Kexec on arm64

[Arun Chandran]

On Wed, Aug 6, 2014 at 7:24 PM, Arun Chandran <achandran at mvista.com> wrote:
> Hi,
>
> On Mon, Aug 4, 2014 at 10:51 PM, Geoff Levand <geoff at infradead.org> wrote:
>> Hi Arun,
>>
>> On Mon, 2014-08-04 at 15:46 +0530, Arun Chandran wrote:
>>> The latest kexec code is working fine in LE/BE mode in UP configuration.
>>>
>>> I had to change kexec-tools code a bit to make sure that "kexec -l"
>>> is not putting dtb at an area where kernel is building its initial page
>>> tables.
>>
>> OK, I'll add in code to handle this.
>>
>>> Now before trying SMP configuration I want to know whether the below "kexec -e"
>>> scenarios are valid(required)?
>>>
>>> 1st stage                                  |               2nd stage
>>> ---------------------------------------------------------------------------------------------
>>> LE UP                                      |               BE UP
>>> LE UP                                      |               BE SMP
>>> LE UP                                      |               LE SMP
>>> LE SMP                                   |               LE UP
>>> LE SMP                                   |               BE UP
>>> LE SMP                                   |               BE SMP
>>
> I am testing "kexec -e" with the script below(it will either
> reboot the board to BE or LE mode randomly).
>
> This test is done without L3 cache. With L3 I have more troubles.
> I think it is best to fix the code without L3 then move
> to testing with L3. That fix may solve the problem with L3
>
> ###############################
> :~$ cat /etc/init.d/S50reboot
> #!/bin/sh
>
> sleep 5
> i=$RANDOM
> j=$(( $i % 2))
>
> if [ $j -eq 0 ] ; then
>         mount /dev/mmcblk0p1 /mnt
>
>         count=`cat /mnt/cnt`
>         echo "KEXEC rebootng to BE count = $count"
>         echo $RANDOM > /mnt/"$count""_BE"
>
>         count=$(( $count + 1 ))
>         echo "$count">/mnt/cnt
>
>         kexec -l /mnt/vmlinux_BE.strip
> --command-line="console=ttyS0,115200 earlyprintk=uart8
> 250-32bit,0x1c020000 debug swiotlb=65536 log_buf_len=2M"
>         umount /mnt
>         kexec -e
> else
>         mount /dev/mmcblk0p1 /mnt
>
>         count=`cat /mnt/cnt`
>         echo "KEXEC rebooting to LE count = $count"
>         echo $RANDOM > /mnt/"$count""_LE"
>
>         count=$(( $count + 1 ))
>         echo "$count">/mnt/cnt
>
>         kexec -l /mnt/vmlinux_LE.strip
> --command-line="console=ttyS0,115200 earlyprintk=uart8
> 250-32bit,0x1c020000 debug swiotlb=65536 log_buf_len=2M"
>         umount /mnt
>         kexec -e
> fi
>
> exit $?
> #############################
>

Some more information:

I am able to run the same test (without L3) for 1.5 hours in case of
no endian switching.

For this testing I used the latest code @
https://git.linaro.org/people/geoff.levand/linux-kexec.git
unlike in the endian switching scenario there is no code change
made for this test.

It ran 370 times for LE-->LE reboots.
It ran 351 times for BE-->BE reboots.

I had to manually stop both tests as it was continuing
without any problem.

I will repeat the same test with L3 cache on

--Arun