On Wed, Jan 14, 2015 at 09:04:24AM -0800, Dan Williams wrote: > On Wed, Jan 14, 2015 at 8:30 AM, Shaohua Li <shli@xxxxxx> wrote: > > > On Wed, Jan 14, 2015 at 12:08:26AM -0800, Dan Williams wrote: > > > On Fri, Jan 9, 2015 at 2:12 PM, Shaohua Li <shli@xxxxxx> wrote: > > > > On Fri, Jan 09, 2015 at 01:59:19PM -0800, Shaohua Li wrote: > > > >> On Fri, Jan 09, 2015 at 04:43:07PM -0500, Tejun Heo wrote: > > > >> > On Fri, Jan 09, 2015 at 10:15:29AM -0800, Shaohua Li wrote: > > > >> > > Ping! > > > >> > > > > >> > I like the idea but it bothers me that we end up with two separate > > > >> > ways of allocating command tags. Any chance the old one can be > > > >> > removed? > > > > > > > > Oh, sorry, my reply truncated. > > > > > > > > So I checked with IPR driver guys, the ipr doesn't use ncq, so we can > > > > always use sata tag 0, but not sure for libsas. > > > > > > > > Maybe Dan can answer if there is a way we can map SCSI tag to SATA tag. > > > > > > > > > > For libsas or for ata drivers? For libsas, iirc, the internal libata > > > tag is ignored and we use the scsi tag for the sas task For ata some > > > drivers want round robin, but others appear to care about using the > > > lowest available: > > https://urldefense.proofpoint.com/v1/url?u=https://bugzilla.kernel.org/show_bug.cgi?id%3D87101&k=ZVNjlDMF0FElm4dQtryO4A%3D%3D%0A&r=3RFlL12A7nwmLRXunVJq2g%3D%3D%0A&m=i5Fz0FbsOgTMAtXW2WeXro9juvX%2BRmkKOK%2Fnl5ZZWMw%3D%0A&s=b8747b79ff51a364a841dd8e2ff51814536fcd63eedb7ad8f1de68e6691e5e98 > > . > > > > > > Is ata using it's own legacy tag ordering scheme getting in the way of > > > other improvements? I'd just as soon recommend letting legacy dogs > > > lie. In other words what do we gain by switching? > > > > > > I need to follow up on bz87101, seems I never reworked the patch as > > > Tejun asked. However, I'm glad a fix like the one proposed in that > > > report can be self-contained to libata and need not worry about > > > supporting ata specific quirks in the block layer tag ordering scheme. > > > > Basically I'd like using block tag instead of an open coded tag > > implementation in libata. The block tag implenmentation is more > > sophisticated, and with blk-mq tag is built-in, so it's great to remove > > tag implementation in libata. > > > Is it great to remove? It's not clear what the cost of maintaining it is. > How would you handle the case where some ata controllers want the lowest > available tag vs others that want round robin. I added a flag to blk-tag/blk-mq tag to implement different tag allocation strategy, for example, round robin and fifo. > > For sata, it's easy to do it. The problem > > is sas, which implements its own scsi driver/tag. My question is if we > > can map SCSI tag (of libsas) to ATA tag. So for example, the ipr sas > > driver doesn't use NCQ, so we can always use ata tag 0 in libata for > > ipr. I'm wondering if we can do some mapping in libsas too, so we can > > completely delete the tag code in libata. > > > > The problem with using the scsi tag for libsas is that the scsi_host may > support 256 commands, but only 32 per ata device. Seems like a problem > blk-mq has already solved (controller + per-endpoint tag queues), but I > have not looked... This is exactly the issue I'm talking about. Is there a way to map the 256 (scsi tag) to ata tag for libsas? Like I said, for ipr sas, I'll do scsi tag -> ata tag 0, since it doesn't use NCQ for SATA. Is there something similar or whatever for libsas? From search, ipr and libsas are the only scsi drivers using libata. Thanks, Shaohua -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
