Re: [PATCH] block: fix splitting segments

From: Ming Lei
Date: Tue Jan 07 2020 - 17:30:55 EST


On Tue, Jan 07, 2020 at 10:11:45AM -0800, Guenter Roeck wrote:
> On Tue, Jan 07, 2020 at 11:23:39PM +0800, Ming Lei wrote:
> > On Tue, Jan 07, 2020 at 04:47:08AM -0800, Guenter Roeck wrote:
> > > Hi,
> > >
> > > On Sun, Dec 29, 2019 at 10:32:30AM +0800, Ming Lei wrote:
> > > > There are two issues in get_max_segment_size():
> > > >
> > > > 1) the default segment boudary mask is bypassed, and some devices still
> > > > require segment to not cross the default 4G boundary
> > > >
> > > > 2) the segment start address isn't taken into account when checking
> > > > segment boundary limit
> > > >
> > > > Fixes the two issues.
> > > >
> > > > Fixes: dcebd755926b ("block: use bio_for_each_bvec() to compute multi-page bvec count")
> > > > Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
> > >
> > > This patch, pushed into mainline as "block: fix splitting segments on
> > > boundary masks", results in the following crash when booting 'versatilepb'
> > > in qemu from disk. Bisect log is attached. Detailed log is at
> > > https://kerneltests.org/builders/qemu-arm-master/builds/1410/steps/qemubuildcommand/logs/stdio
> > >
> > > Guenter
> > >
> > > ---
> > > Crash:
> > >
> > > kernel BUG at block/bio.c:1885!
> > > Internal error: Oops - BUG: 0 [#1] ARM
> >
> > Please apply the following debug patch, and post the log.
> >
>
> Here you are:
>
> max_sectors 2560 max_segs 96 max_seg_size 65536 mask ffffffff
> c738da80: 8c80/0 2416 28672, 0
> total sectors 56
>
> (I replaced %p with %px).
>

Please try the following patch and see if it makes a difference.
If not, replace trace_printk with printk in previous debug patch,
and apply the debug patch only & post the log.

diff --git a/block/blk-merge.c b/block/blk-merge.c
index 347782a24a35..f152bdee9b05 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -159,12 +159,12 @@ static inline unsigned get_max_io_size(struct request_queue *q,

static inline unsigned get_max_segment_size(const struct request_queue *q,
struct page *start_page,
- unsigned long offset)
+ unsigned long long offset)
{
unsigned long mask = queue_segment_boundary(q);

offset = mask & (page_to_phys(start_page) + offset);
- return min_t(unsigned long, mask - offset + 1,
+ return min_t(unsigned long long, mask - offset + 1,
queue_max_segment_size(q));
}


Thanks,
Ming