From: Andreas Herrmann on
On Thu, Nov 12, 2009 at 01:06:36PM +0100, Dmitry Adamushko wrote:
> 2009/11/12 Dmitry Adamushko <dmitry.adamushko(a)>:
> > 2009/11/12 Ingo Molnar <mingo(a)>:
> >>
> >> -tip testing found the following bug - there's a _long_ boot delay of
> >> 58.6 seconds if the CPU family is not supported:
> >>
> >> [ 1.421761] calling microcode_init+0x0/0x137 @ 1
> >> [ 1.426532] platform microcode: firmware: requesting amd-ucode/microcode_amd.bin
> >> [ 61.433126] microcode: failed to load file amd-ucode/microcode_amd.bin
> >> [ 61.439682] microcode: CPU0: AMD CPU family 0xf not supported
> >> [ 61.445441] microcode: CPU1: AMD CPU family 0xf not supported
> >> [ 61.451273] Microcode Update Driver: v2.00 <tigran(a)>, Peter Oruba
> >> [ 61.459116] initcall microcode_init+0x0/0x137 returned 0 after 58625622 usecs
> >>
> >> Where does this delay come from?
> >
> > My guess is that it's comming from
> >
> > static int loading_timeout = 60; /* In seconds */
> >
> > drivers/base/firmware_class.c
> >
> > given that you seem to have MICROCODE build in kernel, so this patch
> >;a=commit;h=d1c84f79a6ba992dc01e312c44a21496303874d6
> >
> > will result in sending a request for a firmware image to user-space
> > (unless that firmware image is also built-in into the kernel) and
> > user-space has not started yet.
> btw., it doesn't make sense for request_firmware() to even try this if
> the system_state != SYSTEM_RUNNING and current == 'init' (it'd perhaps
> make some sense if it's been done in a context of another task -- like
> in case of a parallel boot).

> And perhaps it just makes sense for microcode to use request_firmware_nowait().

That would be asynchronous.

I think I should ensure that microcode_amd.c is compiled into
microcode.o if and only if its built as module. microcode_amd.c
supports only the firmware interface.

Thus I suggest to add below.


From 99cd1e170a30ea81164fd13333a5e5bb9587e4e8 Mon Sep 17 00:00:00 2001
From: Andreas Herrmann <andreas.herrmann3(a)>
Date: Thu, 12 Nov 2009 16:08:38 +0100
Subject: [PATCH] x86, ucode-amd: Provide it only if microcode is compiled as module

microcode_amd.c supports only the firmware interface. Thus it depends
on the udev firmware helper. As we won't compile the micorode patches
into the kernel it also doesn't make sense to compile microcode_amd.c
into kernel.

This also ensures that loading an updated AMD microcode patch
container file is always possible via

Signed-off-by: Andreas Herrmann <andreas.herrmann3(a)>
arch/x86/Kconfig | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 17abcfa..0559ca3 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -959,7 +959,7 @@ config MICROCODE_INTEL

bool "AMD microcode patch loading support"
- depends on MICROCODE
+ depends on MICROCODE=m
select FW_LOADER
If you select this option, microcode patch loading support for AMD

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)
More majordomo info at
Please read the FAQ at
From: Borislav Petkov on
On Thu, Nov 12, 2009 at 04:48:34PM +0100, Dmitry Adamushko wrote:
> request_firmware_nowait() sends an async request which can be
> preserved (and this is an assumption -- I haven't really verified it
> yet) until some latter stage when user-space has been started and is
> capable of handling (cached) firmware-load requests. I may be (and
> perhaps I'm) wrong with the above assumption and the solution is
> either never build such a module into the kernel or only do it with
> built-in firmware blobs.

I don't think built-in blobs is the way we want to go here - in that
case updating the microcode would require rebuilding the kernel, which
is a clear overkill and exactly the opposite of what we should be doing.
Imagine a big supercomputer consisting of several thousand nodes, all
with identical CPUs. Now, everytime there's microcode patch available,
you have to reboot all those machines after having distributed the
updated kernel images just so that all nodes have their microcode
updated. Many admins would go: "Hmm, no!"

What actually got somehow dropped from Andreas' patch and which we
talked about and agreed upon earlier is that the best thing to do would
be to do

$ rmmod microcode
$ modprobe microcode

after having copied the new ucode patch to /lib/firmware without
disturbing the machine execution.

The async _nowait() version sounds good but in that case you're still
going to need to trigger the microcode update somehow (and AFAIK there's
no mechanism for that yet.) So reloading the module is the easiest thing
and it doesn't need any code changes except the Kconfig oneliner.



Operating | Advanced Micro Devices GmbH
System | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. M�nchen, Germany
Research | Gesch�ftsf�hrer: Andrew Bowd, Thomas M. McCoy, Giuliano Meroni
Center | Sitz: Dornach, Gemeinde Aschheim, Landkreis M�nchen
(OSRC) | Registergericht M�nchen, HRB Nr. 43632

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)
More majordomo info at
Please read the FAQ at