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

[James Bottomley <jbottomley@parallels.com>]

On Wed, 2014-07-16 at 04:01 -0700, hch@infradead.org wrote:
> On Sun, Jul 13, 2014 at 08:58:34AM -0400, Martin K. Petersen wrote:
> > >>>>> "KY" == KY Srinivasan <kys@microsoft.com> writes:
> > 
> > KY> Windows hosts do support UNMAP and set the field in the
> > KY> EVPD. However, since the host advertises SPC-2 compliance, Linux
> > KY> does not even query the VPD page.
> >  
> > >> If we want to enable UNMAP in this case I'd prefer a blacklist entry
> > >> than trying UNMAP despite the device not advertising it.
> > 
> > I agree with that. We could do something like the patch below.
> > 
> > However, I do think it's a good idea that you guys are looking into
> > reporting SPC-3.
> 
> KY mentioned that they have a prototype for that now.
> 
> Btw, I looked over sd.c a bit more, and I think I understand why they
> get the WRITE SAME commands now:
> 
> read_capacity_16 calls sd_config_discard(sdkp, SD_LBP_WS16) if the LPBME
> bit is set.  At least older SBC drafts left it wide open if a target
> supports WRITE SAME with UNMAP or UNMAP in this case.  So I think we'd
> still want a patch to use UNMAP instead of WRITE SAME for this case,
> which should also fix hyperv.  Below is the quick hack version of that
> that just checks the host no_write_same flag, as the one on the device
> isn't set yet - I guess we need to refactor some of that logic.
> 
> 
> diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
> index 87566b5..4480fdf 100644
> --- a/drivers/scsi/sd.c
> +++ b/drivers/scsi/sd.c
> @@ -2035,7 +2035,10 @@ static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
>  		if (buffer[14] & 0x40) /* LBPRZ */
>  			sdkp->lbprz = 1;
>  
> -		sd_config_discard(sdkp, SD_LBP_WS16);
> +		if (sdp->host->no_write_same)
> +			sd_config_discard(sdkp, SD_LBP_UNMAP);
> +		else
> +			sd_config_discard(sdkp, SD_LBP_WS16);

Right, I already said this was the problem: that was the reason for my
patch.  However, there are a couple of other cases (including the /sys
entry) which is why I patched sd_config_discard instead.

James