Message ID | ME3P282MB270323F98B97A1A98A50F8F7BBF1A@ME3P282MB2703.AUSP282.PROD.OUTLOOK.COM (mailing list archive) |
---|---|
Headers | show |
Series | net: wwan: t7xx: fw flashing & coredump support | expand |
Tue, Sep 12, 2023 at 11:48:40AM CEST, songjinjian@hotmail.com wrote: >Adds support for t7xx wwan device firmware flashing & coredump collection >using devlink. I don't believe that use of devlink is correct here. It seems like a misfit. IIUC, what you need is to communicate with the modem. Basically a communication channel to modem. The other wwan drivers implement these channels in _ctrl.c files, using multiple protocols. Why can't you do something similar and let devlink out of this please? Until you put in arguments why you really need devlink and why is it a good fit, I'm against this. Please don't send any other versions of this patchset that use devlink. NACK.
Tue, Sep 12, 2023 at 11:48:40AM CEST, songjinjian@hotmail.com wrote: pw-bot: changes-requested
On Wed, 13 Sept 2023 at 11:17, Jiri Pirko <jiri@resnulli.us> wrote: > > Tue, Sep 12, 2023 at 11:48:40AM CEST, songjinjian@hotmail.com wrote: > >Adds support for t7xx wwan device firmware flashing & coredump collection > >using devlink. > > I don't believe that use of devlink is correct here. It seems like a > misfit. IIUC, what you need is to communicate with the modem. Basically > a communication channel to modem. The other wwan drivers implement these > channels in _ctrl.c files, using multiple protocols. Why can't you do > something similar and let devlink out of this please? > > Until you put in arguments why you really need devlink and why is it a > good fit, I'm against this. Please don't send any other versions of this > patchset that use devlink. The t7xx driver already has regular wwan data and control interfaces registered with the wwan framework, making it functional. Here the exposed low level resources are not really wwan/class specific as it is for firmware upgrade and coredump, so I think that is why Jinjian chose the 'feature agnostic' devlink framework. IMHO I think it makes sense to rely on such a framework, or maybe on the devcoredump class. That said, I see the protocol for flashing and doing the coreboot is fastboot, which is already supported on the user side with the fastboot tool, so I'm not sure abstracting it here makes sense. If the protocol is really fasboot compliant, Wouldn't it be simpler to directly expose it as a new device/channel? and rely on a userspace tool for regular fastboot operations (flash, boot, dump). This may require slightly modifying the fastboot tool to detect and support that new transport (in addition to the existing usb and ethernet support). Regards, Loic
Thu, Sep 21, 2023 at 11:36:26AM CEST, loic.poulain@linaro.org wrote: >On Wed, 13 Sept 2023 at 11:17, Jiri Pirko <jiri@resnulli.us> wrote: >> >> Tue, Sep 12, 2023 at 11:48:40AM CEST, songjinjian@hotmail.com wrote: >> >Adds support for t7xx wwan device firmware flashing & coredump collection >> >using devlink. >> >> I don't believe that use of devlink is correct here. It seems like a >> misfit. IIUC, what you need is to communicate with the modem. Basically >> a communication channel to modem. The other wwan drivers implement these >> channels in _ctrl.c files, using multiple protocols. Why can't you do >> something similar and let devlink out of this please? >> >> Until you put in arguments why you really need devlink and why is it a >> good fit, I'm against this. Please don't send any other versions of this >> patchset that use devlink. > >The t7xx driver already has regular wwan data and control interfaces >registered with the wwan framework, making it functional. Here the >exposed low level resources are not really wwan/class specific as it >is for firmware upgrade and coredump, so I think that is why Jinjian >chose the 'feature agnostic' devlink framework. IMHO I think it makes >sense to rely on such a framework, or maybe on the devcoredump class. > >That said, I see the protocol for flashing and doing the coreboot is >fastboot, which is already supported on the user side with the >fastboot tool, so I'm not sure abstracting it here makes sense. If the >protocol is really fasboot compliant, Wouldn't it be simpler to >directly expose it as a new device/channel? and rely on a userspace >tool for regular fastboot operations (flash, boot, dump). This may >require slightly modifying the fastboot tool to detect and support >that new transport (in addition to the existing usb and ethernet >support). Sounds sane. Please let devlink out of this. > >Regards, >Loic