x86: if we cannot calibrate the TSC, we panic.
authorRusty Russell <rusty@rustcorp.com.au>
Tue, 4 Mar 2008 12:07:50 +0000 (23:07 +1100)
committerIngo Molnar <mingo@elte.hu>
Thu, 17 Apr 2008 15:40:52 +0000 (17:40 +0200)
The current tsc_init() clears the TSC feature bit if the TSC khz
cannot be calculated, causing us to panic in
arch/x86/kernel/cpu/bugs.c check_config().  We should simply mark it
unstable.

Frankly, someone should take an axe to this code.  mark_tsc_unstable()
not only marks it unstable, but sets tsc_enabled to 0, which seems
redundant but is actually important here because means it won't be
used by sched_clock() either.  Perhaps a tristate enum "UNUSABLE,
UNSTABLE, OK" would be clearer, and separate mark_tsc_unstable() and
mark_tsc_broken() functions?

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/kernel/tsc_32.c

index c2241e0..68657d8 100644 (file)
@@ -392,13 +392,15 @@ void __init tsc_init(void)
        int cpu;
 
        if (!cpu_has_tsc)
-               goto out_no_tsc;
+               return;
 
        cpu_khz = calculate_cpu_khz();
        tsc_khz = cpu_khz;
 
-       if (!cpu_khz)
-               goto out_no_tsc;
+       if (!cpu_khz) {
+               mark_tsc_unstable("could not calculate TSC khz");
+               return;
+       }
 
        printk("Detected %lu.%03lu MHz processor.\n",
                                (unsigned long)cpu_khz / 1000,
@@ -431,9 +433,4 @@ void __init tsc_init(void)
                tsc_enabled = 1;
 
        clocksource_register(&clocksource_tsc);
-
-       return;
-
-out_no_tsc:
-       setup_clear_cpu_cap(X86_FEATURE_TSC);
 }