[U-Boot] am3517: segmentation fault in Linux after upgrading from arago src to mainline u-boot

[Måns Rullgård]

Yegor Yefremov <yegorslists at googlemail.com> writes:

> On Fri, Dec 6, 2013 at 8:28 AM, Yegor Yefremov
> <yegorslists at googlemail.com> wrote:
>> On Thu, Dec 5, 2013 at 6:28 PM, Måns Rullgård <mans at mansr.com> wrote:
>>> Yegor Yefremov <yegorslists at googlemail.com> writes:
>>>
>>>> On Thu, Nov 28, 2013 at 5:57 PM, Tom Rini <trini at ti.com> wrote:
>>>>> On Thu, Nov 28, 2013 at 05:39:09PM +0100, Yegor Yefremov wrote:
>>>>>
>>>>>> I'm having problems with the new u-boot (November 2013) and am3517
>>>>>> SoC. When I start Debian 7.0 armhf I get lots of Segmentation faults,
>>>>>> I can't even perform apt-get udpate. With the older u-boot (arago
>>>>>> sources from 2010) everything seems to work properly.
>>>>>>
>>>>>> I've checked ATAG_MEM and both u-boot versions deliver same RAM start
>>>>>> address and size. I've compared EMIF4 settings
>>>>>> (arch/arm/cpu/armv7/omap3/emif4.c) and PINMUX and everything is the
>>>>>> same.
>>>>>>
>>>>>> Where else can I look for differences?
>>>>>
>>>>> Can you bisect the arago tree from what it's based on, to what the top
>>>>> of tree is there?  Some errata or another being patched up I guess..
>>>>
>>>> I've compared the x-loader's init routine with the one from SPL. An
>>>> found this workaround in the mainline U-Boot
>>>> (arch/arm/cpu/armv7/omap3/board.c):
>>>>
>>>> static void omap3_setup_aux_cr(void)
>>>> {
>>>>         /* Workaround for Cortex-A8 errata: #454179 #430973
>>>>          *      Set "IBE" bit
>>>>          *      Set "Disable Branch Size Mispredicts" bit
>>>>          * Workaround for erratum #621766
>>>>          *      Enable L1NEON bit
>>>>          * ACR |= (IBE | DBSM | L1NEON) => ACR |= 0xE0
>>>>          */
>>>>         omap3_update_aux_cr_secure(0xE0, 0);
>>>> }
>>>>
>>>> After commenting this routine and I don't see any segmentation faults
>>>> during apt usage.
>>>
>>> On an r1pX Cortex-A8 (as found in the 35xx chips) you absolutely want
>>> those bits set, especially if you're running Thumb2 code.  Otherwise you
>>> will get random crashes.
>>
>> Where can I find info, if am3517 is r1pX? I've looked through "AM35x
>> ARM Microprocessor: Technical Reference Manual", but found nothing
>> similar.
>
> This info was in the Errata.pdf:
>
> AM3517AZCN i.e. silicon revision 1.1 (r1p7 ROM 15:58)

All of those errata are present in r1p7.  The 454179 and 430973 also
need workarounds enabled in the kernel.  Enabling the workarounds on a
CPU without the problems will not break anything, but it will cause a
minor performance drop.

-- 
Måns Rullgård
mans at mansr.com