On 11/10/2010 05:40 PM, James Bottomley wrote:
> Your commit:
>
> [SCSI] host lock push-down
>
> Move the mid-layer's ->queuecommand() invocation from being locked
> with the host lock to being unlocked to facilitate speeding up the
> critical path for drivers who don't need this lock taken anyway.
>
> The patch below presents a simple SCSI host lock push-down as an
> equivalent transformation. No locking or other behavior should change
> with this patch. All existing bugs and locking orders are preserved.
>
> Minimal code disturbance was attempted with this change. Most drivers
> needed only two one-line modifications for their host lock push-down.
>
> Signed-off-by: Jeff Garzik<[email protected]>
> Signed-off-by: James Bottomley<[email protected]>
>
> has been added to the upstream SCSI tree
> You can find it here:
No comments on renaming ->queuecommand to something else?
The consequences are rather dire if this goes unnoticed, yes?
Jeff
In http://marc.info/?l=linux-ide&m=128891665713984&w=2 I wrote:
> An alternate arrangement, not presented by this patch, might
> be preferred: in order to make it clear that queuecommand
> locking has changed, one could s/queuecommand/queuecommand_nl/ in
> Scsi_Host_Template, in order to guarantee that drivers are either
> (a) upgraded or (b) broken at compile time. Compile-time detection of
> new locking may be desirable, and I'll volunteer to change my patch to
> do that, if community members prefer that route instead of below.
On Wed, 2010-11-10 at 18:28 -0500, Jeff Garzik wrote:
> On 11/10/2010 05:40 PM, James Bottomley wrote:
> > Your commit:
> >
> > [SCSI] host lock push-down
> >
> > Move the mid-layer's ->queuecommand() invocation from being locked
> > with the host lock to being unlocked to facilitate speeding up the
> > critical path for drivers who don't need this lock taken anyway.
> >
> > The patch below presents a simple SCSI host lock push-down as an
> > equivalent transformation. No locking or other behavior should change
> > with this patch. All existing bugs and locking orders are preserved.
> >
> > Minimal code disturbance was attempted with this change. Most drivers
> > needed only two one-line modifications for their host lock push-down.
> >
> > Signed-off-by: Jeff Garzik<[email protected]>
> > Signed-off-by: James Bottomley<[email protected]>
> >
> > has been added to the upstream SCSI tree
> > You can find it here:
>
> No comments on renaming ->queuecommand to something else?
What we wondered about doing differently isn't really relevant for a
change log ... that should just really be about what was done (to avoid
confusion).
> The consequences are rather dire if this goes unnoticed, yes?
You mean if there's a missed in-tree driver? Yes, but I took care to
make sure all SCSI drivers were accounted for. For out of tree drivers,
as with the eh lock push down, it's caveat emptor.
James
On 11/10/2010 06:45 PM, James Bottomley wrote:
> On Wed, 2010-11-10 at 18:28 -0500, Jeff Garzik wrote:
>> On 11/10/2010 05:40 PM, James Bottomley wrote:
>>> Your commit:
>>>
>>> [SCSI] host lock push-down
>>>
>>> Move the mid-layer's ->queuecommand() invocation from being locked
>>> with the host lock to being unlocked to facilitate speeding up the
>>> critical path for drivers who don't need this lock taken anyway.
>>>
>>> The patch below presents a simple SCSI host lock push-down as an
>>> equivalent transformation. No locking or other behavior should change
>>> with this patch. All existing bugs and locking orders are preserved.
>>>
>>> Minimal code disturbance was attempted with this change. Most drivers
>>> needed only two one-line modifications for their host lock push-down.
>>>
>>> Signed-off-by: Jeff Garzik<[email protected]>
>>> Signed-off-by: James Bottomley<[email protected]>
>>>
>>> has been added to the upstream SCSI tree
>>> You can find it here:
>>
>> No comments on renaming ->queuecommand to something else?
>
> What we wondered about doing differently isn't really relevant for a
> change log ... that should just really be about what was done (to avoid
> confusion).
Wasn't referring to the changelog (perhaps shouldn't have quoted that);
just asking the question generally.
>> The consequences are rather dire if this goes unnoticed, yes?
>
> You mean if there's a missed in-tree driver? Yes, but I took care to
> make sure all SCSI drivers were accounted for. For out of tree drivers,
> as with the eh lock push down, it's caveat emptor.
Thinking about out-of-tree drivers, yes.
Jeff
On Wed, 2010-11-10 at 18:52 -0500, Jeff Garzik wrote:
> On 11/10/2010 06:45 PM, James Bottomley wrote:
> > On Wed, 2010-11-10 at 18:28 -0500, Jeff Garzik wrote:
> >> On 11/10/2010 05:40 PM, James Bottomley wrote:
> >>> Your commit:
> >>>
> >>> [SCSI] host lock push-down
> >>>
> >>> Move the mid-layer's ->queuecommand() invocation from being locked
> >>> with the host lock to being unlocked to facilitate speeding up the
> >>> critical path for drivers who don't need this lock taken anyway.
> >>>
> >>> The patch below presents a simple SCSI host lock push-down as an
> >>> equivalent transformation. No locking or other behavior should change
> >>> with this patch. All existing bugs and locking orders are preserved.
> >>>
> >>> Minimal code disturbance was attempted with this change. Most drivers
> >>> needed only two one-line modifications for their host lock push-down.
> >>>
> >>> Signed-off-by: Jeff Garzik<[email protected]>
> >>> Signed-off-by: James Bottomley<[email protected]>
> >>>
> >>> has been added to the upstream SCSI tree
> >>> You can find it here:
> >>
> >> No comments on renaming ->queuecommand to something else?
> >
> > What we wondered about doing differently isn't really relevant for a
> > change log ... that should just really be about what was done (to avoid
> > confusion).
>
> Wasn't referring to the changelog (perhaps shouldn't have quoted that);
> just asking the question generally.
>
>
> >> The consequences are rather dire if this goes unnoticed, yes?
> >
> > You mean if there's a missed in-tree driver? Yes, but I took care to
> > make sure all SCSI drivers were accounted for. For out of tree drivers,
> > as with the eh lock push down, it's caveat emptor.
>
> Thinking about out-of-tree drivers, yes.
>
Hi Jeff and James,
Thank you for getting this initial patch merged. I really think this
was and is the best choice moving forward. Also, a seriously big thank
you to all of the other folks who have helped identify LLDs issues for
host_lock less mode for drivers!!
In the next days I will get a atomic_t scsi_host->cmd_serial_number
patch rebased (which is really very minor at this point w/o the
scsi_error.c changes), and merge the current host_lock-less 'scoreboard'
on top of jgarzik's code and tag for .38. So, please let me know if
you would to include minor the atomic_t scsi_host->cmd_serial_number
patch for .37 or if you would rather have this immediately preceed the
first series of "enable host-lock_less for LLD vendor superturbo hba"
for the .38 round.
Best,
--nab