Re: [PATCH] fs: Improve eventpoll logging to stop indicting timerfd

[Thomas Gleixner <tglx@linutronix.de>]

Manish,

On Wed, Mar 24 2021 at 22:18, Manish Varma wrote:
> On Mon, Mar 22, 2021 at 2:40 PM Thomas Gleixner <tglx@linutronix.de> wrote:
>>
>> Not that I expect any real dependencies on it, but as always the devil
>> is in the details.
>
> Right, there are some userspace which depends on "[timerfd]" string
> https://codesearch.debian.net/search?q=%22%5Btimerfd%5D%22&literal=1

Details :)

> So, modifying file descriptor names at-least for timerfd will definitely
> break those.
>
> With that said, I am now thinking about leaving alone the file descriptor
> names as is, and instead, adding those extra information about the
> associated processes (i.e. process name or rather PID of the
> process) along with token ID directly into wakesource name, at the
> time of creating new wakesource i.e. in ep_create_wakeup_source().
>
> So, the wakesource names, that currently named as "[timerfd]", will be
> named something like:
> "epollitem<N>:<PID>.[timerfd]"
>
> Where N is the number of wakesource created since boot.

Where N is a unique ID token. :)

> This way we can still associate the process with the wakesource
> name and also distinguish multiple instances of wakesources using
> the integer identifier.

If that solves your problem and does not make anything else breaks which
relies on the exisitng epollitem naming convention, then I don't see a
problem with that.

Thanks,

        tglx