Previously these were libc functions but were requested to
be made into system calls for atomicity and to coalesce what
might be two entrances into the kernel (signal mask setting
and floating point trap) into one.
A few style nits and comments from bde are also included.
Tested on alpha by: gallatin
to reflect its new location, and add page queue and flag locking.
Notes: (1) alpha, i386, and ia64 had identical implementations
of pmap_collect() in terms of machine-independent interfaces;
(2) sparc64 doesn't require it; (3) powerpc had it as a TODO.
has broken int 12H.
If hw.hasbrokenint12="1" in loader environment, kernel never use BIOS
INT 12 call to determine base memory size.
Otherwise, kernel use INT 12 in old behaviour.
This should fix kernel panic problem caused by 1.544 changes.
MFC after: 1 day
sysctls to MI code; this reduces code duplication and makes all of them
available on sparc64, and the latter two on powerpc.
The semantics by the i386 and pc98 hw.availpages is slightly changed:
previously, holes between ranges of available pages would be included,
while they are excluded now. The new behaviour should be more correct
and brings i386 in line with the other architectures.
Move physmem to vm/vm_init.c, where this variable is used in MI code.
take advantage of the fact that the vm object's list of pages is
now ordered to reduce the overhead of finding the desired set of
pages to be mapped. (See revision 1.215 of vm/vm_page.c.)
remove global variable in_vm86call, set vm86 calling flag in PCB flags.
2.Fix vm86 BIOS calling preempted problem by changing vm86_lock mutex type
from MTX_DEF to MTX_SPIN. vm86pcb is not remembered in thread struct,
when the thread calling vm86 BIOS is preempted by interrupt thread,
and later switching back to the thread would cause incorrect context be
loaded into CPU registers, this leads to kernel crash.
not look like the prerequisites to fill it in properly will be in the tree
for the upcoming release, but it's mostly done, so there is no need for these
to stay around to remind us.
o It turns out that we always need to try to route the interrupts for
the case where the $PIR tells us there can be only one. Some machines
require this, while others fail when we try to do this (bogusly, imho).
Since we have no apriori way of knowing which is which, we always try to
do the routing and hope for the best if things fail.
o Add some additional comments that state the obvious, but amplify it in
non-obvious ways (judging from the questions I've gotten).
This should un-break older laptops that still have to use PCIBIOS to route
interrupts.
Tested by: sam
Use exact width types, since this is a MD file and won't be used elsewhere.
Fix a couple of resulting printf breakages
Bug found by: phk using Flexlint
handling clean and functional as 5.x evolves. This allows some of the
nasty bandaids in the 5.x codepaths to be unwound.
Encapsulate 4.x signal handling under COMPAT_FREEBSD4 (there is an
anti-foot-shooting measure in place, 5.x folks need this for a while) and
finish encapsulating the older stuff under COMPAT_43. Since the ancient
stuff is required on alpha (longjmp(3) passes a 'struct osigcontext *'
to the current sigreturn(2), instead of the 'ucontext_t *' that sigreturn
is supposed to take), add a compile time check to prevent foot shooting
there too. Add uniform COMPAT_43 stubs for ia64/sparc64/powerpc.
Tested on: i386, alpha, ia64. Compiled on sparc64 (a few days ago).
Approved by: re
Try INT 15H/E820H first, then fall back to the old compatibility
method (INT 12H).
This is a workaround for newer machines which have broken INT 12H BIOS
service implementation.
Reviewed by: -current ML
MFC after: 3 days
long doubles at the moment (printf truncates them to doubles).
However, long doubles to appear to work to the ranges listed in this
commit on both -stable (4.5) and -current. There may be some slight
rounding issues with long doubles, but that's an orthogonal issue to
these constants.
I've had this in my local tree for 3 months, and in my company's local
tree for 15 months with no ill effects.
Obtained from: NetBSD
Not likely to like it: bde
so that there is ony one copy of it. Fix that one copy
so that KSEs with no mailbox in a KSE program are not a cause
of page faults (this can legitmatly happen).
Submitted by: (parts) davidxu
This is for the not-quite-ready signal/fpu abi stuff. It may not see
the light of day, but I'm certainly not going to be able to validate it
when getting shot in the foot due to syscall number conflicts.
execve_secure() system call, which permits a process to pass in a label
for a label change during exec. This permits SELinux to change the
label for the resulting exec without a race following a manual label
change on the process. Because this interface uses our general purpose
MAC label abstraction, we call it execve_mac(), and wrap our port of
SELinux's execve_secure() around it with appropriate sid mappings.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories
The primary reason for this is to allow MD code to process machine
specific attributes, segments or sections in the ELF file and
update machine specific state accordingly. An immediate use of this
is in the ia64 port where unwind information is updated to allow
debugging and tracing in/across modules. Note that this commit
does not add the functionality to the ia64 port. See revision 1.9
of ia64/ia64/elf_machdep.c.
Validated on: alpha, i386, ia64
ACL configuration changes, this shouldn't result in different code paths
for file systems not explicitly configured for ACLs by the system
administrator. For UFS1, administrators must still recompile their
kernel to add support for extended attributes; for UFS2, it's sufficient
to enable ACLs using tunefs or at mount-time (tunefs preferred for
reliability reasons). UFS2, for a variety of reasons, including
performance and reliability, is the preferred file system for use with
ACLs.
Approved by: re
2. Update a comment. We now restore much more than RTC updates and
interrupts.
3. Order change. Stop interrupts by writing to RTC_STATUSB,
restore rate bits for the interrupts by writing to RTC_STATUSA,
then enable interrupts again.
This seems to be done perfectly backwards in startrtclock().
Otherwise, the idea for this change was obtained from
startrtclock().
4. Don't stop the clock (RTCB_HALT). We only program some control bits
and don't want to stop the clock.
5. (Not really related.) Add caveats to the comment about timer_restore().
The update is non-atomic since locking is not done.
On locking:
6. rtcin() and writertc() are locked() adequately by splhigh() in RELENG_4,
but this locking is null in -current.
7. Doing things in the correct order in (3) combined with (6) is probably
enough locking for rtcrestore() in RELENG_4. In -current, the
writertc()'s race with rtcintr() unless the BIOS disables RTC interrupts.
Submitted by: bde (including commit message)
MFC after: 1 week
This is most beneficial for vmware client os installs.
Reviewed by: jmallet, iedowse, tlambert2@mindspring.com
MFC After: never, -STABLE does not currently use this instruction
- Begin moving scheduler specific functionality into sched_4bsd.c
- Replace direct manipulation of scheduler data with hooks provided by the
new api.
- Remove KSE specific state modifications and single runq assumptions from
kern_switch.c
Reviewed by: -arch
that add an instance of themselves. The npx(4) driver doesn't even check
the npx 'port' hint but hardcodes IO_NPX instead. The npx(4) driver also
will use isa IRQ 13 (on x86, 8 on pc98) by default if no 'irq' hint is
specified, so we don't need that hint either.
in specific situations. The owner thread must be blocked, and the
borrower can not proceed back to user space with the borrowed KSE.
The borrower will return the KSE on the next context switch where
teh owner wants it back. This removes a lot of possible
race conditions and deadlocks. It is consceivable that the
borrower should inherit the priority of the owner too.
that's another discussion and would be simple to do.
Also, as part of this, the "preallocatd spare thread" is attached to the
thread doing a syscall rather than the KSE. This removes the need to lock
the scheduler when we want to access it, as it's now "at hand".
DDB now shows a lot mor info for threaded proceses though it may need
some optimisation to squeeze it all back into 80 chars again.
(possible JKH project)
Upcalls are now "bound" threads, but "KSE Lending" now means that
other completing syscalls can be completed using that KSE before the upcall
finally makes it back to the UTS. (getting threads OUT OF THE KERNEL is
one of the highest priorities in the KSE system.) The upcall when it happens
will present all the completed syscalls to the KSE for selection.
there are some strange machines that seem to need this.
o delete bogus comment.
o don't use the the bios for read/writing config space. They interact badly
with SMP and being called from ISR. This brings -current in line with
-stable.
# make the latter #ifdef on USE_PCI_BIOS_FOR_READ_WRITE in case we
# need to go back in a hurry.
NB: But it will enable it in all kernels not having options "NO_GEOM"
Put the GEOM related options into the intended order.
Add "options NO_GEOM" to all kernel configs apart from NOTES.
In some order of controlled fashion, the NO_GEOM options will be
removed, architecture by architecture in the coming days.
There are currently three known issues which may force people to
need the NO_GEOM option:
boot0cfg/fdisk:
Tries to update the MBR while it is being used to control
slices. GEOM does not allow this as a direct operation.
SCSI floppy drives:
Appearantly the scsi-da driver return "EBUSY" if no media
is inserted. This is wrong, it should return ENXIO.
PC98:
It is unclear if GEOM correctly recognizes all variants of
PC98 disklabels. (Help Wanted! I have neither docs nor HW)
These issues are all being worked.
Sponsored by: DARPA & NAI Labs.
This will be removed when new versions of syscalls sigreturn()
and sigaction() are added (mini is working on this but is in
the middle of a move).
This should fix the problem of cvsupd dying.
These are still unknown name but these are working as well
as the other ServerWorks chipset.
Description strings should be corrected when the chipsets
are known.
MFC after: 1 week
doesn't give them enough stack to do much before blowing away the pcb.
This adds MI and MD code to allow the allocation of an alternate kstack
who's size can be speficied when calling kthread_create. Passing the
value 0 prevents the alternate kstack from being created. Note that the
ia64 MD code is missing for now, and PowerPC was only partially written
due to the pmap.c being incomplete there.
Though this patch does not modify anything to make use of the alternate
kstack, acpi and usb are good candidates.
Reviewed by: jake, peter, jhb
of 1 so that it is not probed until after acpi0 is probed and attached.
- In legacy_probe(), return ENXIO if acpi0 is around and alive.
- nexus_attach() is now much simpler and just lets its child drivers do
all the work.
and attach routines have succeeded so that if they fail we can still use
the PnP BIOS to find ISA on-board devices. The fact that we do this here
is gross but fixing it properly involves a lot more work.
- nexus no longer has PCI bridges as direct children, so the PCI bus
ivar is no longer used and is removed.
- Don't attach default EISA, ISA, or MCA busses. Instead, if we do not
have an acpi0 device after bus_generic_probe(), add a legacy0 child
device.
- Remove machine/nexusvar.h.
ACPI or for when ACPI support is disabled or not present in the kernel.
Basically, the nexus device is now split into two with some parts
(such as adding default ISA, MCA, and EISA busses if they aren't found
as well as support for PCI bus device ivars) being moved to the legacy
driver.
expand to __attribute__((packed)) and __attribute__((aligned(x)))
respectively. Replace the handful of gcc-ism's that use
__attribute__((aligned(16))) etc around the kernel with __aligned(16).
There are over 400 __attribute__((packed)) to deal with, that can come
later. I just want to use __packed in new code rather than add more
gcc-ism's.
so that it is MI. Allow nfs_mountroot to return an error if the nfs_diskless
struct is not valid, rather than panicing later on. Call nfs_setup_diskless()
from nfs_mountroot if NFS_ROOT is defined, like bootpc_init(). Removed legacy
root mount support for sparc64, and enabled NFS_ROOT by default.
function were put in i386/i386/machdep.c from where it has been
cut and pasted to other architectures with only minor corruption.
Disklabel is really a MI format in many ways, at least it certainly
is when you operate on struct disklabel.
Put bounds_check_with_label() back in subr_disklabel.c where it belongs.
Sponsored by: DARPA & NAI Labs.
MD function is just a wrapper around db_stack_trace_cmd() that prints out
a backtrace of curthread. Currently, this function is only implemented
on i386 and alpha (and the alpha version isn't quite tested yet, will do
that in a bit). Other changes:
- For i386, fix a bug in the raw frame address case. The eip we extract
from the passed in frame address does not match the frame we received.
Thus, instead of printing a bogus frame with the wrong eip, go ahead
and advance frame down to the same frame as the eip we are using.
- For alpha, attempt to add a way of doing a raw trace for alpha. Instead
of passing a frame address in 'addr', pass in a pointer to a structure
containing PC and KSP and use those to start the backtrace. The alpha
db_print_backtrace() uses asm to read in the current PC and KSP values
into such a request.
Tested on: i386
Requested by: many
This patch addresses a bug that can cause a GPF in the kernel - if a
process makes use of i386_set_ldt to install a LDT entry, then loads
a corresponding segment descriptor into %gs, forks, and if the child
execs.
In this scenario, setregs executes user_ldt_free and then determines
how to reset the %gs register:
/* reset %gs as well */
if (pcb == curpcb)
load_gs(_udatasel);
else
pcb->pcb_gs = _udatasel;
This is insufficient in the fork/exec case, since pcb will be equal
to curpcb when the child execs; load_gs will reset %gs to _udatasel
but it doesn't reset pcb->pcb_gs; upon return from the system call,
cpu_switch_load_gs will thus attempt to restore %gs from pcb->pcb_gs
and trigger a GPF since all LDT entries have already been cleared.
The fix is to always reset pcb->pcb_gs to _udatasel.
Submitted by: Christian Zander <zander@minion.de>
Reviewed by: jake
under way to move the remnants of the a.out toolchain to ports. As the
comment in src/Makefile said, this stuff is deprecated and one should not
expect this to remain beyond 4.0-REL. It has already lasted WAY beyond
that.
Notable exceptions:
gcc - I have not touched the a.out generation stuff there.
ldd/ldconfig - still have some code to interface with a.out rtld.
old as/ld/etc - I have not removed these yet, pending their move to ports.
some includes - necessary for ldd/ldconfig for now.
Tested on: i386 (extensively), alpha
- Maintain fpu state across signals.
- Use ucontext_t's to store KSE thread state.
- Synthesize state for the UTS upon each upcall, rather than
saving and copying a trapframe.
- Save and restore FPU state properly in ucontext_t's.
Reviewed by: deischen, julian
Approved by: -arch
next step is to allow > 1 to be allocated per process. This would give
multi-processor threads. (when the rest of the infrastructure is
in place)
While doing this I noticed libkvm and sys/kern/kern_proc.c:fill_kinfo_proc
are diverging more than they should.. corrective action needed soon.
to control the mapping of things like the ACPI and APM into memory.
The problem is that starting X changes these values, so if something
was using the bits of BIOS mapped into memory (say ACPI or APM),
then next time they access this memory the machine would hang.
This patch refuse to change MTRR values it doesn't understand,
unless a new "force" option is given. This means X doesn't change
them by accident but someone can override that if they really want
to.
PR: 28418
Tested by: Christopher Masto <chris@netmonger.net>,
David Bushong <david@bushong.net>,
Santos <casd@myrealbox.com>
MFC after: 1 week
to userland in the signal handler that were not being iflled out before, but
should and can be.
This part of sendsig could be slightly refactored to use an MI interface, or
ideally, *sendsig*() would have an API change to accept a siginfo_t, which
would be filled out by an MI function in the level above sendsig, and said MI
function would make a small call into MD code to fill out the MD parts (some
of which may be bogus, such as the si_addr stuff in some places). This would
eventually make it possible for parts of the kernel sending signals to set up
a siginfo with meaningful information.
Reviewed by: mux
MFC after: 2 weeks
if compiling with I686_CPU as a target. CPU_DISABLE_SSE will prevent
this from happening and will guarantee the code is not compiled in.
I am still not happy with this, but gcc is now generating code that uses
these instructions if you set CPUTYPE to p3/p4 or athlon-4/mp/xp or higher.
route interrupts if the child bus is described in the PCIBIOS interrupt
routing table. For child busses that are in the routing table, they do
not necessarily use a 'swizzle' on their pins on the parent bus to route
interrupts for child devices. If the child bus is an embedded device then
the pins on the child devices can be (and usually are) directly connected
either to a PIC or to a Interrupt Router. This fixes PCIBIOS interrupt
routing across PCI-PCI bridges for embedded devices.
IRQ for an entry in a PCIBIOS interrupt routing ($PIR) table.
- Change pci_cfgintr() to except the current IRQ of a device as a fourth
argument and to use that IRQ for the device if it is valid.
- If an intpin entry in a $PIR entry has a link of 0, it means that that
intpin isn't connected to anything that can trigger an interrupt. Thus,
test the link against 0 to find invalid entries in the table instead of
implicitly relying on the irqs field to be zero. In the machines I have
looked at, intpin entries with a link of 0 often have the bits for all
possible interrupts for PCI devices set.
not the 'entry' member. The entry point is formed from both a base and
a relative entry point. 'entry' is that relative offset. It is perfectly
valid to have an entry point with a relative offset of 0. PCIbios.ventry
is the virtual address of the entry point that takes both 'base' and
'entry' into account, thus it is the proper variable to test to see if we
have an entry point or not.
lnc(4) will attach to AMD PCnet/FAST NICs if pcn(4) does not attach.
I.e. pcn(4) gets first chance. There is a problem however in that pcn(4)
was moved out of the install kernel so that the module would be used.
This however causes bad installs if one has an AMD PCnet/FAST NIC.
sysentvec. Initialized all fields of all sysentvecs, which will allow
them to be used instead of constants in more places. Provided stack
fixup routines for emulations that previously used the default.
Don't attempt to follow null pointers for zombie processes in db_ps().
Style fix: use explicit an comparison with NULL for all null pointer
checks in db_ps() instead of for half of them.
db_interface.c:
Fixed ddb's handling of traps from with ddb on i386's only.
This was mostly fixed in rev.1.27 (by longjmp()'ing back to the top
level) but was completly broken in rev.1.48 (by not unwinding the new
state (mainly db_active) either before or after the longjmp(). This
mostly never worked for other arches, since rev.1.27 has not been ported
and lower level longjmp()'s only handle traps for memory accesses. All
cases should be handled at a lower level to provided better control and
simplify unwinding of state.
Implementation details: don't pretend to maintain db_active in a nested
way -- ddb cannot be reentered in a nested way. Use db_active instead
of the db_global_jmpbuf_valid flag and longjmp()'s return value for things
related to reentering ddb. [re]entering is still not atomic enough.
in the original hardwired sysctl implementation.
The buf size calculator still overflows an integer on machines with large
KVA (eg: ia64) where the number of pages does not fit into an int. Use
'long' there.
Change Maxmem and physmem and related variables to 'long', mostly for
completeness. Machines are not likely to overflow 'int' pages in the
near term, but then again, 640K ought to be enough for anybody. This
comes for free on 32 bit machines, so why not?
alive!" message right as the scsi probe messages happen. This is a bit
nasty, but it seems to work. At the point that we unlock the AP's, briefly
wait till they are all done while we hold the console on their behalf.
These types are unlikely to ever become very MD. They include:
clockid_t, ct_rune_t, fflags_t, intrmask_t, mbstate_t, off_t, pid_t,
rune_t, socklen_t, timer_t, wchar_t, and wint_t.
While moving them, make a few adjustments (submitted by bde):
o __ct_rune_t needs to be precisely `int', not necessarily __int32_t,
since the arg type of the ctype functions is int.
o __rune_t, __wchar_t and __wint_t inherit this via a typedef of
__ct_rune_t.
o Some minor wording changes in the comment blocks for ct_rune_t and
mbstate_t.
Submitted by: bde (partially)
called <machine/_types.h>.
o <machine/ansi.h> will continue to live so it can define MD clock
macros, which are only MD because of gratuitous differences between
architectures.
o Change all headers to make use of this. This mainly involves
changing:
#ifdef _BSD_FOO_T_
typedef _BSD_FOO_T_ foo_t;
#undef _BSD_FOO_T_
#endif
to:
#ifndef _FOO_T_DECLARED
typedef __foo_t foo_t;
#define _FOO_T_DECLARED
#endif
Concept by: bde
Reviewed by: jake, obrien
make a series of modifications to the credential arguments relating
to file read and write operations to cliarfy which credential is
used for what:
- Change fo_read() and fo_write() to accept "active_cred" instead of
"cred", and change the semantics of consumers of fo_read() and
fo_write() to pass the active credential of the thread requesting
an operation rather than the cached file cred. The cached file
cred is still available in fo_read() and fo_write() consumers
via fp->f_cred. These changes largely in sys_generic.c.
For each implementation of fo_read() and fo_write(), update cred
usage to reflect this change and maintain current semantics:
- badfo_readwrite() unchanged
- kqueue_read/write() unchanged
pipe_read/write() now authorize MAC using active_cred rather
than td->td_ucred
- soo_read/write() unchanged
- vn_read/write() now authorize MAC using active_cred but
VOP_READ/WRITE() with fp->f_cred
Modify vn_rdwr() to accept two credential arguments instead of a
single credential: active_cred and file_cred. Use active_cred
for MAC authorization, and select a credential for use in
VOP_READ/WRITE() based on whether file_cred is NULL or not. If
file_cred is provided, authorize the VOP using that cred,
otherwise the active credential, matching current semantics.
Modify current vn_rdwr() consumers to pass a file_cred if used
in the context of a struct file, and to always pass active_cred.
When vn_rdwr() is used without a file_cred, pass NOCRED.
These changes should maintain current semantics for read/write,
but avoid a redundant passing of fp->f_cred, as well as making
it more clear what the origin of each credential is in file
descriptor read/write operations.
Follow-up commits will make similar changes to other file descriptor
operations, and modify the MAC framework to pass both credentials
to MAC policy modules so they can implement either semantic for
revocation.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, NAI Labs
This is an architecture that present a thing message passing interface
to the OS. You can query as to how many ports and what kind are attached
and enable them and so on.
A less grand view is that this is just another way to package SCSI (SPI or
FC) and FC-IP into a one-driver interface set.
This driver support the following hardware:
LSI FC909: Single channel, 1Gbps, Fibre Channel (FC-SCSI only)
LSI FC929: Dual Channel, 1-2Gbps, Fibre Channel (FC-SCSI only)
LSI 53c1020: Single Channel, Ultra4 (320M) (Untested)
LSI 53c1030: Dual Channel, Ultra4 (320M)
Currently it's in fair shape, but expect a lot of changes over the
next few weeks as it stabilizes.
Credits:
The driver is mostly from some folks from Jeff Roberson's company- I've
been slowly migrating it to broader support that I it came to me as.
The hardware used in developing support came from:
FC909: LSI-Logic, Advansys (now Connetix)
FC929: LSI-Logic
53c1030: Antares Microsystems (they make a very fine board!)
MFC after: 3 weeks
missed the pmap_kenter/kremove in this file, which leads to read()/write()
of /dev/mem using stale TLB entries. (gah!) Fortunately, mmap of /dev/mem
wasn't affected, so it wasn't as bad as it could have been. This throws
some light on the 'X server affects stability' thread....
Pointed out by: bde
by pmap_qenter() and pmap_qremove() is pointless. In fact, it probably
leads to unnecessary pmap_page_protect() calls if one of these pages is
paged out after unwiring.
Note: setting PG_MAPPED asserts that the page's pv list may be
non-empty. Since checking the status of the page's pv list isn't any
harder than checking this flag, the flag should probably be eliminated.
Alternatively, PG_MAPPED could be set by pmap_enter() exclusively
rather than various places throughout the kernel.
<stdint.h>. Previously, parts were defined in <machine/ansi.h> and
<machine/limits.h>. This resulted in two problems:
(1) Defining macros in <machine/ansi.h> gets in the way of that
header only defining types.
(2) Defining C99 limits in <machine/limits.h> adds pollution to
<limits.h>.
handler in the kernel at the same time. Also, allow for the
exec_new_vmspace() code to build a different sized vmspace depending on
the executable environment. This is a big help for execing i386 binaries
on ia64. The ELF exec code grows the ability to map partial pages when
there is a page size difference, eg: emulating 4K pages on 8K or 16K
hardware pages.
Flesh out the i386 emulation support for ia64. At this point, the only
binary that I know of that fails is cvsup, because the cvsup runtime
tries to execute code in pages not marked executable.
Obtained from: dfr (mostly, many tweaks from me).
identify this gadget on the CPUID result alone, so I intend to activate
the necessary magic (i8254 frequency for instance) for it based on the
precense of the on-chip host to PCI bridge.
machine will result in approximately a 4.2% loss of performance (buildworld)
and approximately a 5% reduction in power consumption (when idle). Add XXX
note on how to really make hlt work (send an IPI to wakeup HLTed cpus on
a thread-schedule event? Generate an interrupt somehow?).
running with interrupts disabled, other cpus locked down, and only
making a temporary local mapping that we immediately back out again.
Tested by: gallatin
-finstrument-functions instead of -mprofiler-epilogue. The former
works essentially the same as the latter but has a higher overhead
(about 22 more bytes per function for passing unused args to the
profiling functions).
Removed all traces of the IDENT Makefile variable, which had been
reduced to just a place for holding profiling's contribution to CFLAGS
(the IDENT that gives the kernel identity was renamed to KERN_IDENT).
page-zeroing code as well as from the general page-zeroing code and use a
lazy tlb page invalidation scheme based on a callback made at the end
of mi_switch.
A number of people came up with this idea at the same time so credit
belongs to Peter, John, and Jake as well.
Two-way SMP buildworld -j 5 tests (second run, after stabilization)
2282.76 real 2515.17 user 704.22 sys before peter's IPI commit
2266.69 real 2467.50 user 633.77 sys after peter's commit
2232.80 real 2468.99 user 615.89 sys after this commit
Reviewed by: peter, jhb
Approved by: peter
choosethread() in MI C code instead of doing it in in assembly in all the
various cpu_switch() functions. This fixes problems on ia64 and sparc64.
Reviewed by: julian, peter, benno
Tested on: i386, alpha, sparc64
- It actually works this time, honest!
- Fine grained TLB shootdowns for SMP on i386. IPI's are very expensive,
so try and optimize things where possible.
- Introduce ranged shootdowns that can be done as a single IPI.
- PG_G support for i386
- Specific-cpu targeted shootdowns. For example, there is no sense in
globally purging the TLB cache for where we are stealing a page from
the local unshared process on the local cpu. Use pm_active to track
this.
- Add some instrumentation for the tlb shootdown code.
- Rip out SMP code from <machine/cpufunc.h>
- Try and fix some very bogus PG_G and PG_PS interactions that were bad
enough to cause vm86 bios calls to break. vm86 depended on our existing
bugs and this was the cause of the VESA panics last time.
- Fix the silly one-line error that caused the 'panic: bad pte' last time.
- Fix a couple of other silly one-line errors that should have caused more
pain than they did.
Some more work is needed:
- pmap_{zero,copy}_page[_idle]. These can be done without IPI's if we
have a hook in cpu_switch.
- The IPI handlers need some cleanup. I have a bogus %ds load that can
be avoided.
- APTD handling is rather bogus and appears to be a large source of
global TLB IPI shootdowns for no really good reason.
I see speedups of between 1.5% and ~4% on buildworlds in a while 1 loop.
I expect to see a bigger difference when there is significant pageout
activity or the system otherwise has memory shortages.
I have backed out a few optimizations that I had been using over the last
few days in order to be a little more conservative. I'll revisit these
again over the next few days as the dust settles.
New option: DISABLE_PG_G - In case I missed something.
the default) is now the only method for i386.
Remove the paraphanalia that supported critmode. Remove td_critnest, clean
up the assembly, and clean up (mostly remove) the old junk from
cpu_critical_enter() and cpu_critical_exit().
ipl.s except doreti which really belongs in with the exceptions as it's
just the other side of the same coin. Will remove ipl.s in a separate commit.
Agreed by: several including bde@freebsd.org
threaded VM pagezero kthread outside of Giant. For some platforms, this
is really easy since it can just use the direct mapped region. For others,
IPI sending is involved or there are other issues, so grab Giant when
needed.
We still have preemption issues to deal with, but Alan Cox has an
interesting suggestion on how to minimize the problem on x86.
Use Luigi's hack for preserving the (lack of) priority.
Turn the idle zeroing back on since it can now actually do something useful
outside of Giant in many cases.
mappings from the page tables, which were mapped with PG_G! We could
reuse the page table entry for another mapping (pmap_mapdev) but it
would never have cleared any remaining PG_G TLB entries.
pmap_swapin_proc/pmap_swapout_proc functions from the MD pmap code
and use a single equivalent MI version. There are other cleanups
needed still.
While here, use the UMA zone hooks to keep a cache of preinitialized
proc structures handy, just like the thread system does. This eliminates
one dependency on 'struct proc' being persistent even after being freed.
There are some comments about things that can be factored out into
ctor/dtor functions if it is worth it. For now they are mostly just
doing statistics to get a feel of how it is working.
