On Tue, Mar 29, 2016 at 2:10 AM, Shaohua Li <shli@xxxxxxxxxx> wrote: > On Sat, Mar 26, 2016 at 05:46:16PM +0100, Sebastian Roesner wrote: >> Hello Ming, Eric, >> >> Am 26.03.2016 um 16:40 schrieb Ming Lei: >> > From the trace, looks the bio returned from bio_clone_mddev() is NULL, >> >and maybe the source bio is buggy, something like the attached patch >> >might be helpful, could you apply the attached debug patch, reproduce >> >and post the log? >> >> I was able to reproduce it on a non-productive system, but only after >> copying the bcache superblocks/partition starts from the original system, >> with new created ones it worked fine. >> >> Full trace and check_bio output can be found here: >> >> http://pastebin.com/ngvGGHBZ > > 320 bvecs exceeds what bio-clone_set can handle. Could you please try below patch? > The source bio is still a cloned bio which includes 320 bvecs, which is weried. And I guess bio_add_page() isn't used for the initial source bio from bcache. >From the stack trace, looks it is related with prio_io(), which just sets bio size as the bucket size, so could you check what the bucket size is about your bcache device via bcache-super-show or sysfs? If the bucket size is (320 * 4K), looks we should limit the queue's max sector as 1MB. BTW, in my bcache block, the bucket size is 512K, and I can't reproduce the issue, and it is stored in superblock. > commit 92761dad7ff6e1bf25de247e0064dd398e797599 > Author: Shaohua Li <shli@xxxxxx> > Date: Mon Mar 28 10:54:35 2016 -0700 > > block: don't make BLK_DEF_MAX_SECTORS too big > > bio_alloc_bioset() allocates bvecs from bvec_slabs which can only allocate > maximum 256 bvec (eg, 1M for 4k pages). We can't bump BLK_DEF_MAX_SECTORS to > exceed this value otherwise bio_alloc_bioset will fail. > > This fixes commit 30e2bc08b2bb7c069. We probably should make the bvec_slabs > hold bigger bvecs if bigger bio size is required. > > Signed-off-by: Shaohua Li <shli@xxxxxx> > > diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h > index 7e5d7e0..da64325 100644 > --- a/include/linux/blkdev.h > +++ b/include/linux/blkdev.h > @@ -1153,7 +1153,11 @@ extern int blk_verify_command(unsigned char *cmd, fmode_t has_write_perm); > enum blk_default_limits { > BLK_MAX_SEGMENTS = 128, > BLK_SAFE_MAX_SECTORS = 255, > - BLK_DEF_MAX_SECTORS = 2560, > + /* > + * if you change this, please also change bvec_alloc and BIO_MAX_PAGES. > + * Otherwise bio_alloc_bioset will break. > + */ > + BLK_DEF_MAX_SECTORS = BIO_MAX_SECTORS, > BLK_MAX_SEGMENT_SIZE = 65536, > BLK_SEG_BOUNDARY_MASK = 0xFFFFFFFFUL, > }; This change should work. But, when multipage bvecs is ready, I think the above limit can be increased too. I suggest to use (BIO_MAX_PAGES << (PAGE_SHIFT - 9)) instead of BIO_MAX_SECTORS, which is removed in my following patch and there is only one user of this macro: http://marc.info/?l=linux-kernel&m=145862727620599&w=2 Thanks, Ming -- To unsubscribe from this list: send the line "unsubscribe linux-bcache" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
