The time code for x86-64 in 2.5.70 isout of date and wildly unstable,
setting the clock to the year 1,115,117 (!) during boot about 60% of the
time. This subsequently causes other pieces of completely unrelated
userspace software to crash randomly for no obvious reason once the
system comes up.
I've forward-ported Vojtech's time code from 2.4, fixing some locking
along the way. The new code supports using the AMD8111 HPET for time
calculations. It also works stably with the PIT/TSC on every boot,
which is the source of the time problems in current 2.5.
Right now, the only known problem is with the fixup of jiffies if a
timer interrupt is lost, which I've hence turned off. There's
preliminary support for using HPET for the gettimeofday vsyscall, but
since vsyscalls are disabled on x86-64 for now, that's obviously
untested.
<b
arch/x86_64/Kconfig | 12 +
arch/x86_64/kernel/acpi/boot.c | 6
arch/x86_64/kernel/apic.c | 8 +
arch/x86_64/kernel/smpboot.c | 11 +
arch/x86_64/kernel/time.c | 253 ++++++++++++++++++++++++++++++++++----- arch/x86_64/kernel/vsyscall.c | 16 +-
arch/x86_64/vmlinux.lds.S | 6
include/asm-x86_64/fixmap.h | 2
include/asm-x86_64/mc146818rtc.h | 5
include/asm-x86_64/timex.h | 30 ++++
include/asm-x86_64/vsyscall.h | 18 +-
11 files changed, 315 insertions(+), 52 deletions(-)
On Wed, Jun 11, 2003 at 11:50:32AM -0700, Bryan O'Sullivan wrote:
> The time code for x86-64 in 2.5.70 isout of date and wildly unstable,
> setting the clock to the year 1,115,117 (!) during boot about 60% of the
> time. This subsequently causes other pieces of completely unrelated
> userspace software to crash randomly for no obvious reason once the
> system comes up.
Thanks for doing this work.
Does it only look this way or is your white space really broken?
> Right now, the only known problem is with the fixup of jiffies if a
> timer interrupt is lost, which I've hence turned off. There's
> preliminary support for using HPET for the gettimeofday vsyscall, but
> since vsyscalls are disabled on x86-64 for now, that's obviously
> untested.
What makes you think they are disabled? They are used for every 64bit
program that uses gettimeofday or time and enabled by default.
> +static inline void rdtscll_sync(unsigned long *tsc)
> +{
> + sync_core();
> + rdtscll(*tsc);
On UP the sync_core is not really needed, but more reliable. May be worth
it to stick into an #ifdef though.
>
> }
>
> + call++;
> +
What's that?
-Andi
On Wed, 2003-06-11 at 12:18, Andi Kleen wrote:
> Does it only look this way or is your white space really broken?
I generated the patch with diff -b because my original changes fixed up
a lot of broken whitespace, which somewhat obscured the functional
changes. I'll send the plain diff -u instead next time.
> On UP the sync_core is not really needed, but more reliable. May be worth
> it to stick into an #ifdef though.
OK.
> > }
> >
> > + call++;
> > +
>
> What's that?
Oops - a piece of instrumentation that got left behind from trying to
fix an SMP deadlock.
<b
On Wed, 2003-06-11 at 12:56, Bryan O'Sullivan wrote:
> I'll send the plain diff -u instead next time.
Here's a new version of the patch, with problems fixed, and whitespace
fixes included. Boots just fine.
<b
arch/x86_64/Kconfig | 12 +
arch/x86_64/kernel/acpi/boot.c | 6
arch/x86_64/kernel/apic.c | 84 +++++-----
arch/x86_64/kernel/smpboot.c | 11 +
arch/x86_64/kernel/time.c | 305 +++++++++++++++++++++++++++++++-------- arch/x86_64/kernel/vsyscall.c | 28 ++-
arch/x86_64/vmlinux.lds.S | 6
include/asm-x86_64/fixmap.h | 2
include/asm-x86_64/mc146818rtc.h | 5
include/asm-x86_64/timex.h | 30 +++
include/asm-x86_64/vsyscall.h | 18 +-
11 files changed, 383 insertions(+), 124 deletions(-)
On Wed, Jun 11, 2003 at 11:50:32AM -0700, Bryan O'Sullivan wrote:
> The time code for x86-64 in 2.5.70 isout of date and wildly unstable,
> setting the clock to the year 1,115,117 (!) during boot about 60% of the
> time. This subsequently causes other pieces of completely unrelated
> userspace software to crash randomly for no obvious reason once the
> system comes up.
>
> I've forward-ported Vojtech's time code from 2.4, fixing some locking
> along the way. The new code supports using the AMD8111 HPET for time
> calculations. It also works stably with the PIT/TSC on every boot,
> which is the source of the time problems in current 2.5.
Thanks a lot for your work!
> Right now, the only known problem is with the fixup of jiffies if a
> timer interrupt is lost, which I've hence turned off. There's
> preliminary support for using HPET for the gettimeofday vsyscall, but
> since vsyscalls are disabled on x86-64 for now, that's obviously
> untested.
This is interesting. I'll have to test why the number of lost tick is
computed incorrectly ....
> --- linux-2.5/arch/x86_64/Kconfig 2003-06-11 11:43:09.000000000 -0700
> +++ x86-64-2.5/arch/x86_64/Kconfig 2003-06-11 10:39:45.000000000 -0700
> @@ -52,6 +52,18 @@
> klogd/syslogd or the X server. You should normally N here, unless
> you want to debug such a crash.
>
> +config HPET_TIMER
> + bool
> + default n
> + help
> + Use the IA-PC HPET (High Precision Event Timer) to manage
> + time in preference to the PIT and RTC, if a HPET is
> + present. The HPET provides a stable time base on SMP
> + systems, unlike the RTC, but it is more expensive to access,
... probably that should be 'unlike the TSC'? Since RTC does provide a
stable time base on a SMP system, and is very far off chip, too.
> + as it is off-chip. You can find the HPET spec at
> + <http://www.intel.com/labs/platcomp/hpet/hpetspec.htm>.
> +
> + If unsure, say Y.
>
> config GENERIC_ISA_DMA
> bool
>
--
Vojtech Pavlik
SuSE Labs, SuSE CR
On Wed, 2003-06-11 at 11:50, Bryan O'Sullivan wrote:
> I've forward-ported Vojtech's time code from 2.4, fixing some locking
> along the way. The new code supports using the AMD8111 HPET for time
> calculations. It also works stably with the PIT/TSC on every boot,
> which is the source of the time problems in current 2.5.
Oh drat, you beat me to it! I've been working on the very same thing.
This is very much needed, so thanks for doing this.
> Right now, the only known problem is with the fixup of jiffies if a
> timer interrupt is lost, which I've hence turned off. There's
> preliminary support for using HPET for the gettimeofday vsyscall, but
> since vsyscalls are disabled on x86-64 for now, that's obviously
> untested.
I'll start testing your patch and I'll see if I can't fixup the lost
tick compensation.
thanks!
-john
/*
+ * Read the period, compute tick and quotient.
+ */
+
+ id = hpet_readl(HPET_ID);
+
+ if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER) ||
+ !(id & HPET_ID_LEGSUP))
+ return -1;
+
+ hpet_period = hpet_readl(HPET_PERIOD);
+ if (hpet_period < 100000 || hpet_period > 100000000)
+ return -1;
+
Line below seems to be wrong, given hpet period is in fsecs.
+ hpet_tick = (tick_nsec + hpet_period / 2) / hpet_period;
--Mika
On Wed, 2003-06-11 at 14:32, Mika Penttilä wrote:
> Line below seems to be wrong, given hpet period is in fsecs.
I don't believe the HPET code got much testing in 2.4, and my boxes
don't have ACPI table entries for the HPET, so it's troublesome to test
it on them.
<b
On Wed, 2003-06-11 at 13:10, Bryan O'Sullivan wrote:
+ if (offset > tick_usec) {
+ lost = offset / tick_usec;
+ offset %= tick_usec;
+ }
Hmmm. Thats likely part to blame for the lost-ticks code not working. I
believe tick_usec is calculated USER_HZ rather then HZ, so you'll be off
by an order of magnitude. I ran into the exact same problem.
thanks
-john
On Wed, 2003-06-11 at 14:23, john stultz wrote:
> Hmmm. Thats likely part to blame for the lost-ticks code not working. I
> believe tick_usec is calculated USER_HZ rather then HZ, so you'll be off
> by an order of magnitude. I ran into the exact same problem.
Unlikely. On my systems, the offset values are off by three orders of
magnitude, and are always negative. There's a more basic error
somewhere before that test.
The actual impact of lost jiffies is pretty low, though. I've beaten
vigorously on my systems with the time patch, and they don't lose timer
interrupts.
<b
On Wed, 2003-06-11 at 14:36, Bryan O'Sullivan wrote:
> On Wed, 2003-06-11 at 14:23, john stultz wrote:
>
> > Hmmm. Thats likely part to blame for the lost-ticks code not working. I
> > believe tick_usec is calculated USER_HZ rather then HZ, so you'll be off
> > by an order of magnitude. I ran into the exact same problem.
>
> Unlikely. On my systems, the offset values are off by three orders of
> magnitude, and are always negative. There's a more basic error
> somewhere before that test.
Ok, that was just the first thing I noticed. Let me dig in and see what
I can find.
> The actual impact of lost jiffies is pretty low, though. I've beaten
> vigorously on my systems with the time patch, and they don't lose timer
> interrupts.
Hmm. In my (very quick) testing I'm seeing very frequent 10,000us
inconsistencies. This is w/ and w/o the vsyscall code enabled, so I'm
not sure whats causing it yet. I'll let you know if I trip over
anything.
thanks
-john
On Wed, 2003-06-11 at 14:36, Bryan O'Sullivan wrote:
> On Wed, 2003-06-11 at 14:23, john stultz wrote:
>
> > Hmmm. Thats likely part to blame for the lost-ticks code not working. I
> > believe tick_usec is calculated USER_HZ rather then HZ, so you'll be off
> > by an order of magnitude. I ran into the exact same problem.
>
> Unlikely. On my systems, the offset values are off by three orders of
> magnitude, and are always negative. There's a more basic error
> somewhere before that test.
Actually it was related. Right before the lines I quoted you subtract
tick_usec from the offset calculation. Since tick_usec is 10,000 rather
then 1,000, offset goes negative and bad stuff ensues.
This patch applies on top of Bryan's patch, and solves the major time
inconsistencies I was seeing earlier. I don't like that I still have to
have the sanity check on offset to ensure it doesn't go negative
(frequently I get offsets of -1), and that along with the fact that I'm
seeing a steady drift forward on every kernel I've run (2.4/2.5) makes
me think that the timer interrupt frequency may be a bit higher then we
intend. Might just be the dev hardware I'm running on, though.
I'm also got some cleanup changes I'd like to make, but I'll wait until
after things work to polish that stuff up.
Let me know if you have any issues with this patch.
thanks
-john
diff -Nru a/arch/x86_64/kernel/time.c b/arch/x86_64/kernel/time.c
--- a/arch/x86_64/kernel/time.c Wed Jun 11 17:28:07 2003
+++ b/arch/x86_64/kernel/time.c Wed Jun 11 17:28:07 2003
@@ -254,11 +254,14 @@
vxtime.last = offset;
} else {
offset = (((tsc - vxtime.last_tsc) *
- vxtime.tsc_quot) >> 32) - tick_usec;
+ vxtime.tsc_quot) >> 32) - (USEC_PER_SEC/HZ);
+ /* sanity check on offset */
+ if(offset < 0)
+ offset = 0;
- if (offset > tick_usec) {
- lost = offset / tick_usec;
- offset %= tick_usec;
+ if (offset > (USEC_PER_SEC/HZ)) {
+ lost = offset / (USEC_PER_SEC/HZ);
+ offset %= (USEC_PER_SEC/HZ);
}
vxtime.last_tsc = tsc - vxtime.quot * delay / vxtime.tsc_quot;
@@ -275,10 +278,7 @@
"tick(s)! (rip %016lx)\n",
(offset - vxtime.last) / hpet_tick - 1,
regs->rip);
- // XXX The accounting of lost ticks is way off right
- // now. -bos
-
- // jiffies += lost;
+ jiffies += lost;
}
/*
Bryan O'Sullivan <[email protected]> writes:
> On Wed, 2003-06-11 at 14:32, Mika Penttil? wrote:
>
>> Line below seems to be wrong, given hpet period is in fsecs.
>
> I don't believe the HPET code got much testing in 2.4, and my boxes
It got quite some testing on my boxes,
> don't have ACPI table entries for the HPET, so it's troublesome to test
> it on them.
Please tell your board vendor to fix it,
Andreas
--
Andreas Jaeger
SuSE Labs [email protected]
private [email protected]
http://www.suse.de/~aj
On Wed, Jun 11, 2003 at 05:33:55PM -0700, john stultz wrote:
> This patch applies on top of Bryan's patch, and solves the major time
> inconsistencies I was seeing earlier. I don't like that I still have to
> have the sanity check on offset to ensure it doesn't go negative
> (frequently I get offsets of -1),
IMO this is to be expected and not a problem. In the case of offset of
'-1' the time will jump one usec forward on the timer interrupt, but
that's harmless and nothing worse should happen.
> and that along with the fact that I'm
> seeing a steady drift forward on every kernel I've run (2.4/2.5) makes
> me think that the timer interrupt frequency may be a bit higher then we
> intend. Might just be the dev hardware I'm running on, though.
The devel hardware's RTC crystals are quite a bit off the correct
frequency, but still in the range fixable by NTP.
> I'm also got some cleanup changes I'd like to make, but I'll wait until
> after things work to polish that stuff up.
>
> Let me know if you have any issues with this patch.
Looks OK to me.
> thanks
> -john
>
> diff -Nru a/arch/x86_64/kernel/time.c b/arch/x86_64/kernel/time.c
> --- a/arch/x86_64/kernel/time.c Wed Jun 11 17:28:07 2003
> +++ b/arch/x86_64/kernel/time.c Wed Jun 11 17:28:07 2003
> @@ -254,11 +254,14 @@
> vxtime.last = offset;
> } else {
> offset = (((tsc - vxtime.last_tsc) *
> - vxtime.tsc_quot) >> 32) - tick_usec;
> + vxtime.tsc_quot) >> 32) - (USEC_PER_SEC/HZ);
> + /* sanity check on offset */
> + if(offset < 0)
> + offset = 0;
>
> - if (offset > tick_usec) {
> - lost = offset / tick_usec;
> - offset %= tick_usec;
> + if (offset > (USEC_PER_SEC/HZ)) {
> + lost = offset / (USEC_PER_SEC/HZ);
> + offset %= (USEC_PER_SEC/HZ);
> }
>
> vxtime.last_tsc = tsc - vxtime.quot * delay / vxtime.tsc_quot;
> @@ -275,10 +278,7 @@
> "tick(s)! (rip %016lx)\n",
> (offset - vxtime.last) / hpet_tick - 1,
> regs->rip);
> - // XXX The accounting of lost ticks is way off right
> - // now. -bos
> -
> - // jiffies += lost;
> + jiffies += lost;
> }
>
> /*
>
>
>
--
Vojtech Pavlik
SuSE Labs, SuSE CR
On Thu, Jun 12, 2003 at 12:32:01AM +0300, Mika Penttil? wrote:
> /*
> + * Read the period, compute tick and quotient.
> + */
> +
> + id = hpet_readl(HPET_ID);
> +
> + if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER) ||
> + !(id & HPET_ID_LEGSUP))
> + return -1;
> +
> + hpet_period = hpet_readl(HPET_PERIOD);
> + if (hpet_period < 100000 || hpet_period > 100000000)
> + return -1;
> +
>
>
> Line below seems to be wrong, given hpet period is in fsecs.
>
>
> + hpet_tick = (tick_nsec + hpet_period / 2) / hpet_period;
Yes, it should be:
hpet_tick = (1000000000L * (USECS_PER_SEC/HZ) + hpet_period / 2) / hpet_period;
>
>
>
>
>
> --Mika
>
>
>
>
>
--
Vojtech Pavlik
SuSE Labs, SuSE CR
On Wed, Jun 11, 2003 at 02:28:45PM -0700, Bryan O'Sullivan wrote:
> On Wed, 2003-06-11 at 14:32, Mika Penttilä wrote:
>
> > Line below seems to be wrong, given hpet period is in fsecs.
>
> I don't believe the HPET code got much testing in 2.4, and my boxes
> don't have ACPI table entries for the HPET, so it's troublesome to test
> it on them.
You can enable HPET_HACK_ENABLE_DANGEROUS, of course. That will enable
the HPET even when the BIOS did not. But make sure you have a recent
AMD-8111 stepping first.
--
Vojtech Pavlik
SuSE Labs, SuSE CR
On Wed, 2003-06-11 at 17:33, john stultz wrote:
> Let me know if you have any issues with this patch.
Thanks, John. Your updated patch has survived some beating on my test
systems. I've also applied Vojtech's fix to hpet_tick.
The latest grand unified x86_64 time patch against 2.5.70 is attached.
<b
arch/x86_64/Kconfig | 12 +
arch/x86_64/kernel/acpi/boot.c | 6
arch/x86_64/kernel/apic.c | 84 +++++-----
arch/x86_64/kernel/smpboot.c | 11 +
arch/x86_64/kernel/time.c | 306 +++++++++++++++++++++++++++++++-------- arch/x86_64/kernel/vsyscall.c | 28 ++-
arch/x86_64/vmlinux.lds.S | 6
include/asm-x86_64/fixmap.h | 2
include/asm-x86_64/mc146818rtc.h | 5
include/asm-x86_64/timex.h | 30 +++
include/asm-x86_64/vsyscall.h | 18 +-
11 files changed, 384 insertions(+), 124 deletions(-)
On Thu, 2003-06-12 at 10:47, Bryan O'Sullivan wrote:
> On Wed, 2003-06-11 at 17:33, john stultz wrote:
>
> > Let me know if you have any issues with this patch.
>
> Thanks, John. Your updated patch has survived some beating on my test
> systems. I've also applied Vojtech's fix to hpet_tick.
One little tweak, you're still subtracting tick_usec when calculating
offset. Doesn't cause any major problems since you're catching error
with the offset<0 switch. However we won't catch lost interrupts
properly if we're always negative.
thanks
-john
Patch for that should be:
--- 1.22/arch/x86_64/kernel/time.c Thu Jun 12 11:40:33 2003
+++ edited/arch/x86_64/kernel/time.c Thu Jun 12 11:42:43 2003
@@ -254,7 +254,7 @@
vxtime.last = offset;
} else {
offset = (((tsc - vxtime.last_tsc) *
- vxtime.tsc_quot) >> 32) - tick_usec;
+ vxtime.tsc_quot) >> 32) - (USEC_PER_SEC / HZ);
if (offset < 0)
offset = 0;
On Thu, Jun 12, 2003 at 10:47:10AM -0700, Bryan O'Sullivan wrote:
> On Wed, 2003-06-11 at 17:33, john stultz wrote:
>
> > Let me know if you have any issues with this patch.
>
> Thanks, John. Your updated patch has survived some beating on my test
> systems. I've also applied Vojtech's fix to hpet_tick.
>
> The latest grand unified x86_64 time patch against 2.5.70 is attached.
This one looks very good indeed. Andi?
>
> <b
>
> arch/x86_64/Kconfig | 12 +
> arch/x86_64/kernel/acpi/boot.c | 6
> arch/x86_64/kernel/apic.c | 84 +++++-----
> arch/x86_64/kernel/smpboot.c | 11 +
> arch/x86_64/kernel/time.c | 306 +++++++++++++++++++++++++++++++-------- arch/x86_64/kernel/vsyscall.c | 28 ++-
> arch/x86_64/vmlinux.lds.S | 6
> include/asm-x86_64/fixmap.h | 2
> include/asm-x86_64/mc146818rtc.h | 5
> include/asm-x86_64/timex.h | 30 +++
> include/asm-x86_64/vsyscall.h | 18 +-
> 11 files changed, 384 insertions(+), 124 deletions(-)
>
> # This is a BitKeeper generated patch for the following project:
> # Project Name: Linux kernel tree
> # This patch format is intended for GNU patch command version 2.5 or higher.
> # This patch includes the following deltas:
> # ChangeSet 1.1436 -> 1.1440
> # arch/x86_64/vmlinux.lds.S 1.16 -> 1.17
> # arch/x86_64/kernel/apic.c 1.20 -> 1.21
> # include/asm-x86_64/fixmap.h 1.3 -> 1.4
> # arch/x86_64/kernel/acpi/boot.c 1.1 -> 1.2
> # include/asm-x86_64/mc146818rtc.h 1.1 -> 1.2
> # include/asm-x86_64/timex.h 1.6 -> 1.7
> # arch/x86_64/kernel/time.c 1.15 -> 1.18
> # arch/x86_64/kernel/vsyscall.c 1.10 -> 1.11
> # arch/x86_64/Kconfig 1.22 -> 1.23
> # include/asm-x86_64/vsyscall.h 1.5 -> 1.6
> # arch/x86_64/kernel/smpboot.c 1.19 -> 1.20
> #
> # The following is the BitKeeper ChangeSet Log
> # --------------------------------------------
> # 03/06/11 [email protected] 1.1437
> # Forward port 2.4 time code. Optionally uses HPET instead of PIT/RTC for
> # gettimeofday calculations. Far more stable than the current 2.5 code.
> #
> # Current caveat: this code doesn't track lost interrupts properly, so we
> # can very slowly lose a jiffy here or there.
> # --------------------------------------------
> # 03/06/11 [email protected] 1.1438
> # Fix residual bogons.
> # 03/06/12 [email protected] 1.1440
> # Further fixes to 2.5 time code.
> # --------------------------------------------
> #
> diff -Nru a/arch/x86_64/Kconfig b/arch/x86_64/Kconfig
> --- a/arch/x86_64/Kconfig Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/Kconfig Thu Jun 12 10:44:01 2003
> @@ -52,6 +52,18 @@
> klogd/syslogd or the X server. You should normally N here, unless
> you want to debug such a crash.
>
> +config HPET_TIMER
> + bool
> + default y
> + help
> + Use the IA-PC HPET (High Precision Event Timer) to manage
> + time in preference to the PIT and RTC, if a HPET is
> + present. The HPET provides a stable time base on SMP
> + systems, unlike the RTC, but it is more expensive to access,
> + as it is off-chip. You can find the HPET spec at
> + <http://www.intel.com/labs/platcomp/hpet/hpetspec.htm>.
> +
> + If unsure, say Y.
>
> config GENERIC_ISA_DMA
> bool
> diff -Nru a/arch/x86_64/kernel/acpi/boot.c b/arch/x86_64/kernel/acpi/boot.c
> --- a/arch/x86_64/kernel/acpi/boot.c Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/kernel/acpi/boot.c Thu Jun 12 10:44:01 2003
> @@ -244,9 +244,11 @@
> return -1;
> }
>
> - hpet.address = hpet_tbl->addr.addrl | ((long) hpet_tbl->addr.addrh << 32);
> + vxtime.hpet_address = hpet_tbl->addr.addrl |
> + ((long) hpet_tbl->addr.addrh << 32);
>
> - printk(KERN_INFO "acpi: HPET id: %#x base: %#lx\n", hpet_tbl->id, hpet.address);
> + printk(KERN_INFO "acpi: HPET id: %#x base: %#lx\n",
> + hpet_tbl->id, vxtime.hpet_address);
>
> return 0;
> }
> diff -Nru a/arch/x86_64/kernel/apic.c b/arch/x86_64/kernel/apic.c
> --- a/arch/x86_64/kernel/apic.c Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/kernel/apic.c Thu Jun 12 10:44:01 2003
> @@ -46,7 +46,7 @@
> void enable_NMI_through_LVT0 (void * dummy)
> {
> unsigned int v, ver;
> -
> +
> ver = apic_read(APIC_LVR);
> ver = GET_APIC_VERSION(ver);
> v = APIC_DM_NMI; /* unmask and set to NMI */
> @@ -297,7 +297,7 @@
> * Double-check whether this APIC is really registered.
> * This is meaningless in clustered apic mode, so we skip it.
> */
> - if (!clustered_apic_mode &&
> + if (!clustered_apic_mode &&
> !test_bit(GET_APIC_ID(apic_read(APIC_ID)), &phys_cpu_present_map))
> BUG();
>
> @@ -309,7 +309,7 @@
>
> if (!clustered_apic_mode) {
> /*
> - * In clustered apic mode, the firmware does this for us
> + * In clustered apic mode, the firmware does this for us
> * Put the APIC into flat delivery mode.
> * Must be "all ones" explicitly for 82489DX.
> */
> @@ -422,15 +422,15 @@
> value = apic_read(APIC_ESR);
> Dprintk("ESR value after enabling vector: %08x\n", value);
> } else {
> - if (esr_disable)
> - /*
> - * Something untraceble is creating bad interrupts on
> + if (esr_disable)
> + /*
> + * Something untraceble is creating bad interrupts on
> * secondary quads ... for the moment, just leave the
> * ESR disabled - we can't do anything useful with the
> * errors anyway - mbligh
> */
> printk("Leaving ESR disabled.\n");
> - else
> + else
> printk("No ESR for 82489DX.\n");
> }
>
> @@ -580,7 +580,7 @@
> * Detect and enable local APICs on non-SMP boards.
> * Original code written by Keir Fraser.
> * On AMD64 we trust the BIOS - if it says no APIC it is likely
> - * not correctly set up (usually the APIC timer won't work etc.)
> + * not correctly set up (usually the APIC timer won't work etc.)
> */
>
> static int __init detect_init_APIC (void)
> @@ -683,19 +683,25 @@
>
> local_irq_save(flags);
>
> - /* For some reasons this doesn't work on Simics, so fake it for now */
> - if (!strstr(boot_cpu_data.x86_model_id, "Screwdriver")) {
> + /* For some reasons this doesn't work on Simics, so fake it for now */
> + if (!strstr(boot_cpu_data.x86_model_id, "Screwdriver")) {
> __setup_APIC_LVTT(clocks);
> return;
> - }
> + }
>
> /* wait for irq slice */
> - {
> + if (vxtime.hpet_address) {
> + int trigger = hpet_readl(HPET_T0_CMP);
> + while (hpet_readl(HPET_COUNTER) >= trigger)
> + /* do nothing */ ;
> + while (hpet_readl(HPET_COUNTER) < trigger)
> + /* do nothing */ ;
> + } else {
> int c1, c2;
> outb_p(0x00, 0x43);
> c2 = inb_p(0x40);
> c2 |= inb_p(0x40) << 8;
> - do {
> + do {
> c1 = c2;
> outb_p(0x00, 0x43);
> c2 = inb_p(0x40);
> @@ -754,10 +760,10 @@
>
> void __init setup_boot_APIC_clock (void)
> {
> - if (disable_apic_timer) {
> - printk(KERN_INFO "Disabling APIC timer\n");
> - return;
> - }
> + if (disable_apic_timer) {
> + printk(KERN_INFO "Disabling APIC timer\n");
> + return;
> + }
>
> printk(KERN_INFO "Using local APIC timer interrupts.\n");
> using_apic_timer = 1;
> @@ -816,7 +822,7 @@
> if ( (!multiplier) || (calibration_result/multiplier < 500))
> return -EINVAL;
>
> - /*
> + /*
> * Set the new multiplier for each CPU. CPUs don't start using the
> * new values until the next timer interrupt in which they do process
> * accounting. At that time they also adjust their APIC timers
> @@ -856,7 +862,7 @@
> * Interrupts are already masked off at this point.
> */
> per_cpu(prof_counter, cpu) = per_cpu(prof_multiplier, cpu);
> - if (per_cpu(prof_counter, cpu) !=
> + if (per_cpu(prof_counter, cpu) !=
> per_cpu(prof_old_multiplier, cpu)) {
> __setup_APIC_LVTT(calibration_result/
> per_cpu(prof_counter, cpu));
> @@ -928,19 +934,19 @@
> ack_APIC_irq();
>
> #if 0
> - static unsigned long last_warning;
> - static unsigned long skipped;
> + static unsigned long last_warning;
> + static unsigned long skipped;
>
> /* see sw-dev-man vol 3, chapter 7.4.13.5 */
> - if (time_before(last_warning+30*HZ,jiffies)) {
> + if (time_before(last_warning+30*HZ,jiffies)) {
> printk(KERN_INFO "spurious APIC interrupt on CPU#%d, %ld skipped.\n",
> smp_processor_id(), skipped);
> - last_warning = jiffies;
> + last_warning = jiffies;
> skipped = 0;
> - } else {
> - skipped++;
> - }
> -#endif
> + } else {
> + skipped++;
> + }
> +#endif
> irq_exit();
> }
>
> @@ -975,7 +981,7 @@
> irq_exit();
> }
>
> -int disable_apic;
> +int disable_apic;
>
> /*
> * This initializes the IO-APIC and APIC hardware if this is
> @@ -983,11 +989,11 @@
> */
> int __init APIC_init_uniprocessor (void)
> {
> - if (disable_apic) {
> + if (disable_apic) {
> printk(KERN_INFO "Apic disabled\n");
> - return -1;
> + return -1;
> }
> - if (!cpu_has_apic) {
> + if (!cpu_has_apic) {
> disable_apic = 1;
> printk(KERN_INFO "Apic disabled by BIOS\n");
> return -1;
> @@ -1015,18 +1021,18 @@
> return 0;
> }
>
> -static __init int setup_disableapic(char *str)
> -{
> +static __init int setup_disableapic(char *str)
> +{
> disable_apic = 1;
> return 0;
> -}
> +}
>
> -static __init int setup_noapictimer(char *str)
> -{
> +static __init int setup_noapictimer(char *str)
> +{
> disable_apic_timer = 1;
> return 0;
> -}
> +}
>
> -__setup("disableapic", setup_disableapic);
> -__setup("noapictimer", setup_noapictimer);
> +__setup("disableapic", setup_disableapic);
> +__setup("noapictimer", setup_noapictimer);
>
> diff -Nru a/arch/x86_64/kernel/smpboot.c b/arch/x86_64/kernel/smpboot.c
> --- a/arch/x86_64/kernel/smpboot.c Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/kernel/smpboot.c Thu Jun 12 10:44:01 2003
> @@ -67,6 +67,8 @@
> /* Set when the idlers are all forked */
> int smp_threads_ready;
>
> +extern void time_init_smp(void);
> +
> /*
> * Trampoline 80x86 program as an array.
> */
> @@ -760,7 +762,7 @@
> if (APIC_init_uniprocessor())
> printk(KERN_NOTICE "Local APIC not detected."
> " Using dummy APIC emulation.\n");
> - return;
> + goto smp_done;
> }
>
> /*
> @@ -784,7 +786,7 @@
> cpu_online_map = phys_cpu_present_map = 1;
> phys_cpu_present_map = 1;
> disable_apic = 1;
> - return;
> + goto smp_done;
> }
>
> verify_local_APIC();
> @@ -799,7 +801,7 @@
> cpu_online_map = phys_cpu_present_map = 1;
> phys_cpu_present_map = 1;
> disable_apic = 1;
> - return;
> + goto smp_done;
> }
>
> connect_bsp_APIC();
> @@ -883,6 +885,9 @@
> */
> if (cpu_has_tsc && cpucount)
> synchronize_tsc_bp();
> +
> + smp_done:
> + time_init_smp();
> }
>
> /* These are wrappers to interface to the new boot process. Someone
> diff -Nru a/arch/x86_64/kernel/time.c b/arch/x86_64/kernel/time.c
> --- a/arch/x86_64/kernel/time.c Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/kernel/time.c Thu Jun 12 10:44:01 2003
> @@ -24,6 +24,7 @@
> #include <linux/module.h>
> #include <linux/device.h>
> #include <linux/bcd.h>
> +#include <asm/pgtable.h>
> #include <asm/vsyscall.h>
> #include <asm/timex.h>
> #ifdef CONFIG_X86_LOCAL_APIC
> @@ -35,41 +36,61 @@
> extern int using_apic_timer;
>
> spinlock_t rtc_lock = SPIN_LOCK_UNLOCKED;
> +spinlock_t i8253_lock = SPIN_LOCK_UNLOCKED;
>
> extern int using_apic_timer;
> extern void smp_local_timer_interrupt(struct pt_regs * regs);
>
> +#undef HPET_HACK_ENABLE_DANGEROUS
>
> -unsigned int cpu_khz; /* TSC clocks / usec, not used here */
> -unsigned long hpet_period; /* fsecs / HPET clock */
> -unsigned long hpet_tick; /* HPET clocks / interrupt */
> -int hpet_report_lost_ticks; /* command line option */
>
> -struct hpet_data __hpet __section_hpet; /* address, quotient, trigger, hz */
> +unsigned int cpu_khz; /* TSC clocks / usec, not used here */
> +unsigned long hpet_period; /* fsecs / HPET clock */
> +unsigned long hpet_tick; /* HPET clocks / interrupt */
> +unsigned long vxtime_hz = 1193182;
> +int report_lost_ticks; /* command line option */
> +
> +struct vxtime_data __vxtime __section_vxtime; /* for vsyscalls */
>
> volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES;
> unsigned long __wall_jiffies __section_wall_jiffies = INITIAL_JIFFIES;
> struct timespec __xtime __section_xtime;
> struct timezone __sys_tz __section_sys_tz;
>
> +static inline void rdtscll_sync(unsigned long *tsc)
> +{
> +#ifdef CONFIG_SMP
> + sync_core();
> +#endif
> + rdtscll(*tsc);
> +}
> +
> /*
> * do_gettimeoffset() returns microseconds since last timer interrupt was
> * triggered by hardware. A memory read of HPET is slower than a register read
> * of TSC, but much more reliable. It's also synchronized to the timer
> * interrupt. Note that do_gettimeoffset() may return more than hpet_tick, if a
> - * timer interrupt has happened already, but hpet.trigger wasn't updated yet.
> + * timer interrupt has happened already, but vxtime.trigger wasn't updated yet.
> * This is not a problem, because jiffies hasn't updated either. They are bound
> * together by xtime_lock.
> - */
> + */
>
> -inline unsigned int do_gettimeoffset(void)
> +static inline unsigned int do_gettimeoffset_tsc(void)
> {
> unsigned long t;
> - sync_core();
> - rdtscll(t);
> - return (t - hpet.last_tsc) * (1000000L / HZ) / hpet.ticks + hpet.offset;
> + unsigned long x;
> + rdtscll_sync(&t);
> + x = ((t - vxtime.last_tsc) * vxtime.tsc_quot) >> 32;
> + return x;
> +}
> +
> +static inline unsigned int do_gettimeoffset_hpet(void)
> +{
> + return ((hpet_readl(HPET_COUNTER) - vxtime.last) * vxtime.quot) >> 32;
> }
>
> +unsigned int (*do_gettimeoffset)(void) = do_gettimeoffset_tsc;
> +
> /*
> * This version of gettimeofday() has microsecond resolution and better than
> * microsecond precision, as we're using at least a 10 MHz (usually 14.31818
> @@ -87,7 +108,8 @@
> sec = xtime.tv_sec;
> usec = xtime.tv_nsec / 1000;
>
> - t = (jiffies - wall_jiffies) * (1000000L / HZ) + do_gettimeoffset();
> + t = (jiffies - wall_jiffies) * (1000000L / HZ) +
> + do_gettimeoffset();
> usec += t;
>
> } while (read_seqretry(&xtime_lock, seq));
> @@ -169,17 +191,17 @@
> real_seconds = nowtime % 60;
> real_minutes = nowtime / 60;
> if (((abs(real_minutes - cmos_minutes) + 15) / 30) & 1)
> - real_minutes += 30; /* correct for half hour time zone */
> + real_minutes += 30; /* correct for half hour time zone */
> real_minutes %= 60;
>
> if (abs(real_minutes - cmos_minutes) < 30) {
> - BIN_TO_BCD(real_seconds);
> - BIN_TO_BCD(real_minutes);
> + BIN_TO_BCD(real_seconds);
> + BIN_TO_BCD(real_minutes);
> CMOS_WRITE(real_seconds, RTC_SECONDS);
> CMOS_WRITE(real_minutes, RTC_MINUTES);
> } else
> - printk(KERN_WARNING "time.c: can't update CMOS clock from %d to %d\n",
> - cmos_minutes, real_minutes);
> + printk(KERN_WARNING "time.c: can't update CMOS clock "
> + "from %d to %d\n", cmos_minutes, real_minutes);
>
> /*
> * The following flags have to be released exactly in this order, otherwise the
> @@ -198,27 +220,65 @@
> static irqreturn_t timer_interrupt(int irq, void *dev_id, struct pt_regs *regs)
> {
> static unsigned long rtc_update = 0;
> + unsigned long tsc, lost = 0;
> + int delay, offset = 0;
>
> /*
> * Here we are in the timer irq handler. We have irqs locally disabled (so we
> * don't need spin_lock_irqsave()) but we don't know if the timer_bh is running
> * on the other CPU, so we need a lock. We also need to lock the vsyscall
> * variables, because both do_timer() and us change them -arca+vojtech
> - */
> + */
>
> write_seqlock(&xtime_lock);
>
> - {
> - unsigned long t;
> + if (vxtime.hpet_address) {
> + offset = hpet_readl(HPET_T0_CMP) - hpet_tick;
> + delay = hpet_readl(HPET_COUNTER) - offset;
> + } else {
> + spin_lock(&i8253_lock);
> + outb_p(0x00, 0x43);
> + delay = inb_p(0x40);
> + delay |= inb(0x40) << 8;
> + spin_unlock(&i8253_lock);
> + delay = LATCH - 1 - delay;
> + }
> +
> + rdtscll_sync(&tsc);
>
> - sync_core();
> - rdtscll(t);
> - hpet.offset = (t - hpet.last_tsc) * (1000000L / HZ) / hpet.ticks + hpet.offset - 1000000L / HZ;
> - if (hpet.offset >= 1000000L / HZ)
> - hpet.offset = 0;
> - hpet.ticks = min_t(long, max_t(long, (t - hpet.last_tsc) * (1000000L / HZ) / (1000000L / HZ - hpet.offset),
> - cpu_khz * 1000/HZ * 15 / 16), cpu_khz * 1000/HZ * 16 / 15);
> - hpet.last_tsc = t;
> + if (vxtime.mode == VXTIME_HPET) {
> + if (offset - vxtime.last > hpet_tick) {
> + lost = (offset - vxtime.last) / hpet_tick - 1;
> + }
> +
> + vxtime.last = offset;
> + } else {
> + offset = (((tsc - vxtime.last_tsc) *
> + vxtime.tsc_quot) >> 32) - tick_usec;
> +
> + if (offset < 0)
> + offset = 0;
> +
> + if (offset > (USEC_PER_SEC / HZ)) {
> + lost = offset / (USEC_PER_SEC / HZ);
> + offset %= (USEC_PER_SEC / HZ);
> + }
> +
> + vxtime.last_tsc = tsc - vxtime.quot * delay / vxtime.tsc_quot;
> +
> + if ((((tsc - vxtime.last_tsc) *
> + vxtime.tsc_quot) >> 32) < offset)
> + vxtime.last_tsc = tsc -
> + (((long) offset << 32) / vxtime.tsc_quot) - 1;
> + }
> +
> + if (lost) {
> + if (report_lost_ticks)
> + printk(KERN_WARNING "time.c: Lost %ld timer "
> + "tick(s)! (rip %016lx)\n",
> + (offset - vxtime.last) / hpet_tick - 1,
> + regs->rip);
> + jiffies += lost;
> }
>
> /*
> @@ -244,16 +304,16 @@
> * If we have an externally synchronized Linux clock, then update CMOS clock
> * accordingly every ~11 minutes. set_rtc_mmss() will be called in the jiffy
> * closest to exactly 500 ms before the next second. If the update fails, we
> - * don'tcare, as it'll be updated on the next turn, and the problem (time way
> + * don't care, as it'll be updated on the next turn, and the problem (time way
> * off) isn't likely to go away much sooner anyway.
> */
>
> if ((~time_status & STA_UNSYNC) && xtime.tv_sec > rtc_update &&
> - abs(xtime.tv_nsec - 500000000) <= tick_nsec / 2) {
> + abs(xtime.tv_nsec - 500000000) <= tick_nsec / 2) {
> set_rtc_mmss(xtime.tv_sec);
> rtc_update = xtime.tv_sec + 660;
> }
> -
> +
> write_sequnlock(&xtime_lock);
>
> return IRQ_HANDLED;
> @@ -263,6 +323,7 @@
> {
> unsigned int timeout, year, mon, day, hour, min, sec;
> unsigned char last, this;
> + unsigned long flags;
>
> /*
> * The Linux interpretation of the CMOS clock register contents: When the
> @@ -272,7 +333,7 @@
> * standard 8.3 MHz ISA bus.
> */
>
> - spin_lock(&rtc_lock);
> + spin_lock_irqsave(&rtc_lock, flags);
>
> timeout = 1000000;
> last = this = 0;
> @@ -286,28 +347,28 @@
> /*
> * Here we are safe to assume the registers won't change for a whole second, so
> * we just go ahead and read them.
> - */
> + */
>
> - sec = CMOS_READ(RTC_SECONDS);
> - min = CMOS_READ(RTC_MINUTES);
> - hour = CMOS_READ(RTC_HOURS);
> - day = CMOS_READ(RTC_DAY_OF_MONTH);
> - mon = CMOS_READ(RTC_MONTH);
> - year = CMOS_READ(RTC_YEAR);
> + sec = CMOS_READ(RTC_SECONDS);
> + min = CMOS_READ(RTC_MINUTES);
> + hour = CMOS_READ(RTC_HOURS);
> + day = CMOS_READ(RTC_DAY_OF_MONTH);
> + mon = CMOS_READ(RTC_MONTH);
> + year = CMOS_READ(RTC_YEAR);
>
> - spin_unlock(&rtc_lock);
> + spin_unlock_irqrestore(&rtc_lock, flags);
>
> /*
> * We know that x86-64 always uses BCD format, no need to check the config
> * register.
> */
>
> - BCD_TO_BIN(sec);
> - BCD_TO_BIN(min);
> - BCD_TO_BIN(hour);
> - BCD_TO_BIN(day);
> - BCD_TO_BIN(mon);
> - BCD_TO_BIN(year);
> + BCD_TO_BIN(sec);
> + BCD_TO_BIN(min);
> + BCD_TO_BIN(hour);
> + BCD_TO_BIN(day);
> + BCD_TO_BIN(mon);
> + BCD_TO_BIN(year);
>
> /*
> * This will work up to Dec 31, 2069.
> @@ -326,6 +387,32 @@
>
> #define TICK_COUNT 100000000
>
> +static unsigned int __init hpet_calibrate_tsc(void)
> +{
> + int tsc_start, hpet_start;
> + int tsc_now, hpet_now;
> + unsigned long flags;
> +
> + local_irq_save(flags);
> + local_irq_disable();
> +
> + hpet_start = hpet_readl(HPET_COUNTER);
> + rdtscl(tsc_start);
> +
> + do {
> + local_irq_disable();
> + hpet_now = hpet_readl(HPET_COUNTER);
> + sync_core();
> + rdtscl(tsc_now);
> + local_irq_restore(flags);
> + } while ((tsc_now - tsc_start) < TICK_COUNT &&
> + (hpet_now - hpet_start) < TICK_COUNT);
> +
> + return (tsc_now - tsc_start) * 1000000000L
> + / ((hpet_now - hpet_start) * hpet_period / 1000);
> +}
> +
> +
> /*
> * pit_calibrate_tsc() uses the speaker output (channel 2) of
> * the PIT. This is better than using the timer interrupt output,
> @@ -339,10 +426,9 @@
> unsigned long start, end;
> unsigned long flags;
>
> - outb((inb(0x61) & ~0x02) | 0x01, 0x61);
> + spin_lock_irqsave(&i8253_lock, flags);
>
> - local_irq_save(flags);
> - local_irq_disable();
> + outb((inb(0x61) & ~0x02) | 0x01, 0x61);
>
> outb(0xb0, 0x43);
> outb((1193182 / (1000 / 50)) & 0xff, 0x42);
> @@ -353,42 +439,146 @@
> sync_core();
> rdtscll(end);
>
> + spin_unlock_irqrestore(&i8253_lock, flags);
>
> - local_irq_restore(flags);
> -
> return (end - start) / 50;
> }
>
> +static int hpet_init(void)
> +{
> + unsigned int cfg, id;
> +
> + if (!vxtime.hpet_address)
> + return -1;
> + set_fixmap_nocache(FIX_HPET_BASE, vxtime.hpet_address);
> +
> +/*
> + * Read the period, compute tick and quotient.
> + */
> +
> + id = hpet_readl(HPET_ID);
> +
> + if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER) ||
> + !(id & HPET_ID_LEGSUP))
> + return -1;
> +
> + hpet_period = hpet_readl(HPET_PERIOD);
> + if (hpet_period < 100000 || hpet_period > 100000000)
> + return -1;
> +
> + hpet_tick = (1000000000L * (USEC_PER_SEC / HZ) + hpet_period / 2) /
> + hpet_period;
> +
> +/*
> + * Stop the timers and reset the main counter.
> + */
> +
> + cfg = hpet_readl(HPET_CFG);
> + cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
> + hpet_writel(cfg, HPET_CFG);
> + hpet_writel(0, HPET_COUNTER);
> + hpet_writel(0, HPET_COUNTER + 4);
> +
> +/*
> + * Set up timer 0, as periodic with first interrupt to happen at hpet_tick,
> + * and period also hpet_tick.
> + */
> +
> + hpet_writel(HPET_T0_ENABLE | HPET_T0_PERIODIC | HPET_T0_SETVAL |
> + HPET_T0_32BIT, HPET_T0_CFG);
> + hpet_writel(hpet_tick, HPET_T0_CMP);
> + hpet_writel(hpet_tick, HPET_T0_CMP);
> +
> +/*
> + * Go!
> + */
> +
> + cfg |= HPET_CFG_ENABLE | HPET_CFG_LEGACY;
> + hpet_writel(cfg, HPET_CFG);
> +
> + return 0;
> +}
> +
> void __init pit_init(void)
> {
> + unsigned long flags;
> +
> + spin_lock_irqsave(&i8253_lock, flags);
> outb_p(0x34, 0x43); /* binary, mode 2, LSB/MSB, ch 0 */
> outb_p(LATCH & 0xff, 0x40); /* LSB */
> outb_p(LATCH >> 8, 0x40); /* MSB */
> + spin_unlock_irqrestore(&i8253_lock, flags);
> }
>
> int __init time_setup(char *str)
> {
> - hpet_report_lost_ticks = 1;
> + report_lost_ticks = 1;
> return 1;
> }
>
> -static struct irqaction irq0 = { timer_interrupt, SA_INTERRUPT, 0, "timer", NULL, NULL};
> +static struct irqaction irq0 = {
> + timer_interrupt, SA_INTERRUPT, 0, "timer", NULL, NULL
> +};
>
> extern void __init config_acpi_tables(void);
>
> void __init time_init(void)
> {
> + char *timename;
> +
> +#ifdef HPET_HACK_ENABLE_DANGEROUS
> + if (!vxtime.hpet_address) {
> + printk(KERN_WARNING "time.c: WARNING: Enabling HPET base "
> + "manually!\n");
> + outl(0x800038a0, 0xcf8);
> + outl(0xff000001, 0xcfc);
> + outl(0x800038a0, 0xcf8);
> + hpet_address = inl(0xcfc) & 0xfffffffe;
> + printk(KERN_WARNING "time.c: WARNING: Enabled HPET "
> + "at %#lx.\n", hpet_address);
> + }
> +#endif
> +
> xtime.tv_sec = get_cmos_time();
> xtime.tv_nsec = 0;
>
> - pit_init();
> - printk(KERN_INFO "time.c: Using 1.1931816 MHz PIT timer.\n");
> - cpu_khz = pit_calibrate_tsc();
> + if (!hpet_init()) {
> + vxtime_hz = (1000000000000000L + hpet_period / 2) /
> + hpet_period;
> + cpu_khz = hpet_calibrate_tsc();
> + timename = "HPET";
> + } else {
> + pit_init();
> + cpu_khz = pit_calibrate_tsc();
> + timename = "PIT";
> + }
> +
> + printk(KERN_INFO "time.c: Using %ld.%06ld MHz %s timer.\n",
> + vxtime_hz / 1000000, vxtime_hz % 1000000, timename);
> printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n",
> cpu_khz / 1000, cpu_khz % 1000);
> - hpet.ticks = cpu_khz * (1000 / HZ);
> - rdtscll(hpet.last_tsc);
> + vxtime.mode = VXTIME_TSC;
> + vxtime.quot = (1000000L << 32) / vxtime_hz;
> + vxtime.tsc_quot = (1000L << 32) / cpu_khz;
> + vxtime.hz = vxtime_hz;
> + rdtscll_sync(&vxtime.last_tsc);
> setup_irq(0, &irq0);
> +}
> +
> +void __init time_init_smp(void)
> +{
> + char *timetype;
> +
> + if (vxtime.hpet_address) {
> + timetype = "HPET";
> + vxtime.last = hpet_readl(HPET_T0_CMP) - hpet_tick;
> + vxtime.mode = VXTIME_HPET;
> + do_gettimeoffset = do_gettimeoffset_hpet;
> + } else {
> + timetype = "PIT/TSC";
> + vxtime.mode = VXTIME_TSC;
> + }
> + printk(KERN_INFO "time.c: Using %s based timekeeping.\n", timetype);
> }
>
> __setup("report_lost_ticks", time_setup);
> diff -Nru a/arch/x86_64/kernel/vsyscall.c b/arch/x86_64/kernel/vsyscall.c
> --- a/arch/x86_64/kernel/vsyscall.c Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/kernel/vsyscall.c Thu Jun 12 10:44:01 2003
> @@ -33,7 +33,7 @@
> *
> * Add HPET support (port from 2.4). Still needed?
> * Nop out vsyscall syscall to avoid anchor for buffer overflows when sysctl off.
> - *
> + *
> * These are not urgent things that we need to address only before shipping the first
> * production binary kernels.
> */
> @@ -77,14 +77,22 @@
>
> do {
> sequence = read_seqbegin(&__xtime_lock);
> -
> - sync_core();
> - rdtscll(t);
> +
> sec = __xtime.tv_sec;
> usec = (__xtime.tv_nsec / 1000) +
> - (__jiffies - __wall_jiffies) * (1000000 / HZ) +
> - (t - __hpet.last_tsc) * (1000000 / HZ) / __hpet.ticks + __hpet.offset;
> + (__jiffies - __wall_jiffies) * (1000000 / HZ);
>
> + if (__vxtime.mode == VXTIME_TSC) {
> + sync_core();
> + rdtscll(t);
> + usec += ((t - __vxtime.last_tsc) *
> + __vxtime.tsc_quot) >> 32;
> + } else {
> +#if 0
> + usec += ((readl(fix_to_virt(VSYSCALL_HPET) + 0xf0) -
> + __vxtime.last) * __vxtime.quot) >> 32;
> +#endif
> + }
> } while (read_seqretry(&__xtime_lock, sequence));
>
> tv->tv_sec = sec + usec / 1000000;
> @@ -100,7 +108,7 @@
> static force_inline int gettimeofday(struct timeval *tv, struct timezone *tz)
> {
> int ret;
> - asm volatile("syscall"
> + asm volatile("syscall"
> : "=a" (ret)
> : "0" (__NR_gettimeofday),"D" (tv),"S" (tz) : __syscall_clobber );
> return ret;
> @@ -109,7 +117,7 @@
> static int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
> {
> if (unlikely(!__sysctl_vsyscall))
> - return gettimeofday(tv,tz);
> + return gettimeofday(tv,tz);
> if (tv)
> do_vgettimeofday(tv);
> if (tz)
> @@ -119,13 +127,13 @@
>
> static time_t __vsyscall(1) vtime(time_t * t)
> {
> - struct timeval tv;
> + struct timeval tv;
> if (unlikely(!__sysctl_vsyscall))
> gettimeofday(&tv, NULL);
> else
> do_vgettimeofday(&tv);
> if (t)
> - *t = tv.tv_sec;
> + *t = tv.tv_sec;
> return tv.tv_sec;
> }
>
> diff -Nru a/arch/x86_64/vmlinux.lds.S b/arch/x86_64/vmlinux.lds.S
> --- a/arch/x86_64/vmlinux.lds.S Thu Jun 12 10:44:01 2003
> +++ b/arch/x86_64/vmlinux.lds.S Thu Jun 12 10:44:01 2003
> @@ -50,10 +50,10 @@
> .xtime_lock : AT ((LOADADDR(.vsyscall_0) + SIZEOF(.vsyscall_0) + 63) & ~(63)) { *(.xtime_lock) }
> xtime_lock = LOADADDR(.xtime_lock);
> . = ALIGN(16);
> - .hpet : AT ((LOADADDR(.xtime_lock) + SIZEOF(.xtime_lock) + 15) & ~(15)) { *(.hpet) }
> - hpet = LOADADDR(.hpet);
> + .vxtime : AT ((LOADADDR(.xtime_lock) + SIZEOF(.xtime_lock) + 15) & ~(15)) { *(.vxtime) }
> + vxtime = LOADADDR(.vxtime);
> . = ALIGN(16);
> - .wall_jiffies : AT ((LOADADDR(.hpet) + SIZEOF(.hpet) + 15) & ~(15)) { *(.wall_jiffies) }
> + .wall_jiffies : AT ((LOADADDR(.vxtime) + SIZEOF(.vxtime) + 15) & ~(15)) { *(.wall_jiffies) }
> wall_jiffies = LOADADDR(.wall_jiffies);
> . = ALIGN(16);
> .sys_tz : AT ((LOADADDR(.wall_jiffies) + SIZEOF(.wall_jiffies) + 15) & ~(15)) { *(.sys_tz) }
> diff -Nru a/include/asm-x86_64/fixmap.h b/include/asm-x86_64/fixmap.h
> --- a/include/asm-x86_64/fixmap.h Thu Jun 12 10:44:01 2003
> +++ b/include/asm-x86_64/fixmap.h Thu Jun 12 10:44:01 2003
> @@ -35,6 +35,8 @@
> enum fixed_addresses {
> VSYSCALL_LAST_PAGE,
> VSYSCALL_FIRST_PAGE = VSYSCALL_LAST_PAGE + ((VSYSCALL_END-VSYSCALL_START) >> PAGE_SHIFT) - 1,
> + VSYSCALL_HPET,
> + FIX_HPET_BASE,
> #ifdef CONFIG_X86_LOCAL_APIC
> FIX_APIC_BASE, /* local (CPU) APIC) -- required for SMP or not */
> #endif
> diff -Nru a/include/asm-x86_64/mc146818rtc.h b/include/asm-x86_64/mc146818rtc.h
> --- a/include/asm-x86_64/mc146818rtc.h Thu Jun 12 10:44:01 2003
> +++ b/include/asm-x86_64/mc146818rtc.h Thu Jun 12 10:44:01 2003
> @@ -24,6 +24,11 @@
> outb_p((val),RTC_PORT(1)); \
> })
>
> +#ifndef CONFIG_HPET_TIMER
> #define RTC_IRQ 8
> +#else
> +/* Temporary workaround due to IRQ routing problem. */
> +#define RTC_IRQ 0
> +#endif
>
> #endif /* _ASM_MC146818RTC_H */
> diff -Nru a/include/asm-x86_64/timex.h b/include/asm-x86_64/timex.h
> --- a/include/asm-x86_64/timex.h Thu Jun 12 10:44:01 2003
> +++ b/include/asm-x86_64/timex.h Thu Jun 12 10:44:01 2003
> @@ -30,6 +30,34 @@
>
> extern unsigned int cpu_khz;
>
> -extern struct hpet_data hpet;
> +/*
> + * Documentation on HPET can be found at:
> + * http://www.intel.com/ial/home/sp/pcmmspec.htm
> + * ftp://download.intel.com/ial/home/sp/mmts098.pdf
> + */
> +
> +#define HPET_ID 0x000
> +#define HPET_PERIOD 0x004
> +#define HPET_CFG 0x010
> +#define HPET_STATUS 0x020
> +#define HPET_COUNTER 0x0f0
> +#define HPET_T0_CFG 0x100
> +#define HPET_T0_CMP 0x108
> +#define HPET_T0_ROUTE 0x110
> +
> +#define HPET_ID_VENDOR 0xffff0000
> +#define HPET_ID_LEGSUP 0x00008000
> +#define HPET_ID_NUMBER 0x00000f00
> +#define HPET_ID_REV 0x000000ff
> +
> +#define HPET_CFG_ENABLE 0x001
> +#define HPET_CFG_LEGACY 0x002
> +
> +#define HPET_T0_ENABLE 0x004
> +#define HPET_T0_PERIODIC 0x008
> +#define HPET_T0_SETVAL 0x040
> +#define HPET_T0_32BIT 0x100
> +
> +extern struct vxtime_data vxtime;
>
> #endif
> diff -Nru a/include/asm-x86_64/vsyscall.h b/include/asm-x86_64/vsyscall.h
> --- a/include/asm-x86_64/vsyscall.h Thu Jun 12 10:44:01 2003
> +++ b/include/asm-x86_64/vsyscall.h Thu Jun 12 10:44:01 2003
> @@ -15,7 +15,7 @@
>
> #ifdef __KERNEL__
>
> -#define __section_hpet __attribute__ ((unused, __section__ (".hpet"), aligned(16)))
> +#define __section_vxtime __attribute__ ((unused, __section__ (".vxtime"), aligned(16)))
> #define __section_wall_jiffies __attribute__ ((unused, __section__ (".wall_jiffies"), aligned(16)))
> #define __section_jiffies __attribute__ ((unused, __section__ (".jiffies"), aligned(16)))
> #define __section_sys_tz __attribute__ ((unused, __section__ (".sys_tz"), aligned(16)))
> @@ -23,22 +23,24 @@
> #define __section_xtime __attribute__ ((unused, __section__ (".xtime"), aligned(16)))
> #define __section_xtime_lock __attribute__ ((unused, __section__ (".xtime_lock"), aligned(L1_CACHE_BYTES)))
>
> +#define VXTIME_TSC 1
> +#define VXTIME_HPET 2
>
> -struct hpet_data {
> - long address; /* base address */
> +struct vxtime_data {
> + long hpet_address; /* HPET base address */
> unsigned long hz; /* HPET clocks / sec */
> - int trigger; /* value at last interrupt */
> int last;
> - int offset;
> unsigned long last_tsc;
> - long ticks;
> + long quot;
> + long tsc_quot;
> + int mode;
> };
>
> #define hpet_readl(a) readl(fix_to_virt(FIX_HPET_BASE) + a)
> #define hpet_writel(d,a) writel(d, fix_to_virt(FIX_HPET_BASE) + a)
>
> /* vsyscall space (readonly) */
> -extern struct hpet_data __hpet;
> +extern struct vxtime_data __vxtime;
> extern struct timespec __xtime;
> extern volatile unsigned long __jiffies;
> extern unsigned long __wall_jiffies;
> @@ -46,7 +48,7 @@
> extern seqlock_t __xtime_lock;
>
> /* kernel space (writeable) */
> -extern struct hpet_data hpet;
> +extern struct vxtime_data vxtime;
> extern unsigned long wall_jiffies;
> extern struct timezone sys_tz;
> extern int sysctl_vsyscall;
--
Vojtech Pavlik
SuSE Labs, SuSE CR
On Thu, Jun 12, 2003 at 12:39:55PM -0700, john stultz wrote:
> On Thu, 2003-06-12 at 10:47, Bryan O'Sullivan wrote:
> > On Wed, 2003-06-11 at 17:33, john stultz wrote:
> >
> > > Let me know if you have any issues with this patch.
> >
> > Thanks, John. Your updated patch has survived some beating on my test
> > systems. I've also applied Vojtech's fix to hpet_tick.
>
> One little tweak, you're still subtracting tick_usec when calculating
> offset. Doesn't cause any major problems since you're catching error
> with the offset<0 switch. However we won't catch lost interrupts
> properly if we're always negative.
>
> thanks
> -john
>
>
> Patch for that should be:
>
> --- 1.22/arch/x86_64/kernel/time.c Thu Jun 12 11:40:33 2003
> +++ edited/arch/x86_64/kernel/time.c Thu Jun 12 11:42:43 2003
> @@ -254,7 +254,7 @@
> vxtime.last = offset;
> } else {
> offset = (((tsc - vxtime.last_tsc) *
> - vxtime.tsc_quot) >> 32) - tick_usec;
> + vxtime.tsc_quot) >> 32) - (USEC_PER_SEC / HZ);
>
> if (offset < 0)
> offset = 0;
In my opinion the "if (offset < 0)" check above is not needed once the
problem you describe is fixed, since the code below actually expects
that offset can be negative and doesn't do anything in that case.
Thinking more about it, the
if ((((tsc - vxtime.last_tsc) * vxtime.tsc_quot) >> 32) < offset)
statement is comparing unsigned and signed values, and should probably
be fixed to do a signed comparison.
--
Vojtech Pavlik
SuSE Labs, SuSE CR
On Thu, 2003-06-12 at 12:39, john stultz wrote:
> One little tweak, you're still subtracting tick_usec when calculating
> offset.
Well spotted, thanks.
I'm leaving in the test for negative offset for now, in spite of the
fix.
<b
arch/x86_64/Kconfig | 12 +
arch/x86_64/kernel/acpi/boot.c | 6
arch/x86_64/kernel/apic.c | 84 +++++-----
arch/x86_64/kernel/smpboot.c | 11 +
arch/x86_64/kernel/time.c | 306 +++++++++++++++++++++++++++++++-------- arch/x86_64/kernel/vsyscall.c | 28 ++-
arch/x86_64/vmlinux.lds.S | 6
include/asm-x86_64/fixmap.h | 2
include/asm-x86_64/mc146818rtc.h | 5
include/asm-x86_64/timex.h | 30 +++
include/asm-x86_64/vsyscall.h | 18 +-
11 files changed, 384 insertions(+), 124 deletions(-)
On Thu, 2003-06-12 at 14:16, Bryan O'Sullivan wrote:
> On Thu, 2003-06-12 at 12:39, john stultz wrote:
>
> > One little tweak, you're still subtracting tick_usec when calculating
> > offset.
>
> Well spotted, thanks.
Caught another (last instance of tick_usec in time.c). This time in
settimeofday(). This fixes the ltp failure.
thanks
-john
diff -Nru a/arch/x86_64/kernel/time.c b/arch/x86_64/kernel/time.c
--- a/arch/x86_64/kernel/time.c Thu Jun 12 18:42:19 2003
+++ b/arch/x86_64/kernel/time.c Thu Jun 12 18:42:19 2003
@@ -129,7 +129,7 @@
write_seqlock_irq(&xtime_lock);
tv->tv_usec -= do_gettimeoffset() +
- (jiffies - wall_jiffies) * tick_usec;
+ (jiffies - wall_jiffies) * (USEC_PER_SEC/HZ);
while (tv->tv_usec < 0) {
tv->tv_usec += 1000000;