Skip to contentSkip to conversation

latent problem in free_init_pages()

DISCUSSION Jan Beulich 2 participants 2 messages Mar 19, 2010

lore
Share & tools
lore
#0 lore raw

The calls to set_memory_rw() and set_memory_np() that both of you added for 2.6.25 imply that both "begin" and "end" are page aligned. This isn't the case, however, for the initrd and SMP locks. While at present this doesn't appear to do any harm, it partly subverts the debugging effect of the latter change and would result in a real issue if the .smp_locks section was moved into the read-only portion of the kernel image (where it really belongs imo).

Jan

#1 lore raw
Quoted 7 lines · Jan Beulich

The calls to set_memory_rw() and set_memory_np() that both of you added for 2.6.25 imply that both "begin" and "end" are page aligned. This isn't the case, however, for the initrd and SMP locks. While at present this doesn't appear to do any harm, it partly subverts the debugging effect of the latter change and would result in a real issue if the .smp_locks section was moved into the read-only portion of the kernel image (where it really belongs imo).

if you change the linker script to move the section, you should also move it to keep the end page aligned.

solving that at link time (like is done now) is obviously nicer than solving it at runtime ;-)