Re: [PATCH v3 0/7] Venus dynamic debug

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Jun 9, 2020 at 10:49 AM Joe Perches <joe@xxxxxxxxxxx> wrote:
>
> (adding Jim Cromie and comments)
>

thanks for bringing me in...


> On Tue, 2020-06-09 at 09:03 -0700, Randy Dunlap wrote:
> > On 6/9/20 4:13 AM, Matthew Wilcox wrote:
> > > On Tue, Jun 09, 2020 at 01:45:57PM +0300, Stanimir Varbanov wrote:
> > > > Here is the third version of dynamic debug improvements in Venus
> > > > driver.  As has been suggested on previous version by Joe [1] I've
> > > > made the relevant changes in dynamic debug core to handle leveling
> > > > as more generic way and not open-code/workaround it in the driver.
> > > >
> > > > About changes:
> > > >  - added change in the dynamic_debug and in documentation
> > > >  - added respective pr_debug_level and dev_dbg_level
> > >
> > > Honestly, this seems like you want to use tracepoints, not dynamic debug.
>
> Tracepoints are a bit heavy and do not have any class
> or grouping mechanism.
>
> debug_class is likely a better name than debug_level
>
> > Also see this patch series:
> > https://lore.kernel.org/lkml/20200605162645.289174-1-jim.cromie@xxxxxxxxx/
> > [PATCH 00/16] dynamic_debug: cleanups, 2 features
> >
> > It adds/expands dynamic debug flags quite a bit.
>
> Yes, and thanks Randy and Jim and Stanimir
>
> I haven't gone through Jim's proposal enough yet.
> It's unfortunate these patches series conflict.
>
> And for Jim, a link to Stanimir's patch series:
> https://lore.kernel.org/lkml/20200609104604.1594-1-stanimir.varbanov@xxxxxxxxxx/
>
>


As Joe noted, there is a lot of ad-hockery to possibly clean up,
but I dont grok how these levels should be distinguished from
KERN_(WARN|INFO|DEBUG) constants.
Those constants are used by coders, partly to convey how bad things are
As a user, Id be reluctant to disable an EMERG callsite.

are you trying to add a User Bit ? or maybe 7-9 of them ?

I have a patchset which adds a 'u' flag, for user.
An earlier version had x,y,z flags for 3 different user purposes.
I simplified, since only 1 was needed to mark up arbitrary sets of callsites.
Another patchset feature lets u select on that flag.

 #> echo u+p > control

Joe suggested class, I certainly find level confusing.

Is what you want user-flags u[1-7], or driver-flags d]1-7]   ?
and let me distinguish,
your flags are set in code, not modifiable by user, only for filtering
on flag/bit state ?
so theyd be different than [pfmltu_] flags, which are user changed.

my patchset also adds filtering on flag-state,
so that "echo u+p > control " could work.

if you had
     echo 'module venus 1+p; 2+p; 9+p' > control
how far would you get ?

if it covers your needs, then we could add
numerical flags (aka U1, U9) can be distinguished from  [pfmltu_PFMLTU]
and excluded from the mod-flags changes

from there, it shouldnt be hard to add some macro help

DECLARE_DYNDBG_FLAG ( 1, 'x' )
DECLARE_DYNDBG_FLAG ( 2, 'y' )
DECLARE_DYNDBG_FLAG ( 3, 'z' )

DECLARE_DYNDBG_FLAG_INFO ( 4, 'q', "unquiet a programmer defined debug
callsite set" )

also, since Im here, and this is pretty much on-topic,
I perused https://lkml.org/lkml/2020/5/21/399

I see 3 things;
- s / dev_dbg / VDBGL /
- add a bunch of VDBGM calls
- sys_get_prop_image_version  signature change.   (this doesnt really
belong here)

ISTM most of the selection of dyndbg callsites in that part of venus
could be selected by format.

   echo "module venus format error +p" > control

if so, refining your messages will define the logical sets for you ?


thanks
JimC (one of them anyway)



[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [Linux for Sparc]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux