Re: [PATCH v2 1/1] arm64: dts: marvell: armada-ap806: reserve PSCI area

[Gregory CLEMENT <gregory.clement@bootlin.com>]

Hi Emmanuel,
 
 On jeu., janv. 10 2019, Emmanuel Vadot <manu@bidouilliste.com> wrote:

>  Hello all,
>
>  Sorry I forgot to answer this email.
>
> On Wed, 26 Dec 2018 18:09:04 +0100
> Heinrich Schuchardt <xypron.glpk@gmx.de> wrote:
>
>> On 12/26/18 5:16 PM, Gregory CLEMENT wrote:
>> > Hi Heinrich,
>> >  
>> >  On ven., déc. 21 2018, Heinrich Schuchardt <xypron.glpk@gmx.de> wrote:
>> > 
>> >> The memory area [0x4000000-0x4200000[ is occupied by the PSCI firmware. Any
>> >> attempt to access it from Linux leads to an immediate crash.
>> >>
>> >> So let's make the same memory reservation as the vendor kernel.
>> >>
>> >> Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
>> > 
>> > We got a similar patch one month ago and Russell King pointed that it
>> > didn't match waht he saw on his MACCHIATObin:
>> > http://lists.infradead.org/pipermail/linux-arm-kernel/2018-November/612360.html
>> > 
>> > on mine under U-Boot I got:
>> > 
>> > Marvell>> md 0x4000000
>> > 04000000: 00000000 00000007 00000005 00000040    ............@...
>> > 04000010: 00000001 00001000 00000007 00000001    ................
>> > 04000020: 00000008 00000000 00000009 00000000    ................
>> > 04000030: 0000000a 00000000 ffffffff ffffffff    ................
>> > 04000040: ffffffff ffffffff ffffffff ffffffff    ................
>> > 04000050: ffffffff ffffffff ffffffff ffffffff    ................
>> > 04000060: ffffffff ffffffff ffffffff ffffffff    ................
>> > 04000070: ffffffff ffffffff ffffffff ffffffff    ................
>> > 04000080: ffffffff ffffffff ffffffff ffffffff    ................
>> > 04000090: ffffffff ffffffff ffffffff ffffffff    ................
>> > 040000a0: ffffffff ffffffff ffffffff ffffffff    ................
>> > 040000b0: ffffffff ffffffff ffffffff ffffffff    ................
>> > 040000c0: ffffffff ffffffff ffffffff ffffffff    ................
>> > 040000d0: ffffffff ffffffff ffffffff ffffffff    ................
>> > 040000e0: ffffffff ffffffff ffffffff ffffffff    ................
>> > 040000f0: ffffffff ffffffff ffffffff ffffffff    ................
>> > 
>> > In my case I have an old ATF, however this kind of setting should be
>> > done by the bootloader.
>> > 
>> > I am interested by your claim about causing an immediate crash when
>> > accessing this region. How did you trigger it?
>> > 
>> > Gregory
>> 
>> Hello Gregory,
>> 
>> judging from you U-Boot prompt you are not using mainline U-Boot. With
>> the legacy Marvell U-Boot I get the same output as you provide above.
>> But you are looking in the wrong place to confirm that this is ATF:
>> 
>> Marvell>> md 04017200
>> 04017200: 616d6920 74206567 5043206f 474d2031     image to CP1 MG
>> 04017210: 746f6e20 70757320 74726f70 000a6465     not supported..
>> 04017220: 4f525245 20203a52 50435320 324c425f    ERROR:   SCP_BL2
>> 04017230: 6f727720 6920676e 6620676d 616d726f     wrong img forma
>> 04017240: 63282074 745f336d 3d657079 0a296425    t (cm3_type=%d).
>> 04017250: 544f4e00 3a454349 43532020 6d492050    .NOTICE:  SCP Im
>> 04017260: 20656761 73656f64 2074276e 746e6f63    age doesn't cont
>> 04017270: 206e6961 66204d50 776d7269 0a657261    ain PM firmware.
>> 04017280: 00000000 00000000 040144d0 00000000    .........D......
>> 04017290: 00000001 00000019 4f525245 20203a52    ........ERROR:
>> 040172a0: 6e695720 20776f64 203a6425 65736162     Window %d: base
>> 040172b0: 64646120 73736572 616e7520 6e67696c     address unalign
>> 040172c0: 74206465 7830206f 000a7825 67696c41    ed to 0x%x..Alig
>> 040172d0: 7075206e 65687420 73616220 64612065    n up the base ad
>> 040172e0: 73657264 6f742073 25783020 000a786c    dress to 0x%lx..
>> 040172f0: 4f525245 20203a52 6e695720 20776f64    ERROR:   Window
>
>  I do confirm what Heinrich is saying, this is needed (and made for)
> mainline U-Boot.
>
>> Possibly the legacy Marvell U-Boot is running at exception level 3
>> (EL3). This would explain why you cannot reproduce the crash with that
>> U-Boot.
>> 
>> The legacy Marvell U-Boot is in bad shape: their booti command cannot
>> load the initrd.img created by Debian. This is why I use mainline
>> U-Boot. Here you can find my recipe:
>
>  I personnaly uses mainline U-Boot because the efi part is too old and
> buggy on Marvell fork.
>
>> https://github.com/xypron/u-boot-build/blob/macchiatobin/Makefile (for
>> U-Boot v2018.11)
>> 
>> When I issue the md command for the region in mainline U-Boot  I get a
>> crash:
>> 
>> => md 0x4000000
>> 
>> 
>> 04000000:"Synchronous Abort" handler, esr 0x96000210
>> 
>> 
>> elr: 000000000006b878 lr : 000000000006b7f4 (reloc)
>> 
>> 
>> elr: 000000007ff8b878 lr : 000000007ff8b7f4
>> 
>> 
>> x0 : 0000000000000009 x1 : 0000000000000000
>> 
>> 
>> x2 : 000000000000003a x3 : 0000000004000000
>> 
>> 
>> x4 : 0000000000000000 x5 : 000000007ffae306
>> 
>> 
>> x6 : 0000000000000004 x7 : 000000007fb0a1e0
>> 
>> 
>> x8 : 000000007fb0a0a0 x9 : 0000000000000008
>> 
>> 
>> x10: 00000000ffffffd0 x11: 0000000000000006
>> 
>> 
>> x12: 000000000001869f x13: 000000000000440c
>> 
>> 
>> x14: 000000007fb0a43c x15: 0000000000000008
>> 
>> 
>> x16: 000000007ffb49ea x17: 000000007ffb49ea
>> 
>> 
>> x18: 000000007fb0fdd8 x19: 0000000000000040
>> 
>> 
>> x20: 0000000004000000 x21: 0000000004000000
>> 
>> 
>> x22: 000000007ffacff1 x23: 0000000000000008
>> 
>> 
>> x24: 0000000000000004 x25: 0000000000000004
>> 
>> 
>> x26: 0000000000000004 x27: 000000007fb0a268
>> 
>> 
>> x28: 0000000000000000 x29: 000000007fb0a1e0
>> 
>> 
>> 
>> Resetting CPU ...
>> 
>> In Linux 4.20 without my patch I regularly experience crashes like the
>> one below. With the patch I never experience such a crash.
>
>  For me this is 100% reproducible with the FreeBSD kernel, not the same
> kind of crash but a hang as soon as we try to call PSCI/SMCCC functions.
>
>  I don't care personnaly which patch is taken as long as it's applied.

Thanks for your confirmation I will apply it.

Gregory

>
>  Cheers,
>
> -- 
> Emmanuel Vadot <manu@bidouilliste.com> <manu@freebsd.org>

-- 
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com