On 08/19/2011 12:13 PM, Guenter Roeck wrote:
On Fri, 2011-08-19 at 14:44 -0400, David Daney wrote:Current Octeon systems do in fact have holes in their memory zones. We need to select HOLES_IN_ZONE. If we do not, some memory configurations will result in crashes at boot time like this: . . . CPU 6 Unable to handle kernel paging request at virtual address 0000000000700000, epc == ffffffff8118fe00, ra == ffffffff8118fe9c Oops[#1]: Cpu 6 . . . ... Call Trace: [<ffffffff8118fe00>] setup_per_zone_wmarks+0x1b0/0x338 [<ffffffff815cd738>] init_per_zone_wmark_min+0x64/0xd0 [<ffffffff81100438>] do_one_initcall+0x38/0x160 . . . Reported-by: Jason Kwon<jason.kwon@xxxxxxxxxxxx> Signed-off-by: David Daney<david.daney@xxxxxxxxxx> Cc: Jason Kwon<jason.kwon@xxxxxxxxxxxx> --- Jason, can you test this patch? Ralf, if Jason reports that it fixes his problem, it probably is needed for 3.0 and 3.1.Your patch fixes the problem for the board with CN38xx and 2GB RAM that crashed previously. Tested-by: Guenter Roeck<guenter.roeck@xxxxxxxxxxxx> Thanks a lot for looking into this. Guenter
I applied the patch to 3.0.3 and was able to boot the CN58XX system without any memory restrictions. The same patched kernel booted on CN38XX ran into a different problem, which I'm looking into.
Thanks, Jason
