Re: [PATCH -tip v3 1/2] kcov: Make runtime functions noinstr-compatible

[Peter Zijlstra]

On Mon, Jun 15, 2020 at 09:53:06AM +0200, Marco Elver wrote:
> 
> Disabling KCOV for smp_processor_id now moves the crash elsewhere. In
> the case of KASAN into its 'memcpy' wrapper, called after
> __this_cpu_read in fixup_bad_iret. This is making me suspicious,
> because it shouldn't be called from the noinstr functions.

With your .config, objtool complains about exactly that though:

vmlinux.o: warning: objtool: fixup_bad_iret()+0x8e: call to memcpy() leaves .noinstr.text section

The utterly gruesome thing below 'cures' that.

> For KCSAN the crash still happens in check_preemption_disabled, in the
> inlined native_save_fl function (apparently on its 'pushf'). If I turn
> fixup_bad_iret's __this_cpu_read into a raw_cpu_read (to bypass
> check_preemption_disabled), no more crash with KCSAN.

vmlinux.o: warning: objtool: debug_smp_processor_id()+0x0: call to __sanitizer_cov_trace_pc() leaves .noinstr.text section
vmlinux.o: warning: objtool: check_preemption_disabled()+0x1f: call to __sanitizer_cov_trace_pc() leaves .noinstr.text section
vmlinux.o: warning: objtool: __this_cpu_preempt_check()+0x4: call to __sanitizer_cov_trace_pc() leaves .noinstr.text section

That could be either of those I suppose, did you have the NOP patches
on? Let me try... those seem to placate objtool at least.

I do see a fair amount of __kasan_check*() crud though:

vmlinux.o: warning: objtool: rcu_nmi_exit()+0x44: call to __kasan_check_read() leaves .noinstr.text section
vmlinux.o: warning: objtool: rcu_dynticks_eqs_enter()+0x1c: call to __kasan_check_write() leaves .noinstr.text section
vmlinux.o: warning: objtool: rcu_nmi_enter()+0x46: call to __kasan_check_read() leaves .noinstr.text section
vmlinux.o: warning: objtool: rcu_dynticks_eqs_exit()+0x21: call to __kasan_check_write() leaves .noinstr.text section
vmlinux.o: warning: objtool: __rcu_is_watching()+0x1c: call to __kasan_check_read() leaves .noinstr.text section
vmlinux.o: warning: objtool: debug_locks_off()+0x1b: call to __kasan_check_write() leaves .noinstr.text section

That wasn't supported to happen with the __no_sanitize patches on (which
I didn't forget). Aah, I think we've lost a bunch of patches.. /me goes
rummage.

This:

  https://lkml.kernel.org/r/20200603114051.896465666@xxxxxxxxxxxxx

that cures the rcu part of that.

Let me go look at your KCSAN thing now...

---
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index af75109485c26..031a21fb5a741 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -675,6 +675,14 @@ struct bad_iret_stack {
 	struct pt_regs regs;
 };
 
+void __always_inline __badcpy(void *dst, void *src, int nr)
+{
+	unsigned long *d = dst, *s = src;
+	nr /= sizeof(unsigned long);
+	while (nr--)
+		*(d++) = *(s++);
+}
+
 asmlinkage __visible noinstr
 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
 {
@@ -690,13 +698,13 @@ struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
 		(struct bad_iret_stack *)__this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
 
 	/* Copy the IRET target to the temporary storage. */
-	memcpy(&tmp.regs.ip, (void *)s->regs.sp, 5*8);
+	__badcpy(&tmp.regs.ip, (void *)s->regs.sp, 5*8);
 
 	/* Copy the remainder of the stack from the current stack. */
-	memcpy(&tmp, s, offsetof(struct bad_iret_stack, regs.ip));
+	__badcpy(&tmp, s, offsetof(struct bad_iret_stack, regs.ip));
 
 	/* Update the entry stack */
-	memcpy(new_stack, &tmp, sizeof(tmp));
+	__badcpy(new_stack, &tmp, sizeof(tmp));
 
 	BUG_ON(!user_mode(&new_stack->regs));
 	return new_stack;