incorrect usage of callout_reset in vbox 4.2.4 ?
Andriy Gapon
avg at FreeBSD.org
Sun Dec 23 10:33:43 UTC 2012
on 20/12/2012 13:14 Bernhard Fröhlich said the following:
> On Thu, Dec 13, 2012 at 2:08 PM, Andriy Gapon <avg at freebsd.org> wrote:
>>
>> It looks like in timer-r0drv-freebsd.c the code tries to pass absolute time as a
>> timeout parameter to callout_reset while that function actually expects relative
>> time (period).
>>
>> I am testing the following patch, but I am sure that the code can be made more
>> efficient.
>>
>> --- timer-r0drv-freebsd.c.orig 2012-12-12 20:13:27.623202784 +0200
>> +++ timer-r0drv-freebsd.c 2012-12-12 20:19:43.368202795 +0200
>> @@ -172,15 +172,16 @@
>> /*
>> * Calc when it should start firing.
>> */
>> - u64First += RTTimeNanoTS();
>> + const uint64_t u64Now = RTTimeNanoTS();
>> + u64First += u64Now;
>>
>> pTimer->fSuspended = false;
>> pTimer->iTick = 0;
>> pTimer->u64StartTS = u64First;
>> pTimer->u64NextTS = u64First;
>>
>> - tv.tv_sec = u64First / 1000000000;
>> - tv.tv_usec = (u64First % 1000000000) / 1000;
>> + tv.tv_sec = (u64First - u64Now) / 1000000000;
>> + tv.tv_usec = ((u64First - u64Now) % 1000000000) / 1000;
>> callout_reset(&pTimer->Callout, tvtohz(&tv), rtTimerFreeBSDCallback, pTimer);
>>
>> return VINF_SUCCESS;
>> @@ -247,8 +248,8 @@
>> if (pTimer->u64NextTS < u64NanoTS)
>> pTimer->u64NextTS = u64NanoTS + RTTimerGetSystemGranularity() / 2;
>>
>> - tv.tv_sec = pTimer->u64NextTS / 1000000000;
>> - tv.tv_usec = (pTimer->u64NextTS % 1000000000) / 1000;
>> + tv.tv_sec = (pTimer->u64NextTS - u64NanoTS) / 1000000000;
>> + tv.tv_usec = ((pTimer->u64NextTS - u64NanoTS) % 1000000000) / 1000;
>> callout_reset(&pTimer->Callout, tvtohz(&tv), rtTimerFreeBSDCallback, pTimer);
>> }
>
> What is your results from that tests? Is the patch correct so should we include
> it into the port?
I believe that the patch is correct.
I haven't seen any regressions nor any visible improvements from it.
--
Andriy Gapon
More information about the freebsd-emulation
mailing list