Implement crashdump support on NVME

MFC after:	3 days
Sponsored by:	Netflix, Inc.

sys/dev/nvd/nvd.c

@@ -47,6 +47,7 @@ struct nvd_disk;
 
 static disk_ioctl_t nvd_ioctl;
 static disk_strategy_t nvd_strategy;
+static dumper_t nvd_dump;
 
 static void nvd_done(void *arg, const struct nvme_completion *cpl);
 
@@ -226,6 +227,26 @@ nvd_ioctl(struct disk *ndisk, u_long cmd, void *data, int fflag,
 	return (ret);
 }
 
+static int
+nvd_dump(void *arg, void *virt, vm_offset_t phys, off_t offset, size_t len)
+{
+	struct nvd_disk *ndisk;
+	struct disk *dp;
+	int error;
+
+	dp = arg;
+	ndisk = dp->d_drv1;
+
+	if (len > 0) {
+		if ((error = nvme_ns_dump(ndisk->ns, virt, offset, len)) != 0)
+			return (error);
+	} else {
+		/* XXX sync to stable storage */
+	}
+
+	return (0);
+}
+
 static void
 nvd_done(void *arg, const struct nvme_completion *cpl)
 {
@@ -302,6 +323,7 @@ nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg)
 	disk = disk_alloc();
 	disk->d_strategy = nvd_strategy;
 	disk->d_ioctl = nvd_ioctl;
+	disk->d_dump = nvd_dump;
 	disk->d_name = NVD_STR;
 	disk->d_drv1 = ndisk;
 

sys/dev/nvme/nvme.h

@@ -876,6 +876,8 @@ int	nvme_ns_cmd_deallocate(struct nvme_namespace *ns, void *payload,
 			       void *cb_arg);
 int	nvme_ns_cmd_flush(struct nvme_namespace *ns, nvme_cb_fn_t cb_fn,
 			  void *cb_arg);
+int	nvme_ns_dump(struct nvme_namespace *ns, void *virt, off_t offset,
+		     size_t len);
 
 /* Registration functions */
 struct nvme_consumer *	nvme_register_consumer(nvme_cons_ns_fn_t    ns_fn,

sys/dev/nvme/nvme_ns_cmd.c

@@ -151,3 +151,47 @@ nvme_ns_cmd_flush(struct nvme_namespace *ns, nvme_cb_fn_t cb_fn, void *cb_arg)
 
 	return (0);
 }
+
+/* Timeout = 1 sec */
+#define NVD_DUMP_TIMEOUT	100000
+
+int
+nvme_ns_dump(struct nvme_namespace *ns, void *virt, off_t offset, size_t len)
+{
+	struct nvme_completion_poll_status status;
+	struct nvme_request *req;
+	struct nvme_command *cmd;
+	uint64_t lba, lba_count;
+	int i;
+
+	status.done = FALSE;
+	req = nvme_allocate_request_vaddr(virt, len, nvme_completion_poll_cb,
+	    &status);
+	if (req == NULL)
+		return (ENOMEM);
+
+	cmd = &req->cmd;
+	cmd->opc = NVME_OPC_WRITE;
+	cmd->nsid = ns->id;
+
+	lba = offset / nvme_ns_get_sector_size(ns);
+	lba_count = len / nvme_ns_get_sector_size(ns);
+
+	*(uint64_t *)&cmd->cdw10 = lba;
+	cmd->cdw12 = lba_count - 1;
+
+	nvme_ctrlr_submit_io_request(ns->ctrlr, req);
+	if (req->qpair == NULL)
+		return (ENXIO);
+
+	i = 0;
+	while ((i++ < NVD_DUMP_TIMEOUT) && (status.done == FALSE)) {
+		DELAY(10);
+		nvme_qpair_process_completions(req->qpair);
+	}
+
+	if (status.done == FALSE)
+		return (ETIMEDOUT);
+
+	return (0);
+}