x86/delay: Do not use cpu_tss in preemptible ctxt in delay_mwaitx()

From: Borislav Petkov <[email protected]>

So Andy had a good idea about using a cacheline-aligned, seldomly used per-cpu var as the MONITORX target but we can't use it in preemptible context. The first simple idea I have is to disable preemption around us dereffing it.

Better ideas?

---
 arch/x86/lib/delay.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

index e912b2f6d36e..c1810afcd2ea 100644
@@ -92,17 +92,22 @@ static void delay_tsc(unsigned long __loops)
 static void delay_mwaitx(unsigned long __loops)
 {
 	u64 start, end, delay, loops = __loops;
+	struct tss_struct *t;
+
+	/*
+	 * Use cpu_tss as a cacheline-aligned, seldomly accessed per-cpu
+	 * variable as the monitor target.
+	 */
+	preempt_disable();
+	t = this_cpu_ptr(&cpu_tss);
+	preempt_enable();
 
 	start = rdtsc_ordered();
 
 	for (;;) {
 		delay = min_t(u64, MWAITX_MAX_LOOPS, loops);
 
-		/*
-		 * Use cpu_tss as a cacheline-aligned, seldomly
-		 * accessed per-cpu variable as the monitor target.
-		 */
-		__monitorx(this_cpu_ptr(&cpu_tss), 0, 0);
+		__monitorx(t, 0, 0);
 
 		/*
 		 * AMD, like Intel, supports the EAX hint and EAX=0xf

2.3.5

What's the actual problem? Is it the preempt warnings and, if so, would raw_cpu_ptr fix it?

It may pay to move it into the loop, though.

--Andy

Yeah, it is the warning:

[    1.565876] BUG: using smp_processor_id() in preemptible [00000000] code: udevd/312
[    1.566123] caller is delay_mwaitx+0x40/0xa0

and yes, I think so. I don't think we care about being in preemptible context since we're going idle anyway and doesn't matter which cpu_tss we touch.

Yeah, I'll use raw_cpu_ptr...

... and won't need to do that.

Thanks for the idea.

Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.

We do use this_cpu_ptr(&cpu_tss) as a cacheline-aligned, seldomly accessed per-cpu var as the MONITORX target in delay_mwaitx(). However, when called in preemptible context, this_cpu_ptr -> smp_processor_id() -> debug_smp_processor_id() fires:

  BUG: using smp_processor_id() in preemptible [00000000] code: udevd/312
  caller is delay_mwaitx+0x40/0xa0

But we don't care about that check - we only need cpu_tss as a MONITORX target and it doesn't really matter which CPU's var we're touching as we're going idle anyway. Fix that.

---
 arch/x86/lib/delay.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index e912b2f6d36e..2f07c291dcc8 100644
@@ -102,7 +102,7 @@ static void delay_mwaitx(unsigned long __loops)
 		 * Use cpu_tss as a cacheline-aligned, seldomly
 		 * accessed per-cpu variable as the monitor target.
 		 */
-		__monitorx(this_cpu_ptr(&cpu_tss), 0, 0);
+		__monitorx(raw_cpu_ptr(&cpu_tss), 0, 0);
 
 		/*
 		 * AMD, like Intel, supports the EAX hint and EAX=0xf

2.3.5


-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.