> -----Original Message----- > From: Keith Busch [mailto:keith.busch@xxxxxxxxx] > Sent: Friday, November 6, 2015 6:58 PM > Subject: Re: [PATCH 4/5] NVMe: add blk polling support > > On Fri, Nov 06, 2015 at 03:46:07PM -0800, Elliott, Robert wrote: > > > -----Original Message----- > > > From: linux-kernel-owner@xxxxxxxxxxxxxxx [mailto:linux-kernel- > > > owner@xxxxxxxxxxxxxxx] On Behalf Of Jens Axboe > > > Sent: Friday, November 6, 2015 11:20 AM > > ... > > > Subject: [PATCH 4/5] NVMe: add blk polling support > > > > > > @@ -953,6 +953,8 @@ static int nvme_process_cq(struct nvme_queue > *nvmeq) > > > head = 0; > > > phase = !phase; > > > } > > > + if (tag && *tag == cqe.command_id) > > > + *tag = -1; > > > ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn); > > > fn(nvmeq, ctx, &cqe); > > > } > > > > The NVMe completion queue entries are 16 bytes long. Although it's > > most likely to write them from 0..15 in one PCIe Memory Write > > transaction, the NVMe device could write those bytes in any order. > > It could thus update the command identifier before the other bytes, > > causing this code to process invalid stale values in the other > > fields. > > That's a very interesting point. We are okay if we can rely on the phase > bit, which we can by my reading of the spec. Coalescing would not work > if the driver can observe a new phase in a partially written CQE. The Phase bit is in the same Dword as the Command Identifier, so it doesn't help. If that Dword were written first, then the preceding three Dwords might not be valid yet. I'll ask our NVMe representative to propose wording like this: 4.6 Completion Queue Entry ... The controller shall write the Dwords in a CQE from lowest to highest (e.g., if the CQE is 16 bytes, write Dword 0, then Dword 1, then Dword 2, then Dword 3). This ensures that the first three Dwords are valid when the Phase Tag and Command Identifier are valid. Additional Dwords, if any, are not valid at that time. and refer to that rule in 7.1 where host processing of completions is discussed. --- Robert Elliott, HPE Persistent Memory -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
