diff --git a/sys/dev/nvme/nvme_qpair.c b/sys/dev/nvme/nvme_qpair.c --- a/sys/dev/nvme/nvme_qpair.c +++ b/sys/dev/nvme/nvme_qpair.c @@ -536,16 +536,31 @@ int done = 0; bool in_panic = dumping || SCHEDULER_STOPPED(); - qpair->num_intr_handler_calls++; - /* * qpair is not enabled, likely because a controller reset is is in * progress. Ignore the interrupt - any I/O that was associated with - * this interrupt will get retried when the reset is complete. + * this interrupt will get retried when the reset is complete. Any + * pending completions for when we're in startup will be completed + * as soon as initialization is complete and we start sending commands + * to the device. */ if (qpair->recovery_state != RECOVERY_NONE) return (false); + /* + * Sanity check initialization. After we reset the hardware, the phase + * is defined to be 1. So if we get here with zero prior calls and the + * phase is 0, it means that we've lost a race between the + * initialization and the ISR running. With the phase wrong, we'll + * process a bunch of completions that aren't really completions leading + * to a KASSERT below. + */ + KASSERT(!(qpair->num_intr_handler_calls == 0 && qpair->phase == 0), + ("%s: Phase wrong for first interrupt call.", + device_get_nameunit(qpair->ctrlr->dev))); + + qpair->num_intr_handler_calls++; + bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* diff --git a/sys/dev/nvme/nvme_sysctl.c b/sys/dev/nvme/nvme_sysctl.c --- a/sys/dev/nvme/nvme_sysctl.c +++ b/sys/dev/nvme/nvme_sysctl.c @@ -155,8 +155,13 @@ nvme_qpair_reset_stats(struct nvme_qpair *qpair) { + /* + * Reset the values. Due to sanity checks in + * nvme_qpair_process_completions, we reset the number of interrupt + * calls to 1. + */ qpair->num_cmds = 0; - qpair->num_intr_handler_calls = 0; + qpair->num_intr_handler_calls = 1; qpair->num_retries = 0; qpair->num_failures = 0; }