nvme: allow passthru cmd error logging
Subsystems (2)
v8 cover nvme: allow passthru cmd error logging
In nvme_end_req() we only log errors which are for non-passthru commands. Add a helper function nvme_log_err_passthru() that allows us to log error for passthru commands by decoding cdw10-cdw15 values of nvme command.
Below is short testlog :-
* Admin Passsthru error log off, no errors are printed
* Admin Passsthru error log on, errors are printed
* IO Passsthru error log off, no errors are printed
* IO Passsthru error log on, errors are printed
v8:
- Add a logging_enabled flag to device structure.
- Add a passthru_err_log_enabled sysfs attribute for namespaces to allow logging of passthru IO commands.
v7:
- Changed attribute/flag to passthru_err_log_enabled.
- Use kstrtobool rather than kstrtoint.
v6:
- Rebase, retest nvme-6.5 and add test log for admin and I/O passthru error log.
v5:
- Trim down code in the nvme_log_error_passthrough(). Use following to get the disk name as an arg to pr_err_ratelimited() :-
ns ? ns->disk->disk_name : dev_name(nr->ctrl->device),
Use following to get the admin vs I/O opcode string as an arg to
pr_err_ratelimited() :-
ns ? nvme_get_opcode_str(nr->cmd->common.opcode) :
nvme_get_admin_opcode_str(nr->cmd->common.opcode),- Rename nvme_log_error_passthrough() -> nvme_log_err_passthru().
- Remove else and return directly in nvme_passthru_err_log_show().
- Generate error on invalid values of the passthru_enable variable in nvme_passthru_log_store().
- Rename passthrough -> passthru.
- Rename sysfs attr from passthru_admin_err_logging -> passthru_log_err.
v4:
- Change sysfs attribute to passthru_admin_err_logging
- Only log passthrough admin commands. IO passthrough commands will always be logged.
v3:
- Log a passthrough specific message that dumps CDW* contents.
- Enable/disable vis sysfs rather than debugfs.
v2:
- Included Pankaj Raghav's patch 'nvme: ignore starting sector while error logging for passthrough requests'
with a couple changes.
- Moved error_logging flag to nvme_ctrl structure
- The entire nvme-debugfs.c does not need to be guarded by #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS.
- Use IS_ENABLED((CONFIG_NVME_ERROR_LOGGING_DEBUG_FS)) to determine if error logging should be initialized.
- Various other nits.
Testing
-------
* Admin Passsthru error log off, no errors are printed :-
[root@localhost ~]# echo 0 > /sys/class/nvme/nvme0/passthru_err_log_enabled [root@localhost ~]# nvme telemetry-log -o /tmp/test /dev/nvme0 [root@localhost ~]# dmesg
[root@localhost ~]
* Admin Passsthru error log on, errors are printed :-
[root@localhost ~]# echo 1 > /sys/class/nvme/nvme0/passthru_err_log_enabled [root@localhost ~]# nvme telemetry-log -o /tmp/test /dev/nvme0 [root@localhost ~]# dmesg
[ 2364.008105] nvme0: Get Log Page(0x2), Invalid Field in Command (sct 0x0 / sc 0x2) DNR cdw10=0x7f0107 cdw11=0x0 cdw12=0x0 cdw13=0x0 cdw14=0x0 cdw15=0x0
* IO Passsthru error log off, errors are not printed :-
[root@localhost ~]# echo 0 > /sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled [root@localhost ~]# nvme write-zeroes -n 1 -s 0x200000 -c 10 /dev/nvme0
NVMe status: LBA Out of Range: The command references an LBA that exceeds the size of the namespace(0x4080) [root@localhost ~]# dmesg
[ 3131.602986] nvme nvme0: using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!
[root@localhost ~]#
* IO Passsthru error log on, errors are printed :-
[root@localhost ~]# echo 1 > /sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled [root@localhost ~]# nvme write-zeroes -n 1 -s 0x200000 -c 10 /dev/nvme0
NVMe status: LBA Out of Range: The command references an LBA that exceeds the size of the namespace(0x4080) [root@localhost ~]# dmesg
[ 2944.910393] nvme nvme0: using deprecated NVME_IOCTL_IO_CMD ioctl on the char device! [ 2944.910423] nvme0n1: Write Zeroes(0x8), LBA Out of Range (sct 0x0 / sc 0x80) DNR cdw10=0x200000 cdw11=0x0 cdw12=0xa cdw13=0x0 cdw14=0x0 cdw15=0x0
[root@localhost ~]#
Alan Adamson (1):
nvme: allow passthru cmd error logging
drivers/nvme/host/core.c | 52 ++++++++++++++++++++++++++++++++++----- drivers/nvme/host/nvme.h | 2 ++ drivers/nvme/host/sysfs.c | 44 +++++++++++++++++++++++++++++++++ include/linux/device.h | 1 + 4 files changed, 93 insertions(+), 6 deletions(-)
Signature · 1 line
2.39.3
PATCH v8 1/1 nvme: allow passthru cmd error logging
Commit d7ac8dca938c ("nvme: quiet user passthrough command errors") disabled error logging for user passthrough commands. This commit adds the ability to opt-in to passthrough admin error logging. IO commands initiated as passthrough will always be logged.
The logging output for passthrough commands (Admin and IO) has been changed to include CDWXX fields.
nvme0n1: Read(0x2), LBA Out of Range (sct 0x0 / sc 0x80) DNR cdw10=0x0 cdw11=0x1
cdw12=0x70000 cdw13=0x0 cdw14=0x0 cdw15=0x0
Add a helper function nvme_log_err_passthru() which allows us to log error for passthru commands by decoding cdw10-cdw15 values of nvme command.
Add a new sysfs attr passthru_err_log_enabled that allows user to conditionally enable passthrough command logging for either passthrough Admin commands sent to the controller or passthrough IO commands sent to a namespace.
By default, passthrough error logging is disabled.
To enable passthrough admin error logging:
echo 1 > /sys/class/nvme/nvme0/passthru_err_log_enabled
To disable passthrough admin error logging:
echo 0 > /sys/class/nvme/nvme0/passthru_err_log_enabled
To enable passthrough io error logging:
echo 1 > /sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
To disable passthrough io error logging:
echo 0 > /sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
[kch] fix sevaral nits and trim down code, details in cover-letter.
--- drivers/nvme/host/core.c | 52 ++++++++++++++++++++++++++++++++++----- drivers/nvme/host/nvme.h | 2 ++ drivers/nvme/host/sysfs.c | 44 +++++++++++++++++++++++++++++++++ include/linux/device.h | 1 + 4 files changed, 93 insertions(+), 6 deletions(-)
drivers/nvme/host/core.c +46 -6
@@ -337,6 +337,30 @@ static void nvme_log_error(struct request *req) nr->status & NVME_SC_DNR ? "DNR " : ""); } +static void nvme_log_err_passthru (struct request *req) +{ +struct nvme_ns *ns = req->q->queuedata; +struct nvme_request *nr =nvme_req (req); + +pr_err_ratelimited ("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s" +"cdw10=0x%x cdw11=0x%x cdw12=0x%x cdw13=0x%x cdw14=0x%x cdw15=0x%x\n" , + ns ? ns->disk->disk_name :dev_name (nr->ctrl->device), + ns ?nvme_get_opcode_str (nr->cmd->common.opcode) : +nvme_get_admin_opcode_str (nr->cmd->common.opcode), + nr->cmd->common.opcode, +nvme_get_error_status_str (nr->status), + nr->status >>8 &7 ,/* Status Code Type */ + nr->status &0xff ,/* Status Code */ + nr->status &NVME_SC_MORE ?"MORE " :"" , + nr->status &NVME_SC_DNR ?"DNR " :"" , + nr->cmd->common.cdw10, + nr->cmd->common.cdw11, + nr->cmd->common.cdw12, + nr->cmd->common.cdw13, + nr->cmd->common.cdw14, + nr->cmd->common.cdw14); +} +enum nvme_disposition {COMPLETE ,RETRY ,
@@ -381,8 +405,12 @@ static inline void nvme_end_req(struct request *req) {blk_status_t status =nvme_error_status (nvme_req (req)->status); -if (unlikely (nvme_req (req)->status && !(req->rq_flags &RQF_QUIET ))) -nvme_log_error (req); +if (unlikely (nvme_req (req)->status && !(req->rq_flags &RQF_QUIET ))) { +if (blk_rq_is_passthrough (req)) +nvme_log_err_passthru (req); +else +nvme_log_error (req); + }nvme_end_req_zoned (req);nvme_trace_bio_complete (req);if (req->cmd_flags &REQ_NVME_MPATH )
@@ -675,10 +703,19 @@ static inline void nvme_clear_nvme_request(struct request *req)/* initialize a passthrough request */ void nvme_init_request (struct request *req,struct nvme_command *cmd) { -if (req->q->queuedata) +struct nvme_request *nr =nvme_req (req); +struct device *device; + +if (req->q->queuedata) { req->timeout =NVME_IO_TIMEOUT ; -else /* no queuedata implies admin queue */ + device =disk_to_dev (req->q->disk); + }else {/* no queuedata implies admin queue */ req->timeout =NVME_ADMIN_TIMEOUT ; + device = nr->ctrl->device; + } + +if (!device->logging_enabled) + req->rq_flags |=RQF_QUIET ;/* passthru commands should let the driver set the SGL flags */ cmd->common.flags &= ~NVME_CMD_SGL_ALL ;
@@ -687,8 +724,7 @@ void nvme_init_request(struct request *req, struct nvme_command *cmd)if (req->mq_hctx->type ==HCTX_TYPE_POLL ) req->cmd_flags |=REQ_POLLED ;nvme_clear_nvme_request (req); - req->rq_flags |=RQF_QUIET ; -memcpy (nvme_req (req)->cmd, cmd,sizeof (*cmd)); +memcpy (nr->cmd, cmd,sizeof (*cmd)); }EXPORT_SYMBOL_GPL (nvme_init_request);
@@ -3682,6 +3718,7 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)nvme_mpath_add_disk (ns, info->anagrpid);nvme_fault_inject_init (&ns->fault_inject, ns->disk->disk_name); +nvme_sysfs_add_passthru_err_log (disk_to_dev (ns->disk));return ;
@@ -3712,6 +3749,7 @@ static void nvme_ns_remove(struct nvme_ns *ns)clear_bit (NVME_NS_READY , &ns->flags);set_capacity (ns->disk,0 ); +nvme_sysfs_remove_passthru_err_log (disk_to_dev (ns->disk));nvme_fault_inject_fini (&ns->fault_inject);/*
@@ -4423,6 +4461,7 @@ EXPORT_SYMBOL_GPL(nvme_start_ctrl);void nvme_uninit_ctrl (struct nvme_ctrl *ctrl) {nvme_hwmon_exit (ctrl); +nvme_sysfs_remove_passthru_err_log (ctrl->device);nvme_fault_inject_fini (&ctrl->fault_inject);dev_pm_qos_hide_latency_tolerance (ctrl->device);cdev_device_del (&ctrl->cdev, ctrl->device);
@@ -4550,6 +4589,7 @@ int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, min(default_ps_max_latency_us, (unsigned long)S32_MAX));nvme_fault_inject_init (&ctrl->fault_inject,dev_name (ctrl->device)); +nvme_sysfs_add_passthru_err_log (ctrl->device);nvme_mpath_init_ctrl (ctrl); ret =nvme_auth_init_ctrl (ctrl);if (ret)
drivers/nvme/host/nvme.h +2 -0
@@ -1101,6 +1101,8 @@ void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,struct nvme_ctrl *nvme_ctrl_from_file (struct file *file);struct nvme_ns *nvme_find_get_ns (struct nvme_ctrl *ctrl,unsigned nsid);void nvme_put_ns (struct nvme_ns *ns); +int nvme_sysfs_add_passthru_err_log (struct device *dev); +void nvme_sysfs_remove_passthru_err_log (struct device *dev);static inline bool nvme_multi_css (struct nvme_ctrl *ctrl) {
drivers/nvme/host/sysfs.c +44 -0
56 lines not on this page · raw message
include/linux/device.h +1 -0
@@ -805,6 +805,7 @@ struct device {#ifdef CONFIG_DMA_OPS_BYPASS bool dma_ops_bypass : 1;#endif + bool logging_enabled:1; };/**
Signature · 1 line
2.39.3
Christoph Hellwig
maintainerdrivers/nvme/host/sysfs.c const struct attribute_group *nvme_ns_id_attr_groups[] = { · line 212
Quoted 9 lines · Alan Adamson
drivers/nvme/host/sysfs.c +9
-0+int nvme_sysfs_add_passthru_err_log (struct device *dev) +{ +return sysfs_create_group (&dev->kobj, &nvme_log_attr_group); +} + +void nvme_sysfs_remove_passthru_err_log (struct device *dev) +{ +sysfs_remove_group (&dev->kobj, &nvme_log_attr_group); +}
Isn't the normal sysfs convention to have an is_visible method instead of dynamically adding/removing groups?
Otherwise this looks good to me.
Hi all,
Quoted 13 lines · Christoph Hellwig
Review
Quoted 9 lines · Alan Adamson
(quoted diff) +9
-0+int nvme_sysfs_add_passthru_err_log(struct device *dev) +{ + return sysfs_create_group(&dev->kobj, &nvme_log_attr_group); +} + +void nvme_sysfs_remove_passthru_err_log(struct device *dev) +{ + sysfs_remove_group(&dev->kobj, &nvme_log_attr_group); +}Isn't the normal sysfs convention to have an
is_visiblemethod instead of dynamically adding/removing groups?
Quoted 2 lines
Otherwise this looks good to me.
This patch adds a new member to struct device: logging_enabled. There are no other users for that member as of now so I think we should avoid that.
How about following on the top of this patch that adds logging_enabled to struct nvme_ctrl and struct nvme_ns so we can remove struct device:logging_enabled ?
drivers/nvme/host/core.c +0 -0
@@ -127,6 +127,11 @@ static void nvme_remove_invalid_namespaces(struct
nvme_ctrl *ctrl,
static void nvme_update_keep_alive(struct nvme_ctrl *ctrl, struct nvme_command *cmd);
+bool is_nvme_class(const struct class *class)
+{
+ return nvme_class == class;
+}
+
void nvme_queue_scan(struct nvme_ctrl *ctrl)
{
/*
@@ -708,17 +713,19 @@ static inline void nvme_clear_nvme_request(struct request *req)
void nvme_init_request(struct request *req, struct nvme_command *cmd) {
struct nvme_request *nr = nvme_req(req);
- struct device *device;
+ bool logging_enabled;
if (req->q->queuedata) {
+ struct nvme_ns *ns = req->q->disk->private_data; +
+ logging_enabled = ns->logging_enabled;
req->timeout = NVME_IO_TIMEOUT;
- device = disk_to_dev(req->q->disk);
} else { /* no queuedata implies admin queue */ req->timeout = NVME_ADMIN_TIMEOUT;
- device = nr->ctrl->device;
+ logging_enabled = nr->ctrl->logging_enabled;
}
- if (!device->logging_enabled)
+ if (!logging_enabled)
req->rq_flags |= RQF_QUIET;
/* passthru commands should let the driver set the SGL flags */
@@ -3718,6 +3725,13 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
nvme_mpath_add_disk(ns, info->anagrpid); nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
+
+ /*
+ * Set ns->disk->device->driver_data to ns so we can access + * ns->logging_enabled in nvme_passthru_err_log_enabled_store() and + * nvme_passthru_err_log_enabled_show().
+ */
+ dev_set_drvdata(disk_to_dev(ns->disk), ns);
nvme_sysfs_add_passthru_err_log(disk_to_dev(ns->disk));
return;
drivers/nvme/host/nvme.h +4 -1
@@ -20,6 +20,8 @@ #include <trace/events/block.h> +extern bool is_nvme_class (const struct class *class); + extern const struct pr_ops nvme_pr_ops; extern unsigned int nvme_io_timeout;
@@ -385,6 +387,7 @@ struct nvme_ctrl { enum nvme_ctrl_type cntrltype; enum nvme_dctype dctype; + bool logging_enabled; }; enum nvme_iopolicy {
@@ -511,7 +514,7 @@ struct nvme_ns { struct device cdev_device; struct nvme_fault_inject fault_inject; - + bool logging_enabled; }; /* NVMe ns supports metadata actions by the controller (generate/strip) */
drivers/nvme/host/sysfs.c +0 -0
@@ -38,20 +38,36 @@ static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL,
nvme_sysfs_rescan);
static ssize_t nvme_passthru_err_log_enabled_show(struct device *dev, struct device_attribute *attr, char *buf)
{
- return sysfs_emit(buf, dev->logging_enabled ? "on" : "off"); + if (is_nvme_class(dev->class)) {
+ struct nvme_ctrl *c = dev_get_drvdata(dev); +
+ return sysfs_emit(buf, c->logging_enabled ? "on\n" :
"off\n");
+ } else {
+ struct nvme_ns *n = dev_get_drvdata(dev);
+
+ return sysfs_emit(buf, n->logging_enabled ? "on\n" :
"off\n");
+ }
}
static ssize_t nvme_passthru_err_log_enabled_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t
count)
{
- int err;
bool passthru_err_log_enabled;
+ int err;
err = kstrtobool(buf, &passthru_err_log_enabled); if (err)
return -EINVAL;
- dev->logging_enabled = passthru_err_log_enabled; + if (is_nvme_class(dev->class)) {
+ struct nvme_ctrl *ctrl = dev_get_drvdata(dev); +
+ ctrl->logging_enabled = passthru_err_log_enabled; + } else {
+ struct nvme_ns *ns = dev_get_drvdata(dev); +
+ ns->logging_enabled = passthru_err_log_enabled; + }
return count;
}
include/linux/device.h +0 -1
@@ -805,7 +805,6 @@ struct device { #ifdef CONFIG_DMA_OPS_BYPASS bool dma_ops_bypass : 1; #endif - bool logging_enabled:1; }; /**
-ck
Sagi Grimberg
maintainerQuoted 16 lines · Chaitanya Kulkarni
Hi all,
Quoted 13 lines · Christoph Hellwig
Review
Quoted 9 lines · Alan Adamson
(quoted diff) +9
-0+int nvme_sysfs_add_passthru_err_log(struct device *dev) +{ + return sysfs_create_group(&dev->kobj, &nvme_log_attr_group); +} + +void nvme_sysfs_remove_passthru_err_log(struct device *dev) +{ + sysfs_remove_group(&dev->kobj, &nvme_log_attr_group); +}Isn't the normal sysfs convention to have an
is_visiblemethod instead of dynamically adding/removing groups?
Quoted 6 lines
Quoted 2 lines
Otherwise this looks good to me.
This patch adds a new member to struct device:
logging_enabled. There are no other users for that member as of now so I think we should avoid that.
We definitely should.
Quoted 49 lines
How about following on the top of this patch that adds
logging_enabledto structnvme_ctrland structnvme_nsso we can remove struct device:logging_enabled ?drivers/nvme/host/core.c +5
-0@@ -127,6 +127,11 @@ static void nvme_remove_invalid_namespaces(structnvme_ctrl *ctrl, static void nvme_update_keep_alive(struct nvme_ctrl *ctrl, struct nvme_command *cmd); +bool is_nvme_class (const struct class *class) +{ + return nvme_class == class; +} + void nvme_queue_scan(struct nvme_ctrl *ctrl) {/*
drivers/nvme/host/core.c +6
-4@@ -708,17 +713,19 @@ static inline void nvme_clear_nvme_request(struct request *req) void nvme_init_request(struct request *req, struct nvme_command *cmd) { struct nvme_request *nr = nvme_req(req); - struct device *device; + bool logging_enabled; if (req->q->queuedata) { + struct nvme_ns *ns = req->q->disk->private_data; + + logging_enabled = ns->logging_enabled; req->timeout = NVME_IO_TIMEOUT; - device = disk_to_dev(req->q->disk); } else { /* no queuedata implies admin queue */ req->timeout = NVME_ADMIN_TIMEOUT; - device = nr->ctrl->device; + logging_enabled = nr->ctrl->logging_enabled; } - if (!device->logging_enabled) + if (!logging_enabled) req->rq_flags |= RQF_QUIET;/* passthru commands should let the driver set the SGL flags */
drivers/nvme/host/core.c +0
-0@@ -3718,6 +3725,13 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
drivers/nvme/host/core.c static void nvme_alloc_ns(struct nvme_ctrl *ctrl, · line 3735
Quoted 10 lines
drivers/nvme/host/core.c +7
-0nvme_mpath_add_disk(ns, info->anagrpid); nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name); + + /* + * Set ns->disk->device->driver_data to ns so we can access + * ns->logging_enabled in nvme_passthru_err_log_enabled_store() and + * nvme_passthru_err_log_enabled_show(). + */ + dev_set_drvdata(disk_to_dev(ns->disk), ns);
Is this needed?
Quoted 68 lines
drivers/nvme/host/core.c +0
-0nvme_sysfs_add_passthru_err_log(disk_to_dev(ns->disk));return;
drivers/nvme/host/nvme.h +4
-1@@ -20,6 +20,8 @@ #include <trace/events/block.h> +extern bool is_nvme_class (const struct class *class); + extern const struct pr_ops nvme_pr_ops; extern unsigned int nvme_io_timeout;@@ -385,6 +387,7 @@ struct nvme_ctrl { enum nvme_ctrl_type cntrltype; enum nvme_dctype dctype; + bool logging_enabled; }; enum nvme_iopolicy {@@ -511,7 +514,7 @@ struct nvme_ns { struct device cdev_device; struct nvme_fault_inject fault_inject; - + bool logging_enabled; }; /* NVMe ns supports metadata actions by the controller (generate/strip) */drivers/nvme/host/sysfs.c +10
-2@@ -38,20 +38,36 @@ static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan); static ssize_t nvme_passthru_err_log_enabled_show(struct device *dev, struct device_attribute *attr, char *buf) { - return sysfs_emit(buf, dev->logging_enabled ? "on" : "off"); + if (is_nvme_class(dev->class)) { + struct nvme_ctrl *c = dev_get_drvdata(dev); + + return sysfs_emit(buf, c->logging_enabled ? "on\n" :"off\n" ); + } else { + struct nvme_ns *n = dev_get_drvdata(dev); + + return sysfs_emit(buf, n->logging_enabled ? "on\n" :"off\n" ); + } } static ssize_t nvme_passthru_err_log_enabled_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count) { - int err; bool passthru_err_log_enabled; + int err; err = kstrtobool(buf, &passthru_err_log_enabled); if (err) return -EINVAL;
drivers/nvme/host/sysfs.c static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL,
Quoted 10 lines
drivers/nvme/host/sysfs.c +9
-1- dev->logging_enabled = passthru_err_log_enabled; + if (is_nvme_class(dev->class)) { + struct nvme_ctrl *ctrl = dev_get_drvdata(dev); + + ctrl->logging_enabled = passthru_err_log_enabled; + } else { + struct nvme_ns *ns = dev_get_drvdata(dev); + + ns->logging_enabled = passthru_err_log_enabled; + }
Why mix ctrl with ns sysfs handlers?
Quoted 20 lines
return count; }include/linux/device.h +0
-2@@ -805,7 +805,6 @@ struct device { #ifdef CONFIG_DMA_OPS_BYPASS bool dma_ops_bypass : 1; #endif - bool logging_enabled:1; }; /** -ck
Sagi,
Quoted 12 lines
Quoted 11 lines
nvme_mpath_add_disk(ns, info->anagrpid);nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
+
+ /*
+ * Setns->disk->device->driver_datato ns so we can access + *ns->logging_enabledinnvme_passthru_err_log_enabled_store()and
+ *nvme_passthru_err_log_enabled_show().
+ */
+dev_set_drvdata(disk_to_dev(ns->disk), ns);
Quoted 1 line
Is this needed?
Yes see explanation below ..
[...]
Quoted 15 lines
Quoted 15 lines
static
ssize_tnvme_passthru_err_log_enabled_store(struct device *dev,
structdevice_attribute*attr, const char *buf,size_tcount)
{
- int err;
boolpassthru_err_log_enabled;
+ int err;err =
kstrtobool(buf, &passthru_err_log_enabled); if (err)
return-EINVAL;
Quoted 12 lines
Review
Quoted 10 lines
-
dev->logging_enabled=passthru_err_log_enabled; + if (is_nvme_class(dev->class)) {
+ structnvme_ctrl*ctrl =dev_get_drvdata(dev);(quoted diff) +7
-0+ + ctrl->logging_enabled = passthru_err_log_enabled; + } else { + struct nvme_ns *ns = dev_get_drvdata(dev); + + ns->logging_enabled = passthru_err_log_enabled; + }Why mix ctrl with ns sysfs handlers?
sorry I didn't understand your question clearly ...
In the original implementation we get two different struct device objects for following commands in sysfs, that sets struct device:logging_enabled flag,
which is also used to determine the logging in nvme_init_request() :-
echo 1 > /sys/class/nvme/nvme0/passthru_err_log_enabled echo 1 > /sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
nvme_init_ctrl() already sets dev_set_drvdata(ctrl->device, ctrl). In this proposal above change in nvme_alloc_ns() sets the dev_set_drvdata(disk_to_dev(ns->disk), ns).
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we update logging_enabled
flags for either ctrl or ns respectively. In nvme_init_request() I use ctrl->logging_enabled and ns->logging_enabled based on admin or io cmd.
-ck
Sagi Grimberg
maintainerQuoted 62 lines · Chaitanya Kulkarni
Sagi,
Quoted 12 lines
Quoted 11 lines
nvme_mpath_add_disk(ns, info->anagrpid);
nvme_fault_inject_init(&ns->fault_inject,ns->disk->disk_name);
+
+ /*
+ * Setns->disk->device->driver_datato ns so we can access + *ns->logging_enabledinnvme_passthru_err_log_enabled_store()and
+ *nvme_passthru_err_log_enabled_show().
+ */
+dev_set_drvdata(disk_to_dev(ns->disk), ns);Quoted 1 line
Is this needed?
Yes see explanation below ..
[...]
Quoted 15 lines
Quoted 15 lines
static
ssize_tnvme_passthru_err_log_enabled_store(struct device
*dev,
structdevice_attribute*attr, const char *buf,size_t
count)
{
- int err;
boolpassthru_err_log_enabled;
+ int err;err =
kstrtobool(buf, &passthru_err_log_enabled); if (err)
return-EINVAL;Quoted 12 lines
Review
Quoted 10 lines
-
dev->logging_enabled=passthru_err_log_enabled; + if (is_nvme_class(dev->class)) {
+ structnvme_ctrl*ctrl =dev_get_drvdata(dev);(quoted diff) +7
-0+ + ctrl->logging_enabled = passthru_err_log_enabled; + } else { + struct nvme_ns *ns = dev_get_drvdata(dev); + + ns->logging_enabled = passthru_err_log_enabled; + }Why mix ctrl with ns sysfs handlers?
sorry I didn't understand your question clearly ...
In the original implementation we get two different struct device objects for following commands in sysfs, that sets struct device:logging_enabled flag,
which is also used to determine the logging innvme_init_request():-echo 1 >
/sys/class/nvme/nvme0/passthru_err_log_enabledecho 1 >/sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
nvme_init_ctrl()already setsdev_set_drvdata(ctrl->device, ctrl). In this proposal above change innvme_alloc_ns()sets thedev_set_drvdata(disk_to_dev(ns->disk), ns).
Quoted 6 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in nvme_dev_attrs and a separate one in nvme_ns_id_attrs ? Then you don't need the awkward is_nvme_class() ? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?
Quoted 8 lines
Quoted 7 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.
Quoted 7 lines
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?
I did have a version that used nvme_ns_id_attrs but didn't think it was an appropriate place for it. I can go back to that.
Alan
Alan,
Quoted 16 lines · On 1/23/24 09:11, [email protected] wrote:
Quoted 8 lines
Quoted 7 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.Quoted 7 lines
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?
Quoted 5 lines
I did have a version that used
nvme_ns_id_attrsbut didn't think it was an appropriate place for it. I can go back to that.Alan
can you please remove the struct device:logging_enabled and use Sagi's suggestion and post a new version ?
There is a also comment from Christoph to use is_visible method instead of dynamically adding and removing groups, can you also add that to next version ? unless that cannot be done for some reason ...
-ck
Quoted 19 lines · Chaitanya Kulkarni
Alan,
Quoted 16 lines · On 1/23/24 09:11, [email protected] wrote:
Quoted 6 lines
Quoted 6 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.Quoted 7 lines
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?I did have a version that used
nvme_ns_id_attrsbut didn't think it was an appropriate place for it. I can go back to that.
Quoted 8 lines
Quoted 2 lines
Alan
can you please remove the struct device:logging_enabled and use Sagi's suggestion and post a new version ?
There is a also comment from Christoph to use
is_visiblemethod instead of dynamically adding and removing groups, can you also add that to next version ? unless that cannot be done for some reason ...
Yes, I'll post a new version. I don't think the is_visible method is necessary since we will not need to create a new group, just use the existing groups (nvme_ns_id and nvme_dev).
Alan
Christoph Hellwig
maintainerQuoted 10 lines · On Tue, Jan 23, 2024 at 09:11:32AM -0800, [email protected] wrote:
Quoted 10 lines
Quoted 6 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.I was asking why should we have a show/store that operate on both ns and ctrl?
Quoted 7 lines
Quoted 4 lines
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?I did have a version that used
nvme_ns_id_attrsbut didn't think it was an appropriate place for it. I can go back to that.
nvme_ns_id_attrs has been renamed now that we're using it for more than just ids, so adding more attributes if they fit is fine.
Quoted 75 lines · Sagi Grimberg
Quoted 65 lines · Chaitanya Kulkarni
Sagi,
Quoted 12 lines
Quoted 11 lines
nvme_mpath_add_disk(ns, info->anagrpid);nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
+
+ /*
+ * Setns->disk->device->driver_datato ns so we can access + *ns->logging_enabledinnvme_passthru_err_log_enabled_store()and
+ *nvme_passthru_err_log_enabled_show().
+ */
+dev_set_drvdata(disk_to_dev(ns->disk), ns);Quoted 1 line
Is this needed?
Yes see explanation below ..
[...]
Quoted 16 lines
Quoted 16 lines
static
ssize_tnvme_passthru_err_log_enabled_store(struct device *dev,
structdevice_attribute*attr, const char *buf,size_t
count)
{
- int err;
boolpassthru_err_log_enabled;
+ int err;err =
kstrtobool(buf, &passthru_err_log_enabled); if (err)
return-EINVAL;Quoted 12 lines
Review
Quoted 10 lines
-
dev->logging_enabled=passthru_err_log_enabled; + if (is_nvme_class(dev->class)) {
+ structnvme_ctrl*ctrl =dev_get_drvdata(dev);(quoted diff) +7
-0+ + ctrl->logging_enabled = passthru_err_log_enabled; + } else { + struct nvme_ns *ns = dev_get_drvdata(dev); + + ns->logging_enabled = passthru_err_log_enabled; + }Why mix ctrl with ns sysfs handlers?
sorry I didn't understand your question clearly ...
In the original implementation we get two different struct device objects
for following commands in sysfs, that sets struct device:logging_enabled flag,
which is also used to determine the logging innvme_init_request():-echo 1 >
/sys/class/nvme/nvme0/passthru_err_log_enabledecho 1 >/sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
nvme_init_ctrl()already setsdev_set_drvdata(ctrl->device, ctrl). In this
proposal above change innvme_alloc_ns()sets thedev_set_drvdata(disk_to_dev(ns->disk), ns).Quoted 6 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.
Quoted 7 lines
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?
no you are not, I was just trying to minimize the changes in the posted patch guess that is a right way to go forward ...
-ck
Quoted 22 lines
Quoted 15 lines
sorry I didn't understand your question clearly ...
In the original implementation we get two different struct device objects for following commands in sysfs, that sets struct device:logging_enabled flag,
which is also used to determine the logging innvme_init_request():-echo 1 >
/sys/class/nvme/nvme0/passthru_err_log_enabledecho 1 >/sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
nvme_init_ctrl()already setsdev_set_drvdata(ctrl->device, ctrl). In this proposal above change innvme_alloc_ns()sets thedev_set_drvdata(disk_to_dev(ns->disk), ns).Quoted 6 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.
Quoted 7 lines
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?
If we create separate show/stores for the ctrl (nvme_dev_attrs) and ns (nvme_ns_attrs), using the exiting macro: DEVICE_ATTR(), the attribute name (passthru_err_log_enabled) can not be used for both nvme_dev_attrs and nvme_ns_attrs.
/sys/class/nvme/nvme0/adm_passthru_err_log_enabled
/sys/class/nvme/nvme0/nvme0n1/io_passthru_err_log_enabled
If I setup the attributes like:
static struct device_attribute dev_attr_adm_passthru_err_log_enabled = \
__ATTR(passthru_err_log_enabled, S_IRUGO | S_IWUSR, \
nvme_adm_passthru_err_log_enabled_show, nvme_adm_passthru_err_log_enabled_store);
static struct device_attribute dev_attr_io_passthru_err_log_enabled = \
__ATTR(passthru_err_log_enabled, S_IRUGO | S_IWUSR, \
nvme_io_passthru_err_log_enabled_show, nvme_io_passthru_err_log_enabled_store);
Then both attributes can be the same.
/sys/class/nvme/nvme0/passthru_err_log_enabled
/sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
What do we prefer?
Alan
Sagi Grimberg
maintainerQuoted 50 lines
Quoted 22 lines
Quoted 15 lines
sorry I didn't understand your question clearly ...
In the original implementation we get two different struct device objects for following commands in sysfs, that sets struct device:logging_enabled flag,
which is also used to determine the logging innvme_init_request():-echo 1 >
/sys/class/nvme/nvme0/passthru_err_log_enabledecho 1 >/sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabled
nvme_init_ctrl()already setsdev_set_drvdata(ctrl->device, ctrl). In this proposal above change innvme_alloc_ns()sets thedev_set_drvdata(disk_to_dev(ns->disk), ns).Quoted 6 lines
This allows us to get the ctrl or ns object associated with the struct device
we get in the sysfs, then based on the device class we updatelogging_enabled
flags for either ctrl or ns respectively. Innvme_init_request()I usectrl->logging_enabledandns->logging_enabledbased on admin or io cmd.Quoted 7 lines
I was asking why should we have a show/store that operate on both ns and ctrl?
Why not have a show/store in
nvme_dev_attrsand a separate one innvme_ns_id_attrs? Then you don't need the awkwardis_nvme_class()? the ns attrs can access the ns in a normal way like the rest? Or am I missing something?If we create separate show/stores for the ctrl (
nvme_dev_attrs) and ns (nvme_ns_attrs), using the exiting macro:DEVICE_ATTR(), the attribute name (passthru_err_log_enabled) can not be used for bothnvme_dev_attrsandnvme_ns_attrs.
/sys/class/nvme/nvme0/adm_passthru_err_log_enabled
/sys/class/nvme/nvme0/nvme0n1/io_passthru_err_log_enabledIf I setup the attributes like:
static struct
device_attributedev_attr_adm_passthru_err_log_enabled= \
__ATTR(passthru_err_log_enabled,S_IRUGO|S_IWUSR, \
nvme_adm_passthru_err_log_enabled_show,nvme_adm_passthru_err_log_enabled_store);static struct
device_attributedev_attr_io_passthru_err_log_enabled= \
__ATTR(passthru_err_log_enabled,S_IRUGO|S_IWUSR, \
nvme_io_passthru_err_log_enabled_show,nvme_io_passthru_err_log_enabled_store);
Quoted 7 lines
Then both attributes can be the same.
/sys/class/nvme/nvme0/passthru_err_log_enabled
/sys/class/nvme/nvme0/nvme0n1/passthru_err_log_enabledWhat do we prefer?
The second one looks fine to me.
Review
Quoted 7 lines
(quoted diff) +6
-0+static ssize_t nvme_passthru_err_log_enabled_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + return sysfs_emit(buf, dev->logging_enabled ? "on" : "off"); +} +
Can we use following on the top of this as most of the XXX_show() functions in the host/sysfs.c are using "\n" ?
drivers/nvme/host/sysfs.c +0 -0
@@ -38,7 +38,7 @@ static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL,
nvme_sysfs_rescan);
static ssize_t nvme_passthru_err_log_enabled_show(struct device *dev, struct device_attribute *attr, char *buf)
{
- return sysfs_emit(buf, dev->logging_enabled ? "on" : "off"); + return sysfs_emit(buf, dev->logging_enabled ? "on\n" : "off\n");
}
static ssize_t nvme_passthru_err_log_enabled_store(struct device *dev,
-ck