Re: [RFC PATCH v2 0/4] staging: vc04: Drop custom logging

[Stefan Wahren]

Hi Umang,

Am 23.09.23 um 11:39 schrieb Umang Jain:
> Hi Stefan,
>
> On 9/17/23 9:06 PM, Stefan Wahren wrote:
> > Hi,
> >
> > Am 14.09.23 um 08:35 schrieb Dan Carpenter:
> > > On Thu, Sep 14, 2023 at 12:25:24AM +0530, Umang Jain wrote:
> > > > Hello,
> > > >
> > > > This series attempts to restart the discussion on custom logging used
> > > > in VC04. In the last feedback gathered in [1] it seems that the logging
> > > > would rather be moved to use dynamic debug. The series tries to move
> > > > in that direction.
> > > >
> > > > The elephant in the room is the ability of turning on/off log levels,
> > > > which this series just drops. Compensated by a crude strings
> > > > ("error", "warning", "info"... etc) for easier grepping.
> > > >
> > > > The log category are also just strings (which probably can be 
> > > > transformed
> > > > to dynamic debug class names moving forwards?).
> > > >
> > > > To move forwards, I would like feedback on the broader direction.
> > > > There are couple of TODOs in each of the patch (summarised in commit
> > > > messages) which require case-by-case discussion.
> > > >
> > > > Additional high-level questions to move forwards:
> > > > 1. Is loss of log levels by moving to dynamic debug, is actually a
> > > >     concern? Is dynamic debug a valid replacement?
> > >
> > > Dynamic debug is honestly going to be an improvement.  I guess, Greg and
> > > I said this back in Jan.
>
> +1
> > > > 2. Whether debugfs should be dropped as well, found vestigial in [2]
> > >
> > > Yes. The "vchiq/log" should be removed.  Ideally as part of this
> > > patchset so it's easier to understand.
> >
> > Yes, but please do not remote vchiq_debugfs entirely. I'm working on a 
> > patch to move the state dump (debug feature) from the character device 
> > /dev/vchiq to debugfs /sys/kernel/debug/vchiq/dump_state.
>
> Can't the state dump be printed to dev_dbg() ? Will it pollute the 
> kernel log?  Having debugfs for a single dump doesn't seem worthwhile if 
> the state dump can be incorporated to dev_dbg() too.

debugfs was created for a good reason. dev_dbg() is a nice tool, but it 
isn't useful in every situation. In case of communication the usage of 
counter is quite popular and removing this ability would make debugging 
not easier.

Try to make "cat /dev/vchiq" you will see you don't want pollute the 
kernel log with this kind of information.

Actually i don't see a problem with have a single dump in debugfs.

Let me send a draft of my changes for a better discussion ...

Regards
Stefan

> > > > 3. whether vchiq_log_trace() should actually be tracing support for 
> > > > VC04
> > >
> > > That can be done later if people want.  No need to discuss it now.
>
> Thanks Dan.
> > > regards,
> > > dan carpenter