On Tue, Oct 9, 2018 at 3:19 AM Jan Kara <jack@xxxxxxx> wrote:
>
> Currently _PAGE_DEVMAP bit is not preserved in mprotect(2) calls. As a
> result we will see warnings such as:
>
> BUG: Bad page map in process JobWrk0013 pte:800001803875ea25 pmd:7624381067
> addr:00007f0930720000 vm_flags:280000f9 anon_vma: (null) mapping:ffff97f2384056f0 index:0
> file:457-000000fe00000030-00000009-000000ca-00000001_2001.fileblock fault:xfs_filemap_fault [xfs] mmap:xfs_file_mmap [xfs] readpage: (null)
> CPU: 3 PID: 15848 Comm: JobWrk0013 Tainted: G W 4.12.14-2.g7573215-default #1 SLE12-SP4 (unreleased)
> Hardware name: Intel Corporation S2600WFD/S2600WFD, BIOS SE5C620.86B.01.00.0833.051120182255 05/11/2018
> Call Trace:
> dump_stack+0x5a/0x75
> print_bad_pte+0x217/0x2c0
> ? enqueue_task_fair+0x76/0x9f0
> _vm_normal_page+0xe5/0x100
> zap_pte_range+0x148/0x740
> unmap_page_range+0x39a/0x4b0
> unmap_vmas+0x42/0x90
> unmap_region+0x99/0xf0
> ? vma_gap_callbacks_rotate+0x1a/0x20
> do_munmap+0x255/0x3a0
> vm_munmap+0x54/0x80
> SyS_munmap+0x1d/0x30
> do_syscall_64+0x74/0x150
> entry_SYSCALL_64_after_hwframe+0x3d/0xa2
> ...
>
> when mprotect(2) gets used on DAX mappings. Also there is a wide variety
> of other failures that can result from the missing _PAGE_DEVMAP flag
> when the area gets used by get_user_pages() later.
>
> Fix the problem by including _PAGE_DEVMAP in a set of flags that get
> preserved by mprotect(2).
>
> Fixes: 69660fd797c3 ("x86, mm: introduce _PAGE_DEVMAP")
> Fixes: ebd31197931d ("powerpc/mm: Add devmap support for ppc64")
> CC: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Jan Kara <jack@xxxxxxx>
Looks good, do you want me to take this upstream along with the livelock fix?
