nvme: Fix nvme reset command timeout handling
PATCH v2 1/1 nvme: Fix nvme reset command timeout handling
From: Keith Busch <[email protected]>
We need to return an error if a timeout occurs on any NVMe command during initialization. Without this, the nvme reset work will be stuck. A timeout will have a negative error code, meaning we need to stop initializing the controller. All postitive returns mean the controller is still usable.
bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=196325
---
Notes:
Changes in v2:
* Consolidated cleanup path
drivers/nvme/host/core.c | 27 ++++++++++++++++++++------- 1 file changed, 20 insertions(+), 7 deletions(-)
drivers/nvme/host/core.c +20 -7
@@ -1509,7 +1509,7 @@ static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,blk_queue_write_cache (q, vwc, vwc); } -static void nvme_configure_apst (struct nvme_ctrl *ctrl) +static int nvme_configure_apst (struct nvme_ctrl *ctrl) {/* * APST (Autonomous Power State Transition) lets us program a
@@ -1538,16 +1538,16 @@ static void nvme_configure_apst(struct nvme_ctrl *ctrl)* then don't do anything. */ if (!ctrl->apsta) -return ; +return 0 ;if (ctrl->npss >31 ) {dev_warn (ctrl->device,"NPSS is invalid; not using APST\n" ); -return ; +return 0 ; } table =kzalloc (sizeof (*table),GFP_KERNEL );if (!table) -return ; +return 0 ;if (!ctrl->apst_enabled || ctrl->ps_max_latency_us ==0 ) {/* Turn off APST. */
@@ -1629,6 +1629,7 @@ static void nvme_configure_apst(struct nvme_ctrl *ctrl)dev_err (ctrl->device,"failed to set APST feature (%d)\n" , ret);kfree (table); +return ret; }static void nvme_set_latency_tolerance (struct device *dev,s32 val)
@@ -1835,13 +1836,16 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)* In fabrics we need to verify the cntlid matches the * admin connect */ -if (ctrl->cntlid !=le16_to_cpu (id->cntlid)) +if (ctrl->cntlid !=le16_to_cpu (id->cntlid)) { ret = -EINVAL ; + goto out_free; + }if (!ctrl->opts->discovery_nqn && !ctrl->kas) {dev_err (ctrl->device,"keep-alive support is mandatory for fabrics\n" ); ret = -EINVAL ; + goto out_free; } }else { ctrl->cntlid =le16_to_cpu (id->cntlid);
@@ -1856,11 +1860,20 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)else if (!ctrl->apst_enabled && prev_apst_enabled)dev_pm_qos_hide_latency_tolerance (ctrl->device); -nvme_configure_apst (ctrl); -nvme_configure_directives (ctrl); + ret =nvme_configure_apst (ctrl); +if (ret <0 ) +return ret; + + ret =nvme_configure_directives (ctrl); +if (ret <0 ) +return ret; ctrl->identified = true; +return 0 ; + +out_free: +kfree (id);return ret; }EXPORT_SYMBOL_GPL (nvme_init_identify);
Signature · 1 line
2.12.3
Looks good,
Quoted 7 lines · Johannes Thumshirn
From: Keith Busch <[email protected]>
We need to return an error if a timeout occurs on any NVMe command during initialization. Without this, the nvme reset work will be stuck. A timeout will have a negative error code, meaning we need to stop initializing the controller. All postitive returns mean the controller is still usable.
Quoted 6 lines
bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=196325
Signed-off-by Keith Busch <[email protected]>Cc Martin Peres <[email protected]>[jth consolidated cleanup path ]Signed-off-by Johannes Thumshirn <[email protected]>
This looks great, thanks for fixing that up!
I noticed the HMB setting in pci.c can also potentially timeout. We can deal with that later.
Looks good,
p.s. we'll need also to update the configure timestamp configuration.