Re: kern/42455: tstile hang with nfs

["Christoph Egger" <Christoph_Egger%gmx.de@localhost>]

db> ps /w
PID   LID          COMMAND     EMUL  PRI WAIT-MSG    WAIT-CHANNEL
1206    1             sync   netbsd   43 tstile       ffffa00026b52498
327   237          qemu-dm   netbsd   43              0
327     2          qemu-dm   netbsd   43 netio        ffffa0000114f3b8
327     1          qemu-dm   netbsd   34 genput       ffffa000005a0840

db> show lock 0xffffa00026b52498
lock address : 0xffffa00026b52498 type     :     sleep/adaptive
initialized  : 0xffffffff80693b1b
shared holds :                  0 exclusive:                  1
shares wanted:                  0 exclusive:                  1
current cpu  :                  0 last held:                  0
current lwp  : 0xffffa0002320ec00 last held: 0xffffa00026628400
last locked  : 0xffffffff8069161d unlocked : 0xffffffff80691696
owner/count  : 0xffffa00026628400 flags    : 0x0000000000000007

Turnstile chain at 0xffffffff80c383f0.
=> Turnstile at 0xffffa000260de308 (wrq=0xffffa000260de328, rdq=0xffffa000260de3
38).
=> 0 waiting readers:
=> 1 waiting writers: 0xffffa00026b57000

This is the hanging 'sync' process:

db> tr /a 0xffffa00026b57000
trace: pid 1206 lid 1 at 0xffffa00026bffa00
sleepq_block() at netbsd:sleepq_block+0xbf
turnstile_block() at netbsd:turnstile_block+0x2c7
rw_enter() at netbsd:rw_enter+0x1d0
vlockmgr() at netbsd:vlockmgr+0xe2
VOP_LOCK() at netbsd:VOP_LOCK+0x28
vn_lock() at netbsd:vn_lock+0xd5
vget() at netbsd:vget+0xe9
nfs_sync() at netbsd:nfs_sync+0xb9
sys_sync() at netbsd:sys_sync+0xe6
syscall() at netbsd:syscall+0xa8


This is the hanging 'qemu-dm' process:

db> tr /a 0xffffa00026628400
trace: pid 327 lid 1 at 0xffffa0002661e880
sleepq_block() at netbsd:sleepq_block+0xbf
mtsleep() at netbsd:mtsleep+0x128
genfs_do_putpages() at netbsd:genfs_do_putpages+0x7f9
VOP_PUTPAGES() at netbsd:VOP_PUTPAGES+0x30
nfs_flush() at netbsd:nfs_flush+0x27
VOP_FSYNC() at netbsd:VOP_FSYNC+0x34
sys_fsync() at netbsd:sys_fsync+0x53
syscall() at netbsd:syscall+0xa8


So it is qemu-dm who hangs first and any subsequent 'sync'
call hangs, too.

That explains why the Xen DomU suddendly hangs during IO.

The question is why does the 'sync' syscall hang in first
place ?

Christoph