[coreboot] CBMEM Console corruption with fsp_baytrail in coreboot 4.2
gaumless at gmail.com
Fri Nov 20 01:17:16 CET 2015
Hi Ben, I'm still trying to get to this. I don't know if you saw how
different the fsp cbmem route is from other platforms. The fsp copies the
contents of the car stack to the top of memory when it tears down the cache
as ram. The pointer to that gets saved and passed to the cbmem transfer
code. Something has obviously happened to this sequence. I plan on trying
to bisect the change. The latest I'll be able to get to that is this
On Thu, Nov 19, 2015 at 17:07 Ben Gardner <gardner.ben at gmail.com> wrote:
> I've narrowed down where the CBMEM console is getting corrupted and
> found a work around that gets it working again.
> It is getting corrupted in the FSP Early Init function. So in the
> Intel blob, not coreboot.
> I added logs to cbmemc_init() and cbmrmc_reinit() that show the
> console pointer, size, cursor and the first 64 bytes of the console
> (including size and console).
> Right before the call to FspInitApi() in fsp_early_init():
> fsp_early_init: cbm_cons_p: fef00000 size=3064 cur=2606
> fef00000: f8 0b 00 00 7b 0a 00 00 63 62 6d 65 6d 63 5f 69 ........cbmemc_i
> fef00010: 6e 69 74 3a 20 63 62 6d 5f 63 6f 6e 73 5f 70 3a nit: cbm_cons_p:
> fef00020: 20 66 65 66 30 30 30 30 30 20 73 69 7a 65 3d 33 fef00000 size=3
> fef00030: 30 36 34 20 63 75 72 3d 30 0a 66 65 66 30 30 30 064 cur=0.fef000
> And this is at the top of romstage_main_continue():
> romstage_main_continue: cbm_cons_p: 00000b96 size=2052398204 cur=1024945500
> 00000b96: 7c 1c 55 7a bd 6d 17 3d 17 a9 76 69 ee 07 d1 b8 |.Uz.m.=..vi....
> 00000ba6: fa 6b 75 85 ee 7a dd 8f e9 48 59 93 b4 f8 0a ce .ku..z...HY.....
> 00000bb6: 1c dc fe a4 09 d6 ba 1d 88 8f 23 e4 40 cc 88 cd ..........#. at ...
> 00000bc6: 90 91 53 e4 71 6d 90 1d bd e9 fa 13 a8 16 fd b5 ..S.qm..........
> So, the FSP is trashing the CAR variables.
> I looked at the changes to src/arch/x86/car.ld and see that the
> CONSOLE line was moved before _car_data_start.
> That move appears to have been in commit dd6fa93d by Aaron Durbin.
> When I moved that CONSOLE line back to after _car_data_end, the
> console works again.
> That also fixes the timestamp table infinite loop issue.
> So, I have a work-around.
> However, I suspect that the root issue of the FSP trashing CAR
> variables persists.
> My initial email about the timestamp issue is here:
> I'm not sure what to do about this.
> The above change seems reasonable and needed for verstage support, so
> sending my 'fix' upstream probably isn't viable.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the coreboot