| Message ID | 20200917030204.50098-2-snelson@pensando.io |
|---|---|
| State | Changes Requested |
| Delegated to: | David Miller |
| Headers | show |
| Series | ionic: add devlink dev flash support | expand |
On Wed, 16 Sep 2020 20:02:00 -0700 Shannon Nelson wrote: > Add a timeout element to the DEVLINK_CMD_FLASH_UPDATE_STATUS > netlink message for use by a userland utility to show that > a particular firmware flash activity may take a long but > bounded time to finish. Also add a handy helper for drivers > to make use of the new timeout value. > > UI usage hints: > - if non-zero, add timeout display to the end of the status line > [component] status_msg ( Xm Ys : Am Bs ) > using the timeout value for Am Bs and updating the Xm Ys > every second > - if the timeout expires while awaiting the next update, > display something like > [component] status_msg ( timeout reached : Am Bs ) > - if new status notify messages are received, remove > the timeout and start over > > Signed-off-by: Shannon Nelson <snelson@pensando.io> Minor nits, otherwise LGTM: Reviewed-by: Jakub Kicinski <kuba@kernel.org> > @@ -3052,6 +3054,9 @@ static int devlink_nl_flash_update_fill(struct sk_buff *msg, > if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TOTAL, > total, DEVLINK_ATTR_PAD)) > goto nla_put_failure; > + if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TIMEOUT, > + timeout, DEVLINK_ATTR_PAD)) > + goto nla_put_failure; nit: since old kernels don't report this user space has to deal with it not being present so I'd be tempted to only report it if timeout is not 0 > +void devlink_flash_update_timeout_notify(struct devlink *devlink, > + const char *status_msg, > + const char *component, > + unsigned long timeout) > +{ > + __devlink_flash_update_notify(devlink, > + DEVLINK_CMD_FLASH_UPDATE_STATUS, > + status_msg, component, 0, 0, timeout); nit: did we ever report cmd == UPDATE_STATUS and total == 0? could this cause a division by zero in some unsuspecting implementation? Perhaps we should pass 1 here? > +} > +EXPORT_SYMBOL_GPL(devlink_flash_update_timeout_notify);
On 9/16/2020 8:02 PM, Shannon Nelson wrote: > Add a timeout element to the DEVLINK_CMD_FLASH_UPDATE_STATUS > netlink message for use by a userland utility to show that > a particular firmware flash activity may take a long but > bounded time to finish. Also add a handy helper for drivers > to make use of the new timeout value. > > UI usage hints: > - if non-zero, add timeout display to the end of the status line > [component] status_msg ( Xm Ys : Am Bs ) > using the timeout value for Am Bs and updating the Xm Ys > every second > - if the timeout expires while awaiting the next update, > display something like > [component] status_msg ( timeout reached : Am Bs ) > - if new status notify messages are received, remove > the timeout and start over > > Signed-off-by: Shannon Nelson <snelson@pensando.io> > --- This one looks good to me. I think the only other things that I saw from previous discussion are: a) we could convert the internal helper devlink_nl_flash_update_fill and __devlink_flash_update_notify to use structs for their fields, and.. b) Jakub pointed out that most drivers don't currently use the component field so we could remove that from the helpers. However, I don't have strong feelings about those either way, so: Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> > include/net/devlink.h | 4 ++++ > include/uapi/linux/devlink.h | 3 +++ > net/core/devlink.c | 29 +++++++++++++++++++++++------ > 3 files changed, 30 insertions(+), 6 deletions(-) > > diff --git a/include/net/devlink.h b/include/net/devlink.h > index eaec0a8cc5ef..f206accf80ad 100644 > --- a/include/net/devlink.h > +++ b/include/net/devlink.h > @@ -1400,6 +1400,10 @@ void devlink_flash_update_status_notify(struct devlink *devlink, > const char *component, > unsigned long done, > unsigned long total); > +void devlink_flash_update_timeout_notify(struct devlink *devlink, > + const char *status_msg, > + const char *component, > + unsigned long timeout); > > int devlink_traps_register(struct devlink *devlink, > const struct devlink_trap *traps, > diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h > index 40d35145c879..4a6e213cfa04 100644 > --- a/include/uapi/linux/devlink.h > +++ b/include/uapi/linux/devlink.h > @@ -460,6 +460,9 @@ enum devlink_attr { > > DEVLINK_ATTR_PORT_EXTERNAL, /* u8 */ > DEVLINK_ATTR_PORT_CONTROLLER_NUMBER, /* u32 */ > + > + DEVLINK_ATTR_FLASH_UPDATE_STATUS_TIMEOUT, /* u64 */ > + > /* add new attributes above here, update the policy in devlink.c */ > > __DEVLINK_ATTR_MAX, > diff --git a/net/core/devlink.c b/net/core/devlink.c > index 19037f114307..01f855e53e06 100644 > --- a/net/core/devlink.c > +++ b/net/core/devlink.c > @@ -3024,7 +3024,9 @@ static int devlink_nl_flash_update_fill(struct sk_buff *msg, > enum devlink_command cmd, > const char *status_msg, > const char *component, > - unsigned long done, unsigned long total) > + unsigned long done, > + unsigned long total, > + unsigned long timeout) Number of paramters here is getting rather large. Does it make sense to convert this one to a struct now? > { > void *hdr; > > @@ -3052,6 +3054,9 @@ static int devlink_nl_flash_update_fill(struct sk_buff *msg, > if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TOTAL, > total, DEVLINK_ATTR_PAD)) > goto nla_put_failure; > + if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TIMEOUT, > + timeout, DEVLINK_ATTR_PAD)) > + goto nla_put_failure; > > out: > genlmsg_end(msg, hdr); > @@ -3067,7 +3072,8 @@ static void __devlink_flash_update_notify(struct devlink *devlink, > const char *status_msg, > const char *component, > unsigned long done, > - unsigned long total) > + unsigned long total, > + unsigned long timeout) > { > struct sk_buff *msg; > int err; > @@ -3081,7 +3087,7 @@ static void __devlink_flash_update_notify(struct devlink *devlink, > return; > > err = devlink_nl_flash_update_fill(msg, devlink, cmd, status_msg, > - component, done, total); > + component, done, total, timeout); > if (err) > goto out_free_msg; > > @@ -3097,7 +3103,7 @@ void devlink_flash_update_begin_notify(struct devlink *devlink) > { > __devlink_flash_update_notify(devlink, > DEVLINK_CMD_FLASH_UPDATE, > - NULL, NULL, 0, 0); > + NULL, NULL, 0, 0, 0); > } > EXPORT_SYMBOL_GPL(devlink_flash_update_begin_notify); > > @@ -3105,7 +3111,7 @@ void devlink_flash_update_end_notify(struct devlink *devlink) > { > __devlink_flash_update_notify(devlink, > DEVLINK_CMD_FLASH_UPDATE_END, > - NULL, NULL, 0, 0); > + NULL, NULL, 0, 0, 0); > } > EXPORT_SYMBOL_GPL(devlink_flash_update_end_notify); > > @@ -3117,10 +3123,21 @@ void devlink_flash_update_status_notify(struct devlink *devlink, > { > __devlink_flash_update_notify(devlink, > DEVLINK_CMD_FLASH_UPDATE_STATUS, > - status_msg, component, done, total); > + status_msg, component, done, total, 0); > } > EXPORT_SYMBOL_GPL(devlink_flash_update_status_notify); > > +void devlink_flash_update_timeout_notify(struct devlink *devlink, > + const char *status_msg, > + const char *component, > + unsigned long timeout) So we dropped the done and total here since in most cases we expect not to need both a timeout and a done/total. I think that make sense. Most of the time I think a command will benefit from one or the other. > +{ > + __devlink_flash_update_notify(devlink, > + DEVLINK_CMD_FLASH_UPDATE_STATUS, > + status_msg, component, 0, 0, timeout); > +} > +EXPORT_SYMBOL_GPL(devlink_flash_update_timeout_notify); > + > static int devlink_nl_cmd_flash_update(struct sk_buff *skb, > struct genl_info *info) > { >
On 9/17/20 12:46 PM, Jakub Kicinski wrote: > On Wed, 16 Sep 2020 20:02:00 -0700 Shannon Nelson wrote: >> Add a timeout element to the DEVLINK_CMD_FLASH_UPDATE_STATUS >> netlink message for use by a userland utility to show that >> a particular firmware flash activity may take a long but >> bounded time to finish. Also add a handy helper for drivers >> to make use of the new timeout value. >> >> UI usage hints: >> - if non-zero, add timeout display to the end of the status line >> [component] status_msg ( Xm Ys : Am Bs ) >> using the timeout value for Am Bs and updating the Xm Ys >> every second >> - if the timeout expires while awaiting the next update, >> display something like >> [component] status_msg ( timeout reached : Am Bs ) >> - if new status notify messages are received, remove >> the timeout and start over >> >> Signed-off-by: Shannon Nelson <snelson@pensando.io> > Minor nits, otherwise LGTM: > > Reviewed-by: Jakub Kicinski <kuba@kernel.org> Thanks. > >> @@ -3052,6 +3054,9 @@ static int devlink_nl_flash_update_fill(struct sk_buff *msg, >> if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TOTAL, >> total, DEVLINK_ATTR_PAD)) >> goto nla_put_failure; >> + if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TIMEOUT, >> + timeout, DEVLINK_ATTR_PAD)) >> + goto nla_put_failure; > nit: since old kernels don't report this user space has to deal with it > not being present so I'd be tempted to only report it if timeout > is not 0 Hmmm... Unless there are any other comments on this, I think we can be consistent with the other fields here. > >> +void devlink_flash_update_timeout_notify(struct devlink *devlink, >> + const char *status_msg, >> + const char *component, >> + unsigned long timeout) >> +{ >> + __devlink_flash_update_notify(devlink, >> + DEVLINK_CMD_FLASH_UPDATE_STATUS, >> + status_msg, component, 0, 0, timeout); > nit: did we ever report cmd == UPDATE_STATUS and total == 0? > could this cause a division by zero in some unsuspecting > implementation? Perhaps we should pass 1 here? Yes, there are several examples of this. The userland devlink.c does a check for total == 0 before calculating the percentage. sln
On 9/17/20 12:50 PM, Jacob Keller wrote: > On 9/16/2020 8:02 PM, Shannon Nelson wrote: >> Add a timeout element to the DEVLINK_CMD_FLASH_UPDATE_STATUS >> netlink message for use by a userland utility to show that >> a particular firmware flash activity may take a long but >> bounded time to finish. Also add a handy helper for drivers >> to make use of the new timeout value. >> >> UI usage hints: >> - if non-zero, add timeout display to the end of the status line >> [component] status_msg ( Xm Ys : Am Bs ) >> using the timeout value for Am Bs and updating the Xm Ys >> every second >> - if the timeout expires while awaiting the next update, >> display something like >> [component] status_msg ( timeout reached : Am Bs ) >> - if new status notify messages are received, remove >> the timeout and start over >> >> Signed-off-by: Shannon Nelson <snelson@pensando.io> >> --- > This one looks good to me. I think the only other things that I saw from > previous discussion are: > > a) we could convert the internal helper devlink_nl_flash_update_fill and > __devlink_flash_update_notify to use structs for their fields, and.. > > b) Jakub pointed out that most drivers don't currently use the component > field so we could remove that from the helpers. > > However, I don't have strong feelings about those either way, so: > > Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> > > Thanks, sln
diff --git a/include/net/devlink.h b/include/net/devlink.h index eaec0a8cc5ef..f206accf80ad 100644 --- a/include/net/devlink.h +++ b/include/net/devlink.h @@ -1400,6 +1400,10 @@ void devlink_flash_update_status_notify(struct devlink *devlink, const char *component, unsigned long done, unsigned long total); +void devlink_flash_update_timeout_notify(struct devlink *devlink, + const char *status_msg, + const char *component, + unsigned long timeout); int devlink_traps_register(struct devlink *devlink, const struct devlink_trap *traps, diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h index 40d35145c879..4a6e213cfa04 100644 --- a/include/uapi/linux/devlink.h +++ b/include/uapi/linux/devlink.h @@ -460,6 +460,9 @@ enum devlink_attr { DEVLINK_ATTR_PORT_EXTERNAL, /* u8 */ DEVLINK_ATTR_PORT_CONTROLLER_NUMBER, /* u32 */ + + DEVLINK_ATTR_FLASH_UPDATE_STATUS_TIMEOUT, /* u64 */ + /* add new attributes above here, update the policy in devlink.c */ __DEVLINK_ATTR_MAX, diff --git a/net/core/devlink.c b/net/core/devlink.c index 19037f114307..01f855e53e06 100644 --- a/net/core/devlink.c +++ b/net/core/devlink.c @@ -3024,7 +3024,9 @@ static int devlink_nl_flash_update_fill(struct sk_buff *msg, enum devlink_command cmd, const char *status_msg, const char *component, - unsigned long done, unsigned long total) + unsigned long done, + unsigned long total, + unsigned long timeout) { void *hdr; @@ -3052,6 +3054,9 @@ static int devlink_nl_flash_update_fill(struct sk_buff *msg, if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TOTAL, total, DEVLINK_ATTR_PAD)) goto nla_put_failure; + if (nla_put_u64_64bit(msg, DEVLINK_ATTR_FLASH_UPDATE_STATUS_TIMEOUT, + timeout, DEVLINK_ATTR_PAD)) + goto nla_put_failure; out: genlmsg_end(msg, hdr); @@ -3067,7 +3072,8 @@ static void __devlink_flash_update_notify(struct devlink *devlink, const char *status_msg, const char *component, unsigned long done, - unsigned long total) + unsigned long total, + unsigned long timeout) { struct sk_buff *msg; int err; @@ -3081,7 +3087,7 @@ static void __devlink_flash_update_notify(struct devlink *devlink, return; err = devlink_nl_flash_update_fill(msg, devlink, cmd, status_msg, - component, done, total); + component, done, total, timeout); if (err) goto out_free_msg; @@ -3097,7 +3103,7 @@ void devlink_flash_update_begin_notify(struct devlink *devlink) { __devlink_flash_update_notify(devlink, DEVLINK_CMD_FLASH_UPDATE, - NULL, NULL, 0, 0); + NULL, NULL, 0, 0, 0); } EXPORT_SYMBOL_GPL(devlink_flash_update_begin_notify); @@ -3105,7 +3111,7 @@ void devlink_flash_update_end_notify(struct devlink *devlink) { __devlink_flash_update_notify(devlink, DEVLINK_CMD_FLASH_UPDATE_END, - NULL, NULL, 0, 0); + NULL, NULL, 0, 0, 0); } EXPORT_SYMBOL_GPL(devlink_flash_update_end_notify); @@ -3117,10 +3123,21 @@ void devlink_flash_update_status_notify(struct devlink *devlink, { __devlink_flash_update_notify(devlink, DEVLINK_CMD_FLASH_UPDATE_STATUS, - status_msg, component, done, total); + status_msg, component, done, total, 0); } EXPORT_SYMBOL_GPL(devlink_flash_update_status_notify); +void devlink_flash_update_timeout_notify(struct devlink *devlink, + const char *status_msg, + const char *component, + unsigned long timeout) +{ + __devlink_flash_update_notify(devlink, + DEVLINK_CMD_FLASH_UPDATE_STATUS, + status_msg, component, 0, 0, timeout); +} +EXPORT_SYMBOL_GPL(devlink_flash_update_timeout_notify); + static int devlink_nl_cmd_flash_update(struct sk_buff *skb, struct genl_info *info) {
Add a timeout element to the DEVLINK_CMD_FLASH_UPDATE_STATUS netlink message for use by a userland utility to show that a particular firmware flash activity may take a long but bounded time to finish. Also add a handy helper for drivers to make use of the new timeout value. UI usage hints: - if non-zero, add timeout display to the end of the status line [component] status_msg ( Xm Ys : Am Bs ) using the timeout value for Am Bs and updating the Xm Ys every second - if the timeout expires while awaiting the next update, display something like [component] status_msg ( timeout reached : Am Bs ) - if new status notify messages are received, remove the timeout and start over Signed-off-by: Shannon Nelson <snelson@pensando.io> --- include/net/devlink.h | 4 ++++ include/uapi/linux/devlink.h | 3 +++ net/core/devlink.c | 29 +++++++++++++++++++++++------ 3 files changed, 30 insertions(+), 6 deletions(-)