smp_active = 1 used to indicate that the system had frozen previously
started AP's, while smp_active = 0 was "AP's not yet started". I have split
this into smp_started (which is set when the AP's come online), and
smp_active is left for turning on/off AP scheduling.
- We now have enough per-cpu idle context, the real idle loop has been
revived (cpu's halt now with nothing to do).
- Some preliminary support for running some operations outside the
global lock (eg: zeroing "free but not yet zeroed pages") is present
but appears to cause problems. Off by default.
- the smp_active sysctl now behaves differently. It's merely a 'true/false'
option. Setting smp_active to zero causes the AP's to halt in the idle
loop and stop scheduling processes.
- bootstrap is a lot safer. Instead of sharing a statically compiled in
stack a number of times (which has caused lots of problems) and then
abandoning it, we use the idle context to boot the AP's directly. This
should help >2 cpu support since the bootlock stuff was in doubt.
- print physical apic id in traps.. helps identify private pages getting
out of sync. (You don't want to know how much hair I tore out with this!)
More cleanup to follow, this is more of a checkpoint than a
'finished' thing.
Added a new variable, 'bsp_apic_ready', which is set as soon as the bootstrap
CPU has initialized its local APIC. Conditionalize the GENSPLR functions
to call ss_lock ONLY after bsp_apic_ready is TRUE; This should prevent
any problems with races between the time the 1st AP becomes ready and the
time smp_active is set.
Made NEW_STRATEGY default.
Removed misc. old cruft.
Centralized simple locks into mp_machdep.c
Centralized simple lock macros into param.h
More cleanup in the direction of making splxx()/cpl MP-safe.
Several new fine-grained locks.
New FAST_INTR() methods:
- separate simplelock for FAST_INTR, no more giant lock.
- FAST_INTR()s no longer checks ipending on way out of ISR.
sio made MP-safe (I hope).
We now tsleep() in kthread_init() between start_init()
and prepare_usermode() while waiting for ALL the idle_loop()
processes to come online.
Debugged & tested by: "Thomas D. Dean" <tomdean@ix.netcom.com>
Reviewed by: David Greenman <dg@root.com>
Work done by BSDI, Jonathan Lemon <jlemon@americantv.com>,
Mike Smith <msmith@gsoft.com.au>, Sean Eric Fagan <sef@kithrup.com>,
and probably alot of others.
Submitted by: Jnathan Lemon <jlemon@americantv.com>
This code was eliminated when the PEND_INTS algorithm was added. But it was
discovered that PEND_INTS only worsen latency for FAST_INTR() routines,
which can't be marked pending.
Noticed & debugged by: dave adkins <adkin003@gold.tc.umn.edu>
- removed TEST_ALTTIMER.
- removed APIC_PIN0_TIMER.
- removed TIMER_ALL.
mplock.s:
- minor update of try_mplock for new algorithm where a CPU uses try_mplock
instead of get_mplock in the ISRs.
Macros to convert the Lite2 lock manager primitives to the names used
in the kernel proper. This allows us to hide them from the lock
manager till they can be turned on.
smp.h:
declarations for the new simplelock functions.
- s_lock_init()
- s_lock()
- s_lock_try()
- s_unlock()
Created lock for IO APIC and apic_imen (SMP version of imen)
- imen_lock
Code to use imen_lock for access from apic_ipl.s and apic_vector.s.
Moved this code *outside* of mp_lock.
It seems to work!!!
1) Make sure that the region mapped by a 4MB page is
properly aligned.
2) Don't turn on the PG_G flag in locore for SMP. I plan
to do that later in startup anyway.
3) Make sure the 2nd processor has PSE enabled, so that 4MB
pages don't hose it.
We don't use PG_G yet on SMP -- there is work to be done to make that
work correctly. It isn't that important anyway...
of the kernel, and also most of the dynamic parts of the kernel. Additionally,
4MB pages will be allocated for display buffers as appropriate (only.)
The 4MB support for SMP isn't complete, but doesn't interfere with operation
either.
this code is controlled by smptests.h: TEST_CPUSTOP, OFF by default
new code for handling mixed-mode 8259/APIC programming without 'ExtInt'
this code is controlled by smptests.h: TEST_ALTTIMER, ON by default
- TEST_CPUSTOP adds stop_cpus()/restart_cpus(), OFF by default
- TEST_ALTTIMER new method for attaching 8259 PIC to APIC
this method avoids 'ExtInt' programming, ON by default
- TIMER_ALL sends 8259/8254 timer INTs to all CPUs, ON by default
- ASMPOSTCODExxx code to display bytes to POST hardware, OFF by default
General cleanup.
New functions to stop/start CPUs via IPIs:
- int stop_cpus( u_int map );
- int restart_cpus( u_int map );
Turned off by default, enabled via smptests.h:TEST_CPUSTOP.
Current version has a BUG, perhaps a deadlock?
Till now NMIs would be ignored. Now an NMI is caught by the BSP.
APs still ignore NMI, am working on code to allow a CPU to stop other CPUs
via an IPI.
available to the kernel (VM_KMEM_SIZE). The default (32 MB) is too low
when having 512 MB or more physical memory in a server environment. This is
relevant on systems where "panic: kmem_malloc: kmem_map too small" is a
problem.
This eliminates a lot of #ifdef SMP type code. Things like _curproc reside
in a data page that is unique on each cpu, eliminating the expensive macros
like: #define curproc (SMPcurproc[cpunumber()])
There are some unresolved bootstrap and address space sharing issues at
present, but Steve is waiting on this for other work. There is still some
strictly temporary code present that isn't exactly pretty.
This is part of a larger change that has run into some bumps, this part is
standalone so it should be safe. The temporary code goes away when the
full idle cpu support is finished.
Reviewed by: fsmp, dyson
cost since it is only done in cpu_switch(), not for every exception.
The extra state is kept in the pcb, and handled much like the npx state,
with similar deficiencies (the state is not preserved across signal
handlers, and error handling loses state).
Changes to pmap.c for lapic_t lapic && ioapic_t ioapic pointers,
currently equal to apic_base && io_apic_base, will stand alone with the
private page mapping.
apic.h has defines like:
#define lapic__id lapic->id
Once private pages and "known virtual addr" mapping of the APICs is
ready all 'lapic__XXX' will be changed to 'lapic.XXX', and the defines
will be removed.
Changes to smp.h for lapic_t lapic && ioapic_t ioapic pointers,
currently equal to apic_base && io_apic_base, will stand alone with the
private page mapping.
reality. There will be a new call interface, but for now the file
pci_compat.c (which is to be deleted, after all drivers are converted)
provides an emulation of the old PCI bus driver functions. The only
change that might be visible to drivers is, that the type pcici_t
(which had been meant to be just a handle, whose exact definition
should not be relied on), has been converted into a pcicfgregs* .
The Tekram AMD SCSI driver bogusly relied on the definition of pcici_t
and has been converted to just call the PCI drivers functions to access
configuration space register, instead of inventing its own ...
This code is by no means complete, but assumed to be fully operational,
and brings the official code base more in line with my development code.
A new generic device descriptor data type has to be agreed on. The PCI
code will then use that data type to provide new functionality:
1) userconfig support
2) "wired" PCI devices
3) conflicts checking against ISA/EISA
4) maps will depend on the command register enable bits
5) PCI to Anything bridges can be defined as devices,
and are probed like any "standard" PCI device.
The following features are currently missing, but will be added back,
soon:
1) unknown device probe message
2) suppression of "mirrored" devices caused by ancient, broken chip-sets
This code relies on generic shared interrupt support just commited to
kern_intr.c (plus the modifications of isa.c and isa_device.h).
This is now the default, it delays most of the MP startup to the function
machdep.c:cpu_startup(). It should be possible to move the 2 functions
found there (mp_start() & mp_announce()) even further down the path once
we know exactly where that should be...
Help from: Peter Wemm <peter@spinner.dialix.com.au>
- The 1st (preparse_mp_table()) counts the number of cpus, busses, etc. and
records the LOCAL and IO APIC addresses.
- The 2nd pass (parse_mp_table()) does the actual parsing of info and recording
into the incore MP table.
This will allow us to defer the 2nd pass untill malloc() & private pages
are available (but thats for another day!).
panic( "xxxxx\n" );
to:
printf( "xxxxx\n" );
panic( "\n" );
For some as yet undetermined reason the argument to panic() is often NOT
printed, and the system sometimes hangs before reaching the panic printout.
So we hopefully at least print some useful info before the hang, as oppossed to
leaving the user clueless as to what has happened.
all uses of it with the equivalent calls to setbits().
This change incidentally eliminates a problem building ELF kernels
that was caused by SETBITS.
Reviewed by: fsmp, peter
Submitted by: bde
simplifies some assumptions and stops some code compile problems.
This should fix the compile hiccup in PR#3491, but smp kernel profiling
isn't likely to be fixed by this.
- one-liners all become inline.
- multi-liners become functions.
- FAST_IPI defines go away.
re-worked APICIPI_BANDAID code.
- now refered to as DETECT_DEADLOCK.
- on by default.
This code re-numbers PCI busses in the MP table to match PCI semantics
when the MP BIOS fails to do it properly.
Reviewed by: Peter Wemm <peter@spinner.DIALix.COM>
Peter Wemm <peter@spinner.DIALix.COM>, Steve Passe <smp@csn.net>
removed all the IPI_INTS code.
made the XFAST_IPI32 code default, renaming Xfastipi32 to Xinvltlb.
There are various options documented in i386/conf/LINT, there is more to
come over the next few days.
The kernel should run pretty much "as before" without the options to
activate SMP mode.
There are a handful of known "loose ends" that need to be fixed, but
have been put off since the SMP kernel is in a moderately good condition
at the moment.
This commit is the result of the tinkering and testing over the last 14
months by many people. A special thanks to Steve Passe for implementing
the APIC code!
for syscalls, so one frame was lost in backtraces from syscalls.
This is handled better in the kernel by using a different mcount
entry point for profiling before the frame pointer is set up.
Expand RCSID().
Use .p2align instead of the ambiguous .align.
Added idempotency ifdef.
Removed unused macros ALTENTRY(), ALTASENTRY(), ASENTRY(), _MID_ENTRY.
Cleaned up formatting.
Reviewed by: jdp reviewed an old version
Obtained from: parts from NetBSD
have successfully built, booted, and run a number of different ELF
kernel configurations, including GENERIC. LINT also builds and
links cleanly, though I have not tried to boot it.
The impact on developers is virtually nil, except for two things.
All linker sets that might possibly be present in the kernel must be
listed in "sys/i386/i386/setdefs.h". And all C symbols that are
also referenced from assembly language code must be listed in
"sys/i386/include/asnames.h". It so happens that failure to do
these things will have no impact on the a.out kernel. But it will
break the build of the ELF kernel.
The ELF bootloader works, but it is not ready to commit quite yet.
Rename the PT* index KSTK* #defines to UMAX*, since we don't have a kernel
stack there any more..
These are used to calculate VM_MAXUSER_ADDRESS and USRSTACK, and really
do not want to be changed with UPAGES since BSD/OS 2.x binary compatability
depends on it.
space. (!)
Have each process use the kernel stack and pcb in the kvm space. Since
the stacks are at a different address, we cannot copy the stack at fork()
and allow the child to return up through the function call tree to return
to user mode - create a new execution context and have the new process
begin executing from cpu_switch() and go to user mode directly.
In theory this should speed up fork a bit.
Context switch the tss_esp0 pointer in the common tss. This is a lot
simpler since than swithching the gdt[GPROC0_SEL].sd.sd_base pointer
to each process's tss since the esp0 pointer is a 32 bit pointer, and the
sd_base setting is split into three different bit sections at non-aligned
boundaries and requires a lot of twiddling to reset.
The 8K of memory at the top of the process space is now empty, and unmapped
(and unmappable, it's higher than VM_MAXUSER_ADDRESS).
Simplity the pmap code to manage process contexts, we no longer have to
double map the UPAGES, this simplifies and should measuably speed up fork().
The following parts came from John Dyson:
Set PG_G on the UPAGES that are now in kernel context, and invalidate
them when swapping them out.
Move the upages object (upobj) from the vmspace to the proc structure.
Now that the UPAGES (pcb and kernel stack) are out of user space, make
rfork(..RFMEM..) do what was intended by sharing the vmspace
entirely via reference counting rather than simply inheriting the mappings.
convenient and makes life difficult for my next commit. We still need
an i386tss to point to for the tss slot in the gdt, so we use a common
tss shared between all processes.
Note that this is going to break debugging until this series of commits
is finished. core dumps will change again too. :-( we really need
a more modern core dump format that doesn't depend on the pcb/upages.
This change makes VM86 mode harder, but the following commits will remove
a lot of constraints for the VM86 system, including the possibility of
extending the pcb for an IO port map etc.
Obtained from: bde
supports All Cyrix CPUs, IBM Blue Lightning CPU and NexGen (now AMD)
Nx586 CPU, and initialize special registers of Cyrix CPU and msr of
IBM Blue Lightning CPU.
If revision of Cyrix 6x86 CPU < 2.7, CPU cache is enabled in
write-through mode. This can be disabled by kernel configuration
options.
Reviewed by: Bruce Evans <bde@freebsd.org> and
Jordan K. Hubbard <jkh@freebsd.org>
at runtime.
etc/make.conf:
Nuked HAVE_FPU option.
lib/msun/Makefile:
Always build the i387 objects. Copy the i387 source files at build
time so that the i387 objects have different names. This is simpler
than renaming the files in the cvs repository or repeating half of
bsd.lib.mk to add explicit rules.
lib/msun/src/*.c:
Renamed all functions that have an i387-specific version by adding
`__generic_' to their names.
lib/msun/src/get_hw_float.c:
New file for getting machdep.hw_float from the kernel.
sys/i386/include/asmacros.h:
Abuse the ENTRY() macro to generate jump vectors and associated code.
This works much like PIC PLT dynamic initialization. The PIC case is
messy. The old i387 entry points are renamed. Renaming is easier
here because the names are given by macro expansions.
Changed it from 4 to 16 for i386's. It can be anything for i386's,
but compiler options limit it to a power of 2, and assembler and
linker deficiencies limit it to a small power of 2 (<= 16).
We use 16 in the kernel to get smaller tables (see Makefile.i386 and
<machine/asmacros.h>). We still use the default of 4 in user mode.
Use HISTCOUNTER instead of (*kcount) in the definition of KCOUNT()
for consistency with other macros.
complained so it cannot be entirely bad :-)
I include the email that probably explains it for people who already know:
> >Compiling with -O3 inlines functions. However the function that is being
> >inlined in makeinfo.c (add_word_args()) is a vararg function and must not be
> >inlined.
> >
> >The code in question is K&R style, and AFIK, there is no way for the compiler
> >to determine that the function uses vararg. Either change the code to use
> >prototypes, or use stdarg, or add a directive to prevent inlining.
>
> Not declaring a varargs function as varargs before it is used gives
> undefined behaviour.
>
> However, in practice the bug is probably in FreeBSD's <varargs.h>, which
> doesn't use gcc's __builtin_next_arg(). gcc should notice that it is
> used and not inline functions that have it. <stdarg.h.> uses it, but I
> think there's another gcc builtin that it should be using.
Patch attached. The ellipsis causes gcc to flag this as a varargs function,
and the name "__builtin_va_alist" is special cased in gcc to hide the last
argument in the arglist.
Reviewed by: bde & phk
Submitted by: jlemon@americantv.com (Jonathan Lemon)
This will make a number of things easier in the future, as well as (finally!)
avoiding the Id-smashing problem which has plagued developers for so long.
Boy, I'm glad we're not using sup anymore. This update would have been
insane otherwise.
also implies VM_PROT_EXEC. We support it that way for now,
since the break system call by default gives VM_PROT_ALL. Now
we have a better chance of coalesing map entries when mixing
mmap/break type operations. This was contributing to excessive
numbers of map entries on the modula-3 runtime system. The
problem is still not "solved", but the situation makes more
sense.
Eventually, when we work on architectures where VM_PROT_READ
is orthogonal to VM_PROT_EXEC, we will have to visit this
issue carefully (esp. regarding security issues.)
(1) deleted #if 0
pc98/pc98/mse.c
(2) hold per-unit I/O ports in ed_softc
pc98/pc98/if_ed.c
pc98/pc98/if_ed98.h
(3) merge more files by segregating changes into headers.
new file (moved from pc98/pc98):
i386/isa/aic_98.h
deleted:
well, it's already in the commit message so I won't repeat the
long list here ;)
Submitted by: The FreeBSD(98) Development Team
I decided to do this for every hardclock() call instead of lazily
in microtime(). The lazy method is simpler but has more overhead
if microtime() is called a lot.
CPU_THISTICKLEN() is now a no-op and should probably go away.
Previously it did nothing directly but had the side effect of
setting i586_last_tick for CPU_CLOCKUPDATE() and i586_avg_tick for
debugging. CPU_CLOCKUPDATE() now uses a better method and
i586_avg_tick is too much trouble to maintain.
Reduced nesting of #includes in the usual case.
Increased nesting of #includes when CLOCK_HAIR is defined. This
is a kludge to get typedefs for inline functions only when the
inline functions are used. Normally only kern_clock.c defines
this. kern_clock.c can't include the i386 headers directly.
Removed unused LOCORE support.
- use a more accurate and more efficient method of compensating for
overheads. The old method counted too much time against leaf
functions.
- normally use the Pentium timestamp counter if available.
On Pentiums, the times are now accurate to within a couple of cpu
clock cycles per function call in the (unlikely) event that there
are no cache misses in or caused by the profiling code.
- optionally use an arbitrary Pentium event counter if available.
- optionally regress to using the i8254 counter.
- scaled the i8254 counter by a factor of 128. Now the i8254 counters
overflow slightly faster than the TSC counters for a 150MHz Pentium :-)
(after about 16 seconds). This is to avoid fractional overheads.
files.i386:
permon.c temporarily has to be classified as a profiling-routine
because a couple of functions in it may be called from profiling code.
options.i386:
- I586_CTR_GUPROF is currently unused (oops).
- I586_PMC_GUPROF should be something like 0x70000 to enable (but not
use unless prof_machdep.c is changed) support for Pentium event
counters. 7 is a control mode and the counter number 0 is somewhere
in the 0000 bits (see perfmon.h for the encoding).
profile.h:
- added declarations.
- cleaned up separation of user mode declarations.
prof_machdep.c:
Mostly clock-select changes. The default clock can be changed by
editing kmem. There should be a sysctl for this.
subr_prof.c:
- added copyright.
- calibrate overheads for the new method.
- documented new method.
- fixed races and and machine dependencies in start/stop code.
mcount.c:
Use the new overhead compensation method.
gmon.h:
- changed GPROF4 counter type from unsigned to int. Oops, this should
be machine-dependent and/or int32_t.
- reorganized overhead counters.
Submitted by: Pentium event counter changes mostly by wollman
previous snap. Specifically, kern_exit and kern_exec now makes a
call into the pmap module to do a very fast removal of pages from the
address space. Additionally, the pmap module now updates the PG_MAPPED
and PG_WRITABLE flags. This is an optional optimization, but helpful
on the X86.
- fixed a sloppy common-style declaration.
- removed an unused macro.
- moved once-used macros to the one file where they are used.
- removed unused forward struct declarations.
- removed __pure.
- declared inline functions as inline in their prototype as well
as in theire definition (gcc unfortunately allows the prototype
to be inconsistent).
- staticized.
dependent operation, and not really a correct name. invltlb and invlpg
are more descriptive, and in the case of invlpg, a real opcode.
Additionally, fix the tlb management code for 386 machines.
with this quite a while ago when somebody reported a BSD/OS 2.1 binary
that wouldn't run. I'm pretty sure they tried it and I'm pretty sure
they mentioned to me that the patch worked.
comparisons in the inb() and outb() macros. I decided that int args
are OK here. Any type that can hold a u_int16_t without overflow
is correct, and 32-bit types are optimal.
Introduced a few tens of warnings (100 in LINT) for use of pessimized
(short) types for the port arg. Only a few drivers are affected by
this. u_short pessimizations aren't detected.
Added `__extension__' before the statement-expression in inb() so
that it can be compiled without warnings by gcc -pedantic.
(1) Add PC98 support to apm_bios.h and ns16550.h, remove pc98/pc98/ic
(2) Move PC98 specific code out of cpufunc.h (to pc98.h)
(3) Let the boot subtrees look more alike
Submitted by: The FreeBSD(98) Development Team
<freebsd98-hackers@jp.freebsd.org>
Changed i586_ctr_bias from long long to u_int. Only the low 32 bits
are used now that microtime uses a multiplication to do the scaling.
Previously the high 32 bits had to match those of rdtsc() to prevent
overflow traps and invalid timeval adjustments.
problem with the 'shell scripts' was found, but there was a 'strange'
problem found with a 486 laptop that we could not find. This commit
backs the code back to 25-jul, and will be re-entered after the snapshot
in smaller (more easily tested) chunks.
performance issues.
1) The pmap module has had too many inlines, and so the
object file is simply bigger than it needs to be.
Some common code is also merged into subroutines.
2) Removal of some *evil* PHYS_TO_VM_PAGE macro calls.
Unfortunately, a few have needed to be added also.
The removal caused the need for more vm_page_lookups.
I added lookup hints to minimize the need for the
page table lookup operations.
3) Removal of some bogus performance improvements, that
mostly made the code more complex (tracking individual
page table page updates unnecessarily). Those improvements
actually hurt 386 processors perf (not that people who
worry about perf use 386 processors anymore :-)).
4) Changed pv queue manipulations/structures to be TAILQ's.
5) The pv queue code has had some performance problems since
day one. Some significant scalability issues are resolved
by threading the pv entries from the pmap AND the physical
address instead of just the physical address. This makes
certain pmap operations run much faster. This does
not affect most micro-benchmarks, but should help loaded system
performance *significantly*. DG helped and came up with most
of the solution for this one.
6) Most if not all pmap bit operations follow the pattern:
pmap_test_bit();
pmap_clear_bit();
That made for twice the necessary pv list traversal. The
pmap interface now supports only pmap_tc_bit type operations:
pmap_[test/clear]_modified, pmap_[test/clear]_referenced.
Additionally, the modified routine now takes a vm_page_t arg
instead of a phys address. This eliminates a PHYS_TO_VM_PAGE
operation.
7) Several rewrites of routines that contain redundant code to
use common routines, so that there is a greater likelihood of
keeping the cache footprint smaller.
