diff options
| author | Paul Mackerras <paulus@samba.org> | 2006-10-10 13:51:00 +1000 |
|---|---|---|
| committer | Paul Mackerras <paulus@samba.org> | 2006-10-10 13:51:00 +1000 |
| commit | d6a29252ad120457cd544d44b2fbea46a00734a7 (patch) | |
| tree | 7dcc33d03f0e1caf48bc749060d78d80946099c4 /arch | |
| parent | 39e3eb7265b8698e5f607a317af13c9478274736 (diff) | |
[POWERPC] Fix secondary CPU startup on old "powersurge" SMP powermacs
On the old "powersurge" SMP powermacs, the second CPU is started up
by sending it an IPI, which has the side effect of stopping the
timebase clock (so the secondary CPU's timebase can be synchronized
with the primary's). The routine that did this used udelay, which
will hang forever when the timebase is stopped, since udelay now spins
until the timebase reaches a certain value.
The end result is that the kernel would hang when bringing up the
second CPU. This fixes it by using a simple loop which just does a
fixed number of iterations to generate the delay. These old systems
were all clocked at around 200 MHz or so, so a fixed number of
iterations is acceptable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Diffstat (limited to 'arch')
| -rw-r--r-- | arch/powerpc/platforms/powermac/smp.c | 7 |
1 files changed, 6 insertions, 1 deletions
diff --git a/arch/powerpc/platforms/powermac/smp.c b/arch/powerpc/platforms/powermac/smp.c index 574cd205b302..eeb2ae5ffc58 100644 --- a/arch/powerpc/platforms/powermac/smp.c +++ b/arch/powerpc/platforms/powermac/smp.c @@ -328,6 +328,7 @@ static void __init smp_psurge_kick_cpu(int nr) { unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8; unsigned long a; + int i; /* may need to flush here if secondary bats aren't setup */ for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32) @@ -340,7 +341,11 @@ static void __init smp_psurge_kick_cpu(int nr) mb(); psurge_set_ipi(nr); - udelay(10); + /* + * We can't use udelay here because the timebase is now frozen. + */ + for (i = 0; i < 2000; ++i) + barrier(); psurge_clr_ipi(nr); if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354); |