Skip to contentSkip to conversation

Machine check when executing ptc.ga

DISCUSSION Zoltan Menyhart 2 participants 2 messages Nov 13, 2003

lore
Share & tools
lore
#0 lore raw

I've got a machine check when executing ptc.ga inside ia64_global_tlb_purge() on an Intel Tiger box. The CPUs are:

family : Itanium 2
model : 1
revision : 5

The failing instruction is:

       ptc.ga r32,r34;;

and r32 = a0000000 00000000, r34 = 00000000 00000040.

The Intel SDM Vol 3, page 3:200 says about the ptc.* instructions:
"Interruptions: Machine Check abort: Register NaT Consumption fault"

I checked with an ITP just before executing ptc.ga, the NaT bits for r32 and r34 are 0.

What is the reason of this machine check?
Is there something non documented problem with ptc.ga ?

(I'm on a 2.4.20, but a ptc.ga should be the same...) Thank you in advance.

Zoltan Menyhart

#1 lore raw
Quoted 5 lines

The failing instruction is:

       ptc.ga r32,r34;;

and r32 = a0000000 00000000, r34 = 00000000 00000040.

You are flushing 64k bytes at 0xa000000000000000. A default pagesize of 16k would mean that this range includes the per-cpu area (at 0xa000000000000000 + 2*PAGE_SIZE on 2.4.20). The per-cpu area is mapped by an entry pinned into the TLB by DTR[1] ... so the ptc chokes beacause it can only flush TLB cache entries.

-Tony