[RESEND][PATCH] ARM errata, 430973: update the affected revisions

[Tony Lindgren]

* Jeroen Hofstee <linux-arm at myspectrum.nl> [150411 02:26]:
> Hello Russell, +Tony,
> 
> Thanks for taking the time to respond,
> 
> On 11-04-15 09:52, Russell King - ARM Linux wrote:
> >On Fri, Apr 10, 2015 at 11:29:13PM +0200, Jeroen Hofstee wrote:
> >>On 25-02-15 20:36, Jeroen Hofstee wrote:
> >>>On 09-12-14 14:30, Jeroen Hofstee wrote:
> >>>>From: Jeroen Hofstee <linux-arm at myspectrum.nl>
> >>>>
> >>>>Update the list of revisions subject to this errata.
> >>>>
> >>>>Cc: Catalin Marinas <catalin.marinas at arm.com>
> >>>>Cc: Russell King <rmk+kernel at arm.linux.org.uk>
> >>>>Cc: Andreas Bießmann <andreas.devel at googlemail.com>
> >>>>Signed-off-by: Jeroen Hofstee <jhofstee at victronenergy.com>
> >>>>---
> >>>>I don't have access to the AT400/AT401/AT490 document, but
> >>>>Andreas was kind enough to provide this information, see
> >>>>https://www.mail-archive.com/u-boot@lists.denx.de/msg156620.html
> >>>>
> >>>>Resending from an address which is subscribed to the ML...
> >>>>---
> >>>>  arch/arm/Kconfig | 2 +-
> >>>>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>
> >>>>diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> >>>>index 89c4b5c..a2202fa 100644
> >>>>--- a/arch/arm/Kconfig
> >>>>+++ b/arch/arm/Kconfig
> >>>>@@ -1063,7 +1063,7 @@ config ARM_ERRATA_430973
> >>>>      depends on CPU_V7
> >>>>      help
> >>>>        This option enables the workaround for the 430973 Cortex-A8
> >>>>-      (r1p0..r1p2) erratum. If a code sequence containing an ARM/Thumb
> >>>>+      (r1p0..r1p3, r1p7) erratum. If a code sequence containing an
> >>>>ARM/Thumb
> >>>>        interworking branch is replaced with another code sequence at
> >>>>the
> >>>>        same virtual address, whether due to self-modifying code or
> >>>>virtual
> >>>>        to physical address re-mapping, Cortex-A8 does not recover from
> >>>>the
> >>>
> >No.  If you read the discussion that the OMAP people are having, it is
> >unclear whether r3p2 is actually affected by this errata or not.
> I scanned through it, but that is a different issue it seems. This patch
> only adds r1p3 and r1p7 to the documentation, since according to ARM
> they are affected by this errata. Newer revision should not be affected
> by this errata (or ARM reintroduced it).
> 
> The patch is _not_ adding r3p2 as an affected version. And is also not
> about how to deal with cores no longer needing this workaround.

Right.. But it seems that having the aux ctrl register bit enabled
without doing the flush btac part leads into a different set of
issues.
 
> >In the discussion with OMAP people, we've come up with a potentially
> >better solution to this, which is to rearrange the code so that only
> >Cortex-A8 executes this workaround, and the BTB flush is always
> >present (which Tony says fixes the problem.)
> 
> The am3517 / r1p7 is a Cortex-a8 and is affected by this errata. I fail to
> see why making this cortex-a8 specific will help anything about the revision
> not being mentioned in the help text. (unless the CONFIG option gets
> completely
> removed).
> >Meanwhile, OMAP people are seeing about updating uboot to set/clear
> >the auxiliary control register bit appropriately for the revision of
> >the core.
> Well almost appropriately. [1] suggest "the bootloaders can be fixed
> Cortex-A8
> revisions later than r1p2 to not set the IBE bit.". That should have been
> r1p7 as
> well.
> 
> >In any case, adding the patch and suggesting people enable this for
> >more Cortex-A8's (eg, non-OMAP) won't actually do anything in a
> >multiplatform kernel.
> >
> Can you give an example of a r1p7 not needing the workaround?

Do you have some pointer for the documentation mentioning r1p7 is
affected? It is possible that the revision range is wrong.. But I'm
more likely to believe it is correct and we're hitting a different
problem.

I'm hoping that with Russell's patch and my patch in the thread
below, you don't necessarily need to enable 430973. That's these
two patches:

http://www.spinics.net/lists/arm-kernel/msg411433.html
http://www.spinics.net/lists/arm-kernel/msg411038.html

If what Russell and I are saying is correct, with the above two
patches your system should behave properly with 430973 even if
bit 6 in the aux ctrl register is set (or unset) by the bootloader.

If r1p7 is behaves with bit 6 cleared and errata 430973 set, then
we know r1p7 is unaffected by 430973.

Regards,

Tony
 
> http://www.spinics.net/lists/arm-kernel/msg411297.html