git commit 54a01510 ("asmlinkage_protect
replaces prevent_tail_call") causes this build failure on s390:

    AS      arch/s390/kernel/entry64.o
  In file included from arch/s390/kernel/entry64.S:14:
  include/linux/linkage.h:34: error: syntax error in macro parameter list
  make[1]: *** [arch/s390/kernel/entry64.o] Error 1
  make: *** [arch/s390/kernel] Error 2

and some other architectures.  The reason is that some architectures add
the "-traditional" flag to the invocation of $(AS), which disables
variadic macro argument support.

So just surround the new define with an #ifndef __ASSEMBLY__ to prevent
any side effects on asm code.

Cc: Roland McGrath <roland@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: NHeiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Increase the PNP "number of devices" limit.  We currently use an unsigned
char, which limits us to 256 devices per protocol.  This patch changes that to
an unsigned int.

Not all backends can take advantage of this: we limit ISAPNP to 10 devices in
isapnp_cfg_begin(), and PNPBIOS is limited to 256 devices because the BIOS
interfaces use a one-byte device node number.

But there is no limit on the number of PNPACPI devices we may have.  Large HP
Integrity machines have more than 256, which causes the current "unsigned char
number" to wrap around.  This causes errors like this:

    pnp: PnP ACPI init
    kobject_add failed for 00:00 with -EEXIST, don't try to register things with the same name in the same directory.

    Call Trace:
     [<a000000100010720>] show_stack+0x40/0xa0
     [<a0000001000107b0>] dump_stack+0x30/0x60
     [<a0000001001dbdf0>] kobject_add+0x290/0x2c0
     [<a0000001002bfd40>] device_add+0x160/0x860
     [<a0000001002c0470>] device_register+0x30/0x60
     [<a00000010026ba70>] __pnp_add_device+0x130/0x180
     [<a00000010026bb70>] pnp_add_device+0xb0/0xe0
     [<a0000001007f2730>] pnpacpi_add_device+0x510/0x5a0
     [<a0000001007f2810>] pnpacpi_add_device_handler+0x50/0x80

This patch increases the limit to fix this PNPACPI problem.  It should not
have any adverse effect on ISAPNP or PNPBIOS because their limits are still
enforced in the backends.
Signed-off-by: NBjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It's really a pretty ugly thing to need, and some day it will hopefully
be obviated by teaching gcc about the magic calling conventions for the
low-level system call code, but in the meantime we can at least add big
honking comments about why we need these insane and strange macros.

I took my comments from my version of the macro, but I ended up deciding
to just pick Roland's version of the actual code instead (with his
prettier syntax that uses vararg macros).  Thus the previous two commits
that actually implement it.
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The prevent_tail_call() macro works around the problem of the compiler
clobbering argument words on the stack, which for asmlinkage functions
is the caller's (user's) struct pt_regs.  The tail/sibling-call
optimization is not the only way that the compiler can decide to use
stack argument words as scratch space, which we have to prevent.
Other optimizations can do it too.

Until we have new compiler support to make "asmlinkage" binding on the
compiler's own use of the stack argument frame, we have work around all
the manifestations of this issue that crop up.

More cases seem to be prevented by also keeping the incoming argument
variables live at the end of the function.  This makes their original
stack slots attractive places to leave those variables, so the compiler
tends not clobber them for something else.  It's still no guarantee, but
it handles some observed cases that prevent_tail_call() did not.
Signed-off-by: NRoland McGrath <roland@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Don't make smp_{r,w,}mb() interpolate a MEMBAR instruction when CONFIG_SMP=n as
SMP memory barries on UP systems should interpolate a compiler barrier only.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Use traps 120-126 to emulate atomic cmpxchg32, xchg32, and XOR-, OR-, AND-, SUB-
and ADD-to-memory operations for userspace.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Move STACK_TOP_MAX up so that we don't try moving the stack above it as that
causes setup_arg_pages() to malfunction.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Handle update_mmu_cache() being called when current->mm is NULL.

We cache static TLB mappings for the current page table in DAMPR4 and DAMPR5
on the theory that the next data lookup is likely to be in the same general
region, and thus is likely to be mapped by the same page table.  However, we
can't get this information if we can't access the appropriate mm_struct.

If current->mm is NULL, we just clear the cache in the knowledge that the TLB
miss handlers will load it.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This fixes Bugzilla #10384

tcp_simple_retransmit does L increment without any checking
whatsoever for overflowing S+L when Reno is in use.

The simplest scenario I can currently think of is rather
complex in practice (there might be some more straightforward
cases though). Ie., if mss is reduced during mtu probing, it
may end up marking everything lost and if some duplicate ACKs
arrived prior to that sacked_out will be non-zero as well,
leading to S+L > packets_out, tcp_clean_rtx_queue on the next
cumulative ACK or tcp_fastretrans_alert on the next duplicate
ACK will fix the S counter.

More straightforward (but questionable) solution would be to
just call tcp_reset_reno_sack() in tcp_simple_retransmit but
it would negatively impact the probe's retransmission, ie.,
the retransmissions would not occur if some duplicate ACKs
had arrived.

So I had to add reno sacked_out reseting to CA_Loss state
when the first cumulative ACK arrives (this stale sacked_out
might actually be the explanation for the reports of left_out
overflows in kernel prior to 2.6.23 and S+L overflow reports
of 2.6.24). However, this alone won't be enough to fix kernel
before 2.6.24 because it is building on top of the commit
1b6d427b ([TCP]: Reduce sacked_out with reno when purging
write_queue) to keep the sacked_out from overflowing.
Signed-off-by: NIlpo Järvinen <ilpo.jarvinen@helsinki.fi>
Reported-by: NAlessandro Suardi <alessandro.suardi@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

flush_cache_vmap / flush_cache_vunmap were calling flush_cache_all which -
having been deprecated - turned into a nop ...
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>

The 'disable_cb' callback is designed as an optimization to tell the host
we don't need callbacks now.  As it is not reliable, the debug check is
overzealous: it can happen on two CPUs at the same time.  Document this.

Even if it were reliable, the virtio_net driver doesn't disable
callbacks on transmit so the START_USE/END_USE debugging reentrance
protection can be easily tripped even on UP.

Thanks to Balaji Rao for the bug report and testing.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
CC: Balaji Rao <balajirrao@gmail.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

ASM_NOP's for 64-bit kernel with CONFIG_GENERIC_CPU is broken
with the recent x86 nops merge. They were using GENERIC_NOPS
which will truncate the upper 32bits of %rsi, because of the missing
64bit rex prefix.

For now, fall back ASM NOPS for generic cpu to K8 NOPS, similar
to the code before the wrong x86 nop merge.

This should resolve the crash seen by Ingo on a test-system:

BUG: unable to handle kernel paging request at 00000000d80d8ee8
IP: [<ffffffff802121af>] save_i387_ia32+0x61/0xd8
PGD b8e0067 PUD 51490067 PMD 0
Oops: 0000 [1] SMP
CPU 2
Modules linked in:
Pid: 3871, comm: distcc Not tainted 2.6.25-rc7-sched-devel.git-x86-latest.git #359
RIP: 0010:[<ffffffff802121af>]  [<ffffffff802121af>] save_i387_ia32+0x61/0xd8
RSP: 0000:ffff81003abd3cb8  EFLAGS: 00010246
RAX: ffff810082e93400 RBX: 00000000ffc37f84 RCX: ffff8100d80d8ee0
RDX: 0000000000000000 RSI: 00000000d80d8ee0 RDI: ffff810082e93400
RBP: 00000000ffc37fdc R08: 00000000ffc37f88 R09: 0000000000000008
R10: ffff81003abd2000 R11: 0000000000000000 R12: ffff810082e93400
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffff81011fb12dc0(0063) knlGS:00000000f7f1a6c0
CS:  0010 DS: 002b ES: 002b CR0: 0000000080050033
CR2: 00000000d80d8ee8 CR3: 0000000076922000 CR4: 00000000000006e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process distcc (pid: 3871, threadinfo ffff81003abd2000, task ffff8100d80d8ee0)
Stack:  ffff8100bb670380 ffffffff8026de50 0000000000000118 0000000000000002
 0000000000000002 ffff81003abd3e68 ffff81003abd3ed8 ffff81003abd3de8
 ffff81003abd3d18 ffffffff80229785 ffff8100d80d8ee0 ffff810001041280
Call Trace:
 [<ffffffff8026de50>] ? __generic_file_aio_write_nolock+0x343/0x377
 [<ffffffff80229785>] ? update_curr+0x54/0x64
 [<ffffffff80227cd3>] ? ia32_setup_sigcontext+0x125/0x1d2
 [<ffffffff8022839f>] ? ia32_setup_frame+0x73/0x1a5
 [<ffffffff8020b2a5>] ? do_notify_resume+0x1aa/0x7db
 [<ffffffff8024ae8c>] ? getnstimeofday+0x31/0x85
 [<ffffffff80249858>] ? ktime_get_ts+0x17/0x48
 [<ffffffff80249933>] ? ktime_get+0xc/0x41
 [<ffffffff8024973e>] ? hrtimer_nanosleep+0x75/0xd5
 [<ffffffff80249261>] ? hrtimer_wakeup+0x0/0x21
 [<ffffffff8020bfbc>] ? int_signal+0x12/0x17
 [<ffffffff8030e6b3>] ? dummy_file_free_security+0x0/0x1

Code: a6 08 05 00 00 f6 40 14 01 74 34 4c 89 e7 48 0f ae 07 48 8b 86 08 05 00 00 80 78 02 00 79 02 db e2 90 8d b4 26 00 00 00 00 89 f6 <48> 8b 46 08 83 60 14 fe 0f 20 c0 48 83 c8 08 0f 22 c0 eb 07 c6 
Signed-off-by: NSuresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

The effects of cgroup_disable=foo are:

- foo isn't auto-mounted if you mount all cgroups in a single hierarchy
- foo isn't visible as an individually mountable subsystem

As a result there will only ever be one call to foo->create(), at init time;
all processes will stay in this group, and the group will never be mounted on
a visible hierarchy.  Any additional effects (e.g.  not allocating metadata)
are up to the foo subsystem.

This doesn't handle early_init subsystems (their "disabled" bit isn't set be,
but it could easily be extended to do so if any of the early_init systems
wanted it - I think it would just involve some nastier parameter processing
since it would occur before the command-line argument parser had been run.

Hugh said:

  Ballpark figures, I'm trying to get this question out rather than
  processing the exact numbers: CONFIG_CGROUP_MEM_RES_CTLR adds 15% overhead
  to the affected paths, booting with cgroup_disable=memory cuts that back to
  1% overhead (due to slightly bigger struct page).

  I'm no expert on distros, they may have no interest whatever in
  CONFIG_CGROUP_MEM_RES_CTLR=y; and the rest of us can easily build with or
  without it, or apply the cgroup_disable=memory patches.

Unix bench's execl test result on x86_64 was

== just after boot without mounting any cgroup fs.==
mem_cgorup=off : Execl Throughput       43.0     3150.1      732.6
mem_cgroup=on  : Execl Throughput       43.0     2932.6      682.0
==

[lizf@cn.fujitsu.com: fix boot option parsing]
Signed-off-by: NBalbir Singh <balbir@linux.vnet.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Sudhir Kumar <skumar@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NSergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>

The commits:

commit 37a47db8
Author: Balaji Rao <balajirrao@gmail.com>
Date:   Wed Jan 30 13:30:03 2008 +0100

    x86: assign IRQs to HPET timers, fix

and

commit e3f37a54
Author: Balaji Rao <balajirrao@gmail.com>
Date:   Wed Jan 30 13:30:03 2008 +0100

    x86: assign IRQs to HPET timers

have been identified to cause a regression on some platforms due to
the assignement of legacy IRQs which makes the legacy devices
connected to those IRQs disfunctional.

Revert them.

This fixes http://bugzilla.kernel.org/show_bug.cgi?id=10382Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

25-rc* stopped working with CONFIG_X86_VSMP on vSMP machines.

Looks like the vsmp irq ops got accidentally removed during merge of x86_64
pvops in 2.6.25. -- commit 6abcd98f removed
vsmp irq ops.

Tested with both CONFIG_X86_VSMP and without CONFIG_X86_VSMP, on vSMP and non
vSMP x86_64 machines.

Please apply.
Signed-off-by: NRavikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

SAT passthrus don't really fit into ATAPI_MISC class.  SAT passthru
commands always transfer multiple of 512 bytes and variable length
response is not allowed.  This patch creates a separate category -
ATAPI_PASS_THRU - for these.

This fixes HSM violation on "hdparm -I".
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Uninline atapi_cmd_type().  It doesn't really have to be inline and
more case will be added which need to access unexported libata
variable.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

A nasty compile error:

In file included from security/keys/internal.h:16,
                 from security/keys/sysctl.c:14:
include/linux/key-ui.h: In function 'key_permission':
include/linux/key-ui.h:51: error: invalid use of undefined type 'struct task_struct'

apparently the compiler has decided that it needs to know sizeof(task_struct)
so that it can add zero to a task_struct* (which is rather dumb of it).

Getting task_struct in scope in these deeply-nested headers is scary-looking,
so let's just remove the "+ 0".

Cc: David Howells <dhowells@redhat.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Make dma_alloc_coherent respect gfp flags (__GFP_COMP is one that
matters).
Signed-off-by: NIvan Kokshaysky <ink@jurassic.park.msu.ru>
Tested-by: NMichael Cree <mcree@orcon.net.nz>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Jaroslav Kysela <perex@perex.cz>
Cc: Takashi Iwai <tiwai@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Currently include/linux/kvm.h is not considered by make headers_install,
because Kbuild cannot handle " unifdef-$(CONFIG_FOO) += foo.h.  This problem
was introduced by

commit fb56dbb3
Author: Avi Kivity <avi@qumranet.com>
Date:   Sun Dec 2 10:50:06 2007 +0200

    KVM: Export include/linux/kvm.h only if $ARCH actually supports KVM

    Currently, make headers_check barfs due to <asm/kvm.h>, which <linux/kvm.h>
    includes, not existing.  Rather than add a zillion <asm/kvm.h>s, export kvm.
    only if the arch actually supports it.
Signed-off-by: NAvi Kivity <avi@qumranet.com>

which makes this an 2.6.25 regression.

One way of solving the issue is to enhance Kbuild, but Avi and David conviced
me, that changing headers_install is not the way to go.  This patch changes
the definition for linux/kvm.h to unifdef-y.

If  unifdef-y is used for linux/kvm.h "make headers_check" will fail on all
architectures without asm/kvm.h.  Therefore, this patch also provides
asm/kvm.h on all architectures.
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>
Acked-by: NAvi Kivity <avi@qumranet.com>
Cc: Sam Ravnborg <sam@ravnborg.org
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

SLAB_DESTROY_BY_RCU is not a direct substitute for normal call_rcu()
freeing, since it'll page freeing but NOT object freeing. So change
cfq to do the freeing on its own.
Signed-off-by: NFabio Checconi <fabio@gandalf.sssup.it>
Acked-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Work around the CPU clock miscalculation on Au1000DA/HA/HB due the
sys_cpupll register being write-only, i.e. actually do what the comment
before cal_r4off() function advertised for years but the code failed at.
This is achieved by just giving user a chance to define the clock
explicitly  in the board config. via CONFIG_SOC_AU1000_FREQUENCY option,
defaulting to 396 MHz if the option is not given...

The patch is based on the AMD's big unpublished patch, the issue seems to
be an undocumented errata (or feature :-)...
Signed-off-by: NSergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>

Recent driver core change causes references to parent devices being
dropped early, at device_del() time, as opposed to when all children
are freed. This causes oops in evdev with grabbed devices. Take the
reference to the parent input device ourselves to ensure that it
stays around long enough.
Signed-off-by: NDmitry Torokhov <dtor@mail.ru>

Allocate the skb for llc responses with the received packet size by
using the size adjustable llc_frame_alloc.
Don't allocate useless extra payload.
Cleanup magic numbers.

So, this fixes oops.
Reported by Jim Westfall:
kernel: skb_over_panic: text:c0541fc7 len:1000 put:997 head:c166ac00 data:c166ac2f tail:0xc166b017 end:0xc166ac80 dev:eth0
kernel: ------------[ cut here ]------------
kernel: kernel BUG at net/core/skbuff.c:95!
Signed-off-by: NJoonwoo Park <joonwpark81@gmail.com>
Acked-by: NArnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch removes the unused include/asm-sh/floppy.h
(ARCH_MAY_HAVE_PC_FDC was not enabled).
Signed-off-by: NAdrian Bunk <bunk@kernel.org>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Acked-by: NHarvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We need to check for CONFIG_{CRIS,FRV} not {CRIS,FRV}.
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

EH actions are ATA_EH_* not ATA_EHI_*.  Rename ATA_EHI_LPM to
ATA_EH_LPM.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Cc: Kristen Carlson Accardi <kristen.c.accardi@intel.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Comment dev_kfree_skb_irq and dev_kfree_skb_any better.
Signed-off-by: NMatti Linnanvuori <mattilinnanvuori@yahoo.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

kill unnecessary llc_station_mac_sa.
Signed-off-by: NJoonwoo Park <joonwpark81@gmail.com>
Acked-by: NArnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Neither of the headers actually compiles when included from userpsace nor
should it be made available as userspace tools should be using the libraries
or at least headers from e2fsprogs.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Acked-by: N"Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Will replace open-coded variants elsewhere.  Done in the same
style as the 32-bit versions.
Signed-off-by: NHarvey Harrison <harvey.harrison@gmail.com>
Acked-by: NRandy Dunlap <randy.dunlap@oracle.com>
Cc: Johannes Berg <johannes@sipsolutions.net>
Cc: John W. Linville <linville@tuxdriver.com>
Cc: Joe Perches <joe@perches.com>
Cc: Jiri Benc <jbenc@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Discourage people from inappropriately using in_atomic()
Signed-off-by: NJonathan Corbet <corbet@lwn.net>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

... and take it out of ->umount_begin() instances.  Call with all locks
already taken (by do_umount()) and leave calling release_mounts() to
caller (it will do release_mounts() anyway, so we can just put into
the same list).
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

make propagate_mount_busy() exclude references from the vfsmounts
that had been isolated by umount_tree() and are just waiting for
release_mounts() to dispose of their ->mnt_parent/->mnt_mountpoint.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Based upon a lockdep report.

Since ->poll() can be invoked from netpoll with interrupts
disabled, we must not unconditionally enable interrupts
in napi_complete().

Instead we must use local_irq_{save,restore}().

Noticed by Peter Zijlstra:

<irqs disabled>

  netpoll_poll()
    poll_napi()
      spin_trylock(&napi->poll_lock)
      poll_one_napi()
        napi->poll() := sky2_poll()
          napi_complete()
            local_irq_disable()
            local_irq_enable() <--- *BUG*

  <irq>
    irq_exit()
      do_softirq()
        net_rx_action()
          spin_lock(&napi->poll_lock) <--- Deadlock!

Because we still hold the lock....
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Took some cycles to re-read the Lguest Journey end-to-end, fix some
rot and tighten some phrases.

Only comments change.  No new jokes, but a couple of recycled old jokes.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>