RE: [SSI-devel] Can you review and let me know if this is a good fix for shm.c?

[SAMPATHKUMAR K. K. <sam...@in...>]

 Hi Laura,
 
 I have incorporated your patch, built successfully and have booted my
 test machine with this kernel. I have also run my manual test that
 previously exposed this problem. It runs successfully now.
 
 I checked-in the kernel file ipc/shm.c just now.
 
 This should have checked-in much earlier. Sorry for the delay.
 
 NOTE:
 I had checked-in the f_mapping fix quite early on, and all instances
 of those were fixed as well along with that check-in.
 
 Regards,
 - Kishore
 
[ Laura Ramirez wrote ... ]
 
+ 
+ Hi Kishore,
+ 
+ I looked at your patch...you are right about the locking, i mistakenly
+ added the shm_ids locks in the wrong place last august...since its
+ only in the error case path, i would imagine the LTP tests would hit it
+ if there was any contention for the lock. 
+ 
+ I'm attaching a patch that fixes the locking. In the new patch, the checks
+ are done under the lock and only releases the lock if the it needs
+ to get the shm_ids lock for the destroy.  This patch doesnt include your 
+ f_mapping fix, but it should.  There are several places where f_op 
+ gets assigned to cfs_shm_file_operations, please make sure that the
+ f_mapping gets assigned as well. 
+ 
+ laura
+ 
+ 
+ > Hi John,
+ >   
+ > While working on one of the hangs related to shared memory during some
+ > ad-hoc testing I did (this is a differnt hang from the one corresponding
+ > to the test program you sent), I was not able to conclude definitively,
+ > but, I believe the attached patch contains the fix.
+ >   
+ > Can you look through this patch and let me know if this is a good fix?
+ >   
+ > I was convinced the problem that the patch tries is fix is indeed a
+ > serious problem. However, I did not want to jump ahead and conclude. So,
+ > I wanted to know, do you think it is useful to "definitively" reproduce
+ > this problem via a test case? If so, I'll try do that.
+ >   
+ > NOTE: I'll send the fix for the shm hang that can be reproduced by the
+ >       test program you sent.
+ >   
+ > Regards,
+ > - Kishore
+ 
+ Content-Description: shm_kishore.patch
+ 
+ [ Attachment, skipping... ]