If you absolutely hammer the shortcut script with multiple vm cores & a nop program, you can trigger a silent barf, perhaps a deadlock w/ the core nop was running on in the page fault handler (including the mitigation) and the other core
Makefile:
run_be: nop
taskset -c 0 shortcut.sh -be --- ./nop
Shell
for i in $(seq 1 5000); do make run_be; done
Ali thinks this is about an IPI during a tlb shootdown which is prepared on a user stack & inaccessible to the remote core (b/c it doesn't have the user pg table loaded).
If you absolutely hammer the shortcut script with multiple vm cores & a nop program, you can trigger a silent barf, perhaps a deadlock w/ the core nop was running on in the page fault handler (including the mitigation) and the other core
Makefile:
Shell
for i in $(seq 1 5000); do make run_be; doneAli thinks this is about an IPI during a tlb shootdown which is prepared on a user stack & inaccessible to the remote core (b/c it doesn't have the user pg table loaded).