Re: [patch net-next 0/4] net: devlink: sync flash and dev info command

[Jiri Pirko <jiri@resnulli.us>]

Tue, Aug 23, 2022 at 06:29:48PM CEST, m.chetan.kumar@linux.intel.com wrote:
>On 8/23/2022 5:50 PM, Jiri Pirko wrote:
>> Tue, Aug 23, 2022 at 12:09:06PM CEST, m.chetan.kumar@linux.intel.com wrote:
>> > On 8/19/2022 1:55 PM, Jiri Pirko wrote:
>> > > Fri, Aug 19, 2022 at 04:49:40AM CEST, kuba@kernel.org wrote:
>> > > > On Thu, 18 Aug 2022 15:00:38 +0200 Jiri Pirko wrote:
>> > > > > Currently it is up to the driver what versions to expose and what flash
>> > > > > update component names to accept. This is inconsistent. Thankfully, only
>> > > > > netdevsim is currently using components, so it is a good time
>> > > > > to sanitize this.
>> > > > 
>> > > > Please take a look at recently merged code - 5417197dd516 ("Merge branch
>> > > > 'wwan-t7xx-fw-flashing-and-coredump-support'"), I don't see any versions
>> > > > there so I think you're gonna break them?
>> > > 
>> > > Ah, crap. Too late :/ They are passing the string to FW (cmd is
>> > > the component name here):
>> > > static int t7xx_devlink_fb_flash(const char *cmd, struct t7xx_port *port)
>> > > {
>> > >           char flash_command[T7XX_FB_COMMAND_SIZE];
>> > > 
>> > >           snprintf(flash_command, sizeof(flash_command), "%s:%s", T7XX_FB_CMD_FLASH, cmd);
>> > >           return t7xx_devlink_fb_raw_command(flash_command, port, NULL);
>> > > }
>> > > 
>> > > This breaks the pairing with info.versions assumption. Any possibility
>> > > to revert this and let them redo?
>> > > 
>> > > Ccing m.chetan.kumar@linux.intel.com, chandrashekar.devegowda@intel.com,
>> > > soumya.prakash.mishra@intel.com
>> > > 
>> > > Guys, could you expose one version for component you are flashing? We
>> > > need 1:1 mapping here.
>> > 
>> > Thanks for the heads-up.
>> > I had a look at the patch & my understanding is driver is supposed
>> > to expose flash update component name & version details via
>> > devlink_info_version_running_put_ext().
>> 
>> Yes.
>> 
>> > 
>> > Is version value a must ? Internally version value is not used for making any
>> > decision so in case driver/device doesn't support it should be ok to pass
>> > empty string ?
>> 
>> No.
>> 
>> > 
>> > Ex:
>> > devlink_info_version_running_put_ext(req, "fw", "",
>> > DEVLINK_INFO_VERSION_TYPE_COMPONENT);
>> > 
>> > One observation:-
>> > While testing I noticed "flash_components:" is not getting displayed as
>> > mentioned in cover note.
>> 
>> You need iproute2 patch for that which is still in my queue:
>> https://github.com/jpirko/iproute2_mlxsw/commit/e1d36409362257cc42a435f6695d2058ab7ab683
>
>Thanks. After applying this patch "flash_components" details are getting
>displayed.
>
>Another observation is if NULL is passed for version_value there is a crash.

So don't pass NULL :)


>Below is the backtrace.
>
>3187.556637] BUG: kernel NULL pointer dereference, address: 0000000000000000
>[ 3187.556659] #PF: supervisor read access in kernel mode
>[ 3187.556666] #PF: error_code(0x0000) - not-present page
>3187.556791] Call Trace:
>[ 3187.556796]  <TASK>
>[ 3187.556801]  ? devlink_info_version_put+0x112/0x1d0
>[ 3187.556823]  ? __nla_put+0x20/0x30
>[ 3187.556833]  devlink_info_version_running_put_ext+0x1c/0x30
>[ 3187.556851]  t7xx_devlink_info_get+0x37/0x40 [mtk_t7xx]
>[ 3187.556880]  devlink_nl_info_fill.constprop.0+0xa1/0x120
>[ 3187.556892]  devlink_nl_cmd_info_get_dumpit+0xa8/0x140
>[ 3187.556901]  netlink_dump+0x1a3/0x340
>[ 3187.556913]  __netlink_dump_start+0x1d0/0x290
>
>Is driver expected to set version number along with component name ?

Of course.


>
>mtk_t7xx WWAN driver is using the devlink interface for flashing the fw to
>WWAN device. If WWAN device is not capable of supporting the versioning for
>each component how should we handle ? Please suggest.

The user should have a visibility about what version is currently
stored/running in the device. You should expose it.


>
>-- 
>Chetan