Message-ID: <20250319091741.5488592b@akair>
Date: Wed, 19 Mar 2025 09:17:41 +0100
From: Andreas Kemnade <andreas@...nade.info>
To: Tony Lindgren <tony@...mide.com>
Cc: "Sverdlin, Alexander" <alexander.sverdlin@...mens.com>,
 "rogerq@...nel.org" <rogerq@...nel.org>, "aaro.koskinen@....fi"
 <aaro.koskinen@....fi>, "linux-kernel@...r.kernel.org"
 <linux-kernel@...r.kernel.org>, "linux-omap@...r.kernel.org"
 <linux-omap@...r.kernel.org>, "khilman@...libre.com" <khilman@...libre.com>
Subject: Re: [PATCH] Revert "bus: ti-sysc: Probe for l4_wkup and l4_cfg
 interconnect devices first"

Am Wed, 19 Mar 2025 05:56:06 +0200
schrieb Tony Lindgren <tony@...mide.com>:

> * Andreas Kemnade <andreas@...nade.info> [250313 22:01]:
> > Am Thu, 13 Mar 2025 20:42:23 +0000
> > schrieb "Sverdlin, Alexander" <alexander.sverdlin@...mens.com>:
> >   
> > > Hi Andreas!
> > > 
> > > On Thu, 2025-03-13 at 20:21 +0100, Andreas Kemnade wrote:  
> > > > > This reverts commit 4700a00755fb5a4bb5109128297d6fd2d1272ee6.
> > > > > 
> > > > > It brakes target-module@...00050 ("ti,sysc-omap2") probe on AM62x in a case
> > > > > when minimally-configured system tries to network-boot:
> > > > >     
> > > > brakes or breaks? To unterstand the severity of the issue...    
> > > 
> > > Thanks for the correction, it should have been "breaks"...
> > >   
> > > > > [    6.888776] probe of 2b300050.target-module returned 517 after 258 usecs
> > > > > [   17.129637] probe of 2b300050.target-module returned 517 after 708 usecs
> > > > > [   17.137397] platform 2b300050.target-module: deferred probe pending: (reason unknown)
> > > > > [   26.878471] Waiting up to 100 more seconds for network.
> > > > > 
> > > > > Arbitrary 10 deferrals is really not a solution to any problem.    
> > > > 
> > > > So there is a point where no more probe of anything pending are
> > > > triggered and therefore things are not probed?    
> > > 
> > > Because there is a point indeed (if we configure quite minimal set of drivers just
> > > enough to mount NFS) when deferred probes are not triggered any longer.
> > >   
> > > > > Stable mmc enumeration can be achiever by filling /aliases node properly
> > > > > (4700a00755fb commit's rationale).
> > > > >     
> > > > yes, it does not look like a clean solution. And we have the
> > > > proper aliases node in many places. What I am a bit wondering about is
> > > > what kind of sleeping dogs we are going to wake up by this revert. So I
> > > > think this should be tested a lot esp. about possible pm issues.
> > > > 
> > > > Not every dependency in the sysc probe area is properly defined.    
> > > 
> > > But the patch I propose to revert is really not a solution for missing
> > > dependencies on syscons. I'm fine with not propagating this to stable,
> > > but reverting in master should give enough time for older SoCs to test,
> > > WDYT?
> > >  
> > I am not against your revert proposal and not against propagating it
> > to stable, I would just like to see some Tested-Bys before it gets
> > applied to anything. If anything nasty pops up, it should be solved in a
> > cleaner way then with the offending patch.  
> 
> Sounds like for the AM62x problem there is simply some resource missing
> that needs to be configured. Did you track down which resource is causing
> the deferred probe without the revert?
> 
I think you have not understand the real problem here. I guess, that
problem can be provoked on other systems, too, if you just limit the
devices to the absolute minimum required.
The problem is as far as I understand a bit different. The problem is
not a resource is missing totally, it is just the artificial deferral
here. If there are just a minimum devices configured, you can come to a
point where there is nothing to trigger a loop through all the deferred
devices causing them to never probe.
An arbitary, unrelated device with a driver popping up would unstall
that deferral. 

I will just play around with the systems I have access to and if nothing
pops up, I will add a Tested-By/Reviewed-By. If more serious problems
pops up (I do not think so), another clean fix should get in before
getting this reverted.

> Reverting the commit does not really fix the root cause. It just ignores
> the problem of the hierarchy of the interconnect instances. Some of the
> interconnect instances are always-on, and contain devices providing
> resources for the other interconnect devices. So I would not consider
> patching MMC aliases all over the place as an alternative to fixing the
> real problem :)
>
So what is the real problem you wanted to fix? MMC aliases are there at
many places already. So is there anything besides MMC order?

Regards,
Andreas

Powered by blists - more mailing lists