remove prototypes with no matching function
ok mpi@

remove unused global vars
ok deraadt@ miod@

remove unneeded includes in arch/i386
ok krw@ miod@

Constify struct cfattach.

an ugly white space

Push KERNEL_LOCK/UNLOCK() dance inside trapsignal().

ok kettenis@, visa@

Use the MI interrupt enable/distable API instead of the MD one on i386 and
remove the MD API.

ok deraadt@

More steps for i386 Meltdown fix:

- provide struct cpu_info_full
- prepare K-U sections
- reorganize interrupt, trap, syscall entry to use K-U trampoline
- prepare pmap for entering special mappings, the mappings are not
setup yet

This code will already trigger performance issues. We do more tlb
flushes, but we do not unmap the kernel yet. The latter
will be needed to prevent Meltdown.

from hshoexer@; input guenther@; OK mlarkin@ deraadt@

Change ci_dev.dv_xname to ci_dev->dv_xname in debugging printf()
to make it compile when enabled.
from hshoexer@

Recommit preparation for i386 Meltdown fix after OpenBSD 6.3 release.

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

iBackout the preparations for fixing Meltdown on i386. The task was
only halfway done and the current state does not help anybody. For
OpenBSD 6.3 release go back to the original code before 2018/03/13.
This gives us a stable release and the changes will come back later.
discussed with guenther@ deraadt@ hshoexer@

Preparation for i386 Meltdown fix:

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

Protect the calls to trapsignal() and psignal() with a kernel lock
when processing an npx interrupt. This fixes a kernel locked
assertion in postsig_done() during the libc ieeefp/except regression
test.
OK visa@

Delete unnecessary <sys/file.h> includes

ok millert@ krw@

Kill SPINLOCK_SPIN_HOOK, use CPU_BUSY_CYCLE() instead.

ok visa@, kettenis@

deprecate use of sys/lock.h and replace it with sys/atomic.h or
machine/lock.h as appropriate.

It's been a quarter century: we can assume volatile is present with that name.

ok dlg@ mpi@ deraadt@

Copied from amd64..

Use an explicit suffix for the "fld" instruction to shut up clang. The correct
instruction is fldl since we try to load a double-precision value.
GCC actually gets it wrong and emits "flds" (which is harmless).

ok guenther@

remove stale cyrix emc support, and switch to just using intial npxcw
ok kettenis

Revert art@'s moving around of the KERNEL_LOCK()/KERNEL_UNLOCK() calls,
as it causes hangs in some ports, including libsigsegv's configure script

confirmed by krw@, landry@

There is a bunch of places in the kernel entry points where we don't
hold the kernel lock, but still need call one function that needs it.

Instead of grabbing the lock all over the place, move the locks into
the affected functions: trapsignal, scdebug*, ktrsyscall, ktrsysret,
systrace_redirect and ADDUPROF. In the cases we already hold the biglock
we'll just recurse.

kettenis@, beck@ ok

Clean up after P_BIGLOCK removal.
KERNEL_PROC_LOCK -> KERNEL_LOCK
KERNEL_PROC_UNLOCK -> KERNEL_UNLOCK

oga@ ok

When reading MXCSR from userland sigcontext or a ptrace request,
mask out invalid bits to prevent a protect fault.

Original diff by joshe@; further feedback and ok kettenis@

Back out previous, it appears to be broken.

When reading MXCSR from userland sigcontext, mask out invalid bits.

This prevents a protection fault if a userland signal handler
scribbles all over it's struct sigcontext

Help from and ok guenther@ kettenis@

Change the strategy for using the FPU in the kernel. The kernel FPU state
is now shared with all processes/threads. As a result, you can now use the
FPU in true process context (instead of just in kernel threads), but you
need to make sure you restore the default FPU state before calling
fpu_kernel_exit() if you change rounding mode, precision or exception masks.

Lots of discussion with thib@ and Mike Belopuhov.

ok thib@, deraadt@

Don't leak the contents of the FPU and SSE registers between processes.

ok deraadt@

Make sure that the FPU IPIs shoot down the right FPU context and not the FPU
context of some random process that happened to be switched onto the FPU
after the decision was made to send the IPI.

fpu_kernel_{enter,exit}; Functions to allow the use of
the FPU in the kernel.

From Mike Belopuhov; Little bits by myself.

Comments/OK kettenis@

delete balony comment; ok kettenis

Load %mxcsr when initializing the FPU on machines that support SSE.

ok deraadt@

Properly handle XMM exceptions instead of panicing the kernel. Fixes the
issue reported by Slava Pestov.

ok deraadt@

branches: 1.44.2; 1.44.6;
Fix "fp_save ipi didn't" panic, and move i386/amd64 closer in the process.
Positive test results by a handful of people. Ok kettenis@

branches: 1.43.4;
quite a bit of simplification by removing cpu classes.
also assume that 386 cpus are really unknown, and promote them to 486
instead of panic.

Short-circuit the detection of the FPU by checking the CPUID features,
as done in FreeBSD. Fixes problems on AMD Geode LX-800 where our
detection logic doesn't detect the FP exception.

ok deraadt@ kettenis@

ansi/deregister

branches: 1.40.2;
Kill option GPL_MATH_EMULATE.

ok deraadt@ and many others.

from todays freebsd advisory: fxrstor on amd cpu does not restore fip,fdp,fop thus leaking other proc's execution history; deraadt@ ok

time-bind and lower power in lock spinning and a couple of other ipi loops by using pause insn; brad@ ok tedu@ ok and feedback krw@ testing

branches: 1.37.2; 1.37.4;
oops. on 686 we count fpswtch in another function...

count fpu lazy context switches; deraadt@ ok

branches: 1.35.2;
SIGFPE signal codes for SSE.
nice deraadt@

Fix conversion of interrupt frame to trap frame.
ok deraadt@, niklas@

Don't assert that P_INMEM is set in p_flag. The assert is only almost
correct because P_INMEM is for various reasons removed before cpu_swapout
is called and cpu_swapout in turn calls npxsave_proc.

deraadt@ ok

debranch SMP, have fun

Enable FXSR all the time, and cope with NPX/FXSR conversions; from netbsd.
Pass SSE/SSE2/XCRYPT flags out via syctl, and prepare for being able to do
xcrypt-* in userland; ok naddy, tested a lot by pvalchev and jolan, also
works on amd64 in 32bit mode

workaround: during npx exception testing, delay(1). For some reason
amd64 (in 32 bit mode) and the new transmeta cpus want this, and lock
up otherwise. very odd.

No need to have several implementations of {read,write}_eflags(),
put it to cpufunc.h.
ok weingart@ tedu@

let ALWAYS_MATH_EMULATE actually compile

Remove the advertising clause in the UCB license which Berkeley
rescinded 22 July 1999. Proofed by myself and Theo.

i386 changes to move to ELF. asm cleanup. Change MAXDSIZ to 512M for 1Gsep.
DARPA funded work.

First round of __P removal in sys

multi-line string literals are deprecated in gcc 3.0
pr#2379

branches: 1.23.2;
New try at a level based irq system.

Revert niklas's broken ICU interrupt handling changes; deraadt@ ok

Redo the ICU interrupt handling system to better emulate an APIC system.
This is done by representing the current blockings of interrupts with
a priority level instead of an interrupt mask. This makes it possible to
share implementations for spl* functions/macros between both ICU
and APIC (a must when going MP) systems. In this process, assign
soft interrupts their own levels to match the way things will be on
APIC systems where they actually will be real interrupts and not just
bits in a pending mask as they are now.
Heavily inspired by Bill Studenmunds SMP work in NetBSD

Replace inclusion of <vm/foo.h> with the correct <uvm/bar.h> when necessary.
(Look ma, I might have broken the tree)

PMAP_NEW and UVM are no longer optional on i386.

Add emulation of Linux features to procfs; mostly from NetBSD. ok deraadt@

Add explicit inclusions of signalvar.h to files actually using syms defined
there but relying on an indirect inclusion

branches: 1.16.6;
cnt.foo -> uvmexp.bar in uvm

GCC 2.8 Wall

Do interrupt time accounting by running interrupt handlers with a seperate
code segment selector (otherwise identical to the standard kernel code
selector); idea by Dave Richards <richards@zso.dec.com>

Provide more information on FP exceptions

add type & union sigval args to sendsig/trapsignal

add another parameter to trapsignal() and sendsig() -- fault addr to be
delivered with in the siginfo information

Add:
* GPL_MATH_EMULATE: fpemul
* ALWAYS_MATH_EMULATE: causes npx probe to always fail.
* better fp context in struct pcb.

sync

sync with 0504; prototype changes

partial sync with netbsd 960418, more to come

Fix back wrong patches.

Merging w/ NetBSD 021796.
speaker upgraded to the current.
some changes to the VM stuff (ie kern_thread.c added and so).

Pass device name to interrupt establish routines so it can be recorded in
the device interrupt chain structures (isa, pci)

Move interrupt chain structure definition to <machine/psl.h> so vmstat can
get at it (i386)

Remove hack to count interrupts the old way (i386)

from netbsd: Put #includes in the right order

from netbsd:
The IST_* and IPL_* constants are not bus-specific; don't treat them as such.

branches: 1.1.1;
Initial revision

remove unused global vars
ok deraadt@ miod@

remove unneeded includes in arch/i386
ok krw@ miod@

Constify struct cfattach.

an ugly white space

Push KERNEL_LOCK/UNLOCK() dance inside trapsignal().

ok kettenis@, visa@

Use the MI interrupt enable/distable API instead of the MD one on i386 and
remove the MD API.

ok deraadt@

More steps for i386 Meltdown fix:

- provide struct cpu_info_full
- prepare K-U sections
- reorganize interrupt, trap, syscall entry to use K-U trampoline
- prepare pmap for entering special mappings, the mappings are not
setup yet

This code will already trigger performance issues. We do more tlb
flushes, but we do not unmap the kernel yet. The latter
will be needed to prevent Meltdown.

from hshoexer@; input guenther@; OK mlarkin@ deraadt@

Change ci_dev.dv_xname to ci_dev->dv_xname in debugging printf()
to make it compile when enabled.
from hshoexer@

Recommit preparation for i386 Meltdown fix after OpenBSD 6.3 release.

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

iBackout the preparations for fixing Meltdown on i386. The task was
only halfway done and the current state does not help anybody. For
OpenBSD 6.3 release go back to the original code before 2018/03/13.
This gives us a stable release and the changes will come back later.
discussed with guenther@ deraadt@ hshoexer@

Preparation for i386 Meltdown fix:

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

Protect the calls to trapsignal() and psignal() with a kernel lock
when processing an npx interrupt. This fixes a kernel locked
assertion in postsig_done() during the libc ieeefp/except regression
test.
OK visa@

Delete unnecessary <sys/file.h> includes

ok millert@ krw@

Kill SPINLOCK_SPIN_HOOK, use CPU_BUSY_CYCLE() instead.

ok visa@, kettenis@

deprecate use of sys/lock.h and replace it with sys/atomic.h or
machine/lock.h as appropriate.

It's been a quarter century: we can assume volatile is present with that name.

ok dlg@ mpi@ deraadt@

Copied from amd64..

Use an explicit suffix for the "fld" instruction to shut up clang. The correct
instruction is fldl since we try to load a double-precision value.
GCC actually gets it wrong and emits "flds" (which is harmless).

ok guenther@

remove stale cyrix emc support, and switch to just using intial npxcw
ok kettenis

Revert art@'s moving around of the KERNEL_LOCK()/KERNEL_UNLOCK() calls,
as it causes hangs in some ports, including libsigsegv's configure script

confirmed by krw@, landry@

There is a bunch of places in the kernel entry points where we don't
hold the kernel lock, but still need call one function that needs it.

Instead of grabbing the lock all over the place, move the locks into
the affected functions: trapsignal, scdebug*, ktrsyscall, ktrsysret,
systrace_redirect and ADDUPROF. In the cases we already hold the biglock
we'll just recurse.

kettenis@, beck@ ok

Clean up after P_BIGLOCK removal.
KERNEL_PROC_LOCK -> KERNEL_LOCK
KERNEL_PROC_UNLOCK -> KERNEL_UNLOCK

oga@ ok

When reading MXCSR from userland sigcontext or a ptrace request,
mask out invalid bits to prevent a protect fault.

Original diff by joshe@; further feedback and ok kettenis@

Back out previous, it appears to be broken.

When reading MXCSR from userland sigcontext, mask out invalid bits.

This prevents a protection fault if a userland signal handler
scribbles all over it's struct sigcontext

Help from and ok guenther@ kettenis@

Change the strategy for using the FPU in the kernel. The kernel FPU state
is now shared with all processes/threads. As a result, you can now use the
FPU in true process context (instead of just in kernel threads), but you
need to make sure you restore the default FPU state before calling
fpu_kernel_exit() if you change rounding mode, precision or exception masks.

Lots of discussion with thib@ and Mike Belopuhov.

ok thib@, deraadt@

Don't leak the contents of the FPU and SSE registers between processes.

ok deraadt@

Make sure that the FPU IPIs shoot down the right FPU context and not the FPU
context of some random process that happened to be switched onto the FPU
after the decision was made to send the IPI.

fpu_kernel_{enter,exit}; Functions to allow the use of
the FPU in the kernel.

From Mike Belopuhov; Little bits by myself.

Comments/OK kettenis@

delete balony comment; ok kettenis

Load %mxcsr when initializing the FPU on machines that support SSE.

ok deraadt@

Properly handle XMM exceptions instead of panicing the kernel. Fixes the
issue reported by Slava Pestov.

ok deraadt@

branches: 1.44.2; 1.44.6;
Fix "fp_save ipi didn't" panic, and move i386/amd64 closer in the process.
Positive test results by a handful of people. Ok kettenis@

branches: 1.43.4;
quite a bit of simplification by removing cpu classes.
also assume that 386 cpus are really unknown, and promote them to 486
instead of panic.

Short-circuit the detection of the FPU by checking the CPUID features,
as done in FreeBSD. Fixes problems on AMD Geode LX-800 where our
detection logic doesn't detect the FP exception.

ok deraadt@ kettenis@

ansi/deregister

branches: 1.40.2;
Kill option GPL_MATH_EMULATE.

ok deraadt@ and many others.

from todays freebsd advisory: fxrstor on amd cpu does not restore fip,fdp,fop thus leaking other proc's execution history; deraadt@ ok

time-bind and lower power in lock spinning and a couple of other ipi loops by using pause insn; brad@ ok tedu@ ok and feedback krw@ testing

branches: 1.37.2; 1.37.4;
oops. on 686 we count fpswtch in another function...

count fpu lazy context switches; deraadt@ ok

branches: 1.35.2;
SIGFPE signal codes for SSE.
nice deraadt@

Fix conversion of interrupt frame to trap frame.
ok deraadt@, niklas@

Don't assert that P_INMEM is set in p_flag. The assert is only almost
correct because P_INMEM is for various reasons removed before cpu_swapout
is called and cpu_swapout in turn calls npxsave_proc.

deraadt@ ok

debranch SMP, have fun

Enable FXSR all the time, and cope with NPX/FXSR conversions; from netbsd.
Pass SSE/SSE2/XCRYPT flags out via syctl, and prepare for being able to do
xcrypt-* in userland; ok naddy, tested a lot by pvalchev and jolan, also
works on amd64 in 32bit mode

workaround: during npx exception testing, delay(1). For some reason
amd64 (in 32 bit mode) and the new transmeta cpus want this, and lock
up otherwise. very odd.

No need to have several implementations of {read,write}_eflags(),
put it to cpufunc.h.
ok weingart@ tedu@

let ALWAYS_MATH_EMULATE actually compile

Remove the advertising clause in the UCB license which Berkeley
rescinded 22 July 1999. Proofed by myself and Theo.

i386 changes to move to ELF. asm cleanup. Change MAXDSIZ to 512M for 1Gsep.
DARPA funded work.

First round of __P removal in sys

multi-line string literals are deprecated in gcc 3.0
pr#2379

branches: 1.23.2;
New try at a level based irq system.

Revert niklas's broken ICU interrupt handling changes; deraadt@ ok

Redo the ICU interrupt handling system to better emulate an APIC system.
This is done by representing the current blockings of interrupts with
a priority level instead of an interrupt mask. This makes it possible to
share implementations for spl* functions/macros between both ICU
and APIC (a must when going MP) systems. In this process, assign
soft interrupts their own levels to match the way things will be on
APIC systems where they actually will be real interrupts and not just
bits in a pending mask as they are now.
Heavily inspired by Bill Studenmunds SMP work in NetBSD

Replace inclusion of <vm/foo.h> with the correct <uvm/bar.h> when necessary.
(Look ma, I might have broken the tree)

PMAP_NEW and UVM are no longer optional on i386.

Add emulation of Linux features to procfs; mostly from NetBSD. ok deraadt@

Add explicit inclusions of signalvar.h to files actually using syms defined
there but relying on an indirect inclusion

branches: 1.16.6;
cnt.foo -> uvmexp.bar in uvm

GCC 2.8 Wall

Do interrupt time accounting by running interrupt handlers with a seperate
code segment selector (otherwise identical to the standard kernel code
selector); idea by Dave Richards <richards@zso.dec.com>

Provide more information on FP exceptions

add type & union sigval args to sendsig/trapsignal

add another parameter to trapsignal() and sendsig() -- fault addr to be
delivered with in the siginfo information

Add:
* GPL_MATH_EMULATE: fpemul
* ALWAYS_MATH_EMULATE: causes npx probe to always fail.
* better fp context in struct pcb.

sync

sync with 0504; prototype changes

partial sync with netbsd 960418, more to come

Fix back wrong patches.

Merging w/ NetBSD 021796.
speaker upgraded to the current.
some changes to the VM stuff (ie kern_thread.c added and so).

Pass device name to interrupt establish routines so it can be recorded in
the device interrupt chain structures (isa, pci)

Move interrupt chain structure definition to <machine/psl.h> so vmstat can
get at it (i386)

Remove hack to count interrupts the old way (i386)

from netbsd: Put #includes in the right order

from netbsd:
The IST_* and IPL_* constants are not bus-specific; don't treat them as such.

branches: 1.1.1;
Initial revision

remove unneeded includes in arch/i386
ok krw@ miod@

Constify struct cfattach.

an ugly white space

Push KERNEL_LOCK/UNLOCK() dance inside trapsignal().

ok kettenis@, visa@

Use the MI interrupt enable/distable API instead of the MD one on i386 and
remove the MD API.

ok deraadt@

More steps for i386 Meltdown fix:

- provide struct cpu_info_full
- prepare K-U sections
- reorganize interrupt, trap, syscall entry to use K-U trampoline
- prepare pmap for entering special mappings, the mappings are not
setup yet

This code will already trigger performance issues. We do more tlb
flushes, but we do not unmap the kernel yet. The latter
will be needed to prevent Meltdown.

from hshoexer@; input guenther@; OK mlarkin@ deraadt@

Change ci_dev.dv_xname to ci_dev->dv_xname in debugging printf()
to make it compile when enabled.
from hshoexer@

Recommit preparation for i386 Meltdown fix after OpenBSD 6.3 release.

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

iBackout the preparations for fixing Meltdown on i386. The task was
only halfway done and the current state does not help anybody. For
OpenBSD 6.3 release go back to the original code before 2018/03/13.
This gives us a stable release and the changes will come back later.
discussed with guenther@ deraadt@ hshoexer@

Preparation for i386 Meltdown fix:

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

Protect the calls to trapsignal() and psignal() with a kernel lock
when processing an npx interrupt. This fixes a kernel locked
assertion in postsig_done() during the libc ieeefp/except regression
test.
OK visa@

Delete unnecessary <sys/file.h> includes

ok millert@ krw@

Kill SPINLOCK_SPIN_HOOK, use CPU_BUSY_CYCLE() instead.

ok visa@, kettenis@

deprecate use of sys/lock.h and replace it with sys/atomic.h or
machine/lock.h as appropriate.

It's been a quarter century: we can assume volatile is present with that name.

ok dlg@ mpi@ deraadt@

Copied from amd64..

Use an explicit suffix for the "fld" instruction to shut up clang. The correct
instruction is fldl since we try to load a double-precision value.
GCC actually gets it wrong and emits "flds" (which is harmless).

ok guenther@

remove stale cyrix emc support, and switch to just using intial npxcw
ok kettenis

Revert art@'s moving around of the KERNEL_LOCK()/KERNEL_UNLOCK() calls,
as it causes hangs in some ports, including libsigsegv's configure script

confirmed by krw@, landry@

There is a bunch of places in the kernel entry points where we don't
hold the kernel lock, but still need call one function that needs it.

Instead of grabbing the lock all over the place, move the locks into
the affected functions: trapsignal, scdebug*, ktrsyscall, ktrsysret,
systrace_redirect and ADDUPROF. In the cases we already hold the biglock
we'll just recurse.

kettenis@, beck@ ok

Clean up after P_BIGLOCK removal.
KERNEL_PROC_LOCK -> KERNEL_LOCK
KERNEL_PROC_UNLOCK -> KERNEL_UNLOCK

oga@ ok

When reading MXCSR from userland sigcontext or a ptrace request,
mask out invalid bits to prevent a protect fault.

Original diff by joshe@; further feedback and ok kettenis@

Back out previous, it appears to be broken.

When reading MXCSR from userland sigcontext, mask out invalid bits.

This prevents a protection fault if a userland signal handler
scribbles all over it's struct sigcontext

Help from and ok guenther@ kettenis@

Change the strategy for using the FPU in the kernel. The kernel FPU state
is now shared with all processes/threads. As a result, you can now use the
FPU in true process context (instead of just in kernel threads), but you
need to make sure you restore the default FPU state before calling
fpu_kernel_exit() if you change rounding mode, precision or exception masks.

Lots of discussion with thib@ and Mike Belopuhov.

ok thib@, deraadt@

Don't leak the contents of the FPU and SSE registers between processes.

ok deraadt@

Make sure that the FPU IPIs shoot down the right FPU context and not the FPU
context of some random process that happened to be switched onto the FPU
after the decision was made to send the IPI.

fpu_kernel_{enter,exit}; Functions to allow the use of
the FPU in the kernel.

From Mike Belopuhov; Little bits by myself.

Comments/OK kettenis@

delete balony comment; ok kettenis

Load %mxcsr when initializing the FPU on machines that support SSE.

ok deraadt@

Properly handle XMM exceptions instead of panicing the kernel. Fixes the
issue reported by Slava Pestov.

ok deraadt@

branches: 1.44.2; 1.44.6;
Fix "fp_save ipi didn't" panic, and move i386/amd64 closer in the process.
Positive test results by a handful of people. Ok kettenis@

branches: 1.43.4;
quite a bit of simplification by removing cpu classes.
also assume that 386 cpus are really unknown, and promote them to 486
instead of panic.

Short-circuit the detection of the FPU by checking the CPUID features,
as done in FreeBSD. Fixes problems on AMD Geode LX-800 where our
detection logic doesn't detect the FP exception.

ok deraadt@ kettenis@

ansi/deregister

branches: 1.40.2;
Kill option GPL_MATH_EMULATE.

ok deraadt@ and many others.

from todays freebsd advisory: fxrstor on amd cpu does not restore fip,fdp,fop thus leaking other proc's execution history; deraadt@ ok

time-bind and lower power in lock spinning and a couple of other ipi loops by using pause insn; brad@ ok tedu@ ok and feedback krw@ testing

branches: 1.37.2; 1.37.4;
oops. on 686 we count fpswtch in another function...

count fpu lazy context switches; deraadt@ ok

branches: 1.35.2;
SIGFPE signal codes for SSE.
nice deraadt@

Fix conversion of interrupt frame to trap frame.
ok deraadt@, niklas@

Don't assert that P_INMEM is set in p_flag. The assert is only almost
correct because P_INMEM is for various reasons removed before cpu_swapout
is called and cpu_swapout in turn calls npxsave_proc.

deraadt@ ok

debranch SMP, have fun

Enable FXSR all the time, and cope with NPX/FXSR conversions; from netbsd.
Pass SSE/SSE2/XCRYPT flags out via syctl, and prepare for being able to do
xcrypt-* in userland; ok naddy, tested a lot by pvalchev and jolan, also
works on amd64 in 32bit mode

workaround: during npx exception testing, delay(1). For some reason
amd64 (in 32 bit mode) and the new transmeta cpus want this, and lock
up otherwise. very odd.

No need to have several implementations of {read,write}_eflags(),
put it to cpufunc.h.
ok weingart@ tedu@

let ALWAYS_MATH_EMULATE actually compile

Remove the advertising clause in the UCB license which Berkeley
rescinded 22 July 1999. Proofed by myself and Theo.

i386 changes to move to ELF. asm cleanup. Change MAXDSIZ to 512M for 1Gsep.
DARPA funded work.

First round of __P removal in sys

multi-line string literals are deprecated in gcc 3.0
pr#2379

branches: 1.23.2;
New try at a level based irq system.

Revert niklas's broken ICU interrupt handling changes; deraadt@ ok

Redo the ICU interrupt handling system to better emulate an APIC system.
This is done by representing the current blockings of interrupts with
a priority level instead of an interrupt mask. This makes it possible to
share implementations for spl* functions/macros between both ICU
and APIC (a must when going MP) systems. In this process, assign
soft interrupts their own levels to match the way things will be on
APIC systems where they actually will be real interrupts and not just
bits in a pending mask as they are now.
Heavily inspired by Bill Studenmunds SMP work in NetBSD

Replace inclusion of <vm/foo.h> with the correct <uvm/bar.h> when necessary.
(Look ma, I might have broken the tree)

PMAP_NEW and UVM are no longer optional on i386.

Add emulation of Linux features to procfs; mostly from NetBSD. ok deraadt@

Add explicit inclusions of signalvar.h to files actually using syms defined
there but relying on an indirect inclusion

branches: 1.16.6;
cnt.foo -> uvmexp.bar in uvm

GCC 2.8 Wall

Do interrupt time accounting by running interrupt handlers with a seperate
code segment selector (otherwise identical to the standard kernel code
selector); idea by Dave Richards <richards@zso.dec.com>

Provide more information on FP exceptions

add type & union sigval args to sendsig/trapsignal

add another parameter to trapsignal() and sendsig() -- fault addr to be
delivered with in the siginfo information

Add:
* GPL_MATH_EMULATE: fpemul
* ALWAYS_MATH_EMULATE: causes npx probe to always fail.
* better fp context in struct pcb.

sync

sync with 0504; prototype changes

partial sync with netbsd 960418, more to come

Fix back wrong patches.

Merging w/ NetBSD 021796.
speaker upgraded to the current.
some changes to the VM stuff (ie kern_thread.c added and so).

Pass device name to interrupt establish routines so it can be recorded in
the device interrupt chain structures (isa, pci)

Move interrupt chain structure definition to <machine/psl.h> so vmstat can
get at it (i386)

Remove hack to count interrupts the old way (i386)

from netbsd: Put #includes in the right order

from netbsd:
The IST_* and IPL_* constants are not bus-specific; don't treat them as such.

branches: 1.1.1;
Initial revision

Constify struct cfattach.

an ugly white space

Push KERNEL_LOCK/UNLOCK() dance inside trapsignal().

ok kettenis@, visa@

Use the MI interrupt enable/distable API instead of the MD one on i386 and
remove the MD API.

ok deraadt@

More steps for i386 Meltdown fix:

- provide struct cpu_info_full
- prepare K-U sections
- reorganize interrupt, trap, syscall entry to use K-U trampoline
- prepare pmap for entering special mappings, the mappings are not
setup yet

This code will already trigger performance issues. We do more tlb
flushes, but we do not unmap the kernel yet. The latter
will be needed to prevent Meltdown.

from hshoexer@; input guenther@; OK mlarkin@ deraadt@

Change ci_dev.dv_xname to ci_dev->dv_xname in debugging printf()
to make it compile when enabled.
from hshoexer@

Recommit preparation for i386 Meltdown fix after OpenBSD 6.3 release.

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

iBackout the preparations for fixing Meltdown on i386. The task was
only halfway done and the current state does not help anybody. For
OpenBSD 6.3 release go back to the original code before 2018/03/13.
This gives us a stable release and the changes will come back later.
discussed with guenther@ deraadt@ hshoexer@

Preparation for i386 Meltdown fix:

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

Protect the calls to trapsignal() and psignal() with a kernel lock
when processing an npx interrupt. This fixes a kernel locked
assertion in postsig_done() during the libc ieeefp/except regression
test.
OK visa@

Delete unnecessary <sys/file.h> includes

ok millert@ krw@

Kill SPINLOCK_SPIN_HOOK, use CPU_BUSY_CYCLE() instead.

ok visa@, kettenis@

deprecate use of sys/lock.h and replace it with sys/atomic.h or
machine/lock.h as appropriate.

It's been a quarter century: we can assume volatile is present with that name.

ok dlg@ mpi@ deraadt@

Copied from amd64..

Use an explicit suffix for the "fld" instruction to shut up clang. The correct
instruction is fldl since we try to load a double-precision value.
GCC actually gets it wrong and emits "flds" (which is harmless).

ok guenther@

remove stale cyrix emc support, and switch to just using intial npxcw
ok kettenis

Revert art@'s moving around of the KERNEL_LOCK()/KERNEL_UNLOCK() calls,
as it causes hangs in some ports, including libsigsegv's configure script

confirmed by krw@, landry@

There is a bunch of places in the kernel entry points where we don't
hold the kernel lock, but still need call one function that needs it.

Instead of grabbing the lock all over the place, move the locks into
the affected functions: trapsignal, scdebug*, ktrsyscall, ktrsysret,
systrace_redirect and ADDUPROF. In the cases we already hold the biglock
we'll just recurse.

kettenis@, beck@ ok

Clean up after P_BIGLOCK removal.
KERNEL_PROC_LOCK -> KERNEL_LOCK
KERNEL_PROC_UNLOCK -> KERNEL_UNLOCK

oga@ ok

When reading MXCSR from userland sigcontext or a ptrace request,
mask out invalid bits to prevent a protect fault.

Original diff by joshe@; further feedback and ok kettenis@

Back out previous, it appears to be broken.

When reading MXCSR from userland sigcontext, mask out invalid bits.

This prevents a protection fault if a userland signal handler
scribbles all over it's struct sigcontext

Help from and ok guenther@ kettenis@

Change the strategy for using the FPU in the kernel. The kernel FPU state
is now shared with all processes/threads. As a result, you can now use the
FPU in true process context (instead of just in kernel threads), but you
need to make sure you restore the default FPU state before calling
fpu_kernel_exit() if you change rounding mode, precision or exception masks.

Lots of discussion with thib@ and Mike Belopuhov.

ok thib@, deraadt@

Don't leak the contents of the FPU and SSE registers between processes.

ok deraadt@

Make sure that the FPU IPIs shoot down the right FPU context and not the FPU
context of some random process that happened to be switched onto the FPU
after the decision was made to send the IPI.

fpu_kernel_{enter,exit}; Functions to allow the use of
the FPU in the kernel.

From Mike Belopuhov; Little bits by myself.

Comments/OK kettenis@

delete balony comment; ok kettenis

Load %mxcsr when initializing the FPU on machines that support SSE.

ok deraadt@

Properly handle XMM exceptions instead of panicing the kernel. Fixes the
issue reported by Slava Pestov.

ok deraadt@

branches: 1.44.2; 1.44.6;
Fix "fp_save ipi didn't" panic, and move i386/amd64 closer in the process.
Positive test results by a handful of people. Ok kettenis@

branches: 1.43.4;
quite a bit of simplification by removing cpu classes.
also assume that 386 cpus are really unknown, and promote them to 486
instead of panic.

Short-circuit the detection of the FPU by checking the CPUID features,
as done in FreeBSD. Fixes problems on AMD Geode LX-800 where our
detection logic doesn't detect the FP exception.

ok deraadt@ kettenis@

ansi/deregister

branches: 1.40.2;
Kill option GPL_MATH_EMULATE.

ok deraadt@ and many others.

from todays freebsd advisory: fxrstor on amd cpu does not restore fip,fdp,fop thus leaking other proc's execution history; deraadt@ ok

time-bind and lower power in lock spinning and a couple of other ipi loops by using pause insn; brad@ ok tedu@ ok and feedback krw@ testing

branches: 1.37.2; 1.37.4;
oops. on 686 we count fpswtch in another function...

count fpu lazy context switches; deraadt@ ok

branches: 1.35.2;
SIGFPE signal codes for SSE.
nice deraadt@

Fix conversion of interrupt frame to trap frame.
ok deraadt@, niklas@

Don't assert that P_INMEM is set in p_flag. The assert is only almost
correct because P_INMEM is for various reasons removed before cpu_swapout
is called and cpu_swapout in turn calls npxsave_proc.

deraadt@ ok

debranch SMP, have fun

Enable FXSR all the time, and cope with NPX/FXSR conversions; from netbsd.
Pass SSE/SSE2/XCRYPT flags out via syctl, and prepare for being able to do
xcrypt-* in userland; ok naddy, tested a lot by pvalchev and jolan, also
works on amd64 in 32bit mode

workaround: during npx exception testing, delay(1). For some reason
amd64 (in 32 bit mode) and the new transmeta cpus want this, and lock
up otherwise. very odd.

No need to have several implementations of {read,write}_eflags(),
put it to cpufunc.h.
ok weingart@ tedu@

let ALWAYS_MATH_EMULATE actually compile

Remove the advertising clause in the UCB license which Berkeley
rescinded 22 July 1999. Proofed by myself and Theo.

i386 changes to move to ELF. asm cleanup. Change MAXDSIZ to 512M for 1Gsep.
DARPA funded work.

First round of __P removal in sys

multi-line string literals are deprecated in gcc 3.0
pr#2379

branches: 1.23.2;
New try at a level based irq system.

Revert niklas's broken ICU interrupt handling changes; deraadt@ ok

Redo the ICU interrupt handling system to better emulate an APIC system.
This is done by representing the current blockings of interrupts with
a priority level instead of an interrupt mask. This makes it possible to
share implementations for spl* functions/macros between both ICU
and APIC (a must when going MP) systems. In this process, assign
soft interrupts their own levels to match the way things will be on
APIC systems where they actually will be real interrupts and not just
bits in a pending mask as they are now.
Heavily inspired by Bill Studenmunds SMP work in NetBSD

Replace inclusion of <vm/foo.h> with the correct <uvm/bar.h> when necessary.
(Look ma, I might have broken the tree)

PMAP_NEW and UVM are no longer optional on i386.

Add emulation of Linux features to procfs; mostly from NetBSD. ok deraadt@

Add explicit inclusions of signalvar.h to files actually using syms defined
there but relying on an indirect inclusion

branches: 1.16.6;
cnt.foo -> uvmexp.bar in uvm

GCC 2.8 Wall

Do interrupt time accounting by running interrupt handlers with a seperate
code segment selector (otherwise identical to the standard kernel code
selector); idea by Dave Richards <richards@zso.dec.com>

Provide more information on FP exceptions

add type & union sigval args to sendsig/trapsignal

add another parameter to trapsignal() and sendsig() -- fault addr to be
delivered with in the siginfo information

Add:
* GPL_MATH_EMULATE: fpemul
* ALWAYS_MATH_EMULATE: causes npx probe to always fail.
* better fp context in struct pcb.

sync

sync with 0504; prototype changes

partial sync with netbsd 960418, more to come

Fix back wrong patches.

Merging w/ NetBSD 021796.
speaker upgraded to the current.
some changes to the VM stuff (ie kern_thread.c added and so).

Pass device name to interrupt establish routines so it can be recorded in
the device interrupt chain structures (isa, pci)

Move interrupt chain structure definition to <machine/psl.h> so vmstat can
get at it (i386)

Remove hack to count interrupts the old way (i386)

from netbsd: Put #includes in the right order

from netbsd:
The IST_* and IPL_* constants are not bus-specific; don't treat them as such.

branches: 1.1.1;
Initial revision

an ugly white space

Push KERNEL_LOCK/UNLOCK() dance inside trapsignal().

ok kettenis@, visa@

Use the MI interrupt enable/distable API instead of the MD one on i386 and
remove the MD API.

ok deraadt@

More steps for i386 Meltdown fix:

- provide struct cpu_info_full
- prepare K-U sections
- reorganize interrupt, trap, syscall entry to use K-U trampoline
- prepare pmap for entering special mappings, the mappings are not
setup yet

This code will already trigger performance issues. We do more tlb
flushes, but we do not unmap the kernel yet. The latter
will be needed to prevent Meltdown.

from hshoexer@; input guenther@; OK mlarkin@ deraadt@

Change ci_dev.dv_xname to ci_dev->dv_xname in debugging printf()
to make it compile when enabled.
from hshoexer@

Recommit preparation for i386 Meltdown fix after OpenBSD 6.3 release.

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

iBackout the preparations for fixing Meltdown on i386. The task was
only halfway done and the current state does not help anybody. For
OpenBSD 6.3 release go back to the original code before 2018/03/13.
This gives us a stable release and the changes will come back later.
discussed with guenther@ deraadt@ hshoexer@

Preparation for i386 Meltdown fix:

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

Protect the calls to trapsignal() and psignal() with a kernel lock
when processing an npx interrupt. This fixes a kernel locked
assertion in postsig_done() during the libc ieeefp/except regression
test.
OK visa@

Delete unnecessary <sys/file.h> includes

ok millert@ krw@

Kill SPINLOCK_SPIN_HOOK, use CPU_BUSY_CYCLE() instead.

ok visa@, kettenis@

deprecate use of sys/lock.h and replace it with sys/atomic.h or
machine/lock.h as appropriate.

It's been a quarter century: we can assume volatile is present with that name.

ok dlg@ mpi@ deraadt@

Copied from amd64..

Use an explicit suffix for the "fld" instruction to shut up clang. The correct
instruction is fldl since we try to load a double-precision value.
GCC actually gets it wrong and emits "flds" (which is harmless).

ok guenther@

remove stale cyrix emc support, and switch to just using intial npxcw
ok kettenis

Revert art@'s moving around of the KERNEL_LOCK()/KERNEL_UNLOCK() calls,
as it causes hangs in some ports, including libsigsegv's configure script

confirmed by krw@, landry@

There is a bunch of places in the kernel entry points where we don't
hold the kernel lock, but still need call one function that needs it.

Instead of grabbing the lock all over the place, move the locks into
the affected functions: trapsignal, scdebug*, ktrsyscall, ktrsysret,
systrace_redirect and ADDUPROF. In the cases we already hold the biglock
we'll just recurse.

kettenis@, beck@ ok

Clean up after P_BIGLOCK removal.
KERNEL_PROC_LOCK -> KERNEL_LOCK
KERNEL_PROC_UNLOCK -> KERNEL_UNLOCK

oga@ ok

When reading MXCSR from userland sigcontext or a ptrace request,
mask out invalid bits to prevent a protect fault.

Original diff by joshe@; further feedback and ok kettenis@

Back out previous, it appears to be broken.

When reading MXCSR from userland sigcontext, mask out invalid bits.

This prevents a protection fault if a userland signal handler
scribbles all over it's struct sigcontext

Help from and ok guenther@ kettenis@

Change the strategy for using the FPU in the kernel. The kernel FPU state
is now shared with all processes/threads. As a result, you can now use the
FPU in true process context (instead of just in kernel threads), but you
need to make sure you restore the default FPU state before calling
fpu_kernel_exit() if you change rounding mode, precision or exception masks.

Lots of discussion with thib@ and Mike Belopuhov.

ok thib@, deraadt@

Don't leak the contents of the FPU and SSE registers between processes.

ok deraadt@

Make sure that the FPU IPIs shoot down the right FPU context and not the FPU
context of some random process that happened to be switched onto the FPU
after the decision was made to send the IPI.

fpu_kernel_{enter,exit}; Functions to allow the use of
the FPU in the kernel.

From Mike Belopuhov; Little bits by myself.

Comments/OK kettenis@

delete balony comment; ok kettenis

Load %mxcsr when initializing the FPU on machines that support SSE.

ok deraadt@

Properly handle XMM exceptions instead of panicing the kernel. Fixes the
issue reported by Slava Pestov.

ok deraadt@

branches: 1.44.2; 1.44.6;
Fix "fp_save ipi didn't" panic, and move i386/amd64 closer in the process.
Positive test results by a handful of people. Ok kettenis@

branches: 1.43.4;
quite a bit of simplification by removing cpu classes.
also assume that 386 cpus are really unknown, and promote them to 486
instead of panic.

Short-circuit the detection of the FPU by checking the CPUID features,
as done in FreeBSD. Fixes problems on AMD Geode LX-800 where our
detection logic doesn't detect the FP exception.

ok deraadt@ kettenis@

ansi/deregister

branches: 1.40.2;
Kill option GPL_MATH_EMULATE.

ok deraadt@ and many others.

from todays freebsd advisory: fxrstor on amd cpu does not restore fip,fdp,fop thus leaking other proc's execution history; deraadt@ ok

time-bind and lower power in lock spinning and a couple of other ipi loops by using pause insn; brad@ ok tedu@ ok and feedback krw@ testing

branches: 1.37.2; 1.37.4;
oops. on 686 we count fpswtch in another function...

count fpu lazy context switches; deraadt@ ok

branches: 1.35.2;
SIGFPE signal codes for SSE.
nice deraadt@

Fix conversion of interrupt frame to trap frame.
ok deraadt@, niklas@

Don't assert that P_INMEM is set in p_flag. The assert is only almost
correct because P_INMEM is for various reasons removed before cpu_swapout
is called and cpu_swapout in turn calls npxsave_proc.

deraadt@ ok

debranch SMP, have fun

Enable FXSR all the time, and cope with NPX/FXSR conversions; from netbsd.
Pass SSE/SSE2/XCRYPT flags out via syctl, and prepare for being able to do
xcrypt-* in userland; ok naddy, tested a lot by pvalchev and jolan, also
works on amd64 in 32bit mode

workaround: during npx exception testing, delay(1). For some reason
amd64 (in 32 bit mode) and the new transmeta cpus want this, and lock
up otherwise. very odd.

No need to have several implementations of {read,write}_eflags(),
put it to cpufunc.h.
ok weingart@ tedu@

let ALWAYS_MATH_EMULATE actually compile

Remove the advertising clause in the UCB license which Berkeley
rescinded 22 July 1999. Proofed by myself and Theo.

i386 changes to move to ELF. asm cleanup. Change MAXDSIZ to 512M for 1Gsep.
DARPA funded work.

First round of __P removal in sys

multi-line string literals are deprecated in gcc 3.0
pr#2379

branches: 1.23.2;
New try at a level based irq system.

Revert niklas's broken ICU interrupt handling changes; deraadt@ ok

Redo the ICU interrupt handling system to better emulate an APIC system.
This is done by representing the current blockings of interrupts with
a priority level instead of an interrupt mask. This makes it possible to
share implementations for spl* functions/macros between both ICU
and APIC (a must when going MP) systems. In this process, assign
soft interrupts their own levels to match the way things will be on
APIC systems where they actually will be real interrupts and not just
bits in a pending mask as they are now.
Heavily inspired by Bill Studenmunds SMP work in NetBSD

Replace inclusion of <vm/foo.h> with the correct <uvm/bar.h> when necessary.
(Look ma, I might have broken the tree)

PMAP_NEW and UVM are no longer optional on i386.

Add emulation of Linux features to procfs; mostly from NetBSD. ok deraadt@

Add explicit inclusions of signalvar.h to files actually using syms defined
there but relying on an indirect inclusion

branches: 1.16.6;
cnt.foo -> uvmexp.bar in uvm

GCC 2.8 Wall

Do interrupt time accounting by running interrupt handlers with a seperate
code segment selector (otherwise identical to the standard kernel code
selector); idea by Dave Richards <richards@zso.dec.com>

Provide more information on FP exceptions

add type & union sigval args to sendsig/trapsignal

add another parameter to trapsignal() and sendsig() -- fault addr to be
delivered with in the siginfo information

Add:
* GPL_MATH_EMULATE: fpemul
* ALWAYS_MATH_EMULATE: causes npx probe to always fail.
* better fp context in struct pcb.

sync

sync with 0504; prototype changes

partial sync with netbsd 960418, more to come

Fix back wrong patches.

Merging w/ NetBSD 021796.
speaker upgraded to the current.
some changes to the VM stuff (ie kern_thread.c added and so).

Pass device name to interrupt establish routines so it can be recorded in
the device interrupt chain structures (isa, pci)

Move interrupt chain structure definition to <machine/psl.h> so vmstat can
get at it (i386)

Remove hack to count interrupts the old way (i386)

from netbsd: Put #includes in the right order

from netbsd:
The IST_* and IPL_* constants are not bus-specific; don't treat them as such.

branches: 1.1.1;
Initial revision

Push KERNEL_LOCK/UNLOCK() dance inside trapsignal().

ok kettenis@, visa@

Use the MI interrupt enable/distable API instead of the MD one on i386 and
remove the MD API.

ok deraadt@

More steps for i386 Meltdown fix:

- provide struct cpu_info_full
- prepare K-U sections
- reorganize interrupt, trap, syscall entry to use K-U trampoline
- prepare pmap for entering special mappings, the mappings are not
setup yet

This code will already trigger performance issues. We do more tlb
flushes, but we do not unmap the kernel yet. The latter
will be needed to prevent Meltdown.

from hshoexer@; input guenther@; OK mlarkin@ deraadt@

Change ci_dev.dv_xname to ci_dev->dv_xname in debugging printf()
to make it compile when enabled.
from hshoexer@

Recommit preparation for i386 Meltdown fix after OpenBSD 6.3 release.

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

iBackout the preparations for fixing Meltdown on i386. The task was
only halfway done and the current state does not help anybody. For
OpenBSD 6.3 release go back to the original code before 2018/03/13.
This gives us a stable release and the changes will come back later.
discussed with guenther@ deraadt@ hshoexer@

Preparation for i386 Meltdown fix:

- provide a cpu_softc for cpu_attach() etc.
- replace per PCB TSS with per CPU TSS

The first change prepares for cpu_info being embedded in a
cpu_full_info. Therefore during autoconf/cpu_attach we hand down
a softc.

The second change removes the per PCB TSS. We now have one TSS per
CPU, thus in cpu_switchto() we only have to patch the ring 0 stack
pointer instead of loading a new TSS. This also allows for cleaning
up the GDT, so we only have a single slot for the TSS.

from hshoexer@; OK deraadt@

Protect the calls to trapsignal() and psignal() with a kernel lock
when processing an npx interrupt. This fixes a kernel locked
assertion in postsig_done() during the libc ieeefp/except regression
test.
OK visa@

Delete unnecessary <sys/file.h> includes

ok millert@ krw@

Kill SPINLOCK_SPIN_HOOK, use CPU_BUSY_CYCLE() instead.

ok visa@, kettenis@

deprecate use of sys/lock.h and replace it with sys/atomic.h or
machine/lock.h as appropriate.

It's been a quarter century: we can assume volatile is present with that name.

ok dlg@ mpi@ deraadt@

Copied from amd64..

Use an explicit suffix for the "fld" instruction to shut up clang. The correct
instruction is fldl since we try to load a double-precision value.
GCC actually gets it wrong and emits "flds" (which is harmless).

ok guenther@

remove stale cyrix emc support, and switch to just using intial npxcw
ok kettenis

Revert art@'s moving around of the KERNEL_LOCK()/KERNEL_UNLOCK() calls,
as it causes hangs in some ports, including libsigsegv's configure script

confirmed by krw@, landry@

There is a bunch of places in the kernel entry points where we don't
hold the kernel lock, but still need call one function that needs it.

Instead of grabbing the lock all over the place, move the locks into
the affected functions: trapsignal, scdebug*, ktrsyscall, ktrsysret,
systrace_redirect and ADDUPROF. In the cases we already hold the biglock
we'll just recurse.

kettenis@, beck@ ok

Clean up after P_BIGLOCK removal.
KERNEL_PROC_LOCK -> KERNEL_LOCK
KERNEL_PROC_UNLOCK -> KERNEL_UNLOCK

oga@ ok

When reading MXCSR from userland sigcontext or a ptrace request,
mask out invalid bits to prevent a protect fault.

Original diff by joshe@; further feedback and ok kettenis@

Back out previous, it appears to be broken.

When reading MXCSR from userland sigcontext, mask out invalid bits.

This prevents a protection fault if a userland signal handler
scribbles all over it's struct sigcontext

Help from and ok guenther@ kettenis@

Change the strategy for using the FPU in the kernel. The kernel FPU state
is now shared with all processes/threads. As a result, you can now use the
FPU in true process context (instead of just in kernel threads), but you
need to make sure you restore the default FPU state before calling
fpu_kernel_exit() if you change rounding mode, precision or exception masks.

Lots of discussion with thib@ and Mike Belopuhov.

ok thib@, deraadt@

Don't leak the contents of the FPU and SSE registers between processes.

ok deraadt@

Make sure that the FPU IPIs shoot down the right FPU context and not the FPU
context of some random process that happened to be switched onto the FPU
after the decision was made to send the IPI.

fpu_kernel_{enter,exit}; Functions to allow the use of
the FPU in the kernel.

From Mike Belopuhov; Little bits by myself.

Comments/OK kettenis@

delete balony comment; ok kettenis

Load %mxcsr when initializing the FPU on machines that support SSE.

ok deraadt@

Properly handle XMM exceptions instead of panicing the kernel. Fixes the
issue reported by Slava Pestov.

ok deraadt@

branches: 1.44.2; 1.44.6;
Fix "fp_save ipi didn't" panic, and move i386/amd64 closer in the process.
Positive test results by a handful of people. Ok kettenis@

branches: 1.43.4;
quite a bit of simplification by removing cpu classes.
also assume that 386 cpus are really unknown, and promote them to 486
instead of panic.

Short-circuit the detection of the FPU by checking the CPUID features,
as done in FreeBSD. Fixes problems on AMD Geode LX-800 where our
detection logic doesn't detect the FP exception.

ok deraadt@ kettenis@

ansi/deregister

branches: 1.40.2;
Kill option GPL_MATH_EMULATE.

ok deraadt@ and many others.

from todays freebsd advisory: fxrstor on amd cpu does not restore fip,fdp,fop thus leaking other proc's execution history; deraadt@ ok

time-bind and lower power in lock spinning and a couple of other ipi loops by using pause insn; brad@ ok tedu@ ok and feedback krw@ testing

branches: 1.37.2; 1.37.4;
oops. on 686 we count fpswtch in another function...

count fpu lazy context switches; deraadt@ ok

branches: 1.35.2;
SIGFPE signal codes for SSE.
nice deraadt@

Fix conversion of interrupt frame to trap frame.
ok deraadt@, niklas@

Don't assert that P_INMEM is set in p_flag. The assert is only almost
correct because P_INMEM is for various reasons removed before cpu_swapout
is called and cpu_swapout in turn calls npxsave_proc.

deraadt@ ok

debranch SMP, have fun

Enable FXSR all the time, and cope with NPX/FXSR conversions; from netbsd.
Pass SSE/SSE2/XCRYPT flags out via syctl, and prepare for being able to do
xcrypt-* in userland; ok naddy, tested a lot by pvalchev and jolan, also
works on amd64 in 32bit mode

workaround: during npx exception testing, delay(1). For some reason
amd64 (in 32 bit mode) and the new transmeta cpus want this, and lock
up otherwise. very odd.

No need to have several implementations of {read,write}_eflags(),
put it to cpufunc.h.
ok weingart@ tedu@

let ALWAYS_MATH_EMULATE actually compile