[dm-devel] Re: [PATCH resubmit] do_mount: reduce stack consumption

[Neil Brown]

On Monday November 7, heiko.carstens at de.ibm.com wrote:
> > > Ok, I'll dust it off, make sure it seems to work (at the time I first
> > > wrote it, I think it caused 'md' to deadlock) and submit it.
> > > 
> > > NeilBrown
> > > 
> > 
> > For your consideration and testing (it works for me, but I'd like to
> > see it tested a bit more heavily in a variety of configurations).
> 
> Works fine for me. Thank you!
> 

Thanks for the feed-back.

> However I noticed another recursive call while dumping all call traces:
>  ...
>  [<00000000001e1952>] zfcp_scsi_command_async+0xf6/0x1c4
>  [<000000000013cd6a>] scsi_dispatch_cmd+0x1ae/0x37c
>  [<0000000000143dc4>] scsi_request_fn+0x1dc/0x390
>  [<000000000012f65a>] generic_unplug_device+0x2a/0x38
>  [<0000000000181262>] dm_table_unplug_all+0x42/0x58
>  [<000000000017e598>] dm_unplug_all+0x2c/0x44
>  [<0000000000181262>] dm_table_unplug_all+0x42/0x58
>  [<000000000017e598>] dm_unplug_all+0x2c/0x44
>  [<00000000000782aa>] block_sync_page+0x4e/0x68
>  [<00000000000514f8>] sync_page+0x48/0x68
>  [<00000000002ab36e>] __wait_on_bit+0x9a/0xe0
>  ...

Hmm.. This (and a similar one with ->issue_flush_fn) would be a lot
harder to unwind, but it is also much more benign.  These sorts of
calls should always be very shallow.
I don't think there is anything useful that can be done here, and I'm
not convinced there is a real problem.
It is worth noticing it and keeping it in mind though.

Thanks,
NeilBrown