proposal for important kernel patch

-------- Forwarded Message --------
Subject: proposal for important kernel patch
Date: Wed, 20 May 2015 09:04:54 -0000
From: Andrea Bravetti <andreabravetti@gmail.com>
Reply-To: Andrea Bravetti <andreabravetti@gmail.com>
To: Tim Gardner <tim.gardner@canonical.com>

I don't know how to propose an important kernel patch, so I ask you.

The first patch I would like to bring to your attention is this:


"acpi_cpufreq" should be built always as module and NEVER builtin
because it prevent phc-linux being loaded, and also any other
acpi_cpufreq replacement provided by the vendor if any.

"phc-linux" is very important if you need to downvolt some cpu.

Since I can't see any problem building it as a module and it should not
cause any regression it should be made even on 14.04 LTS, and for any
future release.

You can read about linux-phc here: http://www.linux-phc.org/forum

Also having linux-phc in the default kernel would be great but it may
need some work.

It's important, especially for some ThinkPad notebook like the Edge 15
0301-J4G that hang frequently over 100C because of buggy voltage
regulation, even new, clean fan, etc...

Not asking to fix in the mainstream kernel somethink so specific and
hardware bug related, but please please please don't force anyone has
this problem to rebuild a custom kernel because it's very annoying, just
build acpi_cpufreq as module and not builtin.

sincerely,
Andrea

On Wed, May 20, 2015 at 06:27:14AM -0600, Tim Gardner wrote:
> 
> 
> 
> -------- Forwarded Message --------
> Subject: proposal for important kernel patch
> Date: Wed, 20 May 2015 09:04:54 -0000
> From: Andrea Bravetti <andreabravetti@gmail.com>
> Reply-To: Andrea Bravetti <andreabravetti@gmail.com>
> To: Tim Gardner <tim.gardner@canonical.com>
> 
> I don't know how to propose an important kernel patch, so I ask you.
> 
> The first patch I would like to bring to your attention is this:
> 
> --- drivers/ata/libata-core.c   2015-05-11 18:49:06.699847395 +0200
> +++ drivers/ata/libata-core.c   2015-05-11 18:52:23.402510885 +0200
> @@ -4246,6 +4246,8 @@
>         { "WDC WD3000JD-*",             NULL,
> ATA_HORKAGE_WD_BROKEN_LPM },
>         { "WDC WD3200JD-*",             NULL,
> ATA_HORKAGE_WD_BROKEN_LPM },
> +       { "Samsung SSD 840*",           NULL,   ATA_HORKAGE_NO_NCQ_TRIM,
> },
> +
>         /* End Marker */
>         { }
>  };
> 
> Why is explained here:
> https://bugs.launchpad.net/ubuntu/+source/fstrim/+bug/1449005
> 
> Whitout this patch, or "libata.force=noncq", the Samsung SSD 840 is
> unusable with new firmware EXT0DB6Q: We lost a working day for this
> problem.

This first issue appears to be fixed by a couple of upstream HORKAGE
changes which appear to be marked for stable.  I have moved the bug listed
above over to the kernel package and marked it up for those fixes so we
can track them through the stable process.

> 
> The second one is a change in default kernel configuration:
> 
> --- .config     2015-05-20 10:28:41.709948652 +0200
> +++ .config     2015-05-20 09:23:13.332909672 +0200
> @@ -635,5 +635,5 @@
>  #
>  CONFIG_X86_PCC_CPUFREQ=y
> -CONFIG_X86_ACPI_CPUFREQ=y
> +CONFIG_X86_ACPI_CPUFREQ=m
>  CONFIG_X86_ACPI_CPUFREQ_CPB=y
>  CONFIG_X86_POWERNOW_K8=m
> 
> "acpi_cpufreq" should be built always as module and NEVER builtin
> because it prevent phc-linux being loaded, and also any other
> acpi_cpufreq replacement provided by the vendor if any.
> 
> "phc-linux" is very important if you need to downvolt some cpu.
> 
> Since I can't see any problem building it as a module and it should not
> cause any regression it should be made even on 14.04 LTS, and for any
> future release.
> 
> You can read about linux-phc here: http://www.linux-phc.org/forum
> 
> Also having linux-phc in the default kernel would be great but it may
> need some work.
> 
> It's important, especially for some ThinkPad notebook like the Edge 15
> 0301-J4G that hang frequently over 100C because of buggy voltage
> regulation, even new, clean fan, etc...
> 
> Not asking to fix in the mainstream kernel somethink so specific and
> hardware bug related, but please please please don't force anyone has
> this problem to rebuild a custom kernel because it's very annoying, just
> build acpi_cpufreq as module and not builtin.

I think this is built-in becuase it did not use to be autoloadable.  We
should investigate if that is still true.

-apw