lkml.org 
[lkml]   [2010]   [Nov]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [GIT PULL] SCSI queuecommand API change for 2.6.37-rc1
On 11/12/2010 08:42 PM, Linus Torvalds wrote:
> On Fri, Nov 12, 2010 at 3:55 PM, James Bottomley
> <James.Bottomley@suse.de> wrote:
>>
>> This patch set contains a single patch modifying the SCSI queuecommand
>> host template API to go from being called with the host lock held to
>> being called locklessly. The transformation is a directly equivalent
>> one (i.e. the locking is simply pushed into each HBA) but will form the
>> basis for optimising locking in the driver patch for the next merge
>> window.
>
> Ok, so we talked about this patch at the KS, but I never saw it.
>
> And now seeing it, I do detest it.
>
> Why? Because if some driver gets missed for any reason (notably if
> it's currently out of tree, and gets merged later), afaik there will
> be ABSOLUTELY ZERO compiler warnings or anything about it, because you
> kept the "queuecommand" function exactly the same. Whether it's a
> locked or non-locked one, it always is of type
[...]
> So please: when you change the semantics of a function, just change
> the function prototype (or function name) at the same time. Especially
> when it comes to a driver interface, so that the drivers don't get
> taken by surprise.
>
> Type safety and automatic compiler warnings really are our friends.
> Especially when the patch was presumably mostly auto-generated, and
> maybe the script missed something, and missing some conversion has
> such subtle effects.

That is precisely what I mentioned in the original patch:
http://marc.info/?l=linux-ide&m=128891665713984&w=2

And brought this up again when James merged my patch:
http://marc.info/?l=linux-scsi&m=128943169802009&w=2

I even volunteer[ed] to do the work to make it happen...

Jeff


P.S. My patch was not auto-generated. One of the two previous attempts
at this change was scripted w/ coccinelle, and it created obvious
locking problems.



\
 
 \ /
  Last update: 2010-11-13 03:05    [W:0.090 / U:1.228 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site