Skip site navigation (1)Skip section navigation (2) 
Date:      Tue, 26 May 2015 09:14:18 +0000
From:      bugzilla-noreply@freebsd.org
To:        freebsd-standards@FreeBSD.org
Subject:   [Bug 200458] [nvme] nvme_ctrlr_disable does not wait for the READY bit to become zero
Message-ID:  <bug-200458-15@https.bugs.freebsd.org/bugzilla/>
next in thread | raw e-mail | index | archive | help 
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=200458

            Bug ID: 200458
           Summary: [nvme] nvme_ctrlr_disable does not wait for the READY
                    bit to become zero
           Product: Base System
           Version: 10.1-RELEASE
          Hardware: Any
                OS: Any
            Status: New
          Severity: Affects Many People
          Priority: ---
         Component: standards
          Assignee: freebsd-standards@FreeBSD.org
          Reporter: oren.berman@pmcs.com

The function nvme_ctrlr_disable in the nvme driver code which sets the CC
enable bit to 0 does not wait for the READY bit in the CSTS register to become
zero.
When triggering a reset It  means that the nvme_ctrlr_eanble is called
immediately after the disable operation without making sure that the previous
operation was successful. 

My proposed fix is to add a new function called nvme_ctrlr_wait_for_not_ready:
static int nvme_ctrlr_wait_for_not_ready(struct nvme_controller *ctrlr)
{
                int ms_waited;
                union cc_register cc;
                union csts_register csts;


                cc.raw = nvme_mmio_read_4(ctrlr, cc);
                csts.raw = nvme_mmio_read_4(ctrlr, csts);

                if (cc.bits.en) {
                                nvme_printf(ctrlr, "%s called with cc.en =
1\n", __func__);
                                return (ENXIO);
                }

                ms_waited = 0;

                while (csts.bits.rdy) {
                                DELAY(1000);
                                if (ms_waited++ > ctrlr->ready_timeout_in_ms) {
                                                nvme_printf(ctrlr, "controller
did not become ready "
                                                    "within %d ms\n",
ctrlr->ready_timeout_in_ms);
                                                return (ENXIO);
                                }
                                csts.raw = nvme_mmio_read_4(ctrlr, csts);
                }

                return (0);
}

And then modify nvme_ctrlr_disable to call this function:

static void
nvme_ctrlr_disable(struct nvme_controller *ctrlr)
{
                union cc_register cc,cc1;
                union csts_register csts;

                cc.raw = nvme_mmio_read_4(ctrlr, cc);
                csts.raw = nvme_mmio_read_4(ctrlr, csts);
                cc1.raw=0;
                if (cc.bits.en == 1 && csts.bits.rdy == 0)
                                nvme_ctrlr_wait_for_ready(ctrlr);


                cc.bits.en = 0;
                nvme_mmio_write_4(ctrlr, cc, cc.raw);
                DELAY(5000);
                nvme_ctrlr_wait_for_not_ready(ctrlr); -->check if READY bit was
set to zero.

}

The same check is being done in the nvme linux driver.

BR
Oren

-- 
You are receiving this mail because:
You are the assignee for the bug.

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-200458-15>