Re: [PATCH v2 00/12] IB: Replace safe uses for ib_get_dma_mr with pd->local_dma_lkey

[Bart Van Assche <bart.vanassche-XdAiOPVOjttBDgjK7y7TUQ@public.gmane.org>]

On 08/05/2015 12:51 PM, Jason Gunthorpe wrote:
> On Tue, Aug 04, 2015 at 11:41:16PM -0700, David Dillow wrote:
>> On Tue, 2015-08-04 at 12:09 -0600, Jason Gunthorpe wrote:
>>> On Mon, Aug 03, 2015 at 11:33:51AM -0700, Bart Van Assche wrote:
>>>>> Bart, do you know what hardware this workaround is for?
>>>>
>>>> I hope the HW vendors can comment on this. Sorry but I'm not sure which HCA
>>>> models and/or firmware versions do not support FMR mapping with a non-zero
>>>> offset.
>>>
>>> Perhaps David can remember why he added this:
>>>
>>> commit 8f26c9ff9cd0317ad867bce972f69e0c6c2cbe3c
>>
>> It's originally from commit 559ce8f150d7d031c79c4d79173860f1bdfe3ce4,
>> and the list's attempts at code archaeology failed us:
>> http://article.gmane.org/gmane.linux.drivers.rdma/7149
>
> Okay.. So over time we went from a clear target specific bug described
> 9 years ago in 559ce through chinese whispers to a general unspecific
> fear of non-zero offset FMR?
>
> But nobody has described FMR failure in this way in the past 9 years
> with any specificity?
>
> My random guesses would be broken mthca firmware at the start of the
> FMR feature (long since fixed) or a wonky target that is now 10 years
> obsolete..
>
> If it was an HCA bug, I strongly have to think it is fixed now. We use
> FMR all over the place and SRP is the only area I've noticed this
> restriction..
>
> If it is a target bug, then FRWR should trigger it as wel, so we are
> already about to revert that workaround.
>
> I'm inclined to drop this entirely.. What do you think Bart?

Even today I see memory corruption at the initiator side with FMR and 
not with FR if I leave out the alignment check. Since this only occurs 
with FMR and not with FR that excludes the target side as a possible 
cause. I will have a closer look at the srp_map_finish_fmr() function. 
Always passing 0 as the second argument of srp_map_desc() in 
srp_map_finish_fmr() is probably wrong. Before 2011 that second argument 
was the page offset of the first sg-list entry.

Bart.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html