Commit Graph

3609 Commits

Author SHA1 Message Date
John Baldwin
cebc7fb16c Improve the handling of cpuset with interrupts.
- For x86, change the interrupt source method to assign an interrupt source
  to a specific CPU to return an error value instead of void, thus allowing
  it to fail.
- If moving an interrupt to a CPU fails due to a lack of IDT vectors in the
  destination CPU, fail the request with ENOSPC rather than panicing.
- For MSI interrupts on x86 (but not MSI-X), only allow cpuset to be used
  on the first interrupt in a group.  Moving the first interrupt in a group
  moves the entire group.
- Use the icu_lock to protect intr_next_cpu() on x86 instead of the
  intr_table_lock to fix a LOR introduced in the last set of MSI changes.
- Add a new privilege PRIV_SCHED_CPUSET_INTR for using cpuset with
  interrupts.  Previously, binding an interrupt to a CPU only performed a
  privilege check if the interrupt had an interrupt thread.  Interrupts
  without a thread could be bound by non-root users as a result.
- If an interrupt event's assign_cpu method fails, then restore the original
  cpuset mask for the associated interrupt thread.

Approved by:	re (kib)
2009-07-01 17:20:07 +00:00
Alexander Motin
9f23a6caa4 Make algorithm a bit more bulletproof. 2009-06-23 23:16:37 +00:00
Alexander Motin
96c5d068d8 Rework r193814:
While general idea of patch was good, it was not working properly due the way
it was implemented. When we are using same timer interrupt for several of
hard/prof/stat purposes we should not send several IPIs same time to other
CPUs. Sending several IPIs same time leads to terrible accounting/profiling
results due to strong synchronization effect, when the second interrupt
handler accounts processing of the first one.
Interlink timer events in a such way, that no more then one IPI is sent for
any original timer interrupt.
2009-06-23 21:45:33 +00:00
Ariff Abdullah
b65cb1db3c When using i8254 as the only kernel timer source:
- Interpolate stat/prof clock using clkintr() in a similar fashion to
  local APIC timer, since statclock usually run slower.

- Liberate hardclockintr() from taking the burden of handling both stat
  and prof clock interrupt. Instead, send IPIs within clkintr() to handle
  those.
2009-06-09 07:26:52 +00:00
Xin LI
38676b52dd Add line width calculations for 15/16 and 24/32 bit modes in case
the "Get Scan Line Length" function fails, as it does in Parallels
(in Version 2.2, Build 2112 at least).

PR:		i386/127367
Obtained from:	DragonFly
Submitted by:	Pedro Giffuni
MFC after:	1 month
2009-06-09 00:54:57 +00:00
Alexander Motin
1703f2b424 Rename statclock_disable variable to atrtcclock_disable that it actually is,
and hide it inside of atrtc driver. Add new tunable hint.atrtc.0.clock
controlling it. Setting it to 0 disables using RTC clock as stat-/
profclock sources.

Teach i386 and amd64 SMP platforms to emulate stat-/profclocks using i8254
hardclock, when LAPIC and RTC clocks are disabled.

This allows to reduce global interrupt rate of idle system down to about
100 interrupts per core, permitting C3 and deeper C-states provide maximum
CPU power efficiency.
2009-05-03 17:47:21 +00:00
Alexander Motin
a40d9024df Add support for using i8254 and rtc timers as event sources for i386 SMP
system. Redistribute hard-/stat-/profclock events to other CPUs using IPI.
2009-05-02 12:59:47 +00:00
John Baldwin
b9dda9d6fe Fix a few nits in the earlier changes to prevent local information leakage
in AMD FPUs:
- Do not clear the affected state in the case that the FPU registers for
  the thread that already owns the FPU are changed via fpu_setregs().  The
  only local information the thread would see is its own state in that
  case.
- Fix a type mismatch for the dummy variable used in a "fld".  It accepts
  a float, not a double.

Reviewed by:	bde
Approved by:	so (cperciva)
MFC after:	1 month
2009-03-25 22:08:30 +00:00
John Baldwin
63de9515b7 Rename (fpu|npx)_cleanstate to (fpu|npx)_initialstate to better reflect
their purpose.

Inspired by:	bde
MFC after:	1 month
2009-03-25 14:17:08 +00:00
John Baldwin
2ee8325f42 A better fix for handling different FPU initial control words for different
ABIs:
- Store the FPU initial control word in the pcb for each thread.
- When first using the FPU, load the initial control word after restoring
  the clean state if it is not the standard control word.
- Provide a correct control word for Linux/i386 binaries under
  FreeBSD/amd64.
- Adjust the control word returned for fpugetregs()/npxgetregs() when a
  thread hasn't used the FPU yet to reflect the real initial control
  word for the current ABI.
- The Linux/i386 ABI for FreeBSD/i386 now properly sets the right control
  word instead of trashing whatever the current state of the FPU is.

Reviewed by:	bde
2009-03-05 19:42:11 +00:00
John Baldwin
20e9dede5e Some cleanups to the i386 FPU support:
- Remove the control word parameter to npxinit().  It was always set
  to __INITIAL_NPXCW__.
- Remove npx_cleanstate_ready as the cleanstate is always initalized
  when it is used.
- Improve the handling of the case when the FPU isn't present.  Now
  the npx0 device no longer succeeds in its probe so all of npx_attach()
  is skipped.  Also, we allow this case with SMP (though that shouldn't
  actually occur as all i386 systems that support SMP have FPUs) now.
  SMP was only an issue back when we had an FPU emulator which was not
  per-CPU.
- MFamd64: Clear some of the state in npx_cleanstate rather than leaving
  it as garbage.
- MFamd64: When a user thread first uses the FPU, use npx_cleanstate for
  the initial FPU state.

Reviewed by:	bde
2009-03-05 18:32:43 +00:00
Roman Divacky
af83f5d77c Change the functions to ANSI in those cases where it breaks promotion
to int rule. See ISO C Standard: SS6.7.5.3:15.

Approved by:	kib (mentor)
Reviewed by:	warner
Tested by:	silence on -current
2009-02-24 18:09:31 +00:00
John Baldwin
0b7dc0a7c6 Some whitespace and style fixes.
Submitted by:	bde (partly)
2009-02-23 15:39:24 +00:00
Kip Macy
3a6d1fcf9c merge 186535, 186537, and 186538 from releng_7_xen
Log:
 - merge in latest xenbus from dfr's xenhvm
 - fix race condition in xs_read_reply by converting tsleep to mtx_sleep

Log:
 unmask evtchn in bind_{virq, ipi}_to_irq

Log:
 - remove code for handling case of not being able to sleep
 - eliminate tsleep - make sleeps atomic
2008-12-29 06:31:03 +00:00
Warner Losh
0c3d9d1f6b style(9) nit: remove unnecessary {} pair. 2008-10-28 04:32:41 +00:00
Jung-uk Kim
29462bea1e Turn off CPU frequency change notifiers when the TSC is P-state invariant
or it is forced by setting 'kern.timecounter.invariant_tsc' tunable
to non-zero.
2008-10-21 00:38:00 +00:00
Yoshihiro Takahashi
05165e3276 - Add the i386_memio_map_load() function to load I/O address table.
- Add the bus_space_compare macro for bus_space consistency.
- Switch using the bus_space_map_load() in isa_load_resourcev().
2008-09-07 04:44:24 +00:00
John Baldwin
aa7c1c059f Add a very simple dpms(4) driver that uses the VESA BIOS DPMS calls to
turn off the external display during suspend and restore it to its
original state on resume.

MFC after:	2 weeks
2008-08-23 21:00:40 +00:00
Kip Macy
24b7d5cd1a Call in to xen for fpu handling when XEN is set
MFC after:	1 month
2008-08-15 21:43:38 +00:00
John Birrell
f1bd3c150c Add a cyclic hook for DTrace. 2008-05-24 06:27:54 +00:00
Poul-Henning Kamp
9b4a8ab7ba Now that all platforms use genclock, shuffle things around slightly
for better structure.

Much of this is related to <sys/clock.h>, which should really have
been called <sys/calendar.h>, but unless and until we need the name,
the repocopy can wait.

In general the kernel does not know about minutes, hours, days,
timezones, daylight savings time, leap-years and such.  All that
is theoretically a matter for userland only.

Parts of kernel code does however care: badly designed filesystems
store timestamps in local time and RTC chips almost universally
track time in a YY-MM-DD HH:MM:SS format, and sometimes in local
timezone instead of UTC.  For this we have <sys/clock.h>

<sys/time.h> on the other hand, deals with time_t, timeval, timespec
and so on.  These know only seconds and fractions thereof.

Move inittodr() and resettodr() prototypes to <sys/time.h>.
Retain the names as it is one of the few surviving PDP/VAX references.

Move startrtclock() to <machine/clock.h> on relevant platforms, it
is a MD call between machdep.c/clock.c.  Remove references to it
elsewhere.

Remove a lot of unnecessary <sys/clock.h> includes.

Move the machdep.disable_rtc_set sysctl to subr_rtc.c where it belongs.
XXX: should be kern.disable_rtc_set really, it's not MD.
2008-04-22 19:38:30 +00:00
Poul-Henning Kamp
36bff1ebfb Convert amd64 and i386 to share the atrtc device driver. 2008-04-14 08:00:00 +00:00
Poul-Henning Kamp
2946435299 Move i386 to generic RTC handling code.
Make clock_if.m and subr_rtc.c standard on i386

Add hints for "atrtc" driver, for non-PnP, non-ACPI systems.
NB: Make sure to install GENERIC.hints into /boot/device.hints in these!

Nuke MD inittodr(), resettodr() functions.

Don't attach to PHP0B00 in the "attimer" dummy driver any more, and remove
comments that no longer apply for that reason.

Add new "atrtc" device driver, which handles IBM PC AT Real Time
Clock compatible devices using subr_rtc and clock_if.

This driver is not entirely clean: other code still fondles the
hardware to get a statclock interrupt on non-ACPI timer systems.

Wrap some overly long lines.

After it has settled in -current, this will be ported to amd64.

Technically this is MFC'able, but I fail to see a good reason.
2008-04-12 20:46:06 +00:00
Poul-Henning Kamp
dad3b6c6fd Back in the good old days, PC's had random pieces of rock for
frequency generation and what frequency the generated was anyones
guess.

In general the 32.768kHz RTC clock x-tal was the best, because that
was a regular wrist-watch Xtal, whereas the X-tal generating the
ISA bus frequency was much lower quality, often costing as much as
several cents a piece, so it made good sense to check the ISA bus
frequency against the RTC clock.

The other relevant property of those machines, is that they
typically had no more than 16MB RAM.

These days, CPU chips croak if their clocks are not tightly within
specs and all necessary frequencies are derived from the master
crystal by means if PLL's.

Considering that it takes on average 1.5 second to calibrate the
frequency of the i8254 counter, that more likely than not, we will
not actually use the result of the calibration, and as the final
clincher, we seldom use the i8254 for anything besides BEL in
syscons anyway, it has become time to drop the calibration code.

If you need to tell the system what frequency your i8254 runs,
you can do so from the loader using hw.i8254.freq or using the
sysctl kern.timecounter.tc.i8254.frequency.
2008-03-26 22:12:00 +00:00
Poul-Henning Kamp
e465985885 The "free-lance" timer in the i8254 is only used for the speaker
these days, so de-generalize the acquire_timer/release_timer api
to just deal with speakers.

The new (optional) MD functions are:
	timer_spkr_acquire()
	timer_spkr_release()
and
	timer_spkr_setfreq()

the last of which configures the timer to generate a tone of a given
frequency, in Hz instead of 1/1193182th of seconds.

Drop entirely timer2 on pc98, it is not used anywhere at all.

Move sysbeep() to kern/tty_cons.c and use the timer_spkr*() if
they exist, and do nothing otherwise.

Remove prototypes and empty acquire-/release-timer() and sysbeep()
functions from the non-beeping archs.

This eliminate the need for the speaker driver to know about
i8254frequency at all.  In theory this makes the speaker driver MI,
contingent on the timer_spkr_*() functions existing but the driver
does not know this yet and still attaches to the ISA bus.

Syscons is more tricky, in one function, sc_tone(), it knows the hz
and things are just fine.

In the other function, sc_bell() it seems to get the period from
the KDMKTONE ioctl in terms if 1/1193182th second, so we hardcode
the 1193182 and leave it at that.  It's probably not important.

Change a few other sysbeep() uses which obviously knew that the
argument was in terms of i8254 frequency, and leave alone those
that look like people thought sysbeep() took frequency in hertz.

This eliminates the knowledge of i8254_freq from all but the actual
clock.c code and the prof_machdep.c on amd64 and i386, where I think
it would be smart to ask for help from the timecounters anyway [TBD].
2008-03-26 20:09:21 +00:00
Poul-Henning Kamp
ebfbcd612a Rename timer0_max_count to i8254_max_count.
Rename timer0_real_max_count to i8254_real_max_count and make it static.
Rename timer_freq to i8254_freq and make it a loader tunable.
2008-03-26 15:03:24 +00:00
Poul-Henning Kamp
f168bfa529 The RTC related pscnt and psdiv variables have no business being public. 2008-03-26 13:25:27 +00:00
Christian Brueffer
662cac9f23 Fix some "in in" typos in comments.
PR:		121490
Submitted by:	Anatoly Borodin <anatoly.borodin@gmail.com>
Approved by:	rwatson (mentor), jkoshy
MFC after:	3 days
2008-03-26 07:32:08 +00:00
John Baldwin
eb2b0540e5 Explicitly use spinlock_enter/exit rather than locking the icu_lock spin
lock in the 8259A drivers as these drivers are only used on UP systems.
This slightly reduces the penalty of an SMP kernel (such as GENERIC) on
a UP x86 machine.
2008-03-20 21:53:27 +00:00
Robert Watson
237fdd787b In keeping with style(9)'s recommendations on macros, use a ';'
after each SYSINIT() macro invocation.  This makes a number of
lightweight C parsers much happier with the FreeBSD kernel
source, including cflow's prcc and lxr.

MFC after:	1 month
Discussed with:	imp, rink
2008-03-16 10:58:09 +00:00
John Baldwin
c05655bfda Use cpu_spinwait() (i.e., "pause") when spinning on rdtsc during DELAY().
MFC after:	1 week
2008-01-17 18:59:38 +00:00
Bruce Evans
d5c90663b2 Don't use plain "ret" instructions at targets of jump instructions,
since the branch caches on at least Athlon XP through Athlon 64 CPU's
don't understand such instructions and guarantee a cache miss taking
at least 10 cycles.  Use the documented workaround "ret $0" instead
("nop; ret" also works, but "ret $0" is probably faster on old CPUs).

Normal code (even asm code) doesn't branch to "ret", since there is
usually some cleanup to do, but the __mcount, .mcount and .mexitcount
entry points were optimized too well to have the minimum number of
instructions (3 instructions each if profiling is not enabled) and
they did this.  I didn't see a significant number of cache misses for
.mexitcount, but for the shared "ret" for __mcount and .mcount I
observed cache misses costing 26 cycles each.  For a send(2) syscall
that makes about 70 function calls, the cost of these cache misses
alone increased the syscall time from about 4000 cycles to about 7000
cycles.  4000 is for a profiling (GUPROF) kernel with profiling disabled;
after this fix, configuring profiling only costs about 600 cycles in the
4000, which is consistent with almost perfect branch prediction in the
mcounting calls.
2007-11-29 02:01:21 +00:00
Bruce Evans
7e7c8806bf Remove entry points for -finstrument functions since they are currently
unused except to obfuscate disassemblies.  -mprofiler-epilogue is
currently with gcc-4 (it does too little), but -finstrument-functions
is broken in a different way (it does too much).

amd64 version: meger whitespace fixes from i386 version.
2007-11-29 01:15:03 +00:00
Peter Wemm
d556638404 Split /dev/nvram driver out of isa/clock.c for i386 and amd64. I have not
refactored it to be a generic device.
Instead of being part of the standard kernel, there is now a 'nvram' device
for i386/amd64.  It is in DEFAULTS like io and mem, and can be turned off
with 'nodevice nvram'.  This matches the previous behavior when it was
first committed.
2007-10-26 03:23:54 +00:00
Yoshihiro Takahashi
7b226dfaa8 Fix a kernel panic due to a NULL pointer access on pc98.
When any PnP device exists, isa_release_resource() is called with no
activated resource.  So a bushandle is not allocated yet.

Approved by:	re (kensmith)
2007-09-01 12:18:28 +00:00
David Malone
9be70a793e It seems that some i386 mothermoards either do not implement the
day of week field correctly, or they remember bad values that are
written into the day of week field. For this reason, ignore the day
of week field when reading the clock on i386 rather than bailing if
it is set incorrectly.

Problems were seen on a number of platforms, including VMWare, qemu,
EPIA ME6000, Epox-3PTA and ABIT-SL30T.

This is a slightly different fix to that proposed by Ted in his PR,
but the same basic idea.

PR:		111117
Submitted by:	Ted Faber <faber@lunabase.org>
Approved by:	re (rwatson)
MFC after:	3 weeks
2007-07-27 09:34:42 +00:00
David Malone
6d8617d42a If clock_ct_to_ts fails to convert time time from the real time clock,
print a one line error message. Add some comments on not being able to
trust the day of week field (I'll act on these comments in a follow up
commit).

Approved by:	re
MFC after:	3 weeks
2007-07-23 09:42:32 +00:00
Peter Wemm
5915fb72fb Prototype (but functional) Linux-ish /dev/nvram interface to the extra
114 bytes of cmos ram in the PC clock chip.  The big difference between
this and the Linux version is that we do not recalculate the checksums
for bytes 16..31.

We use this at work when cloning identical machines - we can copy the
bios settings as well.  Reading /dev/nvram gives 114 bytes of data but
you can seek/read/write whichever bytes you like.

Yes, this is a "foot, gun, fire!" type of device.
2007-06-15 22:58:14 +00:00
Jeff Roberson
982d11f836 Commit 14/14 of sched_lock decomposition.
- Use thread_lock() rather than sched_lock for per-thread scheduling
   sychronization.
 - Use the per-process spinlock rather than the sched_lock for per-process
   scheduling synchronization.

Tested by:      kris, current@
Tested on:      i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-05 00:00:57 +00:00
David Malone
041b706b2f Despite several examples in the kernel, the third argument of
sysctl_handle_int is not sizeof the int type you want to export.
The type must always be an int or an unsigned int.

Remove the instances where a sizeof(variable) is passed to stop
people accidently cut and pasting these examples.

In a few places this was sysctl_handle_int was being used on 64 bit
types, which would truncate the value to be exported.  In these
cases use sysctl_handle_quad to export them and change the format
to Q so that sysctl(1) can still print them.
2007-06-04 18:25:08 +00:00
John Baldwin
fb610ca1f9 Minor fixes and tweaks to the x86 interrupt code:
- Split the intr_table_lock into an sx lock used for most things, and a
  spin lock to protect intrcnt_index.  Originally I had this as a spin lock
  so interrupt code could use it to lookup sources.  However, we don't
  actually do that because it would add a lot of overhead to interrupts,
  and if we ever do support removing interrupt sources, we can use other
  means to safely do so w/o locking in the interrupt handling code.
- Replace is_enabled (boolean) with is_handlers (a count of handlers) to
  determine if a source is enabled or not.  This allows us to notice when
  a source is no longer in use.  When that happens, we now invoke a new
  PIC method (pic_disable_intr()) to inform the PIC driver that the
  source is no longer in use.  The I/O APIC driver frees the APIC IDT
  vector when this happens.  The MSI driver no longer needs to have a
  hack to clear is_enabled during msi_alloc() and msix_alloc() as a result
  of this change as well.
- Add an apic_disable_vector() to reset an IDT vector back to Xrsvd to
  complement apic_enable_vector() and use it in the I/O APIC and MSI code
  when freeing an IDT vector.
- Add a new nexus hook: nexus_add_irq() to ask the nexus driver to add an
  IRQ to its irq_rman.  The MSI code uses this when it creates new
  interrupt sources to let the nexus know about newly valid IRQs.
  Previously the msi_alloc() and msix_alloc() passed some extra stuff
  back to the nexus methods which then added the IRQs.  This approach is
  a bit cleaner.
- Change the MSI sx lock to a mutex.  If we need to create new sources,
  drop the lock, create the required number of sources, then get the lock
  and try the allocation again.
2007-05-08 21:29:14 +00:00
Nate Lawson
0d4ac62a35 Add an interface for drivers to be notified of changes to CPU frequency.
cpufreq_pre_change is called before the change, giving each driver a chance
to revoke the change.  cpufreq_post_change provides the results of the
change (success or failure).  cpufreq_levels_changed gives the unit number
of the cpufreq device whose number of available levels has changed.  Hook
in all the drivers I could find that needed it.

* TSC: update TSC frequency value.  When the available levels change, take the
highest possible level and notify the timecounter set_cputicker() of that
freq.  This gets rid of the "calcru: runtime went backwards" messages.
* identcpu: updates the sysctl hw.clockrate value
* Profiling: if profiling is active when the clock changes, let the user
know the results may be inaccurate.

Reviewed by:	bde, phk
MFC after:	1 month
2007-03-26 18:03:29 +00:00
Bruce Evans
d78180f8f5 Partial fix for a bug in rev.1.231. If suspend/resume clobbers the
RTC state, then it may clobber the RTC index register, so the index
register must be restored before using it to restore control registers
in rtc_restore().

The following problems remain:
- rtc_restore() is only called if pmtimer is configured.  Buggy
  suspend/resumes are more likely to clobber the index register than
  a control register, so pmtimer is more needed than it used to be.
- pmtimer doesn't exist for amd64.
- Restoring of the RTC state may race with rtcintr().  If an RTC
  interrupt is handled before the state is restored, then rtcin(RTC_INTR)
  in rtcintr() may read from the wrong register, so rtcintr() may spin
  forever.  This may be mitigated by the most common state clobbering
  being to turn off RTC interrupts.
2007-03-05 09:10:17 +00:00
Yoshihiro Takahashi
de038d5ffc style(9). 2007-03-04 04:55:19 +00:00
Paolo Pisati
ef544f6312 o break newbus api: add a new argument of type driver_filter_t to
bus_setup_intr()

o add an int return code to all fast handlers

o retire INTR_FAST/IH_FAST

For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current

Reviewed by: many
Approved by: re@
2007-02-23 12:19:07 +00:00
Bruce Evans
71799af2d5 Cleaned up declaration and initialization of clock_lock. It is only
used by clock code, so don't export it to the world for machdep.c to
initialize.  There is a minor problem initializing it before it is
used, since although clock initialization is split up so that parts
of it can be done early, the first part was never done early enough
to actually work.  Split it up a bit more and do the first part as
late as possible to document the necessary order.  The functions that
implement the split are still bogusly exported.

Cleaned up initialization of the i8254 clock hardware using the new
split.  Actually initialize it early enough, and don't work around it
not being initialized in DELAY() when DELAY() is called early for
initialization of some console drivers.

This unfortunately moves a little more code before the early debugger
breakpoint so that it is harder to debug.  The ordering of console and
related initialization is delicate because we want to do as little as
possible before the breakpoint, but must initialize a console.
2007-01-23 08:01:20 +00:00
Kip Macy
a5c5d4402c Evidently FreeBSD has long relied on the compiler to treat structures
passed by value (trap frames) as if they were in fact being passed by
reference. For better or worse, this incorrect behaviour is no longer
present in gcc 4.1. In this patch I convert all trapframe arguments to
be explicitly pass by reference. I also remove vm86_initflags, pushing
the very little work that it actually does up into vm86_prepcall.

Reviewed by: kan
Tested by: kan
2006-12-17 05:07:01 +00:00
Bruce Evans
b73057227b Optimized RTC accesses by avoiding null writes to the index register
and by only delaying when an RTC register is written to.  The delay
after writing to the data register is now not just a workaround.

This reduces the number of ISA accesses in the usual case from 4 to
1.  The usual case is 2 rtcin()'s for each RTC interrupt.  The index
register is almost always RTC_INTR for this.  The 3 extra ISA accesses
were 1 for writing the index and 2 for delays.  Some delays are needed
in theory, but in practice they now just slow down slow accesses some
more since almost eveyone including us does them wrong so modern systems
enforce sufficient delays in hardware.  I used to have the delays ifdefed
out, but with the index register optimization the delays are rarely
executed so the old magic ones can be kept or even implemented non-
magically without significant cost.

Optimizing RTC interrupt handling is more interesting than it used to
be because RTC interrupts are currently needed to fix the more efficient
apic timer interrupts on some systems.  apic_timer_hz is normally 2000
so the RTC interrupt rate needs to be 2048 to keep the apic timer
firing on such systems.  Without these changes, each RTC interrupt
normally took 10 ISA accesses (2 PIC accesses and 2 sets of 4 RTC
accesses).  Each ISA access takes 1-1.5uS so 10 of then at 2048 Hz
takes 2-3% of a CPU.  Now 4 of them take 0.8-1.2% of a CPU.
2006-12-03 03:49:28 +00:00
Bruce Evans
6a70163fcc Removed some SMP ifdefs so that using the TSC as a cputime clock is
not completely decided at config time.  Just don't default to using
the TSC if there are multiple active CPUs.  Also, don't default to
using the TSC if it is broken.  SMP ifdefs are still used to disallow
using perfmon since perfmon is always broken if SMP is just configured.

This only helps much for SMP kernels running on 1 CPU.  The overheads
for using the i8254 cputime clock were a bit too high on 486/33's, and
now on multi-GHz CPUs they are usually in the 99-99.9% range.  Switching
from the old default of an i8254 clock to the TSC works poorly because
the overheads are not recalibrated.

Use the same condition for declaring perfmon stuff as for using it.
2006-10-29 09:48:44 +00:00
Bruce Evans
3a110062fd Cleaned up includes. <machine/profile.h> was unused. <machine/timerreg.h>
was only used in the GUPROF case, so the messes to get its i386 prerequisites
included shouldn't have been needed.

Fixed some style bugs. Quote #error contents, and don't repeat an #error
directive on amd64.
2006-10-28 06:38:51 +00:00