need to separate the part clearing the tlb batch's arch data out of arch_tlbbatch_flush().
--- arch/riscv/mm/tlbflush.c | 1 - arch/x86/mm/tlb.c | 2 -- mm/rmap.c | 1 + 3 files changed, 1 insertion(+), 3 deletions(-)
arch/riscv/mm/tlbflush.c +0 -1
@@ -234,5 +234,4 @@ void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch) {__flush_tlb_range (&batch->cpumask,FLUSH_TLB_NO_ASID ,0 ,FLUSH_TLB_MAX_SIZE ,PAGE_SIZE ); -cpumask_clear (&batch->cpumask); }
arch/x86/mm/tlb.c +0 -2
@@ -1265,8 +1265,6 @@ void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch)local_irq_enable (); } -cpumask_clear (&batch->cpumask); -put_flush_tlb_info ();put_cpu (); }
mm/rmap.c +1 -0
@@ -649,6 +649,7 @@ void try_to_unmap_flush(void)return ;arch_tlbbatch_flush (&tlb_ubc->arch); +arch_tlbbatch_clear (&tlb_ubc->arch); tlb_ubc->flush_required = false; tlb_ubc->writable = false; }
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 5/12 mm: buddy: make room for a new variable, ugen, in struct page
In reply to Byungchul ParkFunctionally, no change. This is a preparation for luf mechanism that tracks need of tlb flush for each page residing in buddy, using a generation number in struct page.
Fortunately, since the private field in struct page is used only to store page order in buddy, ranging from 0 to MAX_PAGE_ORDER, that can be covered with unsigned short int. So splitted it into two smaller ones, order and ugen, so that the both can be used in buddy at the same time.
--- include/linux/mm_types.h | 40 +++++++++++++++++++++++++++++++++------- mm/internal.h | 4 ++-- mm/page_alloc.c | 13 ++++++++----- 3 files changed, 43 insertions(+), 14 deletions(-)
include/linux/mm_types.h +33 -7
@@ -108,13 +108,25 @@ struct page {pgoff_t index;/* Our offset within mapping. */ unsigned long share;/* share count for fsdax */ }; -/** -* @private: Mapping-private opaque data. -* Usually used for buffer_heads if PagePrivate. -* Used for swp_entry_t if PageSwapCache. -* Indicates order in the buddy system if PageBuddy. -*/ -unsigned long private; +union { +/** +* @private: Mapping-private opaque data. +* Usually used for buffer_heads if PagePrivate. +* Used for swp_entry_t if PageSwapCache. +*/ +unsigned long private; +struct { +/* +* Indicates order in the buddy system if PageBuddy. +*/ +unsigned short int order; +/* +* Tracks need of tlb flush used by luf, +* which stands for lazy unmap flush. +*/ +unsigned short int ugen; + }; + }; };struct {/* page_pool used by netstack */ /**
@@ -521,6 +533,20 @@ static inline void set_page_private(struct page *page, unsigned long private) page->private = private; } +#define page_buddy_order (page) ((page)->order) + +static inline void set_page_buddy_order (struct page *page,unsigned int order) +{ + page->order = (unsigned short int )order; +} + +#define page_buddy_ugen (page) ((page)->ugen) + +static inline void set_page_buddy_ugen (struct page *page,unsigned short int ugen) +{ + page->ugen = ugen; +} +static inline void *folio_get_private (struct folio *folio) {return folio->private;
mm/internal.h +2 -2
@@ -453,7 +453,7 @@ struct alloc_context {static inline unsigned int buddy_order (struct page *page) {/* PageBuddy() must be checked by the caller */ -return page_private (page); +return page_buddy_order (page); }/*
@@ -467,7 +467,7 @@ static inline unsigned int buddy_order(struct page *page)* times, potentially observing different values in the tests and the actual * use of the result. */ -#define buddy_order_unsafe (page) READ_ONCE(page_private(page)) +#define buddy_order_unsafe (page) READ_ONCE(page_buddy_order(page))/* * This function checks whether a page is free && is the buddy
mm/page_alloc.c +8 -5
@@ -565,9 +565,12 @@ void prep_compound_page(struct page *page, unsigned int order)prep_compound_head (page, order); } -static inline void set_buddy_order (struct page *page,unsigned int order) +static inline void set_buddy_order_ugen (struct page *page, +unsigned int order, +unsigned short int ugen) { -set_page_private (page, order); +set_page_buddy_order (page, order); +set_page_buddy_ugen (page, order);__SetPageBuddy (page); }
@@ -834,7 +837,7 @@ static inline void __free_one_page(struct page *page, } done_merging: -set_buddy_order (page, order); +set_buddy_order_ugen (page, order,0 );if (fpi_flags &FPI_TO_TAIL ) to_tail = true;
@@ -1344,7 +1347,7 @@ static inline void expand(struct zone *zone, struct page *page,continue ;__add_to_free_list (&page[size], zone, high, migratetype, false); -set_buddy_order (&page[size], high); +set_buddy_order_ugen (&page[size], high,0 ); nr_added += size; }account_freepages (zone, nr_added, migratetype);
@@ -6802,7 +6805,7 @@ static void break_down_buddy_pages(struct zone *zone, struct page *page,continue ;add_to_free_list (current_buddy, zone, high, migratetype, false); -set_buddy_order (current_buddy, high); +set_buddy_order_ugen (current_buddy, high,0 ); } }
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 6/12 mm: add folio_put_ugen() to deliver unmap generation number to pcp or buddy
In reply to Byungchul ParkIntroduced a new API, folio_put_ugen(), to deliver unmap generation number to pcp or buddy that will be used by luf mechanism to track need of tlb flush for each page residing in pcp or buddy.
For now, the delivery should work for the following call path that is of releasing source folios during migration:
folio_put_ugen() __folio_put_ugen() free_unref_page() free_unref_page_commit() free_one_page() __free_one_page()
The generation number should be handed over properly when pages travel between pcp and buddy, and must do necessary handling on exit from pcp or buddy. This patch doesn't include actual body for tlb flush on the exit, which will be filled by the main patch of luf mechanism.
--- include/linux/mm.h | 22 +++++++ include/linux/sched.h | 1 + mm/compaction.c | 10 +++ mm/internal.h | 70 +++++++++++++++++++- mm/page_alloc.c | 144 ++++++++++++++++++++++++++++++++++-------- mm/page_isolation.c | 6 ++ mm/page_reporting.c | 10 +++ mm/swap.c | 12 +++- 8 files changed, 247 insertions(+), 28 deletions(-)
include/linux/mm.h +22 -0
@@ -1312,6 +1312,7 @@ static inline struct folio *virt_to_folio(const void *x) }void __folio_put (struct folio *folio); +void __folio_put_ugen (struct folio *folio,unsigned short int ugen);void put_pages_list (struct list_head *pages);
@@ -1509,6 +1510,27 @@ static inline void folio_put(struct folio *folio)__folio_put (folio); } +/** +* folio_put_ugen - Decrement the last reference count on a folio. +* @folio: The folio. +* @ugen: The unmap generation # of TLB flush that the folio requires. +* +* The folio's reference count should be one since the only user, folio +* migration code, calls folio_put_ugen() only when the folio has no +* reference else. The memory will be released back to the page +* allocator and may be used by another allocation immediately. Do not +* access the memory or the struct folio after calling folio_put_ugen(). +* +* Context: May be called in process or interrupt context, but not in NMI +* context. May be called while holding a spinlock. +*/ +static inline void folio_put_ugen (struct folio *folio,unsigned short int ugen) +{ +if (WARN_ON (!folio_put_testzero (folio))) +return ; +__folio_put_ugen (folio, ugen); +} +/** * folio_put_refs - Reduce the reference count on a folio. * @folio: The folio.
include/linux/sched.h +1 -0
@@ -1339,6 +1339,7 @@ struct task_struct {#endif struct tlbflush_unmap_batch tlb_ubc; +unsigned short int ugen;/* Cache last used pipe for splice(): */ struct pipe_inode_info *splice_pipe;
mm/compaction.c +10 -0
@@ -701,6 +701,11 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,if (locked)spin_unlock_irqrestore (&cc->zone->lock, flags); +/* +* Check and flush before using the isolated pages. +*/ +check_flush_task_ugen (); +/* * Be careful to not go outside of the pageblock. */
@@ -1673,6 +1678,11 @@ static void fast_isolate_freepages(struct compact_control *cc)spin_unlock_irqrestore (&cc->zone->lock, flags); +/* +* Check and flush before using the isolated pages. +*/ +check_flush_task_ugen (); +/* Skip fast search if enough freepages isolated */ if (cc->nr_freepages >= cc->nr_migratepages)break ;
mm/internal.h +69 -1
80 lines not on this page · raw message
mm/page_alloc.c +118 -26
405 lines not on this page · raw message
mm/page_isolation.c +6 -0
@@ -260,6 +260,12 @@ static void unset_migratetype_isolate(struct page *page, int migratetype) zone->nr_isolate_pageblock--; out:spin_unlock_irqrestore (&zone->lock, flags); + +/* +* Check and flush for the pages that have been isolated. +*/ +if (isolated_page) +check_flush_task_ugen (); } static inline struct page *
mm/page_reporting.c +10 -0
@@ -221,6 +221,11 @@ page_reporting_cycle(struct page_reporting_dev_info *prdev, struct zone *zone,/* release lock before waiting on report processing */ spin_unlock_irq (&zone->lock); +/* +* Check and flush before using the isolated pages. +*/ +check_flush_task_ugen (); +/* begin processing pages in local list */ err = prdev->report (prdev, sgl,PAGE_REPORTING_CAPACITY );
@@ -253,6 +258,11 @@ page_reporting_cycle(struct page_reporting_dev_info *prdev, struct zone *zone,spin_unlock_irq (&zone->lock); +/* +* Check and flush before using the isolated pages. +*/ +check_flush_task_ugen (); +return err; }
mm/swap.c +11 -1
@@ -126,10 +126,20 @@ void __folio_put(struct folio *folio)if (folio_test_large (folio) &&folio_test_large_rmappable (folio))folio_undo_large_rmappable (folio);mem_cgroup_uncharge (folio); -free_unref_page (&folio->page,folio_order (folio)); +free_unref_page (&folio->page,folio_order (folio),0 ); }EXPORT_SYMBOL (__folio_put); +void __folio_put_ugen (struct folio *folio,unsigned short int ugen) +{ +if (WARN_ON (!can_luf_folio (folio))) +return ; + +page_cache_release (folio); +mem_cgroup_uncharge (folio); +free_unref_page (&folio->page,0 , ugen); +} +/** * put_pages_list() - release a list of pages * @pages: list of pages threaded on page->lru
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 7/12 mm: add a parameter, unmap generation number, to free_unref_folios()
In reply to Byungchul ParkUnmap generation number is used by luf mechanism to track need of tlb flush for each page residing in pcp or buddy.
The number should be delivered to pcp or buddy via free_unref_folios() that is for releasing folios that have been unmapped during reclaim in shrink_folio_list().
--- mm/internal.h | 2 +- mm/page_alloc.c | 10 +++++----- mm/swap.c | 6 +++--- mm/vmscan.c | 8 ++++---- 4 files changed, 13 insertions(+), 13 deletions(-)
mm/internal.h +1 -1
@@ -639,7 +639,7 @@ extern bool free_pages_prepare(struct page *page, unsigned int order);extern int user_min_free_kbytes;void free_unref_page (struct page *page,unsigned int order,unsigned short int ugen); -void free_unref_folios (struct folio_batch *fbatch); +void free_unref_folios (struct folio_batch *fbatch,unsigned short int ugen);extern void zone_pcp_reset (struct zone *zone);extern void zone_pcp_disable (struct zone *zone);
mm/page_alloc.c +5 -5
@@ -2706,7 +2706,7 @@ void free_unref_page(struct page *page, unsigned int order,/* * Free a batch of folios */ -void free_unref_folios (struct folio_batch *folios) +void free_unref_folios (struct folio_batch *folios,unsigned short int ugen) {unsigned long __maybe_unused UP_flags;struct per_cpu_pages *pcp =NULL ;
@@ -2729,7 +2729,7 @@ void free_unref_folios(struct folio_batch *folios)*/ if (!pcp_allowed_order (order)) {free_one_page (folio_zone (folio), &folio->page, - pfn, order,FPI_NONE ,0 ); + pfn, order,FPI_NONE , ugen);continue ; } folio->private = (void *)(unsigned long )order;
@@ -2765,7 +2765,7 @@ void free_unref_folios(struct folio_batch *folios)*/ if (is_migrate_isolate (migratetype)) {free_one_page (zone, &folio->page, pfn, - order,FPI_NONE ,0 ); + order,FPI_NONE , ugen);continue ; }
@@ -2778,7 +2778,7 @@ void free_unref_folios(struct folio_batch *folios)if (unlikely (!pcp)) {pcp_trylock_finish (UP_flags);free_one_page (zone, &folio->page, pfn, - order,FPI_NONE ,0 ); + order,FPI_NONE , ugen);continue ; } locked_zone = zone;
@@ -2793,7 +2793,7 @@ void free_unref_folios(struct folio_batch *folios)trace_mm_page_free_batched (&folio->page);free_unref_page_commit (zone, pcp, &folio->page, migratetype, - order,0 ); + order, ugen); }if (pcp) {
mm/swap.c +3 -3
@@ -163,11 +163,11 @@ void put_pages_list(struct list_head *pages)/* LRU flag must be clear because it's passed using the lru */ if (folio_batch_add (&fbatch, folio) >0 )continue ; -free_unref_folios (&fbatch); +free_unref_folios (&fbatch,0 ); }if (fbatch.nr) -free_unref_folios (&fbatch); +free_unref_folios (&fbatch,0 );INIT_LIST_HEAD (pages); }EXPORT_SYMBOL (put_pages_list);
@@ -1029,7 +1029,7 @@ void folios_put_refs(struct folio_batch *folios, unsigned int *refs) folios->nr = j;mem_cgroup_uncharge_folios (folios); -free_unref_folios (folios); +free_unref_folios (folios,0 ); }EXPORT_SYMBOL (folios_put_refs);
mm/vmscan.c +4 -4
@@ -1460,7 +1460,7 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,if (folio_batch_add (&free_folios, folio) ==0 ) {mem_cgroup_uncharge_folios (&free_folios);try_to_unmap_flush (); -free_unref_folios (&free_folios); +free_unref_folios (&free_folios,0 ); }continue ;
@@ -1527,7 +1527,7 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,mem_cgroup_uncharge_folios (&free_folios);try_to_unmap_flush (); -free_unref_folios (&free_folios); +free_unref_folios (&free_folios,0 );list_splice (&ret_folios, folio_list);count_vm_events (PGACTIVATE , pgactivate);
@@ -1869,7 +1869,7 @@ static unsigned int move_folios_to_lru(struct lruvec *lruvec,if (folio_batch_add (&free_folios, folio) ==0 ) {spin_unlock_irq (&lruvec->lru_lock);mem_cgroup_uncharge_folios (&free_folios); -free_unref_folios (&free_folios); +free_unref_folios (&free_folios,0 );spin_lock_irq (&lruvec->lru_lock); }
@@ -1891,7 +1891,7 @@ static unsigned int move_folios_to_lru(struct lruvec *lruvec,if (free_folios.nr) {spin_unlock_irq (&lruvec->lru_lock);mem_cgroup_uncharge_folios (&free_folios); -free_unref_folios (&free_folios); +free_unref_folios (&free_folios,0 );spin_lock_irq (&lruvec->lru_lock); }
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 8/12 mm/rmap: recognize read-only tlb entries during batched tlb flush
In reply to Byungchul ParkFunctionally, no change. This is a preparation for luf mechanism that requires to recognize read-only tlb entries and handle them in a different way. The newly introduced API in this patch, fold_ubc(), will be used by luf mechanism.
--- include/linux/sched.h | 1 + mm/internal.h | 4 ++++ mm/rmap.c | 34 ++++++++++++++++++++++++++++++++-- 3 files changed, 37 insertions(+), 2 deletions(-)
include/linux/sched.h +1 -0
@@ -1339,6 +1339,7 @@ struct task_struct {#endif struct tlbflush_unmap_batch tlb_ubc; +struct tlbflush_unmap_batch tlb_ubc_ro;unsigned short int ugen;/* Cache last used pipe for splice(): */
mm/internal.h +4 -0
@@ -1100,6 +1100,7 @@ extern struct workqueue_struct *mm_percpu_wq;void try_to_unmap_flush (void );void try_to_unmap_flush_dirty (void );void flush_tlb_batched_pending (struct mm_struct *mm); +void fold_ubc (struct tlbflush_unmap_batch *dst,struct tlbflush_unmap_batch *src);#else static inline void try_to_unmap_flush (void ) {
@@ -1110,6 +1111,9 @@ static inline void try_to_unmap_flush_dirty(void)static inline void flush_tlb_batched_pending (struct mm_struct *mm) { } +static inline void fold_ubc (struct tlbflush_unmap_batch *dst,struct tlbflush_unmap_batch *src) +{ +}#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ extern const struct trace_print_flags pageflag_names[];
mm/rmap.c +32 -2
@@ -635,6 +635,28 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio, }#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH + +void fold_ubc (struct tlbflush_unmap_batch *dst, +struct tlbflush_unmap_batch *src) +{ +if (!src->flush_required) +return ; + +/* +* Fold src to dst. +*/ +arch_tlbbatch_fold (&dst->arch, &src->arch); + dst->writable = dst->writable || src->writable; + dst->flush_required = true; + +/* +* Reset src. +*/ +arch_tlbbatch_clear (&src->arch); + src->flush_required = false; + src->writable = false; +} +/* * Flush TLB entries for recently unmapped pages from remote CPUs. It is * important if a PTE was dirty when it was unmapped that it's flushed
@@ -644,7 +666,9 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,void try_to_unmap_flush (void ) {struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; +struct tlbflush_unmap_batch *tlb_ubc_ro = ¤t->tlb_ubc_ro; +fold_ubc (tlb_ubc, tlb_ubc_ro);if (!tlb_ubc->flush_required)return ;
@@ -658,8 +682,9 @@ void try_to_unmap_flush(void)void try_to_unmap_flush_dirty (void ) {struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; +struct tlbflush_unmap_batch *tlb_ubc_ro = ¤t->tlb_ubc_ro; -if (tlb_ubc->writable) +if (tlb_ubc->writable || tlb_ubc_ro->writable)try_to_unmap_flush (); }
@@ -676,13 +701,18 @@ void try_to_unmap_flush_dirty(void)static void set_tlb_ubc_flush_pending (struct mm_struct *mm,pte_t pteval,unsigned long uaddr) { -struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; +struct tlbflush_unmap_batch *tlb_ubc;int batch;bool writable =pte_dirty (pteval);if (!pte_accessible (mm, pteval))return ; +if (pte_write (pteval)) + tlb_ubc = ¤t->tlb_ubc; +else + tlb_ubc = ¤t->tlb_ubc_ro; +arch_tlbbatch_add_pending (&tlb_ubc->arch, mm, uaddr); tlb_ubc->flush_required = true;
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 9/12 mm: implement LUF(Lazy Unmap Flush) defering tlb flush when folios get unmapped
In reply to Byungchul ParkA new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
tlb flush can be defered when folios get unmapped as long as it guarantees to perform tlb flush needed, before the folios actually become used, of course, only if all the corresponding ptes don't have write permission. Otherwise, the system will get messed up.
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through mmap(), give up luf mechanism and perform tlb flush required right away.
No matter what type of workload is used for performance evaluation, the result would be positive thanks to the unconditional reduction of tlb flushes, tlb misses and interrupts. For the test, I picked up one of the most popular and heavy workload, llama.cpp that is a LLM(Large Language Model) inference engine.
The result would depend on memory latency and how often reclaim runs, which implies tlb miss overhead and how many times unmapping happens. In my system, the result shows:
1. tlb flushes are reduced about 95%.
2. tlb misses(itlb) are reduced about 80%.
3. tlb misses(dtlb store) are reduced about 57%.
4. tlb misses(dtlb load) are reduced about 24%.
5. tlb shootdown interrupts are reduced about 95%.
6. The test program runtime is reduced about 5%.
The test environment and the result is like:
Machine: bare metal, x86_64, Intel(R) Xeon(R) Gold 6430
CPU: 1 socket 64 core with hyper thread on Numa: 2 nodes (64 CPUs DRAM 42GB, no CPUs CXL expander 98GB) Config: swap off, numa balancing tiering on, demotion enabled
The test set:
llama.cpp/main -m $(70G_model1) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model2) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model3) -p "who are you?" -s 1 -t 15 -n 20 & wait
where -t: nr of threads, -s: seed used to make the runtime stable, -n: nr of tokens that determines the runtime, -p: prompt to ask, -m: LLM model to use.
Run the test set 10 times successively with caches dropped every run via 'echo 3 > /proc/sys/vm/drop_caches'. Each inference prints its runtime at the end of each.
1. Runtime from the output of llama.cpp:
BEFORE
------
llama_print_timings: total time = 1002461.95 ms / 24 tokens llama_print_timings: total time = 1044978.38 ms / 24 tokens llama_print_timings: total time = 1000653.09 ms / 24 tokens llama_print_timings: total time = 1047104.80 ms / 24 tokens llama_print_timings: total time = 1069430.36 ms / 24 tokens llama_print_timings: total time = 1068201.16 ms / 24 tokens llama_print_timings: total time = 1078092.59 ms / 24 tokens llama_print_timings: total time = 1073200.45 ms / 24 tokens llama_print_timings: total time = 1067136.00 ms / 24 tokens llama_print_timings: total time = 1076442.56 ms / 24 tokens llama_print_timings: total time = 1004142.64 ms / 24 tokens llama_print_timings: total time = 1042942.65 ms / 24 tokens llama_print_timings: total time = 999933.76 ms / 24 tokens llama_print_timings: total time = 1046548.83 ms / 24 tokens llama_print_timings: total time = 1068671.48 ms / 24 tokens llama_print_timings: total time = 1068285.76 ms / 24 tokens llama_print_timings: total time = 1077789.63 ms / 24 tokens llama_print_timings: total time = 1071558.93 ms / 24 tokens llama_print_timings: total time = 1066181.55 ms / 24 tokens llama_print_timings: total time = 1076767.53 ms / 24 tokens llama_print_timings: total time = 1004065.63 ms / 24 tokens llama_print_timings: total time = 1044522.13 ms / 24 tokens llama_print_timings: total time = 999725.33 ms / 24 tokens llama_print_timings: total time = 1047510.77 ms / 24 tokens llama_print_timings: total time = 1068010.27 ms / 24 tokens llama_print_timings: total time = 1068999.31 ms / 24 tokens llama_print_timings: total time = 1077648.05 ms / 24 tokens llama_print_timings: total time = 1071378.96 ms / 24 tokens llama_print_timings: total time = 1066326.32 ms / 24 tokens llama_print_timings: total time = 1077088.92 ms / 24 tokens
AFTER
-----
llama_print_timings: total time = 988522.03 ms / 24 tokens llama_print_timings: total time = 997204.52 ms / 24 tokens llama_print_timings: total time = 996605.86 ms / 24 tokens llama_print_timings: total time = 991985.50 ms / 24 tokens llama_print_timings: total time = 1035143.31 ms / 24 tokens llama_print_timings: total time = 993660.18 ms / 24 tokens llama_print_timings: total time = 983082.14 ms / 24 tokens llama_print_timings: total time = 990431.36 ms / 24 tokens llama_print_timings: total time = 992707.09 ms / 24 tokens llama_print_timings: total time = 992673.27 ms / 24 tokens llama_print_timings: total time = 989285.43 ms / 24 tokens llama_print_timings: total time = 996710.06 ms / 24 tokens llama_print_timings: total time = 996534.64 ms / 24 tokens llama_print_timings: total time = 991344.17 ms / 24 tokens llama_print_timings: total time = 1035210.84 ms / 24 tokens llama_print_timings: total time = 994714.13 ms / 24 tokens llama_print_timings: total time = 984184.15 ms / 24 tokens llama_print_timings: total time = 990909.45 ms / 24 tokens llama_print_timings: total time = 991881.48 ms / 24 tokens llama_print_timings: total time = 993918.03 ms / 24 tokens llama_print_timings: total time = 990061.34 ms / 24 tokens llama_print_timings: total time = 998076.69 ms / 24 tokens llama_print_timings: total time = 997082.59 ms / 24 tokens llama_print_timings: total time = 990677.58 ms / 24 tokens llama_print_timings: total time = 1036054.94 ms / 24 tokens llama_print_timings: total time = 994125.93 ms / 24 tokens llama_print_timings: total time = 982467.01 ms / 24 tokens llama_print_timings: total time = 990191.60 ms / 24 tokens llama_print_timings: total time = 993319.24 ms / 24 tokens llama_print_timings: total time = 992540.57 ms / 24 tokens
2. tlb shootdowns from 'cat /proc/interrupts':
BEFORE
------
TLB:
125553646 141418810 161932620 176853972 186655697 190399283 192143823 196414038 192872439 193313658 193395617 192521416 190788161 195067598 198016061 193607347 194293972 190786732 191545637 194856822 191801931 189634535 190399803 196365922 195268398 190115840 188050050 193194908 195317617 190820190 190164820 185556071 226797214 229592631 216112464 209909495 205575979 205950252 204948111 197999795 198892232 205287952 199344631 195015158 195869844 198858745 195692876 200961904 203463252 205921722 199850838 206145986 199613202 199961345 200129577 203020521 207873649 203697671 197093386 204243803 205993323 200934664 204193128 194435376 TLB shootdowns
AFTER
-----
TLB:
5648092 6610142 7032849 7882308 8088518 8352310
8656536 8705136 8647426 8905583 8985408 8704522
8884344 9026261 8929974 8869066 8877575 8810096
8770984 8754503 8801694 8865925 8787524 8656432
8755912 8682034 8773935 8832925 8797997 8515777
8481240 8891258 10595243 10285973 9756935 9573681
9398968 9069244 9242984 8899009 9310690 9029095
9069758 9105825 9092703 9270202 9460287 9258546
9180415 9232723 9270611 9175020 9490420 9360316
9420818 9057663 9525631 9310152 9152242 8654483
9181804 9050847 8919916 8883856 TLB shootdowns3. tlb numbers from 'perf stat' per test set:
BEFORE
------
3163679332 dTLB-load-misses
2017751856 dTLB-store-misses
327092903 iTLB-load-misses
1357543886 tlb:tlb_flush
AFTER
-----
2394694609 dTLB-load-misses
861144167 dTLB-store-misses
64055579 iTLB-load-misses
69175002 tlb:tlb_flush
--- include/linux/sched.h | 9 ++ mm/internal.h | 43 +++++- mm/memory.c | 8 ++ mm/mmap.c | 8 ++ mm/rmap.c | 308 +++++++++++++++++++++++++++++++++++++++++- 5 files changed, 366 insertions(+), 10 deletions(-)
include/linux/sched.h +9 -0
@@ -1340,8 +1340,17 @@ struct task_struct {struct tlbflush_unmap_batch tlb_ubc;struct tlbflush_unmap_batch tlb_ubc_ro; +struct tlbflush_unmap_batch tlb_ubc_luf;unsigned short int ugen; +#if defined(CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH ) +/* +* whether all the mappings of a folio during unmap are read-only +* so that luf can work on the folio +*/ +bool can_luf; +#endif +/* Cache last used pipe for splice(): */ struct pipe_inode_info *splice_pipe;
mm/internal.h +39 -4
61 lines not on this page · raw message
mm/memory.c +8 -0
@@ -3617,6 +3617,14 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)if (vmf->page) folio =page_folio (vmf->page); +/* +* The folio may or may not be one that is under luf's control +* and might be about to change its permission to writable. +* Conservatively give up deferring tlb flush just in case. +*/ +if (folio) +luf_flush (); +/* * Shared mapping: we are guaranteed to have VM_WRITE and * FAULT_FLAG_WRITE set at this point.
mm/mmap.c +8 -0
@@ -1271,6 +1271,14 @@ unsigned long do_mmap(struct file *file, unsigned long addr, pkey =0 ; } +/* +* This mmap may or may not be mapping to ones that is under +* luf's control. However, conservatively give up deferring tlb +* flush just in case. +*/ +if (prot &PROT_WRITE ) +luf_flush (); +/* Do simple checking here so the lower-level routines won't have * to. we assume access permissions have been handled by the open * of the memory object, so we don't do any here.
mm/rmap.c +302 -6
366 lines not on this page · raw message
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 10/12 mm: separate move/undo parts from migrate_pages_batch()
In reply to Byungchul ParkFunctionally, no change. This is a preparation for luf mechanism that requires to use separated folio lists for its own handling during migration. Refactored migrate_pages_batch() so as to separate move/undo parts from migrate_pages_batch().
--- mm/migrate.c | 134 +++++++++++++++++++++++++++++++-------------------- 1 file changed, 83 insertions(+), 51 deletions(-)
mm/migrate.c +83 -51
@@ -1609,6 +1609,81 @@ static int migrate_hugetlbs(struct list_head *from, new_folio_t get_new_folio,return nr_failed; } +static void migrate_folios_move (struct list_head *src_folios, +struct list_head *dst_folios, +free_folio_t put_new_folio,unsigned long private, +enum migrate_mode mode,int reason, +struct list_head *ret_folios, +struct migrate_pages_stats *stats, +int *retry,int *thp_retry,int *nr_failed, +int *nr_retry_pages) +{ +struct folio *folio, *folio2, *dst, *dst2; +bool is_thp; +int nr_pages; +int rc; + + dst = list_first_entry(dst_folios, struct folio, lru); + dst2 =list_next_entry (dst, lru); +list_for_each_entry_safe (folio, folio2, src_folios, lru) { + is_thp =folio_test_large (folio) &&folio_test_pmd_mappable (folio); + nr_pages =folio_nr_pages (folio); + +cond_resched (); + + rc =migrate_folio_move (put_new_folio, private, + folio, dst, mode, + reason, ret_folios); +/* +* The rules are: +* Success: folio will be freed +* -EAGAIN: stay on the unmap_folios list +* Other errno: put on ret_folios list +*/ +switch (rc) { +case -EAGAIN : + *retry +=1 ; + *thp_retry += is_thp; + *nr_retry_pages += nr_pages; +break ; +case MIGRATEPAGE_SUCCESS : + stats->nr_succeeded += nr_pages; + stats->nr_thp_succeeded += is_thp; +break ; +default : + *nr_failed +=1 ; + stats->nr_thp_failed += is_thp; + stats->nr_failed_pages += nr_pages; +break ; + } + dst = dst2; + dst2 =list_next_entry (dst, lru); + } +} + +static void migrate_folios_undo (struct list_head *src_folios, +struct list_head *dst_folios, +free_folio_t put_new_folio,unsigned long private, +struct list_head *ret_folios) +{ +struct folio *folio, *folio2, *dst, *dst2; + + dst = list_first_entry(dst_folios, struct folio, lru); + dst2 =list_next_entry (dst, lru); +list_for_each_entry_safe (folio, folio2, src_folios, lru) { +int old_page_state =0 ; +struct anon_vma *anon_vma =NULL ; + +__migrate_folio_extract (dst, &old_page_state, &anon_vma); +migrate_folio_undo_src (folio, old_page_state &PAGE_WAS_MAPPED , + anon_vma, true, ret_folios); +list_del (&dst->lru); +migrate_folio_undo_dst (dst, true, put_new_folio, private); + dst = dst2; + dst2 =list_next_entry (dst, lru); + } +} +/* * migrate_pages_batch() first unmaps folios in the from list as many as * possible, then move the unmapped folios.
@@ -1631,7 +1706,7 @@ static int migrate_pages_batch(struct list_head *from,int pass =0 ;bool is_thp = false;bool is_large = false; -struct folio *folio, *folio2, *dst =NULL , *dst2; +struct folio *folio, *folio2, *dst =NULL ;int rc, rc_saved =0 , nr_pages;LIST_HEAD (unmap_folios);LIST_HEAD (dst_folios);
@@ -1790,42 +1865,11 @@ static int migrate_pages_batch(struct list_head *from, thp_retry =0 ; nr_retry_pages =0 ; - dst = list_first_entry(&dst_folios, struct folio, lru); - dst2 =list_next_entry (dst, lru); -list_for_each_entry_safe (folio, folio2, &unmap_folios, lru) { - is_thp =folio_test_large (folio) &&folio_test_pmd_mappable (folio); - nr_pages =folio_nr_pages (folio); - -cond_resched (); - - rc =migrate_folio_move (put_new_folio, private, - folio, dst, mode, - reason, ret_folios); -/* -* The rules are: -* Success: folio will be freed -* -EAGAIN: stay on the unmap_folios list -* Other errno: put on ret_folios list -*/ -switch (rc) { -case -EAGAIN : - retry++; - thp_retry += is_thp; - nr_retry_pages += nr_pages; -break ; -case MIGRATEPAGE_SUCCESS : - stats->nr_succeeded += nr_pages; - stats->nr_thp_succeeded += is_thp; -break ; -default : - nr_failed++; - stats->nr_thp_failed += is_thp; - stats->nr_failed_pages += nr_pages; -break ; - } - dst = dst2; - dst2 =list_next_entry (dst, lru); - } +/* Move the unmapped folios */ +migrate_folios_move (&unmap_folios, &dst_folios, + put_new_folio, private, mode, reason, + ret_folios, stats, &retry, &thp_retry, + &nr_failed, &nr_retry_pages); } nr_failed += retry; stats->nr_thp_failed += thp_retry;
@@ -1834,20 +1878,8 @@ static int migrate_pages_batch(struct list_head *from, rc = rc_saved ? : nr_failed; out:/* Cleanup remaining folios */ - dst = list_first_entry(&dst_folios, struct folio, lru); - dst2 =list_next_entry (dst, lru); -list_for_each_entry_safe (folio, folio2, &unmap_folios, lru) { -int old_page_state =0 ; -struct anon_vma *anon_vma =NULL ; - -__migrate_folio_extract (dst, &old_page_state, &anon_vma); -migrate_folio_undo_src (folio, old_page_state &PAGE_WAS_MAPPED , - anon_vma, true, ret_folios); -list_del (&dst->lru); -migrate_folio_undo_dst (dst, true, put_new_folio, private); - dst = dst2; - dst2 =list_next_entry (dst, lru); - } +migrate_folios_undo (&unmap_folios, &dst_folios, + put_new_folio, private, ret_folios);return rc; }
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 11/12 mm, migrate: apply luf mechanism to unmapping during migration
In reply to Byungchul ParkA new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
Applied the mechanism to unmapping during migration.
--- include/linux/rmap.h | 2 +- mm/migrate.c | 56 ++++++++++++++++++++++++++++++++------------ mm/rmap.c | 9 ++++--- 3 files changed, 48 insertions(+), 19 deletions(-)
include/linux/rmap.h +1 -1
@@ -657,7 +657,7 @@ static inline int folio_try_share_anon_rmap_pmd(struct folio *folio,int folio_referenced (struct folio *,int is_locked,struct mem_cgroup *memcg,unsigned long *vm_flags); -void try_to_migrate (struct folio *folio,enum ttu_flags flags); +bool try_to_migrate (struct folio *folio,enum ttu_flags flags);void try_to_unmap (struct folio *,enum ttu_flags flags); int make_device_exclusive_range(struct mm_struct *mm, unsigned long start,
mm/migrate.c +41 -15
155 lines not on this page · raw message
mm/rmap.c +6 -3
@@ -2600,8 +2600,9 @@ static bool try_to_migrate_one(struct folio *folio, struct vm_area_struct *vma,* * Tries to remove all the page table entries which are mapping this folio and * replace them with special swap entries. Caller must hold the folio lock. +* Return true if all the mappings are read-only, otherwise false. */ -void try_to_migrate (struct folio *folio,enum ttu_flags flags) +bool try_to_migrate (struct folio *folio,enum ttu_flags flags) {struct rmap_walk_control rwc = { .rmap_one = try_to_migrate_one,
@@ -2620,11 +2621,11 @@ void try_to_migrate(struct folio *folio, enum ttu_flags flags)*/ if (WARN_ON_ONCE (flags & ~(TTU_RMAP_LOCKED |TTU_SPLIT_HUGE_PMD |TTU_SYNC |TTU_BATCH_FLUSH ))) -return ; +return false;if (folio_is_zone_device (folio) && (!folio_is_device_private (folio) && !folio_is_device_coherent (folio))) -return ; +return false;/* * During exec, a temporary VMA is setup and later moved.
@@ -2649,6 +2650,8 @@ void try_to_migrate(struct folio *folio, enum ttu_flags flags)fold_ubc (tlb_ubc_luf, tlb_ubc_ro); elsefold_ubc (tlb_ubc ,tlb_ubc_ro ); + +return can_luf; }#ifdef CONFIG_DEVICE_PRIVATE
Signature · 1 line
2.17.1
Byungchul Park
reviewerPATCH v10 12/12 mm, vmscan: apply luf mechanism to unmapping during folio reclaim
In reply to Byungchul ParkA new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
Applied the mechanism to unmapping during folio reclaim.
--- include/linux/rmap.h | 5 +++-- mm/rmap.c | 5 ++++- mm/vmscan.c | 21 ++++++++++++++++++++- 3 files changed, 27 insertions(+), 4 deletions(-)
include/linux/rmap.h +3 -2
@@ -658,7 +658,7 @@ int folio_referenced(struct folio *, int is_locked, struct mem_cgroup *memcg, unsigned long *vm_flags);bool try_to_migrate (struct folio *folio,enum ttu_flags flags); -void try_to_unmap (struct folio *,enum ttu_flags flags); +bool try_to_unmap (struct folio *,enum ttu_flags flags); int make_device_exclusive_range(struct mm_struct *mm, unsigned long start, unsigned long end, struct page **pages,
@@ -777,8 +777,9 @@ static inline int folio_referenced(struct folio *folio, int is_locked,return 0 ; } -static inline void try_to_unmap (struct folio *folio,enum ttu_flags flags) +static inline bool try_to_unmap (struct folio *folio,enum ttu_flags flags) { +return false; }static inline int folio_mkclean (struct folio *folio)
mm/rmap.c +4 -1
@@ -2237,10 +2237,11 @@ static int folio_not_mapped(struct folio *folio)* Tries to remove all the page table entries which are mapping this * folio. It is the caller's responsibility to check if the folio is * still mapped if needed (use TTU_SYNC to prevent accounting races). +* Return true if all the mappings are read-only, otherwise false. * * Context: Caller must hold the folio lock. */ -void try_to_unmap (struct folio *folio,enum ttu_flags flags) +bool try_to_unmap (struct folio *folio,enum ttu_flags flags) {struct rmap_walk_control rwc = { .rmap_one = try_to_unmap_one,
@@ -2265,6 +2266,8 @@ void try_to_unmap(struct folio *folio, enum ttu_flags flags)fold_ubc (tlb_ubc_luf, tlb_ubc_ro); elsefold_ubc (tlb_ubc ,tlb_ubc_ro ); + +return can_luf; }/*
mm/vmscan.c +20 -1
@@ -1031,14 +1031,17 @@ static unsigned int shrink_folio_list(struct list_head *folio_list, struct reclaim_stat *stat, bool ignore_references) {struct folio_batch free_folios; +struct folio_batch free_folios_luf;LIST_HEAD (ret_folios);LIST_HEAD (demote_folios);unsigned int nr_reclaimed =0 ;unsigned int pgactivate =0 ;bool do_demote_pass;struct swap_iocb *plug =NULL ; +unsigned short int ugen;folio_batch_init (&free_folios); +folio_batch_init (&free_folios_luf);memset (stat,0 ,sizeof (*stat));cond_resched (); do_demote_pass =can_demote (pgdat->node_id, sc);
@@ -1050,6 +1053,7 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,enum folio_references references =FOLIOREF_RECLAIM ;bool dirty, writeback;unsigned int nr_pages; +bool can_luf = false;cond_resched ();
@@ -1292,7 +1296,7 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,if (folio_test_large (folio) &&list_empty (&folio->_deferred_list)) flags |=TTU_SYNC ; -try_to_unmap (folio, flags); + can_luf =try_to_unmap (folio, flags);if (folio_mapped (folio)) { stat->nr_unmap_fail += nr_pages; if (!was_swapbacked &&
@@ -1457,6 +1461,18 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,if (folio_test_large (folio) &&folio_test_large_rmappable (folio))folio_undo_large_rmappable (folio); + +if (can_luf) { +if (folio_batch_add (&free_folios_luf, folio) ==0 ) { +mem_cgroup_uncharge_folios (&free_folios_luf); + ugen =try_to_unmap_luf (); +if (!ugen) +try_to_unmap_flush (); +free_unref_folios (&free_folios_luf, ugen); + } +continue ; + } +if (folio_batch_add (&free_folios, folio) ==0 ) {mem_cgroup_uncharge_folios (&free_folios);try_to_unmap_flush ();
@@ -1526,8 +1542,11 @@ static unsigned int shrink_folio_list(struct list_head *folio_list, pgactivate = stat->nr_activate[0 ] + stat->nr_activate[1 ];mem_cgroup_uncharge_folios (&free_folios); +mem_cgroup_uncharge_folios (&free_folios_luf); + ugen =try_to_unmap_luf ();try_to_unmap_flush ();free_unref_folios (&free_folios,0 ); +free_unref_folios (&free_folios_luf, ugen);list_splice (&ret_folios, folio_list);count_vm_events (PGACTIVATE , pgactivate);
Signature · 1 line
2.17.1
Quoted 37 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:https://lore.kernel.org/lkml/[email protected]/
However, it's only for migration through hinting fault. I thought it'd be much better if we have a general mechanism to reduce all the tlb numbers that we can apply to any unmap code, that we normally believe tlb flush should be followed.
I'm suggesting a new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
tlb flush can be defered when folios get unmapped as long as it guarantees to perform tlb flush needed, before the folios actually become used, of course, only if all the corresponding ptes don't have write permission. Otherwise, the system will get messed up.
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
Quoted 6 lines
3. When a writable mapping is created e.g. through
mmap(), give up luf mechanism and perform tlb flush required right away.No matter what type of workload is used for performance evaluation, the result would be positive thanks to the unconditional reduction of tlb flushes, tlb misses and interrupts.
Are there any downsides of the optimization? Will it cause regression for workloads with almost no read-only mappings? Will it cause regression for page allocation?
Quoted 3 lines
For the test, I picked up one of
the most popular and heavy workload, llama.cpp that is a LLM(Large Language Model) inference engine.
IIUC, llama.cpp is a workload with huge read-only mapping.
Quoted 348 lines
The result would depend on memory latency and how often reclaim runs, which implies tlb miss overhead and how many times unmapping happens. In my system, the result shows:
1. tlb flushes are reduced about 95%.
2. tlb misses(itlb) are reduced about 80%.
3. tlb misses(dtlb store) are reduced about 57%.
4. tlb misses(dtlb load) are reduced about 24%.
5. tlb shootdown interrupts are reduced about 95%.
6. The test program runtime is reduced about 5%.The test environment and the result is like:
Machine: bare metal,
x86_64, Intel(R) Xeon(R) Gold 6430CPU: 1 socket 64 core with hyper thread on Numa: 2 nodes (64 CPUs DRAM 42GB, no CPUs CXL expander 98GB) Config: swap off, numa balancing tiering on, demotion enabledThe test set:
llama.cpp/main -m $(70G_model1) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model2) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model3) -p "who are you?" -s 1 -t 15 -n 20 & wait
where -t: nr of threads, -s: seed used to make the runtime stable, -n: nr of tokens that determines the runtime, -p: prompt to ask, -m: LLM model to use.
Run the test set 10 times successively with caches dropped every run via 'echo 3 >
/proc/sys/vm/drop_caches'. Each inference prints its runtime at the end of each.1. Runtime from the output of llama.cpp:
BEFORE
------llama_print_timings: total time = 1002461.95 ms / 24 tokens llama_print_timings: total time = 1044978.38 ms / 24 tokens llama_print_timings: total time = 1000653.09 ms / 24 tokens llama_print_timings: total time = 1047104.80 ms / 24 tokens llama_print_timings: total time = 1069430.36 ms / 24 tokens llama_print_timings: total time = 1068201.16 ms / 24 tokens llama_print_timings: total time = 1078092.59 ms / 24 tokens llama_print_timings: total time = 1073200.45 ms / 24 tokens llama_print_timings: total time = 1067136.00 ms / 24 tokens llama_print_timings: total time = 1076442.56 ms / 24 tokens llama_print_timings: total time = 1004142.64 ms / 24 tokens llama_print_timings: total time = 1042942.65 ms / 24 tokens llama_print_timings: total time = 999933.76 ms / 24 tokens llama_print_timings: total time = 1046548.83 ms / 24 tokens llama_print_timings: total time = 1068671.48 ms / 24 tokens llama_print_timings: total time = 1068285.76 ms / 24 tokens llama_print_timings: total time = 1077789.63 ms / 24 tokens llama_print_timings: total time = 1071558.93 ms / 24 tokens llama_print_timings: total time = 1066181.55 ms / 24 tokens llama_print_timings: total time = 1076767.53 ms / 24 tokens llama_print_timings: total time = 1004065.63 ms / 24 tokens llama_print_timings: total time = 1044522.13 ms / 24 tokens llama_print_timings: total time = 999725.33 ms / 24 tokens llama_print_timings: total time = 1047510.77 ms / 24 tokens llama_print_timings: total time = 1068010.27 ms / 24 tokens llama_print_timings: total time = 1068999.31 ms / 24 tokens llama_print_timings: total time = 1077648.05 ms / 24 tokens llama_print_timings: total time = 1071378.96 ms / 24 tokens llama_print_timings: total time = 1066326.32 ms / 24 tokens llama_print_timings: total time = 1077088.92 ms / 24 tokensAFTER ----- llama_print_timings: total time = 988522.03 ms / 24 tokens llama_print_timings: total time = 997204.52 ms / 24 tokens llama_print_timings: total time = 996605.86 ms / 24 tokens llama_print_timings: total time = 991985.50 ms / 24 tokens llama_print_timings: total time = 1035143.31 ms / 24 tokens llama_print_timings: total time = 993660.18 ms / 24 tokens llama_print_timings: total time = 983082.14 ms / 24 tokens llama_print_timings: total time = 990431.36 ms / 24 tokens llama_print_timings: total time = 992707.09 ms / 24 tokens llama_print_timings: total time = 992673.27 ms / 24 tokens llama_print_timings: total time = 989285.43 ms / 24 tokens llama_print_timings: total time = 996710.06 ms / 24 tokens llama_print_timings: total time = 996534.64 ms / 24 tokens llama_print_timings: total time = 991344.17 ms / 24 tokens llama_print_timings: total time = 1035210.84 ms / 24 tokens llama_print_timings: total time = 994714.13 ms / 24 tokens llama_print_timings: total time = 984184.15 ms / 24 tokens llama_print_timings: total time = 990909.45 ms / 24 tokens llama_print_timings: total time = 991881.48 ms / 24 tokens llama_print_timings: total time = 993918.03 ms / 24 tokens llama_print_timings: total time = 990061.34 ms / 24 tokens llama_print_timings: total time = 998076.69 ms / 24 tokens llama_print_timings: total time = 997082.59 ms / 24 tokens llama_print_timings: total time = 990677.58 ms / 24 tokens llama_print_timings: total time = 1036054.94 ms / 24 tokens llama_print_timings: total time = 994125.93 ms / 24 tokens llama_print_timings: total time = 982467.01 ms / 24 tokens llama_print_timings: total time = 990191.60 ms / 24 tokens llama_print_timings: total time = 993319.24 ms / 24 tokens llama_print_timings: total time = 992540.57 ms / 24 tokens 2. tlb shootdowns from 'cat /proc/interrupts': BEFORE ------ TLB: 125553646 141418810 161932620 176853972 186655697 190399283 192143823 196414038 192872439 193313658 193395617 192521416 190788161 195067598 198016061 193607347 194293972 190786732 191545637 194856822 191801931 189634535 190399803 196365922 195268398 190115840 188050050 193194908 195317617 190820190 190164820 185556071 226797214 229592631 216112464 209909495 205575979 205950252 204948111 197999795 198892232 205287952 199344631 195015158 195869844 198858745 195692876 200961904 203463252 205921722 199850838 206145986 199613202 199961345 200129577 203020521 207873649 203697671 197093386 204243803 205993323 200934664 204193128 194435376 TLB shootdowns AFTER ----- TLB: 5648092 6610142 7032849 7882308 8088518 8352310 8656536 8705136 8647426 8905583 8985408 8704522 8884344 9026261 8929974 8869066 8877575 8810096 8770984 8754503 8801694 8865925 8787524 8656432 8755912 8682034 8773935 8832925 8797997 8515777 8481240 8891258 10595243 10285973 9756935 9573681 9398968 9069244 9242984 8899009 9310690 9029095 9069758 9105825 9092703 9270202 9460287 9258546 9180415 9232723 9270611 9175020 9490420 9360316 9420818 9057663 9525631 9310152 9152242 8654483 9181804 9050847 8919916 8883856 TLB shootdowns 3. tlb numbers from 'perf stat' per test set: BEFORE ------ 3163679332 dTLB-load-misses 2017751856 dTLB-store-misses 327092903 iTLB-load-misses 1357543886 tlb:tlb_flush AFTER ----- 2394694609 dTLB-load-misses 861144167 dTLB-store-misses 64055579 iTLB-load-misses 69175002 tlb:tlb_flush---
Changes from v9:
1. Expand the candidate to apply this mechanism:
BEFORE - The souce folios at any type of migration. AFTER - Any folios that have been unmapped and freed.
2. Change the workload for test:
BEFORE - XSBench
AFTER - llama.cpp (one of the most popluar real workload)3. Change the test environment:
BEFORE - qemu machine, too small DRAM(1GB), large remote mem AFTER - bare metal, real CXL memory, practical memory size
4. Rename the mechanism from MIGRC(Migration Read Copy) to LUF(Lazy Unmap Flush) to reflect the current version of the mechanism can be applied not only to unmap during migration but any unmap code e.g. unmap in
shrink_folio_list().5. Fix build error for riscv. (feedbacked by kernel test bot)
6. Supplement commit messages to describe what this mechanism is for, especially in the patches for arch code. (feedbacked by Thomas Gleixner)
7. Clean up some trivial things.
Changes from v8:
1. Rebase on akpm/mm.git mm-unstable as of April 18, 2024.
2. Supplement comments and commit message.
3. Change the candidate to apply migrc mechanism:BEFORE - The source folios at demotion and promotion. AFTER - The souce folios at any type of migration.
4. Change how migrc mechanism works:
BEFORE - Reduce tlb flushes by deferring
folio_free()for
source folios during demotion and promotion.
AFTER - Reduce tlb flushes by deferring tlb flush until they
actually become used, out of pcp or buddy. The current version of migrc does *not* defer callingfolio_free()but let it go as it is as the same as vanilla kernel, with the folios marked kind of 'need to tlb flush'. And then handle the flush when the page exits from pcp or buddy so as to prevent changing vm stats e.g. free pages.Changes from v7:
1. Rewrite cover letter to explain what 'migrc' mechasism is. (feedbacked by Andrew Morton)
2. Supplement the commit message of a patch 'mm: Add APIs to free a folio directly to the buddy bypassing pcp'. (feedbacked by Andrew Morton)Changes from v6:
1. Fix build errors in case of
CONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSHdisabled by moving migrc_flush_{start,end}() calls from arch code totry_to_unmap_flush()inmm/rmap.c.Changes from v5:
1. Fix build errors in case of
CONFIG_MIGRATIONdisabled orCONFIG_HWPOISON_INJECTmoduled. (feedbacked by kernel test bot and Raymond Jay Golo)
2. Organize migrc code with two kconfigs,CONFIG_MIGRATIONandCONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH.Changes from v4:
1. Rebase on v6.7.
2. Fix build errors in arm64 that is doing nothing for tlb flush but hasCONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH. (reported by kernel test robot)
3. Don't use any page flag. So the system would give up migrc mechanism more often but it's okay. The final improvement is good enough.
4. Instead, optimize full tlb flush(arch_tlbbatch_flush()) by avoiding redundant CPUs from tlb flush.Changes from v3:
1. Don't use the kconfig,
CONFIG_MIGRC, and remove sysctl knob,migrc_enable. (feedbacked by Nadav)
2. Remove the optimization skipping CPUs that have already performed tlb flushes needed by any reason when performing tlb flushes by migrc because I can't tell the performance difference between w/ the optimization and w/o that. (feedbacked by Nadav)
3. Minimize arch-specific code. While at it, move all the migrc declarations and inline functions frominclude/linux/mm.htomm/internal.h(feedbacked by Dave Hansen, Nadav)
4. Separate a part making migrc paused when the system is in high memory pressure to another patch. (feedbacked by Nadav)
5. Rename:
a.arch_tlbbatch_clean()toarch_tlbbatch_clear(), b.tlb_ubc_nowrtotlb_ubc_ro,
c.migrc_try_flush_free_folios()tomigrc_flush_free_folios(), d.migrc_stoptomigrc_pause.
(feedbacked by Nadav)
6. Use ->lrulist_headinstead of introducing a newllist_head. (feedbacked by Nadav)
7. Use non-atomic operations of page-flag when it's safe. (feedbacked by Nadav)
8. Use stack instead of keeping a pointer of 'structmigrc_req' in struct task, which is for manipulating it locally. (feedbacked by Nadav)
9. Replace a lot of simple functions to inline functions placed in a header,mm/internal.h. (feedbacked by Nadav)
10. Add additional sufficient comments. (feedbacked by Nadav)
11. Remove a lot of wrapper functions. (feedbacked by Nadav)Changes from RFC v2:
1. Remove additional occupation in struct page. To do that, unioned with lru field for migrc's list and added a page flag. I know page flag is a thing that we don't like to add but no choice because migrc should distinguish folios under migrc's control from others. Instead, I force migrc to be used only on 64 bit system to mitigate you guys from getting angry.
2. Remove meaningless internal object allocator that I introduced to minimize impact onto the system. However, a ton of tests showed there was no difference.
3. Stop migrc from working when the system is in high memory pressure like about to perform direct reclaim. At the condition where the swap mechanism is heavily used, I found the system suffered from regression without this control.
4. Exclude folios thatpte_dirty()== true from migrc's interest so that migrc can work simpler.
5. Combine several patches that work tightly coupled to one.
6. Add sufficient comments for better review.
7. Manage migrc's request in per-node manner (from globally).
8. Add tlb miss improvement in commit message.
9. Test with moreCPUs(4 -> 16)to see bigger improvement.Changes from RFC:
1. Fix a bug triggered when a destination folio at the previous migration becomes a source folio at the next migration, before the folio gets handled properly so that the folio can play with another migration. There was inconsistency in the folio's state. Fixed it.
2. Split the patch set into more pieces so that the folks can review better. (Feedbacked by Nadav Amit)
3. Fix a wrong usage of barrier e.g.smp_mb__after_atomic(). (Feedbacked by Nadav Amit)
4. Tried to add sufficient comments to explain the patch set better. (Feedbacked by Nadav Amit)Byungchul Park (12):
x86/tlb: add APIs manipulating tlb batch's arch data
arm64: tlbflush: add APIs manipulating tlb batch's arch data riscv, tlb: add APIs manipulating tlb batch's arch data x86/tlb, riscv/tlb, mm/rmap: separatearch_tlbbatch_clear()out of
arch_tlbbatch_flush()
mm: buddy: make room for a new variable, ugen, in struct page
mm: addfolio_put_ugen()to deliver unmap generation number to pcp or
buddy
mm: add a parameter, unmap generation number, tofree_unref_folios()mm/rmap: recognize read-only tlb entries during batched tlb flush
mm: implement LUF(Lazy Unmap Flush) defering tlb flush when folios get
unmapped
mm: separate move/undo parts frommigrate_pages_batch()mm, migrate: apply luf mechanism to unmapping during migration mm, vmscan: apply luf mechanism to unmapping during folio reclaimarch/arm64/include/asm/tlbflush.h | 18 ++ arch/riscv/include/asm/tlbflush.h | 21 ++ arch/riscv/mm/tlbflush.c | 1 - arch/x86/include/asm/tlbflush.h | 18 ++ arch/x86/mm/tlb.c | 2 - include/linux/mm.h | 22 ++ include/linux/mm_types.h | 40 +++- include/linux/rmap.h | 7 +- include/linux/sched.h | 11 + mm/compaction.c | 10 + mm/internal.h | 115 +++++++++- mm/memory.c | 8 + mm/migrate.c | 184 ++++++++++------ mm/mmap.c | 8 + mm/page_alloc.c | 157 +++++++++++--- mm/page_isolation.c | 6 + mm/page_reporting.c | 10 + mm/rmap.c | 345 +++++++++++++++++++++++++++++- mm/swap.c | 18 +- mm/vmscan.c | 29 ++- 20 files changed, 904 insertions(+), 126 deletions(-)
Quoted 1 line
base-commit:
f52bcd4a9f6058704a6f6b6b50418f579defd4fe
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 42 lines · Ying
Quoted 40 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:https://lore.kernel.org/lkml/[email protected]/
However, it's only for migration through hinting fault. I thought it'd be much better if we have a general mechanism to reduce all the tlb numbers that we can apply to any unmap code, that we normally believe tlb flush should be followed.
I'm suggesting a new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
tlb flush can be defered when folios get unmapped as long as it guarantees to perform tlb flush needed, before the folios actually become used, of course, only if all the corresponding ptes don't have write permission. Otherwise, the system will get messed up.
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through
mmap(), give up luf mechanism and perform tlb flush required right away.
Quoted 6 lines
Quoted 3 lines
No matter what type of workload is used for performance evaluation, the result would be positive thanks to the unconditional reduction of tlb flushes, tlb misses and interrupts.
Are there any downsides of the optimization? Will it cause regression for workloads with almost no read-only mappings? Will it cause
IMHO, no. LUF does almost nothing for folios writable mapped.
Quoted 1 line
regression for page allocation?
TLB flush might be added in prep_new_page() if pended, however, that should have been already performed anyway. It's not additional overhead.
The TLB flush even can be skipped by the batched work.
Quoted 5 lines
Quoted 3 lines
For the test, I picked up one of
the most popular and heavy workload, llama.cpp that is a LLM(Large Language Model) inference engine.IIUC, llama.cpp is a workload with huge read-only mapping.
Right. LUF works on read-only mapping. So the more read-only mappings are used, the better LUF works. Fortunately, such a workload with huge read-only mappings that is normal in a light weight inference engine, is quite popular these days in the era of LLM.
Byungchul
Quoted 353 lines
Quoted 348 lines
The result would depend on memory latency and how often reclaim runs, which implies tlb miss overhead and how many times unmapping happens. In my system, the result shows:
1. tlb flushes are reduced about 95%.
2. tlb misses(itlb) are reduced about 80%.
3. tlb misses(dtlb store) are reduced about 57%.
4. tlb misses(dtlb load) are reduced about 24%.
5. tlb shootdown interrupts are reduced about 95%.
6. The test program runtime is reduced about 5%.The test environment and the result is like:
Machine: bare metal,
x86_64, Intel(R) Xeon(R) Gold 6430CPU: 1 socket 64 core with hyper thread on Numa: 2 nodes (64 CPUs DRAM 42GB, no CPUs CXL expander 98GB) Config: swap off, numa balancing tiering on, demotion enabledThe test set:
llama.cpp/main -m $(70G_model1) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model2) -p "who are you?" -s 1 -t 15 -n 20 & llama.cpp/main -m $(70G_model3) -p "who are you?" -s 1 -t 15 -n 20 & wait
where -t: nr of threads, -s: seed used to make the runtime stable, -n: nr of tokens that determines the runtime, -p: prompt to ask, -m: LLM model to use.
Run the test set 10 times successively with caches dropped every run via 'echo 3 >
/proc/sys/vm/drop_caches'. Each inference prints its runtime at the end of each.1. Runtime from the output of llama.cpp:
BEFORE
------llama_print_timings: total time = 1002461.95 ms / 24 tokens llama_print_timings: total time = 1044978.38 ms / 24 tokens llama_print_timings: total time = 1000653.09 ms / 24 tokens llama_print_timings: total time = 1047104.80 ms / 24 tokens llama_print_timings: total time = 1069430.36 ms / 24 tokens llama_print_timings: total time = 1068201.16 ms / 24 tokens llama_print_timings: total time = 1078092.59 ms / 24 tokens llama_print_timings: total time = 1073200.45 ms / 24 tokens llama_print_timings: total time = 1067136.00 ms / 24 tokens llama_print_timings: total time = 1076442.56 ms / 24 tokens llama_print_timings: total time = 1004142.64 ms / 24 tokens llama_print_timings: total time = 1042942.65 ms / 24 tokens llama_print_timings: total time = 999933.76 ms / 24 tokens llama_print_timings: total time = 1046548.83 ms / 24 tokens llama_print_timings: total time = 1068671.48 ms / 24 tokens llama_print_timings: total time = 1068285.76 ms / 24 tokens llama_print_timings: total time = 1077789.63 ms / 24 tokens llama_print_timings: total time = 1071558.93 ms / 24 tokens llama_print_timings: total time = 1066181.55 ms / 24 tokens llama_print_timings: total time = 1076767.53 ms / 24 tokens llama_print_timings: total time = 1004065.63 ms / 24 tokens llama_print_timings: total time = 1044522.13 ms / 24 tokens llama_print_timings: total time = 999725.33 ms / 24 tokens llama_print_timings: total time = 1047510.77 ms / 24 tokens llama_print_timings: total time = 1068010.27 ms / 24 tokens llama_print_timings: total time = 1068999.31 ms / 24 tokens llama_print_timings: total time = 1077648.05 ms / 24 tokens llama_print_timings: total time = 1071378.96 ms / 24 tokens llama_print_timings: total time = 1066326.32 ms / 24 tokens llama_print_timings: total time = 1077088.92 ms / 24 tokensAFTER ----- llama_print_timings: total time = 988522.03 ms / 24 tokens llama_print_timings: total time = 997204.52 ms / 24 tokens llama_print_timings: total time = 996605.86 ms / 24 tokens llama_print_timings: total time = 991985.50 ms / 24 tokens llama_print_timings: total time = 1035143.31 ms / 24 tokens llama_print_timings: total time = 993660.18 ms / 24 tokens llama_print_timings: total time = 983082.14 ms / 24 tokens llama_print_timings: total time = 990431.36 ms / 24 tokens llama_print_timings: total time = 992707.09 ms / 24 tokens llama_print_timings: total time = 992673.27 ms / 24 tokens llama_print_timings: total time = 989285.43 ms / 24 tokens llama_print_timings: total time = 996710.06 ms / 24 tokens llama_print_timings: total time = 996534.64 ms / 24 tokens llama_print_timings: total time = 991344.17 ms / 24 tokens llama_print_timings: total time = 1035210.84 ms / 24 tokens llama_print_timings: total time = 994714.13 ms / 24 tokens llama_print_timings: total time = 984184.15 ms / 24 tokens llama_print_timings: total time = 990909.45 ms / 24 tokens llama_print_timings: total time = 991881.48 ms / 24 tokens llama_print_timings: total time = 993918.03 ms / 24 tokens llama_print_timings: total time = 990061.34 ms / 24 tokens llama_print_timings: total time = 998076.69 ms / 24 tokens llama_print_timings: total time = 997082.59 ms / 24 tokens llama_print_timings: total time = 990677.58 ms / 24 tokens llama_print_timings: total time = 1036054.94 ms / 24 tokens llama_print_timings: total time = 994125.93 ms / 24 tokens llama_print_timings: total time = 982467.01 ms / 24 tokens llama_print_timings: total time = 990191.60 ms / 24 tokens llama_print_timings: total time = 993319.24 ms / 24 tokens llama_print_timings: total time = 992540.57 ms / 24 tokens 2. tlb shootdowns from 'cat /proc/interrupts': BEFORE ------ TLB: 125553646 141418810 161932620 176853972 186655697 190399283 192143823 196414038 192872439 193313658 193395617 192521416 190788161 195067598 198016061 193607347 194293972 190786732 191545637 194856822 191801931 189634535 190399803 196365922 195268398 190115840 188050050 193194908 195317617 190820190 190164820 185556071 226797214 229592631 216112464 209909495 205575979 205950252 204948111 197999795 198892232 205287952 199344631 195015158 195869844 198858745 195692876 200961904 203463252 205921722 199850838 206145986 199613202 199961345 200129577 203020521 207873649 203697671 197093386 204243803 205993323 200934664 204193128 194435376 TLB shootdowns AFTER ----- TLB: 5648092 6610142 7032849 7882308 8088518 8352310 8656536 8705136 8647426 8905583 8985408 8704522 8884344 9026261 8929974 8869066 8877575 8810096 8770984 8754503 8801694 8865925 8787524 8656432 8755912 8682034 8773935 8832925 8797997 8515777 8481240 8891258 10595243 10285973 9756935 9573681 9398968 9069244 9242984 8899009 9310690 9029095 9069758 9105825 9092703 9270202 9460287 9258546 9180415 9232723 9270611 9175020 9490420 9360316 9420818 9057663 9525631 9310152 9152242 8654483 9181804 9050847 8919916 8883856 TLB shootdowns 3. tlb numbers from 'perf stat' per test set: BEFORE ------ 3163679332 dTLB-load-misses 2017751856 dTLB-store-misses 327092903 iTLB-load-misses 1357543886 tlb:tlb_flush AFTER ----- 2394694609 dTLB-load-misses 861144167 dTLB-store-misses 64055579 iTLB-load-misses 69175002 tlb:tlb_flush---
Changes from v9:
1. Expand the candidate to apply this mechanism:
BEFORE - The souce folios at any type of migration. AFTER - Any folios that have been unmapped and freed.
2. Change the workload for test:
BEFORE - XSBench
AFTER - llama.cpp (one of the most popluar real workload)3. Change the test environment:
BEFORE - qemu machine, too small DRAM(1GB), large remote mem AFTER - bare metal, real CXL memory, practical memory size
4. Rename the mechanism from MIGRC(Migration Read Copy) to LUF(Lazy Unmap Flush) to reflect the current version of the mechanism can be applied not only to unmap during migration but any unmap code e.g. unmap in
shrink_folio_list().5. Fix build error for riscv. (feedbacked by kernel test bot)
6. Supplement commit messages to describe what this mechanism is for, especially in the patches for arch code. (feedbacked by Thomas Gleixner)
7. Clean up some trivial things.
Changes from v8:
1. Rebase on akpm/mm.git mm-unstable as of April 18, 2024.
2. Supplement comments and commit message.
3. Change the candidate to apply migrc mechanism:BEFORE - The source folios at demotion and promotion. AFTER - The souce folios at any type of migration.
4. Change how migrc mechanism works:
BEFORE - Reduce tlb flushes by deferring
folio_free()for
source folios during demotion and promotion.
AFTER - Reduce tlb flushes by deferring tlb flush until they
actually become used, out of pcp or buddy. The current version of migrc does *not* defer callingfolio_free()but let it go as it is as the same as vanilla kernel, with the folios marked kind of 'need to tlb flush'. And then handle the flush when the page exits from pcp or buddy so as to prevent changing vm stats e.g. free pages.Changes from v7:
1. Rewrite cover letter to explain what 'migrc' mechasism is. (feedbacked by Andrew Morton)
2. Supplement the commit message of a patch 'mm: Add APIs to free a folio directly to the buddy bypassing pcp'. (feedbacked by Andrew Morton)Changes from v6:
1. Fix build errors in case of
CONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSHdisabled by moving migrc_flush_{start,end}() calls from arch code totry_to_unmap_flush()inmm/rmap.c.Changes from v5:
1. Fix build errors in case of
CONFIG_MIGRATIONdisabled orCONFIG_HWPOISON_INJECTmoduled. (feedbacked by kernel test bot and Raymond Jay Golo)
2. Organize migrc code with two kconfigs,CONFIG_MIGRATIONandCONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH.Changes from v4:
1. Rebase on v6.7.
2. Fix build errors in arm64 that is doing nothing for tlb flush but hasCONFIG_ARCH_WANT_BATCHED_UNMAP_tlb_FLUSH. (reported by kernel test robot)
3. Don't use any page flag. So the system would give up migrc mechanism more often but it's okay. The final improvement is good enough.
4. Instead, optimize full tlb flush(arch_tlbbatch_flush()) by avoiding redundant CPUs from tlb flush.Changes from v3:
1. Don't use the kconfig,
CONFIG_MIGRC, and remove sysctl knob,migrc_enable. (feedbacked by Nadav)
2. Remove the optimization skipping CPUs that have already performed tlb flushes needed by any reason when performing tlb flushes by migrc because I can't tell the performance difference between w/ the optimization and w/o that. (feedbacked by Nadav)
3. Minimize arch-specific code. While at it, move all the migrc declarations and inline functions frominclude/linux/mm.htomm/internal.h(feedbacked by Dave Hansen, Nadav)
4. Separate a part making migrc paused when the system is in high memory pressure to another patch. (feedbacked by Nadav)
5. Rename:
a.arch_tlbbatch_clean()toarch_tlbbatch_clear(), b.tlb_ubc_nowrtotlb_ubc_ro,
c.migrc_try_flush_free_folios()tomigrc_flush_free_folios(), d.migrc_stoptomigrc_pause.
(feedbacked by Nadav)
6. Use ->lrulist_headinstead of introducing a newllist_head. (feedbacked by Nadav)
7. Use non-atomic operations of page-flag when it's safe. (feedbacked by Nadav)
8. Use stack instead of keeping a pointer of 'structmigrc_req' in struct task, which is for manipulating it locally. (feedbacked by Nadav)
9. Replace a lot of simple functions to inline functions placed in a header,mm/internal.h. (feedbacked by Nadav)
10. Add additional sufficient comments. (feedbacked by Nadav)
11. Remove a lot of wrapper functions. (feedbacked by Nadav)Changes from RFC v2:
1. Remove additional occupation in struct page. To do that, unioned with lru field for migrc's list and added a page flag. I know page flag is a thing that we don't like to add but no choice because migrc should distinguish folios under migrc's control from others. Instead, I force migrc to be used only on 64 bit system to mitigate you guys from getting angry.
2. Remove meaningless internal object allocator that I introduced to minimize impact onto the system. However, a ton of tests showed there was no difference.
3. Stop migrc from working when the system is in high memory pressure like about to perform direct reclaim. At the condition where the swap mechanism is heavily used, I found the system suffered from regression without this control.
4. Exclude folios thatpte_dirty()== true from migrc's interest so that migrc can work simpler.
5. Combine several patches that work tightly coupled to one.
6. Add sufficient comments for better review.
7. Manage migrc's request in per-node manner (from globally).
8. Add tlb miss improvement in commit message.
9. Test with moreCPUs(4 -> 16)to see bigger improvement.Changes from RFC:
1. Fix a bug triggered when a destination folio at the previous migration becomes a source folio at the next migration, before the folio gets handled properly so that the folio can play with another migration. There was inconsistency in the folio's state. Fixed it.
2. Split the patch set into more pieces so that the folks can review better. (Feedbacked by Nadav Amit)
3. Fix a wrong usage of barrier e.g.smp_mb__after_atomic(). (Feedbacked by Nadav Amit)
4. Tried to add sufficient comments to explain the patch set better. (Feedbacked by Nadav Amit)Byungchul Park (12):
x86/tlb: add APIs manipulating tlb batch's arch data
arm64: tlbflush: add APIs manipulating tlb batch's arch data riscv, tlb: add APIs manipulating tlb batch's arch data x86/tlb, riscv/tlb, mm/rmap: separatearch_tlbbatch_clear()out of
arch_tlbbatch_flush()
mm: buddy: make room for a new variable, ugen, in struct page
mm: addfolio_put_ugen()to deliver unmap generation number to pcp or
buddy
mm: add a parameter, unmap generation number, tofree_unref_folios()mm/rmap: recognize read-only tlb entries during batched tlb flush
mm: implement LUF(Lazy Unmap Flush) defering tlb flush when folios get
unmapped
mm: separate move/undo parts frommigrate_pages_batch()mm, migrate: apply luf mechanism to unmapping during migration mm, vmscan: apply luf mechanism to unmapping during folio reclaimarch/arm64/include/asm/tlbflush.h | 18 ++ arch/riscv/include/asm/tlbflush.h | 21 ++ arch/riscv/mm/tlbflush.c | 1 - arch/x86/include/asm/tlbflush.h | 18 ++ arch/x86/mm/tlb.c | 2 - include/linux/mm.h | 22 ++ include/linux/mm_types.h | 40 +++- include/linux/rmap.h | 7 +- include/linux/sched.h | 11 + mm/compaction.c | 10 + mm/internal.h | 115 +++++++++- mm/memory.c | 8 + mm/migrate.c | 184 ++++++++++------ mm/mmap.c | 8 + mm/page_alloc.c | 157 +++++++++++--- mm/page_isolation.c | 6 + mm/page_reporting.c | 10 + mm/rmap.c | 345 +++++++++++++++++++++++++++++- mm/swap.c | 18 +- mm/vmscan.c | 29 ++- 20 files changed, 904 insertions(+), 126 deletions(-)Quoted 1 line
base-commit:
f52bcd4a9f6058704a6f6b6b50418f579defd4fe--
Best Regards,
Huang, Ying
Quoted 10 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:
And, I still have interest of the performance impact of commit 7e12beb8ca2a ("migrate_pages: batch flushing TLB"). In the email above, you said that the performance of v6.5-rc5 + 7e12beb8ca2a reverted has better performance than v6.5-rc5. Can you provide more details? For example, the number of TLB flushing IPI for two kernels?
I should have followed up the above email. Sorry about that. Anyway, we should try to fix issue of that commit too.
--
Best Regards,
Huang, Ying
[snip]
Byungchul Park
reviewerQuoted 11 lines · Ying
Quoted 9 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:
Quoted 7 lines
Quoted 1 line
And, I still have interest of the performance impact of commit
7e12beb8ca2a("migrate_pages: batch flushing TLB"). In the email above, you said that the performance of v6.5-rc5 +7e12beb8ca2areverted has better performance than v6.5-rc5. Can you provide more details? For example, the number of TLB flushing IPI for two kernels?
Okay. I will test and share the result with what you asked me now once I get available for the test.
Byungchul
Quoted 8 lines
I should have followed up the above email. Sorry about that. Anyway, we should try to fix issue of that commit too.
--
Best Regards,
Huang, Ying[snip]
Byungchul Park
reviewerQuoted 14 lines · Byungchul Park
Quoted 13 lines · Ying
Quoted 10 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:
Quoted 8 lines
Quoted 5 lines
And, I still have interest of the performance impact of commit
7e12beb8ca2a("migrate_pages: batch flushing TLB"). In the email above, you said that the performance of v6.5-rc5 +7e12beb8ca2areverted has better performance than v6.5-rc5. Can you provide more details? For example, the number of TLB flushing IPI for two kernels?Okay. I will test and share the result with what you asked me now once I get available for the test.
I should admit that the test using qemu is so unstable. While using qemu for the test, kernel with 7e12beb8ca2a applied gave better results sometimes and worse ones sometimes. I should've used a bare metal from the beginning. Sorry for making you confused with the unstable result.
Since I thought you asked me for the test with the same environment in the link above, I used qemu to reproduce the similar result but changed the number of threads for the test from 16 to 14 to get rid of noise that might be introduced by other than the intended test just in case.
As expected, the stats are better with your work:
------------------------------------------
v6.6-rc5 with 7e12beb8ca2a commit reverted
------------------------------------------
1) from output of XSBench
Threads: 14 Runtime: 1127.043 seconds Lookups: 1,700,000,000 Lookups/s: 1,508,371
2) from /proc/vmstat
numa_hit 15580171
numa_miss 1034233
numa_foreign 1034233
numa_interleave 773
numa_local 7927442
numa_other 8686962
numa_pte_updates 24068923
numa_hint_faults 24061125
numa_hint_faults_local 0
numa_pages_migrated 7426480
pgmigrate_success 15407375
pgmigrate_fail 1849
compact_migrate_scanned 4445414
compact_daemon_migrate_scanned 4445414
pgdemote_kswapd 7651061
pgdemote_direct 0
nr_tlb_remote_flush 8080092
nr_tlb_remote_flush_received 109915713
nr_tlb_local_flush_all 53800
nr_tlb_local_flush_one 770466
3) from /proc/interrupts
TLB: 8022927 7840769 123588 7837008 7835967 7839837
7838332 7839886 7837610 7837221 7834524 407260 7430090 7835696 7839081 7712568 TLB shootdowns
4) from 'perf stat -a'
222371217 itlb.itlb_flush
919832520 tlb_flush.dtlb_thread
372223809 tlb_flush.stlb_any
120210808042 dTLB-load-misses
979352769 dTLB-store-misses
3650767665 iTLB-load-misses
-----------------------------------------
v6.6-rc5 with 7e12beb8ca2a commit applied
-----------------------------------------
1) from output of XSBench
Threads: 14 Runtime: 1105.521 seconds Lookups: 1,700,000,000 Lookups/s: 1,537,737
2) from /proc/vmstat
numa_hit 24148399
numa_miss 797483
numa_foreign 797483
numa_interleave 772
numa_local 12214575
numa_other 12731307
numa_pte_updates 24250278
numa_hint_faults 24199756
numa_hint_faults_local 0
numa_pages_migrated 11476195
pgmigrate_success 23634639
pgmigrate_fail 1391
compact_migrate_scanned 3760803
compact_daemon_migrate_scanned 3760803
pgdemote_kswapd 11932217
pgdemote_direct 0
nr_tlb_remote_flush 2151945
nr_tlb_remote_flush_received 29672808
nr_tlb_local_flush_all 124006
nr_tlb_local_flush_one 741165
3) from /proc/interrupts
TLB: 2130784 2120142 2117571 844962 2071766 114675
2117258 2119596 2116816 1205446 2119176 2119209 2116792 2118763 2118773 2117762 TLB shootdowns
4) from 'perf stat -a'
60851902 itlb.itlb_flush
334068491 tlb_flush.dtlb_thread
223732916 tlb_flush.stlb_any
120207083382 dTLB-load-misses
446823059 dTLB-store-misses
1926669373 iTLB-load-misses
---
Byungchul
Hi, Byungchul,
Quoted 125 lines · Byungchul Park
Quoted 14 lines · Byungchul Park
Quoted 13 lines · Ying
Quoted 10 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:Quoted 8 lines
Quoted 5 lines
And, I still have interest of the performance impact of commit
7e12beb8ca2a("migrate_pages: batch flushing TLB"). In the email above, you said that the performance of v6.5-rc5 +7e12beb8ca2areverted has better performance than v6.5-rc5. Can you provide more details? For example, the number of TLB flushing IPI for two kernels?Okay. I will test and share the result with what you asked me now once I get available for the test.
I should admit that the test using qemu is so unstable. While using qemu for the test, kernel with
7e12beb8ca2aapplied gave better results sometimes and worse ones sometimes. I should've used a bare metal from the beginning. Sorry for making you confused with the unstable result.Since I thought you asked me for the test with the same environment in the link above, I used qemu to reproduce the similar result but changed the number of threads for the test from 16 to 14 to get rid of noise that might be introduced by other than the intended test just in case.
As expected, the stats are better with your work:
------------------------------------------ v6.6-rc5 with 7e12beb8ca2a commit reverted ------------------------------------------ 1) from output of XSBench Threads: 14 Runtime: 1127.043 seconds Lookups: 1,700,000,000 Lookups/s: 1,508,371 2) from /proc/vmstat numa_hit 15580171 numa_miss 1034233 numa_foreign 1034233 numa_interleave 773 numa_local 7927442 numa_other 8686962 numa_pte_updates 24068923 numa_hint_faults 24061125 numa_hint_faults_local 0 numa_pages_migrated 7426480 pgmigrate_success 15407375 pgmigrate_fail 1849 compact_migrate_scanned 4445414 compact_daemon_migrate_scanned 4445414 pgdemote_kswapd 7651061 pgdemote_direct 0 nr_tlb_remote_flush 8080092 nr_tlb_remote_flush_received 109915713 nr_tlb_local_flush_all 53800 nr_tlb_local_flush_one 770466 3) from /proc/interrupts TLB: 8022927 7840769 123588 7837008 7835967 7839837 7838332 7839886 7837610 7837221 7834524 407260 7430090 7835696 7839081 7712568 TLB shootdowns 4) from 'perf stat -a' 222371217 itlb.itlb_flush 919832520 tlb_flush.dtlb_thread 372223809 tlb_flush.stlb_any 120210808042 dTLB-load-misses 979352769 dTLB-store-misses 3650767665 iTLB-load-misses ----------------------------------------- v6.6-rc5 with 7e12beb8ca2a commit applied ----------------------------------------- 1) from output of XSBench Threads: 14 Runtime: 1105.521 seconds Lookups: 1,700,000,000 Lookups/s: 1,537,737 2) from /proc/vmstat numa_hit 24148399 numa_miss 797483 numa_foreign 797483 numa_interleave 772 numa_local 12214575 numa_other 12731307 numa_pte_updates 24250278 numa_hint_faults 24199756 numa_hint_faults_local 0 numa_pages_migrated 11476195 pgmigrate_success 23634639 pgmigrate_fail 1391 compact_migrate_scanned 3760803 compact_daemon_migrate_scanned 3760803 pgdemote_kswapd 11932217 pgdemote_direct 0 nr_tlb_remote_flush 2151945 nr_tlb_remote_flush_received 29672808 nr_tlb_local_flush_all 124006 nr_tlb_local_flush_one 741165 3) from /proc/interrupts TLB: 2130784 2120142 2117571 844962 2071766 114675 2117258 2119596 2116816 1205446 2119176 2119209 2116792 2118763 2118773 2117762 TLB shootdowns
Quoted 9 lines
4) from 'perf stat -a' 60851902 itlb.itlb_flush 334068491 tlb_flush.dtlb_thread 223732916 tlb_flush.stlb_any 120207083382 dTLB-load-misses 446823059 dTLB-store-misses 1926669373 iTLB-load-misses
Thanks a lot for test results!
From your test results, the TLB shootdown IPI can be reduced effectively with commit 7e12beb8ca2a. So that the benchmark score improved a little.
And, your changes will reduce the TLB shootdown IPI further, right? Do you have the number?
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 139 lines · Ying
Hi, Byungchul,
Quoted 125 lines · Byungchul Park
Quoted 14 lines · Byungchul Park
Quoted 13 lines · Ying
Quoted 10 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:Quoted 8 lines
Quoted 5 lines
And, I still have interest of the performance impact of commit
7e12beb8ca2a("migrate_pages: batch flushing TLB"). In the email above, you said that the performance of v6.5-rc5 +7e12beb8ca2areverted has better performance than v6.5-rc5. Can you provide more details? For example, the number of TLB flushing IPI for two kernels?Okay. I will test and share the result with what you asked me now once I get available for the test.
I should admit that the test using qemu is so unstable. While using qemu for the test, kernel with
7e12beb8ca2aapplied gave better results sometimes and worse ones sometimes. I should've used a bare metal from the beginning. Sorry for making you confused with the unstable result.Since I thought you asked me for the test with the same environment in the link above, I used qemu to reproduce the similar result but changed the number of threads for the test from 16 to 14 to get rid of noise that might be introduced by other than the intended test just in case.
As expected, the stats are better with your work:
------------------------------------------ v6.6-rc5 with 7e12beb8ca2a commit reverted ------------------------------------------ 1) from output of XSBench Threads: 14 Runtime: 1127.043 seconds Lookups: 1,700,000,000 Lookups/s: 1,508,371 2) from /proc/vmstat numa_hit 15580171 numa_miss 1034233 numa_foreign 1034233 numa_interleave 773 numa_local 7927442 numa_other 8686962 numa_pte_updates 24068923 numa_hint_faults 24061125 numa_hint_faults_local 0 numa_pages_migrated 7426480 pgmigrate_success 15407375 pgmigrate_fail 1849 compact_migrate_scanned 4445414 compact_daemon_migrate_scanned 4445414 pgdemote_kswapd 7651061 pgdemote_direct 0 nr_tlb_remote_flush 8080092 nr_tlb_remote_flush_received 109915713 nr_tlb_local_flush_all 53800 nr_tlb_local_flush_one 770466 3) from /proc/interrupts TLB: 8022927 7840769 123588 7837008 7835967 7839837 7838332 7839886 7837610 7837221 7834524 407260 7430090 7835696 7839081 7712568 TLB shootdowns 4) from 'perf stat -a' 222371217 itlb.itlb_flush 919832520 tlb_flush.dtlb_thread 372223809 tlb_flush.stlb_any 120210808042 dTLB-load-misses 979352769 dTLB-store-misses 3650767665 iTLB-load-misses ----------------------------------------- v6.6-rc5 with 7e12beb8ca2a commit applied ----------------------------------------- 1) from output of XSBench Threads: 14 Runtime: 1105.521 seconds Lookups: 1,700,000,000 Lookups/s: 1,537,737 2) from /proc/vmstat numa_hit 24148399 numa_miss 797483 numa_foreign 797483 numa_interleave 772 numa_local 12214575 numa_other 12731307 numa_pte_updates 24250278 numa_hint_faults 24199756 numa_hint_faults_local 0 numa_pages_migrated 11476195 pgmigrate_success 23634639 pgmigrate_fail 1391 compact_migrate_scanned 3760803 compact_daemon_migrate_scanned 3760803 pgdemote_kswapd 11932217 pgdemote_direct 0 nr_tlb_remote_flush 2151945 nr_tlb_remote_flush_received 29672808 nr_tlb_local_flush_all 124006 nr_tlb_local_flush_one 741165 3) from /proc/interrupts TLB: 2130784 2120142 2117571 844962 2071766 114675 2117258 2119596 2116816 1205446 2119176 2119209 2116792 2118763 2118773 2117762 TLB shootdownsQuoted 9 lines
4) from 'perf stat -a' 60851902 itlb.itlb_flush 334068491 tlb_flush.dtlb_thread 223732916 tlb_flush.stlb_any 120207083382 dTLB-load-misses 446823059 dTLB-store-misses 1926669373 iTLB-load-misses
Quoted 7 lines
Thanks a lot for test results!
Quoted 1 line
From your test results, the TLB shootdown IPI can be reduced effectively
with commit
7e12beb8ca2a. So that the benchmark score improved a little.And, your changes will reduce the TLB shootdown IPI further, right? Do
Yes, right. LUF(Lazy Unmap Flush) reduces TLB shootdown IPI further.
Quoted 1 line
you have the number?
You can find the number obtained from llama.cpp in this cover letter:
https://lore.kernel.org/lkml/[email protected]/
If you meant the number from the same test above, XSBench + qemu, I will re-test with mm-unstable branch of mm tree and share the result shortly.
Byungchul
Byungchul Park
reviewerQuoted 150 lines · Byungchul Park
Quoted 139 lines · Ying
Hi, Byungchul,
Quoted 125 lines · Byungchul Park
Quoted 14 lines · Byungchul Park
Quoted 13 lines · Ying
Quoted 10 lines · Byungchul Park
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:Quoted 8 lines
Quoted 5 lines
And, I still have interest of the performance impact of commit
7e12beb8ca2a("migrate_pages: batch flushing TLB"). In the email above, you said that the performance of v6.5-rc5 +7e12beb8ca2areverted has better performance than v6.5-rc5. Can you provide more details? For example, the number of TLB flushing IPI for two kernels?Okay. I will test and share the result with what you asked me now once I get available for the test.
I should admit that the test using qemu is so unstable. While using qemu for the test, kernel with
7e12beb8ca2aapplied gave better results sometimes and worse ones sometimes. I should've used a bare metal from the beginning. Sorry for making you confused with the unstable result.Since I thought you asked me for the test with the same environment in the link above, I used qemu to reproduce the similar result but changed the number of threads for the test from 16 to 14 to get rid of noise that might be introduced by other than the intended test just in case.
As expected, the stats are better with your work:
------------------------------------------ v6.6-rc5 with 7e12beb8ca2a commit reverted ------------------------------------------ 1) from output of XSBench Threads: 14 Runtime: 1127.043 seconds Lookups: 1,700,000,000 Lookups/s: 1,508,371 2) from /proc/vmstat numa_hit 15580171 numa_miss 1034233 numa_foreign 1034233 numa_interleave 773 numa_local 7927442 numa_other 8686962 numa_pte_updates 24068923 numa_hint_faults 24061125 numa_hint_faults_local 0 numa_pages_migrated 7426480 pgmigrate_success 15407375 pgmigrate_fail 1849 compact_migrate_scanned 4445414 compact_daemon_migrate_scanned 4445414 pgdemote_kswapd 7651061 pgdemote_direct 0 nr_tlb_remote_flush 8080092 nr_tlb_remote_flush_received 109915713 nr_tlb_local_flush_all 53800 nr_tlb_local_flush_one 770466 3) from /proc/interrupts TLB: 8022927 7840769 123588 7837008 7835967 7839837 7838332 7839886 7837610 7837221 7834524 407260 7430090 7835696 7839081 7712568 TLB shootdowns 4) from 'perf stat -a' 222371217 itlb.itlb_flush 919832520 tlb_flush.dtlb_thread 372223809 tlb_flush.stlb_any 120210808042 dTLB-load-misses 979352769 dTLB-store-misses 3650767665 iTLB-load-misses ----------------------------------------- v6.6-rc5 with 7e12beb8ca2a commit applied ----------------------------------------- 1) from output of XSBench Threads: 14 Runtime: 1105.521 seconds Lookups: 1,700,000,000 Lookups/s: 1,537,737 2) from /proc/vmstat numa_hit 24148399 numa_miss 797483 numa_foreign 797483 numa_interleave 772 numa_local 12214575 numa_other 12731307 numa_pte_updates 24250278 numa_hint_faults 24199756 numa_hint_faults_local 0 numa_pages_migrated 11476195 pgmigrate_success 23634639 pgmigrate_fail 1391 compact_migrate_scanned 3760803 compact_daemon_migrate_scanned 3760803 pgdemote_kswapd 11932217 pgdemote_direct 0 nr_tlb_remote_flush 2151945 nr_tlb_remote_flush_received 29672808 nr_tlb_local_flush_all 124006 nr_tlb_local_flush_one 741165 3) from /proc/interrupts TLB: 2130784 2120142 2117571 844962 2071766 114675 2117258 2119596 2116816 1205446 2119176 2119209 2116792 2118763 2118773 2117762 TLB shootdownsQuoted 9 lines
4) from 'perf stat -a' 60851902 itlb.itlb_flush 334068491 tlb_flush.dtlb_thread 223732916 tlb_flush.stlb_any 120207083382 dTLB-load-misses 446823059 dTLB-store-misses 1926669373 iTLB-load-missesQuoted 7 lines
Thanks a lot for test results!
Quoted 1 line
From your test results, the TLB shootdown IPI can be reduced effectively
with commit
7e12beb8ca2a. So that the benchmark score improved a little.And, your changes will reduce the TLB shootdown IPI further, right? Do
Yes, right. LUF(Lazy Unmap Flush) reduces TLB shootdown IPI further.
Quoted 8 lines
Quoted 1 line
you have the number?
You can find the number obtained from llama.cpp in this cover letter:
https://lore.kernel.org/lkml/[email protected]/If you meant the number from the same test above, XSBench + qemu, I will re-test with mm-unstable branch of mm tree and share the result shortly.
I retested the same test but based on a recent mm-unstable branch of mm tree instead of v6.6-rc5. The result changed because of the different base, from v6.6-rc5 to a recent mm-unstable branch of mm tree.
----------------------------------------------------
mm-unstable branch with 7e12beb8ca2a commit reverted
----------------------------------------------------
1) from output of XSBench
Threads: 14 Runtime: 1067.771 seconds Lookups: 1,700,000,000 Lookups/s: 1,592,101
2) from /proc/vmstat
numa_hit 11502876
numa_miss 1130877
numa_foreign 1130877
numa_interleave 115
numa_local 5879006
numa_other 6754747
numa_pte_updates 19390661
numa_hint_faults 19319467
numa_hint_faults_local 0
numa_pages_migrated 5472749
pgmigrate_success 11593079
pgmigrate_fail 549666
compact_migrate_scanned 5408404
compact_daemon_migrate_scanned 5408404
pgdemote_kswapd 5610705
pgdemote_direct 0
nr_tlb_remote_flush 6200106
nr_tlb_remote_flush_received 84362539
nr_tlb_local_flush_all 39202
nr_tlb_local_flush_one 760046
3) from /proc/interrupts
TLB: 3812782 3840646 4806989 5235846 5127512 5048603
6012100 6022642 5088907 5212207 4076329 6014857 6017060 6014964 6009362 6018368 TLB shootdowns
4) from 'perf stat -a'
180449546 itlb.itlb_flush
768913454 tlb_flush.dtlb_thread
304745973 tlb_flush.stlb_any
119589742349 dTLB-load-misses
826525376 dTLB-store-misses
2950724801 iTLB-load-misses
---------------------------------------------------
mm-unstable branch with 7e12beb8ca2a commit applied
---------------------------------------------------
1) from output of XSBench
Threads: 14 Runtime: 1043.972 seconds Lookups: 1,700,000,000 Lookups/s: 1,628,395
2) from /proc/vmstat
numa_hit 16865880
numa_miss 1129958
numa_foreign 1129958
numa_interleave 115
numa_local 8565072
numa_other 9430766
numa_pte_updates 19240583
numa_hint_faults 19239948
numa_hint_faults_local 0
numa_pages_migrated 8159078
pgmigrate_success 17000781
pgmigrate_fail 1410437
compact_migrate_scanned 5075605
compact_daemon_migrate_scanned 5075605
pgdemote_kswapd 8297460
pgdemote_direct 0
nr_tlb_remote_flush 1516807
nr_tlb_remote_flush_received 20938785
nr_tlb_local_flush_all 95801
nr_tlb_local_flush_one 740597
3) from /proc/interrupts
TLB: 927080 567584 840684 1484285 1495859 1408641
1496227 1493909 1359465 1227623 1265431 1496361 1392337 1489451 1495799 1494700 TLB shootdowns
4) from 'perf stat -a'
43564429 itlb.itlb_flush
272921880 tlb_flush.dtlb_thread
175495467 tlb_flush.stlb_any
119602211976 dTLB-load-misses
355190881 dTLB-store-misses
1539926469 iTLB-load-misses
---------------------------------------------------------
mm-unstable branch with 7e12beb8ca2a commit applied + LUF
---------------------------------------------------------
1) from output of XSBench
Threads: 14 Runtime: 1033.973 seconds Lookups: 1,700,000,000 Lookups/s: 1,644,144
2) from /proc/vmstat
numa_hit 18617127
numa_miss 1075467
numa_foreign 1075467
numa_interleave 115
numa_local 9440134
numa_other 10252460
numa_pte_updates 19473883
numa_hint_faults 19470143
numa_hint_faults_local 0
numa_pages_migrated 8978959
pgmigrate_success 18675500
pgmigrate_fail 1577460
compact_migrate_scanned 5465414
compact_daemon_migrate_scanned 5465414
pgdemote_kswapd 9172431
pgdemote_direct 0
nr_tlb_remote_flush 85818
nr_tlb_remote_flush_received 1036316
nr_tlb_local_flush_all 34674
nr_tlb_local_flush_one 740870
3) from /proc/interrupts
TLB: 55328 31254 44449 72887 73407 73775
73353 73658 35802 68184 70998 73504 74072 64700 73718 73862 TLB shootdowns
4) from 'perf stat -a'
2054390 itlb.itlb_flush
150073902 tlb_flush.dtlb_thread
135630767 tlb_flush.stlb_any
117880065362 dTLB-load-misses
217521760 dTLB-store-misses
908338035 iTLB-load-misses
---
The result looks incredible. You can also see the result if you try to test a workload triggering reclaim or migration with LUF.
Byungchul
Quoted 5 lines · Byungchul Park
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
Is this just _pure_ unmapping (like MADV_DONTNEED), or does it apply to changing the memory map, like munmap() itself?
Quoted 6 lines
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through
mmap(), give up luf mechanism and perform tlb flush required right away.
Let's say you do this:
fd = open("/some/file", O_RDONLY);
ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...);
foo1 = *ptr1;You now have a read-only PTE pointing to the first page of /some/file. Let's say try_to_unmap() comes along and decides it can_luf_folio(). The page gets pulled out of the page cache and freed, the PTE is zeroed.
But the TLB is never flushed.
Now, someone does:
fd2 = open("/some/other/file", O_RDONLY);
ptr2 = mmap(ptr1, size, PROT_READ, MAP_FIXED, fd, ...);
foo2 = *ptr2;and they overwrite the old VMA. Does foo2 have the contents of the new "/some/other/file" or the old "/some/file"? How does the new mmap() know that there was something to flush?
BTW, the same thing could happen without a new mmap(). Someone could modify the file in the middle, maybe even from another process.
fd = open("/some/file", O_RDONLY);
ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...);
foo1 = *ptr1;
// LUF happens here
// "/some/file" changes
foo2 = *ptr1; // Does this see the change?Byungchul Park
reviewerQuoted 9 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
Is this just _pure_ unmapping (like
MADV_DONTNEED), or does it apply to changing the memory map, likemunmap()itself?
I think it can be applied to any unmapping of ro ones but LUF for now is working only with unmapping during folio migrion and reclaim.
Quoted 18 lines
Quoted 6 lines
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through
mmap(), give up luf mechanism and perform tlb flush required right away.Let's say you do this:
fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;You now have a read-only PTE pointing to the first page of /some/file. Let's say
try_to_unmap()comes along and decides itcan_luf_folio(). The page gets pulled out of the page cache and freed, the PTE is zeroed.
But the TLB is never flushed.
Quoted 8 lines
Now, someone does:
fd2 = open("/some/other/file", O_RDONLY); ptr2 = mmap(ptr1, size, PROT_READ, MAP_FIXED, fd, ...); foo2 = *ptr2;and they overwrite the old VMA. Does foo2 have the contents of the new "/some/other/file" or the old "/some/file"? How does the new
mmap()
Good point. It should've give up LUF at the 2nd mmap() in this case. I will fix it by introducing a new flag in task_struct indicating if LUF has left stale maps for the task so that LUF can give up and flush right away in mmap().
Quoted 4 lines
know that there was something to flush?
BTW, the same thing could happen without a new
mmap(). Someone could modify the file in the middle, maybe even from another process.
Thank you for the pointing out. I will fix it too by introducing a new flag in inode or something to make LUF aware if updating the file has been tried so that LUF can give up and flush right away in the case.
Plus, I will add another give-up at code changing the permission of vma to writable.
Thank you very much.
Byungchul
Quoted 6 lines
fd = open("/some/file", O_RDONLY);
ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...);
foo1 = *ptr1;
// LUF happens here
// "/some/file" changes
foo2 = *ptr1; // Does this see the change?On 5/26/24 18:57, Byungchul Park wrote:
...
Quoted 2 lines
Plus, I will add another give-up at code changing the permission of vma to writable.
I suspect you have a much more general problem on your hands. Just tweaking the VFS or mmap() code likely isn't going to cut it.
I guess we'll see what you come up with next, but this email was really just the result of Vlastimil and I chatting on IRC for five minutes about this set.
It has absolutely not been tested nor reviewed enough. <fud>I hope the performance gains stick around once more of the bugs are gone.</fud>
Byungchul Park
reviewerQuoted 7 lines · Dave Hansen
On 5/26/24 18:57, Byungchul Park wrote:
...Quoted 2 lines
Plus, I will add another give-up at code changing the permission of vma to writable.
I suspect you have a much more general problem on your hands. Just tweaking the VFS or
mmap()code likely isn't going to cut it.
LUF is interested in limited folios that are migratable or reclaimable in lru for now. So, IMHO, fixing a few things is going to cut it.
Quoted 6 lines
I guess we'll see what you come up with next, but this email was really just the result of Vlastimil and I chatting on IRC for five minutes about this set.
It has absolutely not been tested nor reviewed enough. <fud>I hope the performance gains stick around once more of the bugs are gone.</fud>
Sure. It should be.
Byungchul
Byungchul Park
reviewerQuoted 8 lines · Byungchul Park
Quoted 7 lines · Dave Hansen
On 5/26/24 18:57, Byungchul Park wrote:
...Quoted 2 lines
Plus, I will add another give-up at code changing the permission of vma to writable.
I suspect you have a much more general problem on your hands. Just tweaking the VFS or
mmap()code likely isn't going to cut it.
What a stupid idiot I am.
I already discuss the exact cases with Nadav Amit at the very beginning around v1. I didn't remember it when I was answering to you.
mmap() or changing the permission by user already performs TLB flush needed within that code, which LUF never touch.
Worth noting currently LUF touchs only unmapping during migration or reclaim. Other updating mapping would perform TLB flush it needs, as is. I guess updating page cache is also already perform TLB flush needed. I need to check it. Probably, it would already do.
Byungchul
Quoted 13 lines
LUF is interested in limited folios that are migratable or reclaimable in lru for now. So, IMHO, fixing a few things is going to cut it.
Quoted 6 lines
I guess we'll see what you come up with next, but this email was really just the result of Vlastimil and I chatting on IRC for five minutes about this set.
It has absolutely not been tested nor reviewed enough. <fud>I hope the performance gains stick around once more of the bugs are gone.</fud>
Sure. It should be.
Byungchul
Byungchul Park
reviewerQuoted 15 lines · Byungchul Park
Quoted 8 lines · Byungchul Park
Quoted 7 lines · Dave Hansen
On 5/26/24 18:57, Byungchul Park wrote:
...Quoted 2 lines
Plus, I will add another give-up at code changing the permission of vma to writable.
I suspect you have a much more general problem on your hands. Just tweaking the VFS or
mmap()code likely isn't going to cut it.What a stupid idiot I am.
I already discuss the exact cases with Nadav Amit at the very beginning around v1. I didn't remember it when I was answering to you.
Quoted 6 lines
mmap()or changing the permission by user already performs TLB flush needed within that code, which LUF never touch.Worth noting currently LUF touchs only unmapping during migration or reclaim. Other updating mapping would perform TLB flush it needs, as is. I guess updating page cache is also already perform TLB flush needed.
This may not be the case tho.. I might need to work on page cache.
Byungchul
Quoted 17 lines
I need to check it. Probably, it would already do.
Byungchul
Quoted 13 lines
LUF is interested in limited folios that are migratable or reclaimable in lru for now. So, IMHO, fixing a few things is going to cut it.
Quoted 6 lines
I guess we'll see what you come up with next, but this email was really just the result of Vlastimil and I chatting on IRC for five minutes about this set.
It has absolutely not been tested nor reviewed enough. <fud>I hope the performance gains stick around once more of the bugs are gone.</fud>
Sure. It should be.
Byungchul
Byungchul Park
reviewerQuoted 1 line · Dave Hansen
It has absolutely not been tested nor reviewed enough. <fud>I hope the
It has been tested enough on my side, and it should be reviewed enough for sure. I will respin after rebasing the current mm-unstble and working on vfs shortly.
Byungchul
Quoted 1 line
performance gains stick around once more of the bugs are gone.</fud>
Quoted 5 lines · Byungchul Park
Quoted 1 line · Dave Hansen
It has absolutely not been tested nor reviewed enough. <fud>I hope the
It has been tested enough on my side, and it should be reviewed enough for sure.
I believe that you have tested and reviewed the patchset by yourself. But there are some other cases that you haven't thought about enough before, as Dave pointed out.
So, I suggest you to try to find out more possible weakness of your patchset. Begin with what Dave and David pointed out.
Quoted 6 lines
I will respin after rebasing the current mm-unstble and working on vfs shortly.
Byungchul
Quoted 1 line
performance gains stick around once more of the bugs are gone.</fud>
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 8 lines · Ying
Quoted 5 lines · Byungchul Park
Quoted 1 line · Dave Hansen
It has absolutely not been tested nor reviewed enough. <fud>I hope the
It has been tested enough on my side, and it should be reviewed enough for sure.
Quoted 6 lines
I believe that you have tested and reviewed the patchset by yourself. But there are some other cases that you haven't thought about enough before, as Dave pointed out.
So, I suggest you to try to find out more possible weakness of your patchset. Begin with what Dave and David pointed out.
I will.
Byungchul
Quoted 10 lines
Quoted 6 lines
I will respin after rebasing the current mm-unstble and working on vfs shortly.
Byungchul
Quoted 1 line
performance gains stick around once more of the bugs are gone.</fud>
--
Best Regards,
Huang, Ying
Quoted 51 lines · Byungchul Park
Quoted 9 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
Is this just _pure_ unmapping (like
MADV_DONTNEED), or does it apply to changing the memory map, likemunmap()itself?I think it can be applied to any unmapping of ro ones but LUF for now is working only with unmapping during folio migrion and reclaim.
Quoted 18 lines
Quoted 6 lines
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through
mmap(), give up luf mechanism and perform tlb flush required right away.Let's say you do this:
fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;You now have a read-only PTE pointing to the first page of /some/file. Let's say
try_to_unmap()comes along and decides itcan_luf_folio(). The page gets pulled out of the page cache and freed, the PTE is zeroed.
But the TLB is never flushed.Quoted 8 lines
Now, someone does:
fd2 = open("/some/other/file", O_RDONLY); ptr2 = mmap(ptr1, size, PROT_READ, MAP_FIXED, fd, ...); foo2 = *ptr2;and they overwrite the old VMA. Does foo2 have the contents of the new "/some/other/file" or the old "/some/file"? How does the new
mmap()Good point. It should've give up LUF at the 2nd
mmap()in this case. I will fix it by introducing a new flag intask_structindicating if LUF has left stale maps for the task so that LUF can give up and flush right away inmmap().Quoted 4 lines
know that there was something to flush?
BTW, the same thing could happen without a new
mmap(). Someone could modify the file in the middle, maybe even from another process.
Quoted 6 lines
Thank you for the pointing out. I will fix it too by introducing a new flag in inode or something to make LUF aware if updating the file has been tried so that LUF can give up and flush right away in the case.
Plus, I will add another give-up at code changing the permission of vma to writable.
I guess that you need a framework similar as "flush_tlb_batched_pending()" to deal with interaction with other TLB related operations.
--
Best Regards,
Huang, Ying
Quoted 10 lines
Thank you very much.
Byungchul
Quoted 6 lines
fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1; // LUF happens here // "/some/file" changes foo2 = *ptr1; // Does this see the change?
Byungchul Park
reviewerQuoted 57 lines · Ying
Quoted 55 lines · Byungchul Park
Quoted 9 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
Is this just _pure_ unmapping (like
MADV_DONTNEED), or does it apply to changing the memory map, likemunmap()itself?I think it can be applied to any unmapping of ro ones but LUF for now is working only with unmapping during folio migrion and reclaim.
Quoted 18 lines
Quoted 6 lines
2. When any non-writable ptes change to writable e.g. through fault handler, give up luf mechanism and perform tlb flush required right away.
3. When a writable mapping is created e.g. through
mmap(), give up luf mechanism and perform tlb flush required right away.Let's say you do this:
fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;You now have a read-only PTE pointing to the first page of /some/file. Let's say
try_to_unmap()comes along and decides itcan_luf_folio(). The page gets pulled out of the page cache and freed, the PTE is zeroed.
But the TLB is never flushed.Quoted 8 lines
Now, someone does:
fd2 = open("/some/other/file", O_RDONLY); ptr2 = mmap(ptr1, size, PROT_READ, MAP_FIXED, fd, ...); foo2 = *ptr2;and they overwrite the old VMA. Does foo2 have the contents of the new "/some/other/file" or the old "/some/file"? How does the new
mmap()Good point. It should've give up LUF at the 2nd
mmap()in this case. I will fix it by introducing a new flag intask_structindicating if LUF has left stale maps for the task so that LUF can give up and flush right away inmmap().Quoted 4 lines
know that there was something to flush?
BTW, the same thing could happen without a new
mmap(). Someone could modify the file in the middle, maybe even from another process.Thank you for the pointing out. I will fix it too by introducing a new flag in inode or something to make LUF aware if updating the file has been tried so that LUF can give up and flush right away in the case.
Quoted 6 lines
Quoted 2 lines
Plus, I will add another give-up at code changing the permission of vma to writable.
I guess that you need a framework similar as "
flush_tlb_batched_pending()" to deal with interaction with other TLB related operations.
Thank you. I will check it.
Byungchul
Quoted 14 lines
--
Best Regards,
Huang, YingQuoted 10 lines
Thank you very much.
Byungchul
Quoted 6 lines
fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1; // LUF happens here // "/some/file" changes foo2 = *ptr1; // Does this see the change?
Quoted 9 lines · Ying
Quoted 6 lines
Thank you for the pointing out. I will fix it too by introducing a new flag in inode or something to make LUF aware if updating the file has been tried so that LUF can give up and flush right away in the case.
Plus, I will add another give-up at code changing the permission of vma to writable.
I guess that you need a framework similar as "
flush_tlb_batched_pending()" to deal with interaction with other TLB related operations.
Where "other TLB related operations" includes both things that traditionally invalidate TLBs (like going Present 1=>0) and things like fault-in that go Present 0=>1 that can result in TLB population.
It's actually a really crummy problem to solve. We don't have _any_ machinery to say, "Hey, you know that PTE you wanted to install? There was something there before you and we haven't flushed it yet. Can you be a doll and do a flush before _populating_ that PTE?"
To solve it generically, I suspect you'll need some kind of special non-present PTE to say:
There _was_ a PTE here that wasn't flushed.
Sure, you can add gunk to the VMA to track when this happens. But that'll penalize anyone populating a PTE anywhere in the VMA at least once. If there were other threads faulting in pages to the same VMA, they'll just end up doing the flush that LUF tried to avoid in the first place.
Byungchul Park
reviewerQuoted 11 lines · Dave Hansen
Quoted 9 lines · Ying
Quoted 6 lines
Thank you for the pointing out. I will fix it too by introducing a new flag in inode or something to make LUF aware if updating the file has been tried so that LUF can give up and flush right away in the case.
Plus, I will add another give-up at code changing the permission of vma to writable.
I guess that you need a framework similar as "
flush_tlb_batched_pending()" to deal with interaction with other TLB related operations.
Quoted 8 lines
Where "other TLB related operations" includes both things that traditionally invalidate TLBs (like going Present 1=>0) and things like fault-in that go Present 0=>1 that can result in TLB population.
It's actually a really crummy problem to solve. We don't have _any_ machinery to say, "Hey, you know that PTE you wanted to install? There was something there before you and we haven't flushed it yet. Can you be a doll and do a flush before _populating_ that PTE?"
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Byungchul
Quoted 10 lines
To solve it generically, I suspect you'll need some kind of special non-present PTE to say:
There _was_ a PTE here that wasn't flushed.
Sure, you can add gunk to the VMA to track when this happens. But that'll penalize anyone populating a PTE anywhere in the VMA at least once. If there were other threads faulting in pages to the same VMA, they'll just end up doing the flush that LUF tried to avoid in the first place.
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;
There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the try_to_unmap() paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.
Then, the user might do:
munmap(ptr1, PAGE_SIZE);
Which will _eventually_ wind up in the zap_pte_range() loop. But that loop will only see pte_none(). It doesn't do _anything_ to the 'struct mmu_gather'.
The munmap() then lands in tlb_flush_mmu_tlbonly() where it looks at the 'struct mmu_gather':
if (!(tlb->freed_tables || tlb->cleared_ptes ||
tlb->cleared_pmds || tlb->cleared_puds ||
tlb->cleared_p4ds))
return;But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Byungchul Park
reviewerQuoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;
Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that mmu_gather mechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?
Thanks again.
Byungchul
Byungchul Park
reviewerQuoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?
I will check the existing optimization of TLB flsuh more in arch level and suggest a better way.
Byungchul
Quoted 3 lines
Thanks again.
Byungchul
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as flush_tlb_batched_pending().
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?
Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().
Right. I'm thinking to add a related code to flush_tlb_batched_pending().
Byungchul
Quoted 3 lines
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?
Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().
While reading flush_tlb_batched_pending(mm), I found it already performs TLB flush for the target mm, if set_tlb_ubc_flush_pending(mm) has been hit at least once since the last flush_tlb_batched_pending(mm).
Since LUF also relies on set_tlb_ubc_flush_pending(mm), it's going to perform TLB flush required, in flush_tlb_batched_pending(mm) during munmap(). So it looks safe to me with regard to munmap() already.
Is there something that I'm missing?
JFYI, regarding to mmap(), I have reworked on fault handler to give up luf when needed in a better way.
Byungchul
Quoted 3 lines
--
Best Regards,
Huang, Ying
Quoted 60 lines · Byungchul Park
Quoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().While reading
flush_tlb_batched_pending(mm), I found it already performs TLB flush for the target mm, ifset_tlb_ubc_flush_pending(mm)has been hit at least once since the lastflush_tlb_batched_pending(mm).
Quoted 8 lines
Since LUF also relies on
set_tlb_ubc_flush_pending(mm), it's going to perform TLB flush required, inflush_tlb_batched_pending(mm)duringmunmap(). So it looks safe to me with regard tomunmap()already.Is there something that I'm missing?
JFYI, regarding to
mmap(), I have reworked on fault handler to give up luf when needed in a better way.
If TLB flush is always enforced during munmap(), then your solution can only avoid TLB flushing for page reclaiming and migration, not unmap. Or do I miss something?
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 66 lines · Ying
Quoted 64 lines · Byungchul Park
Quoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().While reading
flush_tlb_batched_pending(mm), I found it already performs TLB flush for the target mm, ifset_tlb_ubc_flush_pending(mm)has been hit at least once since the lastflush_tlb_batched_pending(mm).Since LUF also relies on
set_tlb_ubc_flush_pending(mm), it's going to perform TLB flush required, inflush_tlb_batched_pending(mm)duringmunmap(). So it looks safe to me with regard tomunmap()already.
Quoted 7 lines
Quoted 4 lines
Is there something that I'm missing?
JFYI, regarding to
mmap(), I have reworked on fault handler to give up luf when needed in a better way.If TLB flush is always enforced during
munmap(), then your solution can only avoid TLB flushing for page reclaiming and migration, not unmap.
^ munmap()?
Do you mean munmap()? IIUC, yes. LUF only works for page reclaiming and migration, but not for munmap(). When munmap()ing, LUF rather needs to give up and perform tlb flush pended.
LUF should not optimize tlb flushes for mappings that users explicitly change e.g. through mmap() and munmap().
Byungchul
Quoted 5 lines
Or do I miss something?
--
Best Regards,
Huang, Ying
Quoted 2 lines · Byungchul Park
LUF should not optimize tlb flushes for mappings that users explicitly change e.g. through
mmap()andmunmap().
We are thoroughly going around in circles at this point.
I'm not quite sure what to do. Ying and I see a problem that we've tried to explain a couple of times. We've tried to show the connection between a LUF-elided TLB flush and how that could affect a later munmap() or mmap(MAP_FIXED).
But these responses seem to keep going back to the fact that LUF doesn't directly affect munmap(), which is true, but quite irrelevant to the problem being described.
So we're at an impasse.
Byungchul, perhaps you should spin another series and maybe Ying and I have to write up a test case to show the bug that we see. Or perhaps someone else can jump into the thread and bridge the communication gap.
Byungchul Park
reviewerQuoted 4 lines · Dave Hansen
Quoted 2 lines · Byungchul Park
LUF should not optimize tlb flushes for mappings that users explicitly change e.g. through
mmap()andmunmap().
Quoted 8 lines
We are thoroughly going around in circles at this point.
I'm not quite sure what to do. Ying and I see a problem that we've tried to explain a couple of times. We've tried to show the connection between a LUF-elided TLB flush and how that could affect a later
munmap()ormmap(MAP_FIXED).But these responses seem to keep going back to the fact that LUF doesn't
I just wanted to understand exactly what Ying meant. My answer might be done in a wrong way if I wrongly got him.
Quoted 6 lines
directly affect
munmap(), which is true, but quite irrelevant to the problem being described.So we're at an impasse.
Byungchul, perhaps you should spin another series and maybe Ying and I
I don't think the current implementation is perfect. I just wanted to know what I'm missing now but.. yes. It would be much better to communicate with a real bug if existing.
I will respin the next version shortly.
Byungchul
Quoted 2 lines
have to write up a test case to show the bug that we see. Or perhaps someone else can jump into the thread and bridge the communication gap.
Byungchul Park
reviewerQuoted 66 lines · Ying
Quoted 64 lines · Byungchul Park
Quoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().While reading
flush_tlb_batched_pending(mm), I found it already performs TLB flush for the target mm, ifset_tlb_ubc_flush_pending(mm)has been hit at least once since the lastflush_tlb_batched_pending(mm).Since LUF also relies on
set_tlb_ubc_flush_pending(mm), it's going to perform TLB flush required, inflush_tlb_batched_pending(mm)duringmunmap(). So it looks safe to me with regard tomunmap()already.
Quoted 7 lines
Quoted 4 lines
Is there something that I'm missing?
JFYI, regarding to
mmap(), I have reworked on fault handler to give up luf when needed in a better way.If TLB flush is always enforced during
munmap(), then your solution can only avoid TLB flushing for page reclaiming and migration, not unmap.
I'm not sure if I understand what you meant. Could you explain it in more detail?
LUF works for only *unmapping* that happens during page reclaiming and migration. Other unmappings than page reclaiming and migration are not what LUF works for. That's why I thought flush_tlb_batched_pending() could handle the pending tlb flushes in the case.
It'd be appreciated if you explain what you meant more.
Byungchul
Quoted 5 lines
Or do I miss something?
--
Best Regards,
Huang, Ying
Quoted 78 lines · Byungchul Park
Quoted 66 lines · Ying
Quoted 64 lines · Byungchul Park
Quoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().While reading
flush_tlb_batched_pending(mm), I found it already performs TLB flush for the target mm, ifset_tlb_ubc_flush_pending(mm)has been hit at least once since the lastflush_tlb_batched_pending(mm).Since LUF also relies on
set_tlb_ubc_flush_pending(mm), it's going to perform TLB flush required, inflush_tlb_batched_pending(mm)duringmunmap(). So it looks safe to me with regard tomunmap()already.Quoted 7 lines
Quoted 4 lines
Is there something that I'm missing?
JFYI, regarding to
mmap(), I have reworked on fault handler to give up luf when needed in a better way.If TLB flush is always enforced during
munmap(), then your solution can only avoid TLB flushing for page reclaiming and migration, not unmap.I'm not sure if I understand what you meant. Could you explain it in more detail?
Quoted 7 lines
LUF works for only *unmapping* that happens during page reclaiming and migration. Other unmappings than page reclaiming and migration are not what LUF works for. That's why I thought
flush_tlb_batched_pending()could handle the pending tlb flushes in the case.It'd be appreciated if you explain what you meant more.
In the following email, you have claimed that LUF can avoid TLB flushing for munmap()/mmap().
https://lore.kernel.org/linux-mm/[email protected]/
Now, you said it can only avoid TLB flushing for page reclaiming and migration.
So, to avoid confusion, I suggest you to send out a new series and make it explicit that it can only optimize page reclaiming and migration, but not munmap(). And it would be good too to add some words about how it interact with other TLB flushing mechanisms.
--
Best Regards,
Huang, Ying
Byungchul Park
reviewerQuoted 85 lines · Ying
Quoted 83 lines · Byungchul Park
Quoted 66 lines · Ying
Quoted 64 lines · Byungchul Park
Quoted 51 lines · Ying
Quoted 43 lines · Byungchul Park
Quoted 34 lines · Dave Hansen
Quoted 5 lines · Byungchul Park
All the code updating ptes already performs TLB flush needed in a safe way if it's inevitable e.g. munmap. LUF which controls when to flush in a higer level than arch code, just leaves stale ro tlb entries that are currently supposed to be in use. Could you give a scenario that you are concering?
Let's go back this scenario:
(quoted diff) +0
-0fd = open("/some/file", O_RDONLY); ptr1 = mmap(-1, size, PROT_READ, ..., fd, ...); foo1 = *ptr1;There's a read-only PTE at 'ptr1'. Right? The page being pointed to is eligible for LUF via the
try_to_unmap()paths. In other words, the page might be reclaimed at any time. If it is reclaimed, the PTE will be cleared.Then, the user might do:
munmap(ptr1, PAGE_SIZE);Which will _eventually_ wind up in the
zap_pte_range()loop. But that loop will only seepte_none(). It doesn't do _anything_ to the 'structmmu_gather'.The
munmap()then lands intlb_flush_mmu_tlbonly()where it looks at the 'structmmu_gather':if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds || tlb->cleared_puds || tlb->cleared_p4ds)) return;Quoted 7 lines
But since there were no cleared PTEs (or anything else) during the unmap, this just returns and doesn't flush the TLB.
We now have an address space with a stale TLB entry at 'ptr1' and not even a VMA there. There's nothing to stop a new VMA from going in, installing a *new* PTE, but getting data from the stale TLB entry that still hasn't been flushed.
Quoted 5 lines
Thank you for the explanation. I got you. I think I could handle the case through a new flag in vma or something indicating LUF has deferred necessary TLB flush for it during unmapping so that
mmu_gathermechanism can be aware of it. Of course, the performance change should be checked again. Thoughts?Quoted 3 lines
I suggest you to start with the simple case. That is, only support page reclaiming and migration. A TLB flushing can be enforced during unmap with something similar as
flush_tlb_batched_pending().While reading
flush_tlb_batched_pending(mm), I found it already performs TLB flush for the target mm, ifset_tlb_ubc_flush_pending(mm)has been hit at least once since the lastflush_tlb_batched_pending(mm).Since LUF also relies on
set_tlb_ubc_flush_pending(mm), it's going to perform TLB flush required, inflush_tlb_batched_pending(mm)duringmunmap(). So it looks safe to me with regard tomunmap()already.Quoted 7 lines
Quoted 4 lines
Is there something that I'm missing?
JFYI, regarding to
mmap(), I have reworked on fault handler to give up luf when needed in a better way.If TLB flush is always enforced during
munmap(), then your solution can only avoid TLB flushing for page reclaiming and migration, not unmap.I'm not sure if I understand what you meant. Could you explain it in more detail?
LUF works for only *unmapping* that happens during page reclaiming and migration. Other unmappings than page reclaiming and migration are not what LUF works for. That's why I thought
flush_tlb_batched_pending()could handle the pending tlb flushes in the case.
Quoted 5 lines
Quoted 2 lines
It'd be appreciated if you explain what you meant more.
In the following email, you have claimed that LUF can avoid TLB flushing for munmap()/mmap().
My bad. Sorry for that confusing expression.
"give up LUF at mmap()" doesn't mean giving up applying LUF to mmap().
"give up LUF at mmap()" means giving up the pending that has been induced by LUF, in other words, giving up the benefit by LUF because we are going through mmap() / munmap().
I will be more careful in expressing these things.
Quoted 4 lines
https://lore.kernel.org/linux-mm/[email protected]/
Now, you said it can only avoid TLB flushing for page reclaiming and migration.
This is true.
Byungchul
Quoted 8 lines
So, to avoid confusion, I suggest you to send out a new series and make it explicit that it can only optimize page reclaiming and migration, but not
munmap(). And it would be good too to add some words about how it interact with other TLB flushing mechanisms.--
Best Regards,
Huang, Ying
Am 10.05.24 um 08:51 schrieb Byungchul Park:
Quoted 27 lines
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:https://lore.kernel.org/lkml/[email protected]/
However, it's only for migration through hinting fault. I thought it'd be much better if we have a general mechanism to reduce all the tlb numbers that we can apply to any unmap code, that we normally believe tlb flush should be followed.
I'm suggesting a new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
tlb flush can be defered when folios get unmapped as long as it guarantees to perform tlb flush needed, before the folios actually become used, of course, only if all the corresponding ptes don't have write permission. Otherwise, the system will get messed up.
Quoted 5 lines
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
Trying to understand the impact: Effectively, a CPU could still read data from a page that has already been freed, until that page gets reallocated again.
The important part I can see is
1) PCP/buddy must not change page content (e.g., poison, init_on_free), otherwise an app might read wrong content.
2) If we mess up the flush-before-realloc, an app might observe data written by whoever allocated the page.
3) We must reliably detect+handle any read-only PTEs for which we didn't flush the TLB yet, otherwise an app could see its memory writes getting lost. I recall that at least uffd-wp might defer TLB flushes (see comment in do_wp_page()). Not sure about other pte_wrprotect() callers that flush the TLB after processing multiple page tables, whereby rmap code might succeed in unmapping a page before the TLB flush happened.
Any other possible issues you stumbled over that are worth mentioning?
Signature · 3 lines
Thanks, David / dhildenb
Byungchul Park
reviewerQuoted 34 lines · David Hildenbrand
Am 10.05.24 um 08:51 schrieb Byungchul Park:
Quoted 27 lines
Hi everyone,
While I'm working with a tiered memory system e.g. CXL memory, I have been facing migration overhead esp. tlb shootdown on promotion or demotion between different tiers. Yeah.. most tlb shootdowns on migration through hinting fault can be avoided thanks to Huang Ying's work, commit
4d4b6d66db("mm,unmap: avoid flushing tlb in batch if PTE is inaccessible"). See the following link for more information:https://lore.kernel.org/lkml/[email protected]/
However, it's only for migration through hinting fault. I thought it'd be much better if we have a general mechanism to reduce all the tlb numbers that we can apply to any unmap code, that we normally believe tlb flush should be followed.
I'm suggesting a new mechanism, LUF(Lazy Unmap Flush), defers tlb flush until folios that have been unmapped and freed, eventually get allocated again. It's safe for folios that had been mapped read-only and were unmapped, since the contents of the folios don't change while staying in pcp or buddy so we can still read the data through the stale tlb entries.
tlb flush can be defered when folios get unmapped as long as it guarantees to perform tlb flush needed, before the folios actually become used, of course, only if all the corresponding ptes don't have write permission. Otherwise, the system will get messed up.
Quoted 5 lines
To achieve that:
1. For the folios that map only to non-writable tlb entries, prevent tlb flush during unmapping but perform it just before the folios actually become used, out of buddy or pcp.
Quoted 8 lines
Trying to understand the impact: Effectively, a CPU could still read data from a page that has already been freed, until that page gets reallocated again.
The important part I can see is
1) PCP/buddy must not change page content (e.g., poison,
init_on_free), otherwise an app might read wrong content.
Exactly. I will take them into account. Thank you.
Quoted 2 lines
2) If we mess up the flush-before-realloc, an app might observe data written by whoever allocated the page.
Yes. However, appropiate TLB flush is performed in prep_new_page(). Basically you are right. I need to pay enough attention to it.
Quoted 8 lines
3) We must reliably detect+handle any read-only PTEs for which we didn't flush the TLB yet, otherwise an app could see its memory writes getting lost. I recall that at least uffd-wp might defer TLB flushes (see comment in
do_wp_page()). Not sure about otherpte_wrprotect()callers that flush the TLB after processing multiple page tables, whereby rmap code might succeed in unmapping a page before the TLB flush happened.Any other possible issues you stumbled over that are worth mentioning?
You mentioned all that I'm concerning but in a clear way.
Byungchul
Quoted 5 lines
Signature · 4 lines
-- Thanks, David / dhildenb