kernel-ark/include
Nick Piggin 4827bbb06e i386: remove bogus comment about memory barrier
The comment being removed by this patch is incorrect and misleading.

In the following situation:

	1. load  ...
	2. store 1 -> X
	3. wmb
	4. rmb
	5. load  a <- Y
	6. store ...

4 will only ensure ordering of 1 with 5.
3 will only ensure ordering of 2 with 6.

Further, a CPU with strictly in-order stores will still only provide that
2 and 6 are ordered (effectively, it is the same as a weakly ordered CPU
with wmb after every store).

In all cases, 5 may still be executed before 2 is visible to other CPUs!

The additional piece of the puzzle that mb() provides is the store/load
ordering, which fundamentally cannot be achieved with any combination of
rmb()s and wmb()s.

This can be an unexpected result if one expected any sort of global ordering
guarantee to barriers (eg. that the barriers themselves are sequentially
consistent with other types of barriers).  However sfence or lfence barriers
need only provide an ordering partial ordering of memory operations -- Consider
that wmb may be implemented as nothing more than inserting a special barrier
entry in the store queue, or, in the case of x86, it can be a noop as the store
queue is in order. And an rmb may be implemented as a directive to prevent
subsequent loads only so long as their are no previous outstanding loads (while
there could be stores still in store queues).

I can actually see the occasional load/store being reordered around lfence on
my core2. That doesn't prove my above assertions, but it does show the comment
is wrong (unless my program is -- can send it out by request).

So:
   mb() and smp_mb() always have and always will require a full mfence
   or lock prefixed instruction on x86.  And we should remove this comment.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Paul McKenney <paulmck@us.ibm.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-09-29 09:13:59 -07:00
..
acpi ACPI: CONFIG_ACPI_SLEEP=n power off regression in 2.6.23-rc8 (NOT in rc7) 2007-09-25 17:58:52 -04:00
asm-alpha
asm-arm
asm-avr32
asm-blackfin Blackfin arch: fix some bugs in lib/string.h functions found by our string testing modules 2007-09-12 16:30:15 +08:00
asm-cris
asm-frv
asm-generic Define termios_1 functions for powerpc, s390, avr32 and frv 2007-09-12 09:08:05 -07:00
asm-h8300
asm-i386 i386: remove bogus comment about memory barrier 2007-09-29 09:13:59 -07:00
asm-ia64 [IA64] Cleanup HPSIM code (was: Re: Enable early console for Ski simulator) 2007-09-01 02:52:25 -07:00
asm-m32r m32r: Rename STI/CLI macros 2007-09-06 11:10:56 +09:00
asm-m68k m68k(nommu): add missing syscalls 2007-09-11 17:21:20 -07:00
asm-m68knommu m68k(nommu): add missing syscalls 2007-09-11 17:21:20 -07:00
asm-mips [MIPS] Fix CONFIG_BUILD_ELF64 kernels with symbols in CKSEG0. 2007-09-27 23:19:16 +01:00
asm-parisc
asm-powerpc [POWERPC] Fix timekeeping on PowerPC 601 2007-09-19 15:26:34 +10:00
asm-ppc
asm-s390
asm-sh
asm-sh64
asm-sparc
asm-sparc64 [SPARC64]: Fix lockdep, particularly on SMP. 2007-09-16 11:51:15 -07:00
asm-um UML: Fix ELF_CORE_COPY_REGS build botch 2007-09-10 18:58:05 -07:00
asm-v850
asm-x86_64 Revert "x86-64: Disable local APIC timer use on AMD systems with C1E" 2007-09-26 15:43:41 -07:00
asm-xtensa
crypto
keys
linux Revert "[PATCH] x86-64: fix x86_64-mm-sched-clock-share" 2007-09-26 15:52:17 -07:00
math-emu
media V4L/DVB (6220a): fix build error for et61x251 driver 2007-09-14 13:39:12 -03:00
mtd
net [TCP]: Fix MD5 signature handling on big-endian. 2007-09-28 15:18:35 -07:00
pcmcia
rdma
rxrpc
scsi
sound
video
xen
Kbuild