Re: svn commit: r219676 - head/sys/x86/x86

2011-03-24 Thread Bruce Evans
On Thu, 24 Mar 2011, David Malone wrote: If atomic 64-bit read/write is possible and it is absolutely necessary, it has to be set via machdep.foo_freq, not kern.timecounter.foo.frequency. Not really. The i386 TSC frequency variable can be 32 bits. That works in almost all cases now, and you

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-24 Thread David Malone
> If atomic 64-bit read/write is possible and it is absolutely > necessary, it has to be set via machdep.foo_freq, not > kern.timecounter.foo.frequency. A number of people have asked me about adjusting the frequency of other time counters - it would seem to make sense to allow this at the timeco

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-24 Thread Jung-uk Kim
On Thursday 24 March 2011 09:58 am, David Malone wrote: > > No new ones. Just the old ones of updating the frequency and all > > related variables without races. > > Is something simple, line an atomic write, an option for this? I > guess we don't have 64-bit atomic writes on all platforms, but we

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-24 Thread David Malone
> No new ones. Just the old ones of updating the frequency and all related > variables without races. Is something simple, line an atomic write, an option for this? I guess we don't have 64-bit atomic writes on all platforms, but we could support it on some platforms. David.

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-21 Thread Bruce Evans
On Sun, 20 Mar 2011, David Malone wrote: On Wed, Mar 16, 2011 at 04:34:10PM +1100, Bruce Evans wrote: Please revert. I use this for setting timecounters from userland. The sysctl is not a hack, but the primary user interface for managing the TSC frequency. The kern.timecounter.tc.TSC.frequen

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-20 Thread David Malone
On Wed, Mar 16, 2011 at 04:34:10PM +1100, Bruce Evans wrote: > Please revert. I use this for setting timecounters from userland. The > sysctl is not a hack, but the primary user interface for managing the > TSC frequency. The kern.timecounter.tc.TSC.frequency sysctl is > secondary, and is readon

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-15 Thread Bruce Evans
On Tue, 15 Mar 2011, Jung-uk Kim wrote: On Tuesday 15 March 2011 05:01 pm, John Baldwin wrote: On Tuesday, March 15, 2011 3:47:20 pm Jung-uk Kim wrote: Author: jkim Date: Tue Mar 15 19:47:20 2011 New Revision: 219676 URL: http://svn.freebsd.org/changeset/base/219676 Log: Do not let machdep.

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-15 Thread Jung-uk Kim
On Tuesday 15 March 2011 05:01 pm, John Baldwin wrote: > On Tuesday, March 15, 2011 3:47:20 pm Jung-uk Kim wrote: > > Author: jkim > > Date: Tue Mar 15 19:47:20 2011 > > New Revision: 219676 > > URL: http://svn.freebsd.org/changeset/base/219676 > > > > Log: > > Do not let machdep.tsc_freq modify

Re: svn commit: r219676 - head/sys/x86/x86

2011-03-15 Thread John Baldwin
On Tuesday, March 15, 2011 3:47:20 pm Jung-uk Kim wrote: > Author: jkim > Date: Tue Mar 15 19:47:20 2011 > New Revision: 219676 > URL: http://svn.freebsd.org/changeset/base/219676 > > Log: > Do not let machdep.tsc_freq modify tsc_freq itself. It is bad for i386 as > it does not operate atomic

svn commit: r219676 - head/sys/x86/x86

2011-03-15 Thread Jung-uk Kim
Author: jkim Date: Tue Mar 15 19:47:20 2011 New Revision: 219676 URL: http://svn.freebsd.org/changeset/base/219676 Log: Do not let machdep.tsc_freq modify tsc_freq itself. It is bad for i386 as it does not operate atomically. Actually, it serves no purpose. Noticed by: bde Modified: