arm/patch: call flush_icache ASAP after writing new instruction.
PATCH 1/1 arm/patch: call flush_icache ASAP after writing new instruction.
The patch moves flush_icache before the patch_unmap call. The change avoids the possibility of the CPU seeing partially-patched instructions if a function from patch_unmap call tree has just been patched. That HAS been observed in practice, leading to kernel panic or freezing in early boot:
https://bugzilla.kernel.org/show_bug.cgi?id=219089
Specifically, the patch_unmap invokes _raw_spin_unlock_irqrestore (could be non-inlined) and that function is being patched during the ftrace_init, so the original code would run the patched code BEFORE flushing the icache.
Note, some arches are more careful about flushing icache early. E.g. arch/riscv/kernel/patch.c:
__patch_insn_set and __patch_insn_write call the local_flush_icache_range before the patch_unmap and have an explicit comment about this.arch/x86/kernel/alternative.c:
text_poke_early calls sync_core (flushes icache) before local_irq_restore.
--- arch/arm/kernel/patch.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
arch/arm/kernel/patch.c +9 -3
@@ -99,13 +99,19 @@ void __kprobes __patch_text_real(void *addr, unsigned int insn, bool remap) size =sizeof (u32); } -if (waddr != addr) { +if (waddr != addr)flush_kernel_vmap_range (waddr, twopage ? size /2 : size); -patch_unmap (FIX_TEXT_POKE0 , &flags); - }flush_icache_range ((uintptr_t )(addr), (uintptr_t )(addr) + size); + +/* Can only call 'patch_unmap' after flushing dcache and icache, +* because it calls 'raw_spin_unlock_irqrestore', but that may +* happen to be the very function we're currently patching +* (as it happens during the ftrace init). +*/ +if (waddr != addr) +patch_unmap (FIX_TEXT_POKE0 , &flags); }static int __kprobes patch_text_stop_machine (void *data)
Signature · 1 line
2.34.1