I earlier reported, that the error handler for ide-scsi exits prematurely if modprobed
from rc.sysinit. I put in some debug prints to apprehend the culprit responsible for
sending the SIGHUP signal that causes the exit.

This is what my log captured:

Jan  1 12:20:13 fallguy kernel: Process 223 [modprobe] starting scsi error handler
Jan  1 12:20:13 fallguy kernel: Wake up parent of scsi_eh_2, pid 224
Jan  1 12:20:13 fallguy kernel: Signals pending for scsi_eh_2: 00000000 00000000
Jan  1 12:20:13 fallguy kernel: Error handler scsi_eh_2 sleeping
Jan  1 12:20:13 fallguy kernel: scsi2 : SCSI host adapter emulation for IDE ATAPI devices
[detected devices skipped]
Jan  1 12:20:14 fallguy kernel: Signal 15 sent from 181 [rc.sysinit] to 182 [getkey]
Jan  1 12:20:14 fallguy kernel: Signal 1 sent from 22 [init] to 22 [init]
Jan  1 12:20:14 fallguy kernel: Signal 18 sent from 22 [init] to 22 [init]
Jan  1 12:20:14 fallguy kernel: Signal 1 sent from 22 [init] to 22 [init]
Jan  1 12:20:14 fallguy kernel: Signal 1 sent from 22 [init] to 24 [initlog]
Jan  1 12:20:14 fallguy kernel: Signal 1 sent from 22 [init] to 78 [khubd]
Jan  1 12:20:14 fallguy kernel: Signal 1 sent from 22 [init] to 224 [scsi_eh_2]
Jan  1 12:20:14 fallguy kernel: Signals pending for scsi_eh_2: 00000001 00000000
Jan  1 12:20:14 fallguy kernel: Error handler scsi_eh_2 exiting

Here is a snapshot of some processes made during rc.sysinit:

  F   UID   PID  PPID PRI  NI   VSZ  RSS WCHAN  STAT TTY        TIME COMMAND
100     0     1     0  15   0  1332  420 schedu S    ?          0:05 init
...
040     0    22     1  16   0  1332  388 wait4  S    tty1       0:00 init
000     0    23    22  15   0  4116 1316 wait4  S    tty1       0:00 /bin/bash /
040     0    24    23  16   0  2160 1364 schedu S    tty1       0:00 /sbin/initl
...

Init must have forked to exec bash to exec rc.sysinit which then gets re-executed
through initlog. When rc.sysinit ends, the last thing it does is send that TERM
signal from sub-process 181 to getkey (process 182) -- the 'Signal 15 ...' line 
above.

As the forked init (process 22) exits, it sends a flurry of signals to all surviving
processes created from it. That looks like standard "if I am to die I need to take
all my offspring down with me" behavior -- do you agree?

Since we want error handlers to survive, IMHO that means that the choice of signal
for error handler exit is unfortunate. The source of scsi_error suggests SIGPWR
might be a worthy alternative. I think that is true. From inspecting init source,
it is not capable of sending SIGPWR. SIGPWR should never be sent by dying processes
(its sole use should be from a power daemon _to_ init to shut the system down when
the juice is running out).

So I suggest the following changes to hosts.c and scsi_error.c:

Hi -

This simple patch adds scsi_level to the scsi_device sysfs attributes,
mainly for use with Doug Gilbert's lsscsi program.

The attachment modifies the SG_IO ioctl that is in
the block layer to:
   - convey the SCSI status value back via the sg_io_hdr
     structure
   - in the event of CHECK CONDITION, convey the sense
     buffer back via the sg_io_hdr structure
   - set "output" fields in sg_io_hdr structure to sane
     values
   - modify the errno behaviour to be more like the
     SCSI generic driver's SG_IO ioctl

This patch was presented around lk 2.5.51 but fell between
the cracks. The only modification is to take account of
the changes to scsi/scsi_lib.c since then.

to properly de-activate it and make the child_tid logic work
correctly.

Clear %fs/%gs in deactivate_mm() on x86, since our LDT will no
longer be valid after this.

Update mm_release() to deactivate MM state before releasing,
and avoid the expensive child_tid FUTEX if we're the last user
of the MM.

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

This patch, which depends on the previous %ebx -> %ebp patch, removes
all pop instruction in the sysenter return path.

This leaks the thread_info address to user mode but this shouldn't be
a security problem.

This is what happens to the various registers:

	%eax: return value from system call: already in place
	%ebx, %esi, %edi: saved by the C compiler
	%ecx, %edx, %ebp: restored by user mode, fixed values by kernels
	%esp, eip: copied to %ecx/%edx and restored by sysexit
	%ds, %es: initialized to __USER_DS on kernel entry
	%cs, %ss: restored by sysexit based on msr
	%fs, %gs: not modified by the kernel (saved around context switch)
	eflags: not preserved, iopl saved around context switch
	FP, XMM: any code that modifies them must save/restore them

Note that while it is possible to change %ebx, %esi, %edi, %ecx, %edx
or %ebp via struct pt_regs, anything that does should set TIF_IRET or
another work flag (and it hopefully already does).

This patch changes assembly code that accesses thread_info to use %ebp
rather than %ebx.

This allows me to take advantage of the fact that %ebp is restored by
user mode in the sysenter register pop removal patch.

vm86() direct return code updated to match [ Linus ]

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

into groveronline.com:/root/bk/linux-acpi

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

This also cleans up the machine check code generally.

into samba.org:/home/paulus/kernel/for-linus-ppc

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

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

Supply the old termios, along with the max and min acceptable baud
rate to uart_get_baud_rate().  uart_get_baud_rate() will now try to
find a baud rate that satisfies the max and min constraint out of
(requested rate, old rate, 9600 baud).

We remove the code which performed a similar act in uart_get_divisor()
and pass an appropriate min and max baud rate to uart_get_baud_rate()
based on the UART clock rate.

into kernel.bkbits.net:/home/davem/sparc-2.5

into kernel.bkbits.net:/home/davem/net-2.5

- Merged AES code from Adam J. Richter <adam@yggdrasil.com>
- Add kconfig help and test vector code from
  Martin Clausen <martin@ostenfeld.dk>
- Minor cleanups: removed EXPORT_NO_SYMBOLS (not needed for 2.5),
  removed debugging code etc.
- Documentation updates.

into kroah.com:/home/linux/linux/BK/gregkh-2.5

Since we killed off the broken 486 strings copies,
the config item isn't needed any longer.

Next in line for the -W cleanups is this patch from Willy
which I've been carrying since 2.5.29 or so. Should be no
functional differences, just no more warnings when we compile
with -W

s/boundry/boundary/

Without this fix, compiling with CONFIG_DEBUG_SPINLOCK
dies with an unknown variable error..