- i've extended the scheduler context-switch mechanism with the following
  per-arch defines:

        prepare_arch_schedule(prev_task);
        finish_arch_schedule(prev_task);
        prepare_arch_switch(rq);
        finish_arch_switch(rq);

- plus switch_to() takes 3 parameters again:

        switch_to(prev, next, last);

- schedule_tail() has the 'prev' task parameter again, it must be passed
  over in switch_to() and passed in to the fork() startup path.

architectures that need to unlock the runqueue before doing the switch can
do the following:

 #define prepare_arch_schedule(prev)             task_lock(prev)
 #define finish_arch_schedule(prev)              task_unlock(prev)
 #define prepare_arch_switch(rq)                 spin_unlock(&(rq)->lock)
 #define finish_arch_switch(rq)                  __sti()

this way the task-lock makes sure that a task is not scheduled on some
other CPU before the switch-out finishes, but the runqueue lock is
dropped. (Local interrupts are kept disabled in this variant, just to
exclude things like TLB flushes - if that matters.)

architectures that can hold the runqueue lock during context-switch can do
the following simplification:

 #define prepare_arch_schedule(prev)             do { } while(0)
 #define finish_arch_schedule(prev)              do { } while(0)
 #define prepare_arch_switch(rq)                 do { } while(0)
 #define finish_arch_switch(rq)                  spin_unlock_irq(&(rq)->lock)

further optimizations possible in the 'simple' variant:

- an architecture does not have to handle the 'last' parameter in
  switch_to() if the 'prev' parameter is unused in finish_arch_schedule().
  This way the inlined return value of context_switch() too gets optimized
  away at compile-time.

- an architecture does not have to pass the 'prev' pointer to
  schedule_tail(), if the 'prev' parameter is unused in
  finish_arch_schedule().

the x86 architecture makes use of these optimizations.

Via this solution we have a reasonably flexible context-switch setup which
falls back to the current (faster) code on x86, but on other platforms the
runqueue lock can be dropped before doing the context-switch as well.

        Ingo

NOTE: i have coded and tested the 'complex' variant on x86 as well to make
      sure it works for you on Sparc64 - but since x86's switch_mm() is
      not too subtle it can use the simpler variant. [ The following
      things had to be done to make x86 arch use the complex variant: the
      4 complex macros have to be used in system.h, entry.S has to
      'pushl %ebx' and 'addl $4, %esp' around the call to schedule_tail(),
      and switch_to() had to be reverted to the 3-parameter variant
      present in the 2.4 kernels.

NOTE2: prepare_to_switch() functionality has been moved into the
       prepare_arch_switch() macro.

NOTE3: please use macros for prepare|finish_arch_switch() so that we can
       keep the scheduler data structures internal to sched.c.

	This patch  fixes the Compaq WL210 problems I've been having.

The symptoms were that no interupts were ever received from the card;
the BIOS reported it had assigned INTA to IRQ 10; but the yenta
driver reported IRQ 17.

The fix is not to rely on the BIOS's initialisation of interrupt
routing in the TI1410, but to do it explicitly.

This may break laptops that actually use the parallel ISA IRQ
mechanism.

on non-x86 architectures)

in structures and unions (and why not other variables too..)

Many old ISA drivers are not 64bit clean. To avoid dealing with them on
x86-64 I guarded them with CONFIG_ISA. This is also helpful for other
archs (including modern x86) that do not have ISA slots anymore.
The patch has been reviewed by Linux-Kernel, with some additional
cleanups by me. It misses some directories like ISDN or sound drivers.

This fixes a potential security issue on i386. FXRSTOR raises an #GP
when the MXCSR field in the FXSAVE frame contains illegal values. By
putting an signal frame just at a border to an unmapped page a malicious
user could potentially cause an exception in the context switch using this.
When the EFAULT occurs in the last bytes of the signal frame the MXCSR
check would be skipped in the old code, but the rogue value would be already
in task_struct, causing problems later.

Here is the minimal fix for this.

Originally discovered by Andrea while doing x86-64 work. It was fixed in
a different way a bit more intrusive way on x86-64 - all FXRSTOR in the
kernel are guarded by exception handlers.

It's nearly impossible to get the correct CPU names for AMD CPUs in
the MP table (would need BIOS calls etc.). The decoding of CPU names form
the mptable doesn't look too useful anyways, because the correct CPU name
from CPUID is printed only a few lines later. This patch just removes
the name table and replaces it with a printout of the hex value.
This mirrors a change in x86-64

Remove myself as networking maintainer. I haven't done much for networking
for a long time and it doesn't look likely it'll change anytime soon.
Also David/Alexey are doing a fine job without me.

Here is the big 2.5.21 x86-64 sync patch. It only touches arch/x86_64
and include/asm-x86_64. It requires a few other changes that I'm sending
in separate mail.

Changes:
- merge wit 2.5.21
- merge from 2.5.21/i386 (new PCI code, new LDT code etc.)
- sync with 2.4-x86_64 tree.
- minor updates to 32bit emulation
- better early console; including serial support.
- now set up dummy PDA for booting to avoid problems
- Fix GS reloading in context switch one instruction race
- Remove hardcoded names from mpparse code
- Fix inline assembly for RAID-5 xor (similar change needed for i386)
- Real per cpu data support based on PDA field
- Cleanup of offset.c generation requested by Kai: it only puts structure
  offsets into offset.h now.
- Fix i387 fxsave signal frame problems.
- Add uname emulation via personality ("linux32")
- New SSE optimized checksum-copy, copy*user, memcpy, clear_page, copy_page
  functions. Other tunings/cleanups in checksum and other user memory
  access function.
- check if exception table is really sorted
- Cleanups in page table handling in preparation of non executable pages
  support.
- Cleanup PDA access to not require offset.h (thanks to kai for kicking me
  to this)
- use long long for u64/s64 to avoid more warnings
- remove CONFIG_ISA
- fix various bugs and other cleanups

into home.transmeta.com:/home/torvalds/v2.5/linux

  enable/disable preemption to not spend too much time with
  preemption disabled. wait_task_inactive() can take quite some
  time occasionally ...

  set p->thread_info->cpu directly if the task is not running
  and is not on any runqueue.

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

Recent header cleanup broke this driver. Needs #include <linux/init.h>

ACPI interpreter update
Change non-interpreter code to account for the interpreter changes.

This update gets things compiling again after a device model change.

This adds preliminary support for _TRA resource on PCI root bridges (Bjorn
Helgaas)

into home.transmeta.com:/home/torvalds/v2.5/linux

end_request(int) turned to end_request(req, int); all old callers
are converted to end_request(CURRENT, value).  Now we can start
killing CURRENT/QUEUE - end_request() was the last obstacle to
that.

ditto for DEVICE_INTR/CLEAR_INTR.  do_mfm/do_acsi/do_hd/... are
explicitly declared in the drivers that use DEVICE_INTR stuff;
DEVICE_INTR/CLEAR_INTR is gone from blk.h.

#define DEVICE_NAME ... removed from the drivers that never use
that macro.

monstrous cascade of #ifdefs is gone from blk.h.  This is the first
step of cleanup - we move #define in question into the code that
includes blk.h (i.e. DEVICE_NR, DEVICE_NAME and DEVICE_INTR are
explicitly set by drivers).

->s_dev is switched to dev_t.  Everything that uses it uses it as
a number - i.e. all instances are either minor() or kdev_t_to_nr().

->j_dev is gone from reiserfs journal.  All its uses had been replaced
with use of ->j_dev_bd; it's simply not used anymore.

get_super() split in two functions - get_super(bdev) and
user_get_super(dev_t).  Callers that used get_super() to get superblock
by (dev_t) syscall argument switched to the latter; the rest had
block_device in question and switched to passing it.

ROOT_DEV converted to dev_t.  There's no reason for it to be kdev_t -
it's both used as a number and set as a number.

do_getattr() renamed to vfs_getattr(), exported and used by
loop.c and nfsd - both blindly accessed ->i_dev/->i_ino of
random inodes when they ought to be calling ->getattr() to
get that information.

misc gratitious kdev_t uses fixed (janitorial stuff - e.g. quota
patches had reintroduced kdevname(sb->s_dev) in places where we'd
switched to sb->s_id, etc.)

added bdev_read_only() - analog of is_read_only() using block_device.
Almost all callers of is_read_only() converted.