Hi Vladimir,
On Fri, Oct 30, 2020 at 02:20:31PM +0200, Vladimir Zapolskiy wrote:
> Hello Gao Xiang,
>
> On 10/22/20 5:57 PM, Gao Xiang via Linux-erofs wrote:
> > From: Gao Xiang <hsiangkao@xxxxxxxxxx>
> >
> > pcluster should be only set up for all managed pages instead of
> > temporary pages. Since it currently uses page->mapping to identify,
> > the impact is minor for now.
> >
> > Fixes: 5ddcee1f3a1c ("erofs: get rid of __stagingpage_alloc helper")
> > Cc: <stable@xxxxxxxxxxxxxxx> # 5.5+
> > Signed-off-by: Gao Xiang <hsiangkao@xxxxxxxxxx>
>
> I was looking exactly at this problem recently, my change is one-to-one
> to your fix, thus I can provide a tag:
>
> Tested-by: Vladimir Zapolskiy <vladimir@xxxxxxxxxx>
Many thanks for confirming this!
I found this when I was killing magical stagingpage page->mapping,
it's somewhat late :-)
>
>
> The fixed problem is minor, but the kernel log becomes polluted, if
> a page allocation debug option is enabled:
>
> % md5sum ~/erofs/testfile
> BUG: Bad page state in process kworker/u9:0 pfn:687de
> page:0000000057b8bcb4 refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x687de
> flags: 0x4000000000002000(private)
> raw: 4000000000002000 dead000000000100 dead000000000122 0000000000000000
> raw: 0000000000000000 ffff888066758690 00000000ffffffff 0000000000000000
> page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
> Modules linked in:
> CPU: 1 PID: 602 Comm: kworker/u9:0 Not tainted 5.9.1 #2
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
> Workqueue: erofs_unzipd z_erofs_decompressqueue_work
> Call Trace:
> dump_stack+0x84/0xba
> bad_page.cold+0xac/0xb1
> check_free_page_bad+0xb0/0xc0
> free_pcp_prepare+0x2c8/0x2d0
> free_unref_page+0x18/0xf0
> put_pages_list+0x11a/0x120
> z_erofs_decompressqueue_work+0xc9/0x110
> ? z_erofs_decompress_pcluster.isra.0+0xf10/0xf10
> ? read_word_at_a_time+0x12/0x20
> ? strscpy+0xc7/0x1a0
> process_one_work+0x30c/0x730
> worker_thread+0x91/0x640
> ? __kasan_check_read+0x11/0x20
> ? rescuer_thread+0x8a0/0x8a0
> kthread+0x1dd/0x200
> ? kthread_unpark+0xa0/0xa0
> ret_from_fork+0x1f/0x30
> Disabling lock debugging due to kernel taint
Yeah, I can make a pull-request to Linus if you need this to be in master
now, or I can post it for v5.11-rc1 since 5.4 LTS isn't effected (and it
would be only a print problem with debugging option.)
Thanks,
Gao Xiang
>
> --
> Best wishes,
> Vladimir
>
