Simon Munton wrote:
>
> Hi,
>
> I've found a problem in the issue() function in sanei_scsi.c, that results
> in the queue in sg filling up, and prevents any more scsi commands being
> issued until the device is closed.
>
> There is a test on line 1690 which can erroneously try to reissue a scsi
> command:
> if (rp == fdp->sane_qhead && errno == EAGAIN)
>
> The first time through the enclosing while loop, rp == fdp->sane_qhead. But
> if the write() call issuing the command was successful, errno is left
> unchanged, and if errno happens to contain EAGAIN, then the command is
> issued again, and again... until the command queue in sg is full.
Simon,
This may explain the strange behaviour that I saw on a
Umax 1220S and reported to Oliver Rauch. At the time we
couldn't resolve it. Excerpt from a post on 2000/05/04
follows:
---------------------------------------------------------
I was watching the progress of a scan in sane 1.0.2 (xsane 0.57
I think) via debug in a recent 2.3.99 linux kernel.
This output comes from: 'cat /proc/scsi/sg/debug' :
dev_max(currently)=11 max_active_device=5 (origin 1)
scsi_dma_free_sectors=5680 sg_pool_secs_aval=320 def_reserved_size=32768
>>> device=sg1 scsi0 chan=0 id=5 lun=0 em=0 sg_tablesize=50 excl=1
FD(1): timeout=600000ms bufflen=131072 (res)sgat=4 low_dma=1
cmd_q=1 f_packid=0 k_orphan=0 closed=0
rcv: id=95 blen=0 dur=10ms sgat=0 op=0x31
rcv: id=95 blen=0 dur=260ms sgat=0 op=0x31
rcv: id=95 blen=0 dur=500ms sgat=0 op=0x31
act: id=95 blen=0 t_o/elap=10000/690ms sgat=0 op=0x31
act: id=95 blen=0 t_o/elap=10000/670ms sgat=0 op=0x31
act: id=95 blen=0 t_o/elap=10000/470ms sgat=0 op=0x31
act: id=95 blen=0 t_o/elap=10000/210ms sgat=0 op=0x31
prior: id=95 blen=0 t_o/elap=10000/0ms sgat=0 op=0x31
It indicates 4 queued up OBJECT POSITION (op=0x31)
commands [act], one waiting to get on the queue [prior]
and 3 finished [rcv] awaiting read()s . So that's 8
outstanding write()s.
---------------------------------------------------------
Oliver said that command should only have been issued once.
Repeating the test with more SANE debug on made the problem
disappear :-) Now if the Umax driver uses that issue()
function we may have an answer.
Doug Gilbert
-- Source code, list archive, and docs: http://www.mostang.com/sane/ To unsubscribe: echo unsubscribe sane-devel | mail majordomo@mostang.com
This archive was generated by hypermail 2b29 : Thu May 25 2000 - 21:03:48 PDT