Bad commit backported on the v5.15.y branch ?

Bad commit backported on the v5.15.y branch ?

[Joel GUITTET]

Hello,

I faced some issue related to scaling frequency on ZynqMP device using v5.15.167 kernel.
As an exemple setting the scaling frequency below show it's not properly set:

cat /sys/devices/system/cpu/cpufreq/policy0/scaling_available_frequencies
299999 399999 599999 1199999

echo 399999 > /sys/devices/system/cpu/cpufreq/policy0/scaling_setspeed

cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq
399999

cat /sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq
299999 ====> Should be 399999

After analysis of this issue with the help of Xilinx, it appears that a commit was backported on the 5.15.y branch, but probably it should not, or not as is. The commit is 9117fc44fd3a9538261e530ba5a022dfc9519620 modifying drivers/clk/zynqmp/divider.c.

Is anybody reading this message able to answer why it was backported ? The information I have until now is that it is intended recent kernel version. Dependencies for this modification is currently under clarification with Xilinx (maybe another commit to backport).

By the way, reverting this commit fix the issue shown above.

Thanks for the feedback,
Joel

Re: Bad commit backported on the v5.15.y branch ?

[Thorsten Leemhuis]

On 11.10.24 10:48, Joel GUITTET wrote:
> 
> I faced some issue related to scaling frequency on ZynqMP device
> using v5.15.167 kernel. As an exemple setting the scaling frequency
> below show it's not properly set:
> 
> cat /sys/devices/system/cpu/cpufreq/policy0/
> scaling_available_frequencies 299999 399999 599999 1199999
> 
> echo 399999 > /sys/devices/system/cpu/cpufreq/policy0/
> scaling_setspeed
> 
> cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq 399999
> 
> cat /sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq 299999
> ====> Should be 399999
> 
> After analysis of this issue with the help of Xilinx, it appears
> that a commit was backported on the 5.15.y branch, but probably it
> should not, or not as is. The commit is
> 9117fc44fd3a9538261e530ba5a022dfc9519620 modifying drivers/clk/
> zynqmp/divider.c.

FWIW, that is 1fe15be1fb6135 ("drivers: clk: zynqmp: update divider
round rate logic") [v6.8-rc1].

> Is anybody reading this message able to answer why it was
> backported ?

Looks like because it fixes a bug. I CCed the original author and those
that handled the patch, maybe they can help us out and tell us what's
the best strategy forward here.

> The information I have until now is that it is intended
> recent kernel version. Dependencies for this modification is
> currently under clarification with Xilinx (maybe another commit to
> backport).
> 
> By the way, reverting this commit fix the issue shown above.
Does 6.12-rc work fine for you? Because if not, we should fix the
problem there.

Ciao, Thorsten

RE: Bad commit backported on the v5.15.y branch ?

[Joel GUITTET]

Thanks for the reply Thorsten.

So is anybody able to indicate why this commit 1fe15be1fb6135 has been backported to 5.15.y?
Actually this creates a bug on v5.15 (see commands executed in my original message).
I don't know for 6.8 or 6.12 release, I'm not able to update my target with such gap.

Regards
Joel

________________________________________
De : Thorsten Leemhuis <regressions@leemhuis.info>
Envoyé : lundi 14 octobre 2024 08h45
À : Jay Buddhabhatti <jay.buddhabhatti@amd.com>
Cc : Greg KH <gregkh@linuxfoundation.org>; Linux kernel regressions list <regressions@lists.linux.dev>; Sasha Levin <sashal@kernel.org>; Stephen Boyd <sboyd@kernel.org>; Joel GUITTET <jguittet.opensource@witekio.com>; linux-kernel@vger.kernel.org <linux-kernel@vger.kernel.org>; stable@vger.kernel.org <stable@vger.kernel.org>
Objet : Re: Bad commit backported on the v5.15.y branch ?
 
[Vous ne recevez pas souvent de courriers de regressions@leemhuis.info. Découvrez pourquoi ceci est important à https://aka.ms/LearnAboutSenderIdentification ]

On 11.10.24 10:48, Joel GUITTET wrote:
>
> I faced some issue related to scaling frequency on ZynqMP device
> using v5.15.167 kernel. As an exemple setting the scaling frequency
> below show it's not properly set:
>
> cat /sys/devices/system/cpu/cpufreq/policy0/
> scaling_available_frequencies 299999 399999 599999 1199999
>
> echo 399999 > /sys/devices/system/cpu/cpufreq/policy0/
> scaling_setspeed
>
> cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq 399999
>
> cat /sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq 299999
> ====> Should be 399999
>
> After analysis of this issue with the help of Xilinx, it appears
> that a commit was backported on the 5.15.y branch, but probably it
> should not, or not as is. The commit is
> 9117fc44fd3a9538261e530ba5a022dfc9519620 modifying drivers/clk/
> zynqmp/divider.c.

FWIW, that is 1fe15be1fb6135 ("drivers: clk: zynqmp: update divider
round rate logic") [v6.8-rc1].

> Is anybody reading this message able to answer why it was
> backported ?

Looks like because it fixes a bug. I CCed the original author and those
that handled the patch, maybe they can help us out and tell us what's
the best strategy forward here.

> The information I have until now is that it is intended
> recent kernel version. Dependencies for this modification is
> currently under clarification with Xilinx (maybe another commit to
> backport).
>
> By the way, reverting this commit fix the issue shown above.
Does 6.12-rc work fine for you? Because if not, we should fix the
problem there.

Ciao, Thorsten

Re: Bad commit backported on the v5.15.y branch ?

[Greg KH]

On Wed, Oct 16, 2024 at 09:23:10AM +0000, Joel GUITTET wrote:
> Thanks for the reply Thorsten.
> 
> So is anybody able to indicate why this commit 1fe15be1fb6135 has been backported to 5.15.y?

Because it has a "Fixes:" tag on it.  Is that tag not correct?

> Actually this creates a bug on v5.15 (see commands executed in my original message).

What "original message"?

> I don't know for 6.8 or 6.12 release, I'm not able to update my target with such gap.

It's already in the following releases:
	5.10.209 5.15.148 6.1.75 6.6.14 6.7.2 6.8
so if it needs to be reverted somewhere, please send the reverts with
the reason why.

thanks,

greg k-h

RE: Bad commit backported on the v5.15.y branch ?

[Joel GUITTET]

> Because it has a "Fixes:" tag on it.  Is that tag not correct?

understood why it's backported. Probably yes. Or maybe another commit should be introduced to not create a new issue.

> What "original message"?

My original message in this mailing list, see below commands that show the problem (last line is wrong):

cat /sys/devices/system/cpu/cpufreq/policy0/scaling_available_frequencies
299999 399999 599999 1199999

echo 399999 > /sys/devices/system/cpu/cpufreq/policy0/scaling_setspeed

cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq
399999

cat /sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq
299999 ====> Should be 399999

Re: Bad commit backported on the v5.15.y branch ?

[Greg KH]

On Wed, Oct 16, 2024 at 01:13:42PM +0000, Joel GUITTET wrote:
> > Because it has a "Fixes:" tag on it.  Is that tag not correct?
> 
> understood why it's backported. Probably yes. Or maybe another commit should be introduced to not create a new issue.

Perhaps, but I have no clue what is going on here, sorry.

Is this an issue in Linus's tree?  Is it in the other stable releases?
If not, are we missing a change?  Or does the issue show up there as
well?

> > What "original message"?
> 
> My original message in this mailing list, see below commands that show the problem (last line is wrong):
> 
> cat /sys/devices/system/cpu/cpufreq/policy0/scaling_available_frequencies
> 299999 399999 599999 1199999
> 
> echo 399999 > /sys/devices/system/cpu/cpufreq/policy0/scaling_setspeed
> 
> cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq
> 399999
> 
> cat /sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq
> 299999 ====> Should be 399999

Ok, so someone needs to tell me what to do here as I'm confused :)

thanks,

greg k-h