[PATCH 4/8] Drivers: scsi: storvsc: Filter WRITE_SAME_16

[KY Srinivasan]

> -----Original Message-----
> From: Martin K. Petersen [mailto:martin.petersen at oracle.com]
> Sent: Friday, July 11, 2014 5:54 AM
> To: hch at infradead.org
> Cc: James Bottomley; KY Srinivasan; linux-kernel at vger.kernel.org;
> devel at linuxdriverproject.org; apw at canonical.com; stable at vger.kernel.org;
> linux-scsi at vger.kernel.org; ohering at suse.com; jasowang at redhat.com
> Subject: Re: [PATCH 4/8] Drivers: scsi: storvsc: Filter WRITE_SAME_16
> 
> >>>>> "hch" == hch at infradead org <hch at infradead.org> writes:
> 
> (Back from vacation: Bear with me while I'm catching up on two weeks of
> linux-scsi stuff...)
> 
> hch> I think the problem is a differnet one.  If we have the logical
> hch> provisioning EVPD it configures what method to use, but if we don't
> hch> have one we simply check for a max unmap blocks field, and if
> hch> that's not present use WRITE SAME.
> 
> Yeah, that was done to accommodate the devices out there that predate the
> LBP VPD. There sadly are several still. And it's hard to motivate people to
> update their storage array firmware even when updates are readily available.
> 
> hch> The patch checks the no_write_same flag before doing that, for
> hch> which we also have to do the write_same setup before the discard
> hch> setup in sd_revalidate_disk.
> 
> The no_write_same was introduced to disable the REQ_WRITE_SAME use
> case where we have no INQUIRY/READ CAPACITY/VPD flags to key off of to
> determine whether it is safe to send the commands.
> 
> no_write_same does not preclude the REQ_DISCARD variants of
> WRITE_SAME. Those are gated by the discard heuristics exclusively.
> 
> So first of all I'd like to know whether it's a WRITE SAME(16) or a WRITE
> SAME(16) with the UNMAP bit set that's coming down.
> 
> hch> Ky: does hyperv support UNMAP?  If so any idea why it doesn't set
> hch> the maximum unmap block count field in the EVPD?
> 
> We shouldn't be sending down WRITE SAME(16) with the UNMAP bit set
> unless the device sets LBPME=1 in the READ CAPACITY(16) response.
> 
> So what does the storsvc report as its thin provisioning capabilities?
Given that our current Host (ws2012 r2) advertises SPC-2 compliance, Linux does not even query this page. We support UNMAP.
We just prototyped a host where we advertised SPC-3 compliance and Linux queries the EVPD page and does not use WRITE_SAME_16

K. Y