On 1/8/20 4:52 PM, Ferruh Yigit wrote: > On 1/8/2020 1:25 PM, Thomas Monjalon wrote: >> 08/01/2020 14:20, Ferruh Yigit: >>> On 1/8/2020 1:06 PM, Thomas Monjalon wrote: >>>> 08/01/2020 13:59, Ferruh Yigit: >>>>> But for dev_ops, instead of having another capabilities indicator, which >>>>> requires PMDs to keep this synchronized, I think it is better if we can >>>>> self >>>>> contain this information within dev_ops, like not implementing dev_ops >>>>> would >>>>> mean it is not supported, this way it is easier to maintain and less >>>>> error prone. >>>> >>>> It means the dev_ops is resetted at init if a device does not support the >>>> feature. >>>> It is against having const dev_ops. >>> >>> I didn't get your comment. >>> For example getting FW version, I am saying instead of keeping another >>> piece of >>> information to say if it is supported by device/driver, better to grasp this >>> from if the driver implemented 'fw_version_get' dev_ops or not. >> >> I like this approach. >> Capabilities should be expressed by setting the function pointer or not >> (NULL). >> But a driver may support a feature for a subset of devices. > > In that case dev_ops itself can return the '-ENOTSUP', since application > interaction will be through the ethdev API, either API send '-ENOTSUP' because > the dev_ops is NULL or dev_ops itself send the '-ENOTSUP' because of the > underlying version of the device, for application it will be clear that that > feature is not supported.
I think it is a good illustration why deriving the capability from dev_ops pointer is not that good idea. >> If a device does not support a feature, the function pointer must be set to >> NULL. >> The only issue is having dev_ops as a const struct. > > Not sure about changing the dev_ops on runtime, it can be very hard to debug. I hope it was just an idea to copy dev_ops and adjust in accordance with the device capabilities on register. I.e. not fully dynamic changes in runtime.
