| Message ID | 20201108141113.65450-1-songmuchun@bytedance.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Free some vmemmap pages of hugetlb page | expand |
Thanks for continuing to work this Muchun! On 11/8/20 6:10 AM, Muchun Song wrote: ... > For tail pages, the value of compound_head is the same. So we can reuse > first page of tail page structs. We map the virtual addresses of the > remaining 6 pages of tail page structs to the first tail page struct, > and then free these 6 pages. Therefore, we need to reserve at least 2 > pages as vmemmap areas. > > When a hugetlbpage is freed to the buddy system, we should allocate six > pages for vmemmap pages and restore the previous mapping relationship. > > If we uses the 1G hugetlbpage, we can save 4095 pages. This is a very > substantial gain. Is that 4095 number accurate? Are we not using two pages of struct pages as in the 2MB case? Also, because we are splitting the huge page mappings in the vmemmap additional PTE pages will need to be allocated. Therefore, some additional page table pages may need to be allocated so that we can free the pages of struct pages. The net savings may be less than what is stated above. Perhaps this should mention that allocation of additional page table pages may be required? ... > Because there are vmemmap page tables reconstruction on the freeing/allocating > path, it increases some overhead. Here are some overhead analysis. > > 1) Allocating 10240 2MB hugetlb pages. > > a) With this patch series applied: > # time echo 10240 > /proc/sys/vm/nr_hugepages > > real 0m0.166s > user 0m0.000s > sys 0m0.166s > > # bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; } kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > Attaching 2 probes... > > @latency: > [8K, 16K) 8360 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > [16K, 32K) 1868 |@@@@@@@@@@@ | > [32K, 64K) 10 | | > [64K, 128K) 2 | | > > b) Without this patch series: > # time echo 10240 > /proc/sys/vm/nr_hugepages > > real 0m0.066s > user 0m0.000s > sys 0m0.066s > > # bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; } kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > Attaching 2 probes... > > @latency: > [4K, 8K) 10176 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > [8K, 16K) 62 | | > [16K, 32K) 2 | | > > Summarize: this feature is about ~2x slower than before. > > 2) Freeing 10240 @MB hugetlb pages. > > a) With this patch series applied: > # time echo 0 > /proc/sys/vm/nr_hugepages > > real 0m0.004s > user 0m0.000s > sys 0m0.002s > > # bpftrace -e 'kprobe:__free_hugepage { @start[tid] = nsecs; } kretprobe:__free_hugepage /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > Attaching 2 probes... > > @latency: > [16K, 32K) 10240 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > > b) Without this patch series: > # time echo 0 > /proc/sys/vm/nr_hugepages > > real 0m0.077s > user 0m0.001s > sys 0m0.075s > > # bpftrace -e 'kprobe:__free_hugepage { @start[tid] = nsecs; } kretprobe:__free_hugepage /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > Attaching 2 probes... > > @latency: > [4K, 8K) 9950 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > [8K, 16K) 287 |@ | > [16K, 32K) 3 | | > > Summarize: The overhead of __free_hugepage is about ~2-4x slower than before. > But according to the allocation test above, I think that here is > also ~2x slower than before. > > But why the 'real' time of patched is smaller than before? Because > In this patch series, the freeing hugetlb is asynchronous(through > kwoker). > > Although the overhead has increased. But the overhead is not on the > allocating/freeing of each hugetlb page, it is only once when we reserve > some hugetlb pages through /proc/sys/vm/nr_hugepages. Once the reservation > is successful, the subsequent allocating, freeing and using are the same > as before (not patched). So I think that the overhead is acceptable. Thank you for benchmarking. There are still some instances where huge pages are allocated 'on the fly' instead of being pulled from the pool. Michal pointed out the case of page migration. It is also possible for someone to use hugetlbfs without pre-allocating huge pages to the pool. I remember the use case pointed out in commit 099730d67417. It says, "I have a hugetlbfs user which is never explicitly allocating huge pages with 'nr_hugepages'. They only set 'nr_overcommit_hugepages' and then let the pages be allocated from the buddy allocator at fault time." In this case, I suspect they were using 'page fault' allocation for initialization much like someone using /proc/sys/vm/nr_hugepages. So, the overhead may not be as noticeable.
On Wed, Nov 11, 2020 at 3:23 AM Mike Kravetz <mike.kravetz@oracle.com> wrote: > > > Thanks for continuing to work this Muchun! > > On 11/8/20 6:10 AM, Muchun Song wrote: > ... > > For tail pages, the value of compound_head is the same. So we can reuse > > first page of tail page structs. We map the virtual addresses of the > > remaining 6 pages of tail page structs to the first tail page struct, > > and then free these 6 pages. Therefore, we need to reserve at least 2 > > pages as vmemmap areas. > > > > When a hugetlbpage is freed to the buddy system, we should allocate six > > pages for vmemmap pages and restore the previous mapping relationship. > > > > If we uses the 1G hugetlbpage, we can save 4095 pages. This is a very > > substantial gain. > > Is that 4095 number accurate? Are we not using two pages of struct pages > as in the 2MB case? Oh, yeah, here should be 4094 and subtract page tables. For a 1GB HugeTLB page, it should be 4086 pages. Thanks for pointing out this problem. > > Also, because we are splitting the huge page mappings in the vmemmap > additional PTE pages will need to be allocated. Therefore, some additional > page table pages may need to be allocated so that we can free the pages > of struct pages. The net savings may be less than what is stated above. > > Perhaps this should mention that allocation of additional page table pages > may be required? Yeah, you are right. In the later patch, I will rework the analysis here. Make it more clear and accurate. > > ... > > Because there are vmemmap page tables reconstruction on the freeing/allocating > > path, it increases some overhead. Here are some overhead analysis. > > > > 1) Allocating 10240 2MB hugetlb pages. > > > > a) With this patch series applied: > > # time echo 10240 > /proc/sys/vm/nr_hugepages > > > > real 0m0.166s > > user 0m0.000s > > sys 0m0.166s > > > > # bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; } kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > > Attaching 2 probes... > > > > @latency: > > [8K, 16K) 8360 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > > [16K, 32K) 1868 |@@@@@@@@@@@ | > > [32K, 64K) 10 | | > > [64K, 128K) 2 | | > > > > b) Without this patch series: > > # time echo 10240 > /proc/sys/vm/nr_hugepages > > > > real 0m0.066s > > user 0m0.000s > > sys 0m0.066s > > > > # bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; } kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > > Attaching 2 probes... > > > > @latency: > > [4K, 8K) 10176 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > > [8K, 16K) 62 | | > > [16K, 32K) 2 | | > > > > Summarize: this feature is about ~2x slower than before. > > > > 2) Freeing 10240 @MB hugetlb pages. > > > > a) With this patch series applied: > > # time echo 0 > /proc/sys/vm/nr_hugepages > > > > real 0m0.004s > > user 0m0.000s > > sys 0m0.002s > > > > # bpftrace -e 'kprobe:__free_hugepage { @start[tid] = nsecs; } kretprobe:__free_hugepage /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > > Attaching 2 probes... > > > > @latency: > > [16K, 32K) 10240 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > > > > b) Without this patch series: > > # time echo 0 > /proc/sys/vm/nr_hugepages > > > > real 0m0.077s > > user 0m0.001s > > sys 0m0.075s > > > > # bpftrace -e 'kprobe:__free_hugepage { @start[tid] = nsecs; } kretprobe:__free_hugepage /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' > > Attaching 2 probes... > > > > @latency: > > [4K, 8K) 9950 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| > > [8K, 16K) 287 |@ | > > [16K, 32K) 3 | | > > > > Summarize: The overhead of __free_hugepage is about ~2-4x slower than before. > > But according to the allocation test above, I think that here is > > also ~2x slower than before. > > > > But why the 'real' time of patched is smaller than before? Because > > In this patch series, the freeing hugetlb is asynchronous(through > > kwoker). > > > > Although the overhead has increased. But the overhead is not on the > > allocating/freeing of each hugetlb page, it is only once when we reserve > > some hugetlb pages through /proc/sys/vm/nr_hugepages. Once the reservation > > is successful, the subsequent allocating, freeing and using are the same > > as before (not patched). So I think that the overhead is acceptable. > > Thank you for benchmarking. There are still some instances where huge pages > are allocated 'on the fly' instead of being pulled from the pool. Michal > pointed out the case of page migration. It is also possible for someone to > use hugetlbfs without pre-allocating huge pages to the pool. I remember the > use case pointed out in commit 099730d67417. It says, "I have a hugetlbfs > user which is never explicitly allocating huge pages with 'nr_hugepages'. > They only set 'nr_overcommit_hugepages' and then let the pages be allocated > from the buddy allocator at fault time." In this case, I suspect they were > using 'page fault' allocation for initialization much like someone using > /proc/sys/vm/nr_hugepages. So, the overhead may not be as noticeable. Thanks for pointing out this using case. > > -- > Mike Kravetz
Hi all, This patch series will free some vmemmap pages(struct page structures) associated with each hugetlbpage when preallocated to save memory. Nowadays we track the status of physical page frames using `struct page` arranged in one or more arrays. And here exists one-to-one mapping between the physical page frame and the corresponding `struct page`. The hugetlbpage support is built on top of multiple page size support that is provided by most modern architectures. For example, x86 CPUs normally support 4K and 2M (1G if architecturally supported) page sizes. Every hugetlbpage has more than one `struct page`. The 2M hugetlbpage has 512 `struct page` and 1G hugetlbpage has 4096 `struct page`. But in the core of hugetlbpage only uses the first 4 `struct page` to store metadata associated with each hugetlbpage. The rest of the `struct page` are usually read the compound_head field which are all the same value. If we can free some struct page memory to buddy system so that we can save a lot of memory. When the system boot up, every 2M hugetlbpage has 512 `struct page` which is 8 pages(sizeof(struct page) * 512 / PAGE_SIZE). hugetlbpage struct pages(8 pages) page frame(8 pages) +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+ | | | 0 | -------------> | 0 | | | | 1 | -------------> | 1 | | | | 2 | -------------> | 2 | | | | 3 | -------------> | 3 | | | | 4 | -------------> | 4 | | 2M | | 5 | -------------> | 5 | | | | 6 | -------------> | 6 | | | | 7 | -------------> | 7 | | | +-----------+ +-----------+ | | | | +-----------+ When a hugetlbpage is preallocated, we can change the mapping from above to bellow. hugetlbpage struct pages(8 pages) page frame(8 pages) +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+ | | | 0 | -------------> | 0 | | | | 1 | -------------> | 1 | | | | 2 | -------------> +-----------+ | | | 3 | -----------------^ ^ ^ ^ ^ | | | 4 | -------------------+ | | | | 2M | | 5 | ---------------------+ | | | | | 6 | -----------------------+ | | | | 7 | -------------------------+ | | +-----------+ | | | | +-----------+ For tail pages, the value of compound_head is the same. So we can reuse first page of tail page structs. We map the virtual addresses of the remaining 6 pages of tail page structs to the first tail page struct, and then free these 6 pages. Therefore, we need to reserve at least 2 pages as vmemmap areas. When a hugetlbpage is freed to the buddy system, we should allocate six pages for vmemmap pages and restore the previous mapping relationship. If we uses the 1G hugetlbpage, we can save 4095 pages. This is a very substantial gain. On our server, run some SPDK/QEMU applications which will use 1000GB hugetlbpage. With this feature enabled, we can save ~16GB(1G hugepage)/~11GB(2MB hugepage) memory. Because there are vmemmap page tables reconstruction on the freeing/allocating path, it increases some overhead. Here are some overhead analysis. 1) Allocating 10240 2MB hugetlb pages. a) With this patch series applied: # time echo 10240 > /proc/sys/vm/nr_hugepages real 0m0.166s user 0m0.000s sys 0m0.166s # bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; } kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' Attaching 2 probes... @latency: [8K, 16K) 8360 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [16K, 32K) 1868 |@@@@@@@@@@@ | [32K, 64K) 10 | | [64K, 128K) 2 | | b) Without this patch series: # time echo 10240 > /proc/sys/vm/nr_hugepages real 0m0.066s user 0m0.000s sys 0m0.066s # bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; } kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' Attaching 2 probes... @latency: [4K, 8K) 10176 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [8K, 16K) 62 | | [16K, 32K) 2 | | Summarize: this feature is about ~2x slower than before. 2) Freeing 10240 @MB hugetlb pages. a) With this patch series applied: # time echo 0 > /proc/sys/vm/nr_hugepages real 0m0.004s user 0m0.000s sys 0m0.002s # bpftrace -e 'kprobe:__free_hugepage { @start[tid] = nsecs; } kretprobe:__free_hugepage /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' Attaching 2 probes... @latency: [16K, 32K) 10240 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| b) Without this patch series: # time echo 0 > /proc/sys/vm/nr_hugepages real 0m0.077s user 0m0.001s sys 0m0.075s # bpftrace -e 'kprobe:__free_hugepage { @start[tid] = nsecs; } kretprobe:__free_hugepage /@start[tid]/ { @latency = hist(nsecs - @start[tid]); delete(@start[tid]); }' Attaching 2 probes... @latency: [4K, 8K) 9950 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [8K, 16K) 287 |@ | [16K, 32K) 3 | | Summarize: The overhead of __free_hugepage is about ~2-4x slower than before. But according to the allocation test above, I think that here is also ~2x slower than before. But why the 'real' time of patched is smaller than before? Because In this patch series, the freeing hugetlb is asynchronous(through kwoker). Although the overhead has increased. But the overhead is not on the allocating/freeing of each hugetlb page, it is only once when we reserve some hugetlb pages through /proc/sys/vm/nr_hugepages. Once the reservation is successful, the subsequent allocating, freeing and using are the same as before (not patched). So I think that the overhead is acceptable. changelog in v3: 1. Rename some helps function name. Thanks Mike. 2. Rework some code. Thanks Mike and Oscar. 3. Remap the tail vmemmap page with PAGE_KERNEL_RO instead of PAGE_KERNEL. Thanks Matthew. 4. Add some overhead analysis in the cover letter. 5. Use vmemap pmd table lock instead of a hugetlb specific global lock. changelog in v2: 1. Fix do not call dissolve_compound_page in alloc_huge_page_vmemmap(). 2. Fix some typo and code style problems. 3. Remove unused handle_vmemmap_fault(). 4. Merge some commits to one commit suggested by Mike. Muchun Song (21): mm/memory_hotplug: Move bootmem info registration API to bootmem_info.c mm/memory_hotplug: Move {get,put}_page_bootmem() to bootmem_info.c mm/hugetlb: Introduce a new config HUGETLB_PAGE_FREE_VMEMMAP mm/hugetlb: Introduce nr_free_vmemmap_pages in the struct hstate mm/hugetlb: Introduce pgtable allocation/freeing helpers mm/bootmem_info: Introduce {free,prepare}_vmemmap_page() mm/bootmem_info: Combine bootmem info and type into page->freelist mm/vmemmap: Initialize page table lock for vmemmap mm/hugetlb: Free the vmemmap pages associated with each hugetlb page mm/hugetlb: Defer freeing of hugetlb pages mm/hugetlb: Allocate the vmemmap pages associated with each hugetlb page mm/hugetlb: Introduce remap_huge_page_pmd_vmemmap helper mm/hugetlb: Use PG_slab to indicate split pmd mm/hugetlb: Support freeing vmemmap pages of gigantic page mm/hugetlb: Add a BUILD_BUG_ON to check if struct page size is a power of two mm/hugetlb: Set the PageHWPoison to the raw error page mm/hugetlb: Flush work when dissolving hugetlb page mm/hugetlb: Add a kernel parameter hugetlb_free_vmemmap mm/hugetlb: Merge pte to huge pmd only for gigantic page mm/hugetlb: Gather discrete indexes of tail page mm/hugetlb: Add BUILD_BUG_ON to catch invalid usage of tail struct page Documentation/admin-guide/kernel-parameters.txt | 9 + Documentation/admin-guide/mm/hugetlbpage.rst | 3 + arch/x86/include/asm/hugetlb.h | 17 + arch/x86/include/asm/pgtable_64_types.h | 8 + arch/x86/mm/init_64.c | 7 +- fs/Kconfig | 16 + include/linux/bootmem_info.h | 79 +++ include/linux/hugetlb.h | 45 ++ include/linux/hugetlb_cgroup.h | 15 +- include/linux/memory_hotplug.h | 27 - include/linux/mm.h | 49 ++ mm/Makefile | 1 + mm/bootmem_info.c | 124 ++++ mm/hugetlb.c | 806 +++++++++++++++++++++++- mm/memory_hotplug.c | 116 ---- mm/sparse-vmemmap.c | 31 + mm/sparse.c | 5 +- 17 files changed, 1181 insertions(+), 177 deletions(-) create mode 100644 include/linux/bootmem_info.h create mode 100644 mm/bootmem_info.c