[gentoo-amd64] Microcode-ctl Problems

[gentoo-amd64] Microcode-ctl Problems

[Frank Peters]

After failing to receive any kind of answer from various sources, including
the developers themselves, I though that I would bring up the issue here.

Does the microcode_ctl utility work on the x86_64 processor, and in particular
the Core 2 quad?  If so, then how can it be successfully implemented?

To those who may not know, microcode_ctl is described at this link:
http://www.urbanmyth.org/microcode

My kernel is compiled to include microcode support and the proper device
(i.e. /dev/cpu/microcode) is in place.  But every time I load the microcode
module, using the command "modprobe microcode," the module takes about
three minutes to be loaded which seems quite abnormal.  Then, executing
"microcode_ctl -f microcode.dat" will report that the new firmware has
been loaded when actually it is not loaded.  Executing the same command
a second time, which should report that the new firmware is already loaded,
only again reports that new firmware has been loaded.  This behavior
indicates that something is very wrong.

Microcode_ctl is available for Gentoo AMD64 and there are no indications
anywhere that it does _not_ work on a 64-bit system.  What needs to
be done to get it to load the new firmware and why does the module
take three minutes to load?

Microcode_ctl is far from being a popular or even essential utility
but someone must have some knowledge of its function.  I recall that
it used to work nicely on a 32-bit machine but since I switched to pure
64-bit I can no longer use it.

Frank Peters

[gentoo-amd64] Re: Microcode-ctl Problems

[Nikos Chantziaras]

On 12/11/2009 09:31 PM, Frank Peters wrote:
> After failing to receive any kind of answer from various sources, including
> the developers themselves, I though that I would bring up the issue here.
>
> Does the microcode_ctl utility work on the x86_64 processor, and in particular
> the Core 2 quad?  If so, then how can it be successfully implemented?
>
> [...]
>
> Microcode_ctl is available for Gentoo AMD64 and there are no indications
> anywhere that it does _not_ work on a 64-bit system.  What needs to
> be done to get it to load the new firmware and why does the module
> take three minutes to load?
>
> Microcode_ctl is far from being a popular or even essential utility
> but someone must have some knowledge of its function.  I recall that
> it used to work nicely on a 32-bit machine but since I switched to pure
> 64-bit I can no longer use it.

I can't offer much help other than the assurance that it works on 
64-bit.  I've been using it to keep my CPU's microcode updated for quite 
a while.  It's a Core 2 Duo though, not Quad, running Gentoo ~amd64.

Re: [gentoo-amd64] Re: Microcode-ctl Problems

[Frank Peters]

On Sat, 12 Dec 2009 02:30:18 +0200
Nikos Chantziaras <realnc@arcor.de> wrote:

> 
> I can't offer much help other than the assurance that it works on 
> 64-bit.  I've been using it to keep my CPU's microcode updated for quite 
> a while.  It's a Core 2 Duo though, not Quad, running Gentoo ~amd64.
> 
> 

Recently I installed Gentoo on another machine that has a Pentium D
processor with dual cores.  As a test, I tried to update the microcode
on this other machine and experienced the exact same behavior.  The microcode
module required about five minutes to load and the update did not occur.

This was quite a surprise to me, but it also tends to indicate that
the fault is with the module and the kernel.  The module should load
immediately and there should be no delay.

Possibly there is some additional kernel option that needs to
be set but I cannot imagine what it could be.

Frank Peters

Re: [gentoo-amd64] Re: Microcode-ctl Problems

[Frank Peters]

On Sat, 12 Dec 2009 02:30:18 +0200
Nikos Chantziaras <realnc@arcor.de> wrote:

> 
> I can't offer much help other than the assurance that it works on 
> 64-bit.  I've been using it to keep my CPU's microcode updated for quite 
> a while.  It's a Core 2 Duo though, not Quad, running Gentoo ~amd64.
> 
> 

Could you possibly check your kernel log?

Here is the log from my machine after loading the microcode module:

Dec 11 23:24:25 (none) kernel: microcode: CPU0 sig=0x6fb, pf=0x10, revision=0xb6
Dec 11 23:24:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
Dec 11 23:25:25 (none) kernel: microcode: CPU1 sig=0x6fb, pf=0x10, revision=0xb6
Dec 11 23:25:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
Dec 11 23:26:25 (none) kernel: microcode: CPU2 sig=0x6fb, pf=0x10, revision=0xb6
Dec 11 23:26:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
Dec 11 23:27:25 (none) kernel: microcode: CPU3 sig=0x6fb, pf=0x10, revision=0xb6
Dec 11 23:27:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
Dec 11 23:28:25 (none) kernel: Microcode Update Driver: v2.00 <tigran@aivazian.fsnet.co.uk>, Peter Oruba

The loading process occurs over four minutes as each cpu seems
to make some sort of request for intel microcode.  There seems
to be a request and then a long wait.

I don't fully understand what is happening but the firmware drivers
seem to play some role.  Maybe my kernel is not configured
correctly for handling firmware.

Frank Peters

[gentoo-amd64] Re: Microcode-ctl Problems

[Nikos Chantziaras]

On 12/12/2009 06:43 AM, Frank Peters wrote:
> On Sat, 12 Dec 2009 02:30:18 +0200
> Nikos Chantziaras<realnc@arcor.de>  wrote:
>
>>
>> I can't offer much help other than the assurance that it works on
>> 64-bit.  I've been using it to keep my CPU's microcode updated for quite
>> a while.  It's a Core 2 Duo though, not Quad, running Gentoo ~amd64.
>
> Could you possibly check your kernel log?
>
> Here is the log from my machine after loading the microcode module:
>
> Dec 11 23:24:25 (none) kernel: microcode: CPU0 sig=0x6fb, pf=0x10, revision=0xb6
> Dec 11 23:24:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
> Dec 11 23:25:25 (none) kernel: microcode: CPU1 sig=0x6fb, pf=0x10, revision=0xb6
> Dec 11 23:25:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
> Dec 11 23:26:25 (none) kernel: microcode: CPU2 sig=0x6fb, pf=0x10, revision=0xb6
> Dec 11 23:26:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
> Dec 11 23:27:25 (none) kernel: microcode: CPU3 sig=0x6fb, pf=0x10, revision=0xb6
> Dec 11 23:27:25 (none) kernel: platform microcode: firmware: requesting intel-ucode/06-0f-0b
> Dec 11 23:28:25 (none) kernel: Microcode Update Driver: v2.00<tigran@aivazian.fsnet.co.uk>, Peter Oruba

Well, I don't load it manually but have /etc/init.d/microcode_ctl in my 
boot runlevel, but I guess this doesn't make any difference.  My logs 
look the same except for the time (there's no delay) and the microcode 
version (probably due to different CPU used.)


> I don't fully understand what is happening but the firmware drivers
> seem to play some role.  Maybe my kernel is not configured
> correctly for handling firmware.

Are you sure you have "Intel microcode patch loading support" 
(CONFIG_MICROCODE_INTEL) enabled in the kernel and "AMD microcode patch 
loading support" (CONFIG_MICROCODE_AMD) disabled?

Also make sure "Userspace firmware loading support" (CONFIG_FW_LOADER) 
is enabled.  It should be enabled automatically by 
CONFIG_MICROCODE_INTEL though.

Re: [gentoo-amd64] Re: Microcode-ctl Problems

[Frank Peters]

On Sat, 12 Dec 2009 14:17:35 +0200
Nikos Chantziaras <realnc@arcor.de> wrote:

> 
> Well, I don't load it manually but have /etc/init.d/microcode_ctl in my 
> boot runlevel, but I guess this doesn't make any difference.  My logs 
> look the same except for the time (there's no delay) and the microcode 
> version (probably due to different CPU used.)
> 

Thanks for looking.

The closest report I could find to my problem (actually the only
report) is at this link:

http://lkml.indiana.edu/hypermail/linux/kernel/0703.3/2196.html

My kernel has all that hotplug and ACPI stuff turned off.  That
should not make any difference but possibly there is a kernel
bug somewhere that has been overlooked.  It is doubtful that
many people use microcode updating and so problems like this
could easily slip past.

I will search for a while longer and then file a bug report with
the Linux kernel mailing list.

Incidentally, the Gentoo boot script for microcode_ctl is not correct.
It uses a "-d" option to specify the microcode device:

microcode_ctl -d /dev/cpu/microcode

But the "-d" option does not exist.  The device is specified at
compile time.

Frank Peters

[gentoo-amd64] Re: Microcode-ctl Problems

[Nikos Chantziaras]

On 12/12/2009 06:08 PM, Frank Peters wrote:
>[...]
> Incidentally, the Gentoo boot script for microcode_ctl is not correct.
> It uses a "-d" option to specify the microcode device:
>
> microcode_ctl -d /dev/cpu/microcode
>
> But the "-d" option does not exist.  The device is specified at
> compile time.

It does exist.  It's just not documented.  If you run "microcode_ctl -d" 
you get "option requires an argument -- 'd'".