Re: [PATCH v4 3/5] mm: Speedup mremap on 1GB or larger regions

From: Kalesh Singh
Date: Thu Dec 17 2020 - 13:16:51 EST


On Thu, Dec 17, 2020 at 12:28 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>
> On Wed, Oct 14, 2020 at 12:53:08AM +0000, Kalesh Singh wrote:
> > Android needs to move large memory regions for garbage collection.
> > The GC requires moving physical pages of multi-gigabyte heap
> > using mremap. During this move, the application threads have to
> > be paused for correctness. It is critical to keep this pause as
> > short as possible to avoid jitters during user interaction.
> >
> > Optimize mremap for >= 1GB-sized regions by moving at the PUD/PGD
> > level if the source and destination addresses are PUD-aligned.
> > For CONFIG_PGTABLE_LEVELS == 3, moving at the PUD level in effect moves
> > PGD entries, since the PUD entry is “folded back” onto the PGD entry.
> > Add HAVE_MOVE_PUD so that architectures where moving at the PUD level
> > isn't supported/tested can turn this off by not selecting the config.
> >
> > Fix build test error from v1 of this series reported by
> > kernel test robot in [1].
> >
> > [1] https://lists.01.org/hyperkitty/list/kbuild-all@xxxxxxxxxxxx/thread/CKPGL4FH4NG7TGH2CVYX2UX76L25BTA3/
> >
> > Signed-off-by: Kalesh Singh <kaleshsingh@xxxxxxxxxx>
> > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > Acked-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>
> I thought I reported it, but maybe I didn't. This patch causes all
> 'parisc' qemu emulations to fail. Typical log:
>
> Freeing unused kernel memory: 604K
> Write protected read-only-after-init data: 2k
> rodata_test: end of .rodata is not page size aligned
> Run /sbin/init as init process
> process '/bin/busybox' started with executable stack
> Failed to execute /sbin/init (error -12)
>
> Reverting this patch fixes the problem.
>
> Bisect log from linux-next below. The patch (and thus the problem)
> are now in mainline Linux.

Hi Guenter,

Thanks for reporting. We enabled this on x86 and arm64. Investigating
what the root cause is now.

Kalesh
>
> Guenter
>
> ---
> # bad: [7bba37a1591369e2e506d599b8f5d7d0516b2dbc] Add linux-next specific files for 20201214
> # good: [0477e92881850d44910a7e94fc2c46f96faa131f] Linux 5.10-rc7
> git bisect start 'HEAD' 'v5.10-rc7'
> # good: [fe5c40ab90a1f82ba97294637eaf875cfdd7a05f] Merge remote-tracking branch 'nand/nand/next'
> git bisect good fe5c40ab90a1f82ba97294637eaf875cfdd7a05f
> # good: [674a0d6de8bd290671f7dff405205871a70300b3] Merge remote-tracking branch 'spi/for-next'
> git bisect good 674a0d6de8bd290671f7dff405205871a70300b3
> # good: [8623dae312f73a2ea3230b1c648d3004cfc224ce] Merge remote-tracking branch 'vfio/next'
> git bisect good 8623dae312f73a2ea3230b1c648d3004cfc224ce
> # good: [dd26635f54bcd8e5d4e875a209f82a0423ba9c08] Merge remote-tracking branch 'gpio/for-next'
> git bisect good dd26635f54bcd8e5d4e875a209f82a0423ba9c08
> # good: [86e9c9a734889fe437442e0a35eb4c61d319cb47] Merge remote-tracking branch 'memblock/for-next'
> git bisect good 86e9c9a734889fe437442e0a35eb4c61d319cb47
> # bad: [3452331fda80b1cb5e121e6718ca6c07264382b2] userfaultfd: add user-mode only option to unprivileged_userfaultfd sysctl knob
> git bisect bad 3452331fda80b1cb5e121e6718ca6c07264382b2
> # bad: [19f468d54fcffc3f98b71e3e12ff23726767d953] mremap: don't allow MREMAP_DONTUNMAP on special_mappings and aio
> git bisect bad 19f468d54fcffc3f98b71e3e12ff23726767d953
> # good: [d89f3ababcac54493a4cb0582c61eb5f426b44e3] mm: remove pagevec_lookup_range_nr_tag()
> git bisect good d89f3ababcac54493a4cb0582c61eb5f426b44e3
> # good: [eba8373dcb40d30952f31d5fc0cff56b78f46273] mm/mlock: remove __munlock_isolate_lru_page()
> git bisect good eba8373dcb40d30952f31d5fc0cff56b78f46273
> # good: [8831d3f3564beba0f3f1b5291c88b35725bc45c9] xen/unpopulated-alloc: consolidate pgmap manipulation
> git bisect good 8831d3f3564beba0f3f1b5291c88b35725bc45c9
> # bad: [b8d53d70851821d8a2040ddca3aa6ee88fc8aaec] mm/page_vma_mapped.c: add colon to fix kernel-doc markups error for check_pte
> git bisect bad b8d53d70851821d8a2040ddca3aa6ee88fc8aaec
> # bad: [e77846c3da1862faa25c08e186a62b03e98c862f] x86: mremap speedup - Enable HAVE_MOVE_PUD
> git bisect bad e77846c3da1862faa25c08e186a62b03e98c862f
> # bad: [72ad8951bac1c559ea1b691a0b035fb339e4d71d] mm: speedup mremap on 1GB or larger regions
> git bisect bad 72ad8951bac1c559ea1b691a0b035fb339e4d71d
> # good: [fa94bfe31609787501a1ff8d7659ade5734ec4e5] kselftests: vm: add mremap tests
> git bisect good fa94bfe31609787501a1ff8d7659ade5734ec4e5
> # first bad commit: [72ad8951bac1c559ea1b691a0b035fb339e4d71d] mm: speedup mremap on 1GB or larger regions
>
> --
> To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@xxxxxxxxxxx.
>