contributions or ideas from Stephen McKay <syssgm@devetir.qld.gov.au>,
Alan Cox <alc@cs.rice.edu>, David Greenman <davidg@freebsd.org> and me:
More usage of the TAILQ macros. Additional minor fix to queue.h.
Performance enhancements to the pageout daemon.
Addition of a wait in the case that the pageout daemon
has to run immediately.
Slightly modify the pageout algorithm.
Significant revamp of the pmap/fork code:
1) PTE's and UPAGES's are NO LONGER in the process's map.
2) PTE's and UPAGES's reside in their own objects.
3) TOTAL elimination of recursive page table pagefaults.
4) The page directory now resides in the PTE object.
5) Implemented pmap_copy, thereby speeding up fork time.
6) Changed the pv entries so that the head is a pointer
and not an entire entry.
7) Significant cleanup of pmap_protect, and pmap_remove.
8) Removed significant amounts of machine dependent
fork code from vm_glue. Pushed much of that code into
the machine dependent pmap module.
9) Support more completely the reuse of already zeroed
pages (Page table pages and page directories) as being
already zeroed.
Performance and code cleanups in vm_map:
1) Improved and simplified allocation of map entries.
2) Improved vm_map_copy code.
3) Corrected some minor problems in the simplify code.
Implemented splvm (combo of splbio and splimp.) The VM code now
seldom uses splhigh.
Improved the speed of and simplified kmem_malloc.
Minor mod to vm_fault to avoid using pre-zeroed pages in the case
of objects with backing objects along with the already
existant condition of having a vnode. (If there is a backing
object, there will likely be a COW... With a COW, it isn't
necessary to start with a pre-zeroed page.)
Minor reorg of source to perhaps improve locality of ref.
Macroize locore.s' page table setup even more, now it's almost readable.
Rename PG_U to PG_A (so that I can...)
Rename PG_u to PG_U. "PG_u" was just too ugly...
Remove some unused vars in pmap.c
Remove PG_KR and PG_KW
Remove SSIZE
Remove SINCR
Remove BTOPKERNBASE
This concludes my spring cleaning, modulus any bug fixes for messes I
have made on the way.
(Funny to be back here in pmap.c, that's where my first significant
contribution to 386BSD was... :-)
time. The results are currently ignored unless certain temporary options
are used.
Added sysctls to support reading and writing the clock frequency variables
(not the frequencies themselves). Writing is supposed to atomically
adjust all related variables.
machdep.c:
Fixed spelling of a function name in a comment so that I can log this
message which should have been with the previous commit.
Initialize `cpu_class' earlier so that it can be used in startrtclock()
instead of in calibrate_cyclecounter() (which no longer exists).
Removed range checking of `cpu'. It is always initialized to CPU_XXX
so it is less likely to be out of bounds than most variables.
clock.h:
Removed I586_CYCLECTR(). Use rdtsc() instead.
clock.c:
TIMER_FREQ is now a variable timer_freq that defaults to the old value of
TIMER_FREQ. #define'ing TIMER_FREQ should still work and may be the best
way of setting the frequency.
Calibration involves counting cycles while watching the RTC for one second.
This gives values correct to within (a few ppm) + (the innaccuracy of the
RTC) on my systems.
page dir+table index.
pmap.h: remove NUPDE, it was wrong and not used. Sanitize KSTKPTEOFF.
vmparam.h: Calculate virtual addr from PDI+PTI from pmap.h rather than
using magic math. Remove UPDT, not used.
regarding apm to LINT
- Disabled the statistics clock on machines which have an APM BIOS and
have the options "APM_BROKEN_STATCLOCK" enabled (which is default
in GENERIC now)
- move around some of the code in clock.c dealing with the rtc to make
it more obvios the effects of disabling the statistics clock
Reviewed by: bde
in a suboptimal manner. I had also noticed some panics that appeared
to be at least superficially caused by this problem. Also, included
are some minor mods to support more general handling of page table page
faulting. More details in a future commit.
Always delay using one inb(0x84) after each i/o in rtcin() - don't
do this conditional on the bogus option DUMMY_NOPS not being defined.
If you want an optionally slightly faster rtcin() again, then inline
it and use a better named option or sysctl variable. It only needs
to be fast in rtcintr().
(This code is as yet untested; to come after man page is written.)
This also adds inlines to cpufunc.h for the RDTSC, RDMSR, WRMSR, and RDPMC
instructions. The user-mode interface is via a subdevice of mem.c;
there is also a kernel-size interface which might be used to aid
profiling.
netscape-2.0 for Linux running all the Java stuff. The scrollbars are now
working, at least on my machine. (whew! :-)
I'm uncomfortable with the size of this commit, but it's too
inter-dependant to easily seperate out.
The main changes:
COMPAT_LINUX is *GONE*. Most of the code has been moved out of the i386
machine dependent section into the linux emulator itself. The int 0x80
syscall code was almost identical to the lcall 7,0 code and a minor tweak
allows them to both be used with the same C code. All kernels can now
just modload the lkm and it'll DTRT without having to rebuild the kernel
first. Like IBCS2, you can statically compile it in with "options LINUX".
A pile of new syscalls implemented, including getdents(), llseek(),
readv(), writev(), msync(), personality(). The Linux-ELF libraries want
to use some of these.
linux_select() now obeys Linux semantics, ie: returns the time remaining
of the timeout value rather than leaving it the original value.
Quite a few bugs removed, including incorrect arguments being used in
syscalls.. eg: mixups between passing the sigset as an int, vs passing
it as a pointer and doing a copyin(), missing return values, unhandled
cases, SIOC* ioctls, etc.
The build for the code has changed. i386/conf/files now knows how
to build linux_genassym and generate linux_assym.h on the fly.
Supporting changes elsewhere in the kernel:
The user-mode signal trampoline has moved from the U area to immediately
below the top of the stack (below PS_STRINGS). This allows the different
binary emulations to have their own signal trampoline code (which gets rid
of the hardwired syscall 103 (sigreturn on BSD, syslog on Linux)) and so
that the emulator can provide the exact "struct sigcontext *" argument to
the program's signal handlers.
The sigstack's "ss_flags" now uses SS_DISABLE and SS_ONSTACK flags, which
have the same values as the re-used SA_DISABLE and SA_ONSTACK which are
intended for sigaction only. This enables the support of a SA_RESETHAND
flag to sigaction to implement the gross SYSV and Linux SA_ONESHOT signal
semantics where the signal handler is reset when it's triggered.
makesyscalls.sh no longer appends the struct sysentvec on the end of the
generated init_sysent.c code. It's a lot saner to have it in a seperate
file rather than trying to update the structure inside the awk script. :-)
At exec time, the dozen bytes or so of signal trampoline code are copied
to the top of the user's stack, rather than obtaining the trampoline code
the old way by getting a clone of the parent's user area. This allows
Linux and native binaries to freely exec each other without getting
trampolines mixed up.
pmap_activate since it's not used anymore. Changed cpu_fork so that
it uses one line of inline assembly rather than calling mvesp() to
get the current stack pointer. Removed mvesp() since it is no longer
being used.
clock interrupts.
Keep a 1-in-16 smoothed average of the length of each tick. If the
CPU speed is correctly diagnosed, this should give experienced users
enough information to figure out a more suitable value for `tick'.
fd and wt drivers need bounce buffers, so this normally saves 32K-1K
of kernel memory.
Keep track of which DMA channels are busy. isa_dmadone() must now be
called when DMA has finished or been aborted.
Panic for unallocated and too-small (required) bounce buffers.
fd.c:
There will be new warnings about isa_dmadone() not being called after
DMA has been aborted.
sound/dmabuf.c:
isa_dmadone() needs more parameters than are available, so temporarily
use a new interface isa_dmadone_nobounce() to avoid having to worry
about panics for fake parameters. Untested.
looking at a high resolution clock for each of the following events:
function call, function return, interrupt entry, interrupt exit,
and interesting branches. The differences between the times of
these events are added at appropriate places in a ordinary histogram
(as if very fast statistical profiling sampled the pc at those
places) so that ordinary gprof can be used to analyze the times.
gmon.h:
Histogram counters need to be 4 bytes for microsecond resolutions.
They will need to be larger for the 586 clock.
The comments were vax-centric and wrong even on vaxes. Does anyone
disagree?
gprof4.c:
The standard gprof should support counters of all integral sizes
and the size of the counter should be in the gmon header. This
hack will do until then. (Use gprof4 -u to examine the results
of non-statistical profiling.)
config/*:
Non-statistical profiling is configured with `config -pp'.
`config -p' still gives ordinary profiling.
kgmon/*:
Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b'
still enables ordinary profiling (and distables non-statistical
profiling) if non-statistical profiling is configured.
bzero.
Deprecated blkclr (removed it).
Removed some old cruft from cpufunc.h.
The optimized bzero was submitted by Torbjorn Granlund <tege@matematik.su.se>
The kernel adaption and other changes by me.
overflows.
It sure would be nice if there was an unmapped page between the PCB and
the stack (and that the size of the stack was configurable!). With the
way things are now, the PCB will get clobbered before the double fault
handler gets control, making somewhat of a mess of things. Despite this,
it is still fairly easy to poke around in the overflowed stack to figure
out the cause.
Protected them with `#ifdef KERNEL' so that <sys/queue.h> is valid C++.
Added the necessary #includes of <sys/queue.h>.
These functions are bogus and should be replaced by the queue macros.
- Don't print out meaningless iCOMP numbers, those are for droids.
- Use a shorter wait to determine clock rate to avoid deficiencies
in DELAY().
- Use a fixed-point representation with 8 bits of fraction to store
the rate and rationalize the variable name. It would be
possible to use even more fraction if it turns out to be
worthwhile (I rather doubt it).
The question of source code arrangement remains unaddressed.
misplaced extern declarations (mostly prototypes of interrupt handlers)
that this exposed. The prototypes should be moved back to the driver
sources when the functions are staticalized.
Added idempotency guards to <machine/conf.h>. "ioconf.h" can't be
included when building LKMs so define a wart in bsd.kmod.mk to help
guard against including it.
free-run and doing a subtract in microtime() rather than resetting the
counter to zero at every clock tick. In combination with the changes to
kern_clock.c, this should eliminate all the immediately obvious sources
of systematic jitter in timekeeping on Pentium machines.
changes to allow devices that don't probe (e.g. /dev/mem)
to create devfs entries
this required giving 'configure' its own SYSINIT entry
so we could duck in just before it with a DEVFS init
and some device inits..
my devfs now looks like:
./misc
./misc/speaker
./misc/mem
./misc/kmem
./misc/null
./misc/zero
./misc/io
./misc/console
./misc/pcaudio
./misc/pcaudioctl
./disks
./disks/rfloppy
./disks/rfloppy/fd0.1440
./disks/rfloppy/fd1.1200
./disks/floppy
./disks/floppy/fd0.1440
./disks/floppy/fd1.1200
also some sligt cleanups.. DEVFS needs a lot of work
but I'm getting back to it..
didn't work are somewhat bogusly optimized away before the constraint
is checked. We still expect constants passed to inline functions to
remain constant, but if the compiler ever decides that they aren't
constant then it will just generate slightly slower code instead of
an error.
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
include/signal.h:
There was massive namespace pollution from including <sys/types.h>.
POSIX functions were declared even when _ANSI_SOURCE is defined.
sys.sys/signal.h:
NSIG was declared even if _ANSI_SOURCE or _POSIX_SOURCE is defined.
sig_atomic_t wasn't declared if _POSIX_SOURCE is defined.
Declare a typedef for signal handling functions and use it to
unobfuscate declarations and to avoid half-baked function types
that cause unwanted compiler warnings at certain warning levels.
Fix confusing comment about SA_RESTART.
sys/i386/include/signal.h:
This has to be included to get the declaration of sig_atomic_t even
when _ANSI_SOURCE is defined, so be more careful about polluting
the ANSI namespace.
Uniformize idempotency ifdefs.
in machdep.c (it should use the global nmbclusters). Moved the calculation
of nmbclusters into conf/param.c (same place where nmbclusters has always
been assigned), and made the calculation include an extra amount based
on "maxusers". NMBCLUSTERS can still be overrided in the kernel config
file as always, but this change will make that generally unnecessary. This
fixes the "bug" reports from people who have misconfigured kernels seeing
the network hang when the mbuf cluster pool runs out.
Reviewed by: John Dyson
attempted to check for insecure and fatal eflags and segment
selectors, but missed many cases and got the IOPL check back to
front. The other syscalls didn't check at all.
sys_process.c, machdep.c:
Only allow PT_WRITE_U to write to the registers (ordinary and FP).
psl.h, locore.s, machdep.c:
Eliminate PSL_MBZ, PSL_MBO and PSL_USERCLR. We are not supposed
to assume anything about the reserved bits. Use PSL_USERCHANGE
and PSL_KERNEL instead. Rename PSL_USERSET to PSL_USER.
exception.s:
Define a private label for use by doreti when returning to user
mode fails.
machdep.c:
In syscalls, allow changing only the eflags that can be changed on
486's in user mode (no longer attempt to allow benign IOPL changes;
allow changing the nasty PSL_NT; don't allow changing the i586
bits).
Don't attempt to check all the cases involving invalid selectors
and %eip's. Just check for privilege violations and let the invalid
things cause a trap.
procfs_machdep.c:
Call the ptrace register functions to do all the work for reading
and writing ordinary registers and for single stepping.
trap.c:
Ignore traps caused by PSL_NT being set. Previously, users could
cause a fatal trap in user mode by setting PSL_NT and executing an
iret, and a fatal trap in kernel mode by setting PSL_NT and making
a syscall. PSL_NT was cleared too late and not in enough modes to
fix the problem.
Make all traps in user mode (except T_NMI) nonfatal.
Recover from traps caused by attempting to load invalid user
registers in doreti by restarting the traps so that they appear to
occur in user mode.
---
Fix bogons that I noticed while fixing the above:
psl.h:
Fix some comments.
Uniformize idempotency ifdef.
exception.s, machdep.c:
Remove rsvd[0-14]. rsvd0 hasn't been reserved since the 486 came
out. Replace rsvd0 by `align'. rsvd[0-11] used wrong (magic
non-unique) trap numbers. Replace rsvd[1-14] by rsvd.
locore.s:
Enable alignment check flag on 486's and 586's.
machdep.c:
Use a better type for kstack[].
Use TFREGP() to find the registers.
Reformat ptrace functions from SEF to something closer to KNF.
procfs_machdep.c:
The wrong pointer to the registers got fixed as a side effect.
Implement reading and writing of FP registers.
/proc/*/*regs now work (only) for processes that are in memory.
Clean up comments.
trap.c, trap.h:
Remove unused trap types.
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
Keep track of interrupt nesting level. It is normally 0
for syscalls and traps, but is fudged to 1 for their exit
processing in case they metamorphose into an interrupt
handler.
i386/genassym.c;
Remove support for the obsolete pcb_iml and pcb_cmap2.
Add support for pcb_inl.
i386/swtch.s:
Fudge the interrupt nesting level across context switches and in
the idle loop so that the work for preemptive context switches
gets counted as interrupt time, the work for voluntary context
switches gets counted mostly as system time (the part when
curproc == 0 gets counted as interrupt time), and only truly idle
time gets counted as idle time.
Remove obsolete support (commented out and otherwise) for pcb_iml.
Load curpcb just before curproc instead of just after so that
curpcb is always valid if curproc is. A few more changes like
this may fix tracing through context switches.
Remove obsolete function swtch_to_inactive().
include/cpu.h:
Use the new interrupt nesting level variable to implement a
non-fake CLF_INTR() so that accounting for the interrupt state
works.
You can use top, iostat or (best) an up to date systat to see
interrupt overheads. I see the expected huge interrupt overheads
for ISA devices (on a 486DX/33, about 55% for an IDE drive
transferring 1250K/sec and the same for a WD8013EBT network card
transferring 1100K/sec). The huge interrupt overheads for serial
devices are unfortunately normally invisible.
include/pcb.h:
Remove the obsolete pcb_iml and pcb_cmap2. Replace them by
padding to preserve binary compatibility.
Use part of the new padding for pcb_inl.
isa/icu.s:
isa/vector.s:
Keep track of interrupt nesting level.
Alphabetize.
Write all i/o functions in sleep so that we don't use anything from
NetBSD.
Restore the correct type of u_int for ports. This saves a whole cycle
per i/o on 486's.
Change `inline' back to __inline to avoid compiler warnings with
-Wreally-all.
Don't implement bdb() unless BDE_DEBUGGER is defined. Declare bdb_exists
outside the function to avoid hundreds of compiler warnings.
Let the compiler pick the register in asms if possible.
Implement ffs() using inline asm(). gcc provides a slightly different
one. It was broken in gcc-2.4.5 but works now. Declaring a correct
version inline ensures getting a correct version. FreeBSD-1.1.5 has
an slow inline version but FreeBSD-2.0 has a library version (which
probably never gets used).
Do inb() and outb() without using %edx for constant ports below 0x100.
Remove casts to the same type in queue functions.
Declare prototypes for everything implemented i386/*.s and also for
everything that is normally implemented as an inline here (I don't
like the current complete dependency on gcc). Ifdef out the prototypes
that are declared elsewhere. THere should be a separate header to
declare things implemented in i386/*.s, but then it would be harder
to override declarations with inlines.
${UII}
Partly support BDE_DEBUGGER. Still broken by conflict with APM. Does
nothing if BDE_DEBUGGER is not defined.
Clean up prototypes and data declarations. Declare most of the segment
functions that are implemented in support.s. Make data private in
machdep.c if possible.
Parenthesize expressions in macros properly!
${Uniformize idempotency ifdef}.
to avoid compiler warnings.
Clean up prototypes: alphabetize; don't use redundant `extern' or
meaningless `extern inline'.
Uniformize idempotency ifdef.
for it is incomplete and buggy. There is no problem unless Xintr0()
is reentered or should be reentered, but high clock interrupt
frequencies for pcaudio cause Xintr0() to be reentered (or clock
ticks to be lost when Xintr0() should have been reentered but
wasn't), and we lose little by delaying the call to softclock().
Move declarations related to the clock driver to clock.h.
Move declarations related to the npx driver to npx.h.
Clean up the remaining declarations.
I know that many of these entries are bogus and need to be revisited,
but let's get the tree working again for now and then do a pass through
looking at all the __FreeBSD__ entries, shall we?
