On 2012-06-11 12:07, Avi Kivity wrote:
> On 06/06/2012 05:28 PM, Jan Kiszka wrote:
>> Due to a offset between the clock used to generate the in-kernel
>> count_load_time (CLOCK_MONOTONIC) and the clock used for processing this
>> in userspace (vm_clock), reading back the output of PIT channel 2 via
>> port 0x61 was broken. One use cases that suffered from it was the CPU
>> frequency calibration of SeaBIOS, which also affected IDE/AHCI timeouts.
>>
>> This fixes it by calibrating the offset between both clocks on
>> kvm_pit_get and adjusting the kernel value before saving it in the
>> userspace state. As the calibration only works while the vm_clock is
>> running, we cache the in-kernel state across stopped phases.
>
> Applied, thanks.
>
>> + clock_offset = LLONG_MAX;
>
> INT64_MAX would me more strictly correct, but in practice it makes no
> difference.
Was looking for this, just not long enough. Need to print some cheat
sheet. However, let's clean this up immediately:
---8<---
clock_offset is int64_t.
Signed-off-by: Jan Kiszka <jan.kiszka@xxxxxxxxxxx>
---
hw/kvm/i8254.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/hw/kvm/i8254.c b/hw/kvm/i8254.c
index 04333d8..c5d3711 100644
--- a/hw/kvm/i8254.c
+++ b/hw/kvm/i8254.c
@@ -62,7 +62,7 @@ static void kvm_pit_get(PITCommonState *pit)
* kvm_pit_channel_state::count_load_time, and vm_clock. Take the
* minimum of several samples to filter out scheduling noise.
*/
- clock_offset = LLONG_MAX;
+ clock_offset = INT64_MAX;
for (i = 0; i < CALIBRATION_ROUNDS; i++) {
offset = qemu_get_clock_ns(vm_clock);
clock_gettime(CLOCK_MONOTONIC, &ts);
--
1.7.3.4
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
