MacBook users: possible fix for the SMP problem
Maxim Sobolev
sobomax at FreeBSD.org
Tue Nov 6 02:05:57 PST 2007
By the way, it seems that you don't need this patch for the amd64 kernel
- 1st gen MacBooks don't have 64-bit mode, while 2nd gens don't have
this problem. I've just checked - the stock 7.0-BETA2 AMd64 kernel boots
just fine into SMP mode on my new Core 2 Duo based MacBook.
-Maxim
Maxim Sobolev wrote:
> I've just tested it on my 1st gen MacBook Pro. Yes, indeed, it solves
> both problems - one with CPU frequency detection and another one with
> the AP startup. Thanks!
>
> -Maxim
>
> Rui Paulo wrote:
>> Hi,
>> I've been contacted by Marco Trillo and I think he has found the
>> source of the SMP problem.
>> The problem seems to rely on Intel ICH7. Basically we need to disable
>> the "LEGACY_USB" bit before we calibrate the clocks.
>> "LEGACY_USB", according to Marco (I don't have the ICH7 spec at hand),
>> "causes legacy USB circuit to generate SMIs".
>>
>> Please try the following patch:
>> --- sys/amd64/isa/clock.c.orig 2007-11-04 20:31:09.000000000 +0000
>> +++ sys/amd64/isa/clock.c 2007-11-04 20:34:59.000000000 +0000
>> @@ -577,6 +577,8 @@ startrtclock()
>> writertc(RTC_STATUSA, rtc_statusa);
>> writertc(RTC_STATUSB, RTCSB_24HR);
>>
>> + outl(0x430, inl(0x430) & ~0x8);
>> +
>> freq = calibrate_clocks();
>> #ifdef CLK_CALIBRATION_LOOP
>> if (bootverbose) {
>> --- sys/i386/isa/clock.c.orig 2007-11-04 20:34:03.000000000 +0000
>> +++ sys/i386/isa/clock.c 2007-11-04 20:34:30.000000000 +0000
>> @@ -621,6 +621,8 @@ startrtclock()
>> writertc(RTC_STATUSA, rtc_statusa);
>> writertc(RTC_STATUSB, RTCSB_24HR);
>>
>> + outl(0x430, inl(0x430) & ~0x8);
>> +
>> freq = calibrate_clocks();
>> #ifdef CLK_CALIBRATION_LOOP
>> if (bootverbose) {
>>
>>
>> This should probably fix two issues:
>> 1) The second core should start without any trick (e.g. key press)
>> 2) We should be able to run with HZ=1000 (the default) without any
>> problem. To check if this is indeed the case, try booting with HZ=1000
>> (loader.conf variable kern.hz) and check if your CPU clock shows up
>> correctly in the dmesg. After that, please also check if 'time sleep
>> 1' takes one second (not more and not less).
>>
>> Also, please test if there are any USB problems.
>>
>> Note: this is still a hack. I'm still thinking about a way to
>> correctly identify on which systems we need to apply this fix.
>>
>> Regards.
>>
>
> _______________________________________________
> freebsd-current at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"
>
More information about the freebsd-i386
mailing list