Re: [dm-devel] [PATCH V15 00/18] block: support multi-page bvec
From: Bart Van Assche
Date: Tue Feb 19 2019 - 11:28:28 EST
On Sun, 2019-02-17 at 21:11 +-0800, Ming Lei wrote:
+AD4 The following patch should fix this issue:
+AD4
+AD4
+AD4 diff --git a/block/blk-merge.c b/block/blk-merge.c
+AD4 index bed065904677..066b66430523 100644
+AD4 --- a/block/blk-merge.c
+AD4 +-+-+- b/block/blk-merge.c
+AD4 +AEAAQA -363,13 +-363,15 +AEAAQA static unsigned int +AF8AXw-blk+AF8-recalc+AF8-rq+AF8-segments(struct request+AF8-queue +ACo-q,
+AD4 struct bio+AF8-vec bv, bvprv +AD0 +AHs NULL +AH0AOw
+AD4 int prev +AD0 0+ADs
+AD4 unsigned int seg+AF8-size, nr+AF8-phys+AF8-segs+ADs
+AD4 - unsigned front+AF8-seg+AF8-size +AD0 bio-+AD4-bi+AF8-seg+AF8-front+AF8-size+ADs
+AD4 +- unsigned front+AF8-seg+AF8-size+ADs
+AD4 struct bio +ACo-fbio, +ACo-bbio+ADs
+AD4 struct bvec+AF8-iter iter+ADs
+AD4
+AD4 if (+ACE-bio)
+AD4 return 0+ADs
+AD4
+AD4 +- front+AF8-seg+AF8-size +AD0 bio-+AD4-bi+AF8-seg+AF8-front+AF8-size+ADs
+AD4 +-
+AD4 switch (bio+AF8-op(bio)) +AHs
+AD4 case REQ+AF8-OP+AF8-DISCARD:
+AD4 case REQ+AF8-OP+AF8-SECURE+AF8-ERASE:
Hi Ming,
With this patch applied test nvmeof-mp/002 fails as follows:
+AFs 694.700400+AF0 kernel BUG at lib/sg+AF8-pool.c:103+ACE
+AFs 694.705932+AF0 invalid opcode: 0000 +AFsAIw-1+AF0 PREEMPT SMP KASAN
+AFs 694.708297+AF0 CPU: 2 PID: 349 Comm: kworker/2:1H Tainted: G B 5.0.0-rc6-dbg+- +ACM-2
+AFs 694.711730+AF0 Hardware name: QEMU Standard PC (i440FX +- PIIX, 1996), BIOS 1.10.2-1 04/01/2014
+AFs 694.715113+AF0 Workqueue: kblockd blk+AF8-mq+AF8-run+AF8-work+AF8-fn
+AFs 694.716894+AF0 RIP: 0010:sg+AF8-alloc+AF8-table+AF8-chained+-0xe5/0xf0
+AFs 694.758222+AF0 Call Trace:
+AFs 694.759645+AF0 nvme+AF8-rdma+AF8-queue+AF8-rq+-0x2aa/0xcc0 +AFs-nvme+AF8-rdma+AF0
+AFs 694.764915+AF0 blk+AF8-mq+AF8-try+AF8-issue+AF8-directly+-0x2a5/0x4b0
+AFs 694.771779+AF0 blk+AF8-insert+AF8-cloned+AF8-request+-0x11e/0x1c0
+AFs 694.778417+AF0 dm+AF8-mq+AF8-queue+AF8-rq+-0x3d1/0x770
+AFs 694.793400+AF0 blk+AF8-mq+AF8-dispatch+AF8-rq+AF8-list+-0x5fc/0xb10
+AFs 694.798386+AF0 blk+AF8-mq+AF8-sched+AF8-dispatch+AF8-requests+-0x2f7/0x300
+AFs 694.803180+AF0 +AF8AXw-blk+AF8-mq+AF8-run+AF8-hw+AF8-queue+-0xd6/0x180
+AFs 694.808933+AF0 blk+AF8-mq+AF8-run+AF8-work+AF8-fn+-0x27/0x30
+AFs 694.810315+AF0 process+AF8-one+AF8-work+-0x4f1/0xa40
+AFs 694.813178+AF0 worker+AF8-thread+-0x67/0x5b0
+AFs 694.814487+AF0 kthread+-0x1cf/0x1f0
+AFs 694.819134+AF0 ret+AF8-from+AF8-fork+-0x24/0x30
The code in sg+AF8-pool.c that triggers the BUG() statement is as follows:
int sg+AF8-alloc+AF8-table+AF8-chained(struct sg+AF8-table +ACo-table, int nents,
struct scatterlist +ACo-first+AF8-chunk)
+AHs
int ret+ADs
BUG+AF8-ON(+ACE-nents)+ADs
+AFs ... +AF0
Bart.