openwrt/staging/blogic.git
16 years agox86: remove arch specific segment headers
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:09 +0000 (13:31 +0100)]
x86: remove arch specific segment headers

This file puts the remainder of the arch specificic segment
headers in segment.h.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify common parts of segment.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:09 +0000 (13:31 +0100)]
x86: unify common parts of segment.h

Although segment handling in i386 and x86_64 are very different,
there's a common part. Put them in segment.h instead of arch specific
headers

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: put get_kernel_rpl in a common location
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:09 +0000 (13:31 +0100)]
x86: put get_kernel_rpl in a common location

This macro is useful for both i386 and x86_64, so put it in a common
location, where both arches can grab it.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: support for branch trace store(BTS)
Markus Metzger [Wed, 30 Jan 2008 12:31:09 +0000 (13:31 +0100)]
x86, ptrace: support for branch trace store(BTS)

Resend using different mail client

Changes to the last version:
- split implementation into two layers: ds/bts and ptrace
- renamed TIF's
- save/restore ds save area msr in __switch_to_xtra()
- make block-stepping only look at BTF bit

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoUML: change sigcontext fields to match x86
Jeff Dike [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
UML: change sigcontext fields to match x86

git-x86, in commit 70aa1bd3839e3ec74ce65316528a82570e8de666, changed
a lot of the sigcontext field names.  This patch changes UML usage to
match.

I also changed includes of generic headers from "" to <>.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify system.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: unify system.h

This patch finishes the unification of system.h file.
i386 needs a constant to be defined, and it is defined inside an ifdef.

Other than that, pretty much nothing but includes are left in the arch
specific headers, and they are deleted.

[ mingo@elte.hu: 64-bit needs the cr8 access inlines. ]

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: move switch_to macro to system.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: move switch_to macro to system.h

This patch moves the switch_to() macro to system.h

As those macros are fundamentally different between i386 and x86_64,
they are enclosed around an ifdef.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify smp parts of system.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: unify smp parts of system.h

The memory barrier parts of system.h are not very different between
i386 and x86_64, the main difference being the availability of
instructions, which we handle with the use of ifdefs.

They are consolidated in system.h file, and then removed from
the arch-specific headers.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove unused macro
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: remove unused macro

Mr. Grep says warn_if_not_ulong() is not used anymore anywhere
in the code. So, we remove it.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify paravirt parts of system.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: unify paravirt parts of system.h

This patch moves the i386 control registers manipulation functions,
wbinvd, and clts functions to system.h. They are essentially the same
as in x86_64.

With this, system.h paravirt comes for free in x86_64.

[ mingo@elte.hu: reintroduced the cr8 bits - needed for resume images ]

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify load_segment macro
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: unify load_segment macro

This patch unifies the load_segment() macro, making them equal in both
x86_64 and i386 architectures. The common version goes to system.h,
and the old are deleted.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: put together equal pieces of system.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: put together equal pieces of system.h

This patch puts together pieces of system_{32,64}.h that
looks like the same. It's the first step towards integration
of this file.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove volatile keyword from clflush.
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: remove volatile keyword from clflush.

the p parameter is an explicit memory reference, and is
enough to prevent gcc to being nasty here. The volatile
seems completely not needed.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: some whitespace cleanups in paging code
Joerg Roedel [Wed, 30 Jan 2008 12:31:08 +0000 (13:31 +0100)]
x86: some whitespace cleanups in paging code

This patch does some whitespace cleanups in the paging code to fix some
checkpatch.pl warnings of my formerly merged cleanup patches.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: PIE executable randomization, uninlining
Andrew Morton [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: PIE executable randomization, uninlining

Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: PIE executable randomization, checkpatch fixes
Andrew Morton [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: PIE executable randomization, checkpatch fixes

#39: FILE: arch/ia64/ia32/binfmt_elf32.c:229:
+elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused)

WARNING: no space between function name and open parenthesis '('
#39: FILE: arch/ia64/ia32/binfmt_elf32.c:229:
+elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused)

WARNING: line over 80 characters
#67: FILE: arch/x86/kernel/sys_x86_64.c:80:
+ new_begin = randomize_range(*begin, *begin + 0x02000000, 0);

ERROR: use tabs not spaces
#110: FILE: arch/x86/kernel/sys_x86_64.c:185:
+ ^I        mm->cached_hole_size = 0;$

ERROR: use tabs not spaces
#111: FILE: arch/x86/kernel/sys_x86_64.c:186:
+ ^I^Imm->free_area_cache = mm->mmap_base;$

ERROR: use tabs not spaces
#112: FILE: arch/x86/kernel/sys_x86_64.c:187:
+ ^I}$

ERROR: use tabs not spaces
#141: FILE: arch/x86/kernel/sys_x86_64.c:216:
+ ^I^I/* remember the largest hole we saw so far */$

ERROR: use tabs not spaces
#142: FILE: arch/x86/kernel/sys_x86_64.c:217:
+ ^I^Iif (addr + mm->cached_hole_size < vma->vm_start)$

ERROR: use tabs not spaces
#143: FILE: arch/x86/kernel/sys_x86_64.c:218:
+ ^I^I        mm->cached_hole_size = vma->vm_start - addr;$

ERROR: use tabs not spaces
#157: FILE: arch/x86/kernel/sys_x86_64.c:232:
+  ^Imm->free_area_cache = TASK_UNMAPPED_BASE;$

ERROR: need a space before the open parenthesis '('
#291: FILE: arch/x86/mm/mmap_64.c:101:
+ } else if(mmap_is_legacy()) {

WARNING: braces {} are not necessary for single statement blocks
#302: FILE: arch/x86/mm/mmap_64.c:112:
+ if (current->flags & PF_RANDOMIZE) {
+ mm->mmap_base += ((long)rnd) << PAGE_SHIFT;
+ }

WARNING: line over 80 characters
#314: FILE: fs/binfmt_elf.c:48:
+static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long);

WARNING: no space between function name and open parenthesis '('
#314: FILE: fs/binfmt_elf.c:48:
+static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long);

WARNING: line over 80 characters
#429: FILE: fs/binfmt_elf.c:438:
+    eppnt, elf_prot, elf_type, total_size);

ERROR: need space after that ',' (ctx:VxV)
#480: FILE: fs/binfmt_elf.c:939:
+ elf_prot, elf_flags,0);
                     ^

total: 9 errors, 7 warnings, 461 lines checked
Your patch has style problems, please review.  If any of these errors
are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.

Please run checkpatch prior to sending patches

Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: PIE executable randomization
Jiri Kosina [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: PIE executable randomization

main executable of (specially compiled/linked -pie/-fpie) ET_DYN binaries
onto a random address (in cases in which mmap() is allowed to perform a
randomization).

The code has been extraced from Ingo's exec-shield patch
http://people.redhat.com/mingo/exec-shield/

[akpm@linux-foundation.org: fix used-uninitialsied warning]
[kamezawa.hiroyu@jp.fujitsu.com: fixed ia32 ELF on x86_64 handling]

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify include/asm-x86/linkage_[32|64].h
Harvey Harrison [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: unify include/asm-x86/linkage_[32|64].h

Remove definitions of FASTCALL/fastcall from linkage_32 as compiled with
-regparm=3 by default since 2.6.20 and should no longer be needed.

CONFIG X86_64 and CONFIG_X86_ALIGNMENT_16 are mutually exclusive as found
in Kconfig.cpu so it should be fine to test them separately.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: integrate 32-bit and 64-bit code in msr.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: integrate 32-bit and 64-bit code in msr.h

This patches proceeds with the integration of msr.h, making
the code unified, instead of having a version for each architecture.
We stick with the native_* functions, and then paravirt comes for free.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: make fixups wordsize agnostic
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: make fixups wordsize agnostic

This patch uses the _ASM_ALIGN and _ASM_PTR macros
to make the fixups in native_read/write_msr_safe look the same
for x86_64 and i386. Besides using this macros, we also have to
take the explicit instruction suffixes out. It's okay
because all this instructions uses registers, and can be sized by
them.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: change write msr functions interface
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: change write msr functions interface

This patche changes the native_write_msr() and friends interface
to explicitly take 2 32-bit registers instead of a 64-bit value.
The change will ease the merge with 64-bit code. As the 64-bit
value will be passed as two registers anyway in i386,
the PVOP_CALL interface has to account for that and use low/high parameters
It would force the x86_64 version to be different.

The change does not make i386 generated code less efficient. As said above,
it would get the values from two registers anyway.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: change rdpmc interface
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:07 +0000 (13:31 +0100)]
x86: change rdpmc interface

the rdpmc instruction gets a counter argument in rcx. However,
the i386 version was ignoring it. To make both x86_64 and i386 versions
the same, as well as to comply with the instruction semantics, this
parameter is added in the i386 version

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: introduce native_read_tscp
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:06 +0000 (13:31 +0100)]
x86: introduce native_read_tscp

Targetting paravirt, this patch introduces native_read_tscp, in
place of rdtscp() macro. When in a paravirt guest, this will
involve a function call, and thus, cannot be done in the vdso area.
These users then have to call the native version directly

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify cpuid functions
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: unify cpuid functions

cpuid is not very different between i386 and x86_64.
We move away the x86_64 version from msr.h, and
unify them at processor.h, where they belong.

cpuid() paravirt then comes for free.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: split get_cycles_sync
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: split get_cycles_sync

This patch splits get_cycles_sync() into  __get_cycles_sync(),
and the rdtscll part. Paravirt guests cannot issue rdtscl directly,
as it involves a function call in vdso area.

So, using the __get_cycles_sync() base, we introduce vget_cycles_sync,
which then calls the native version of rdtscll. Ideally, however, a guest
should define its own clocksource, together with a vread function

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: allow sched clock to be overridden by paravirt
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: allow sched clock to be overridden by paravirt

This patch turns the sched_clock into native_sched_clock.
sched clock becomes a weak symbol, which can then give its
place to a paravirt definition.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify msr smp funcs
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: unify msr smp funcs

The functions under #ifdef CONFIG_SMP in msr.h are the same
for both x86_64 and i386, and this patches removes one of them,
putting them in a single location

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: wipe out traditional opt from x86_64 Makefile
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: wipe out traditional opt from x86_64 Makefile

Among other things, using -traditional as a gcc option stops us from
using macro token pasting, which is a feature we heavily rely on.

There was still a use of -traditional in arch/x86/kernel/Makefile_64,
which this patch removes.

I don't see any problems building kernels in my x86_64 box without
-traditional.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use def_bool where possible in Kconfig.cpu
Harvey Harrison [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: use def_bool where possible in Kconfig.cpu

x86: Use def_bool where possible in Kconfig.cpu

Change occurances of:
bool
default X

to:
def_bool X

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: clean up process_32/64.c
Hiroshi Shimamoto [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: clean up process_32/64.c

White space and coding style clean up.
Make process_32/64.c similar.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use def_bool where possible
Harvey Harrison [Wed, 30 Jan 2008 12:31:03 +0000 (13:31 +0100)]
x86: use def_bool where possible

Change occurances of:
bool
default X

to:
def_bool X

Change ocurances of:
bool "Foo"
default X

to:
def_bool X
prompt "Foo"

Shows no difference in generated config for allmodconfig/allyesconfig.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use __PAGE_KERNEL* instead of _KERNPG_TABLE
Joerg Roedel [Wed, 30 Jan 2008 12:31:02 +0000 (13:31 +0100)]
x86: use __PAGE_KERNEL* instead of _KERNPG_TABLE

This minor cleanup replaces _KERNPG_TABLE with the __PAGE_KERNEL* for 2MB PTEs
in the 64-bit memory initialization code. The __PAGE_KERNEL* defines are more
appropriate for PTEs.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: define all _PAGE_* in terms of _PAGE_BIT_* on 64-bit
Joerg Roedel [Wed, 30 Jan 2008 12:31:02 +0000 (13:31 +0100)]
x86: define all _PAGE_* in terms of _PAGE_BIT_* on 64-bit

This patch defines the _PAGE_* paging attributes in pgtable_64.h in terms of
the former defined _PAGE_BIT_* values.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: 32-bit IOAPIC: de-fang IRQ compression
Len Brown [Wed, 30 Jan 2008 12:31:02 +0000 (13:31 +0100)]
x86: 32-bit IOAPIC: de-fang IRQ compression

commit c434b7a6aedfe428ad17cd61b21b125a7b7a29ce
(x86: avoid wasting IRQs for PCI devices)
created a concept of "IRQ compression" on i386
to conserve IRQ numbers on systems with many
sparsely populated IO APICs.

The same scheme was also added to x86_64,
but later removed when x86_64 recieved an IRQ over-haul
that made it unnecessary -- including per-CPU
IRQ vectors that greatly increased the IRQ capacity
on the machine.

i386 has not received the analogous over-haul,
and thus a previous attempt to delete IRQ compression
from i386 was rejected on the theory that there may
exist machines that actually need it.  The fact is
that the author of IRQ compression patch was unable
to confirm the actual existence of such a system.

As a result, all i386 kernels with IOAPIC support
pay the following:

1. confusion

IRQ compression re-names the traditional IOAPIC
pin numbers (aka ACPI GSI's) into sequential IRQ #s:

ACPI: PCI Interrupt 0000:00:1c.0[A] -> GSI 20 (level, low) -> IRQ 16
ACPI: PCI Interrupt 0000:00:1c.1[B] -> GSI 21 (level, low) -> IRQ 17
ACPI: PCI Interrupt 0000:00:1c.2[C] -> GSI 22 (level, low) -> IRQ 18
ACPI: PCI Interrupt 0000:00:1c.3[D] -> GSI 23 (level, low) -> IRQ 19
ACPI: PCI Interrupt 0000:00:1c.4[A] -> GSI 20 (level, low) -> IRQ 16

This makes /proc/interrupts look different
depending on system configuration and device probe order.
It is also different than the x86_64 kernel running
on the exact same system.  As a result, programmers
get confused when comparing systems.

2. complexity

The IRQ code in Linux is already overly complex,
and IRQ compression makes it worse.  There have
already been two bug workarounds related to IRQ
compression -- the IRQ0 timer workaround and
the VIA PCI IRQ workaround.

3. size

All i386 kernels with IOAPIC support contain an int[4096] --
a 4 page array to contain the renamed IRQs.

So while the irq compression code on i386 should really
be deleted -- even before merging the x86_64 irq-overhaul,
this patch simply disables it on all high volume systems
to avoid problems #1 and #2 on most all i386 systems.

A large system with pin numbers >=64 will still have compression
to conserve limited IRQ numbers for sparse IOAPICS.  However,
the vast majority of the planet, those with only pin numbers < 64
will use an identity GSI -> IRQ mapping.

Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
16 years agox86: use generic register name in the thread and tss structures
H. Peter Anvin [Wed, 30 Jan 2008 12:31:02 +0000 (13:31 +0100)]
x86: use generic register name in the thread and tss structures

This changes size-specific register names (eip/rip, esp/rsp, etc.) to
generic names in the thread and tss structures.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace merge removals
Roland McGrath [Wed, 30 Jan 2008 12:31:02 +0000 (13:31 +0100)]
x86: x86 ptrace merge removals

This removes the old separate 64-bit and ia32 ptrace source files.
They are no longer used.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace merge complete
Roland McGrath [Wed, 30 Jan 2008 12:31:02 +0000 (13:31 +0100)]
x86: x86 ptrace merge complete

This switches over the 64-bit build to use the shared ptrace code,
instead of the old ptrace_64.c and arch/x86/ia32/ptrace32.c code.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ia32 ptrace arch merge
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: x86 ia32 ptrace arch merge

This moves the sys32_ptrace code into arch/x86/kernel/ptrace.c,
verbatim except for a few hard-coded sizes replaced with sizeof.
Here this code can use the shared local functions in this file.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ia32 ptrace getreg/putreg merge
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: x86 ia32 ptrace getreg/putreg merge

This reimplements the 64-bit IA32-emulation register access
functions in arch/x86/kernel/ptrace.c, where they can share
some guts with the native access functions directly.

These functions are not used yet, but this paves the way to move
IA32 ptrace support into this file to share its local functions.

[akpm@linuxfoundation.org: Build fix]

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace merge syscall trace
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: x86 ptrace merge syscall trace

This moves the 64-bit syscall tracing functions into ptrace.c,
so that ptrace_64.c becomes entirely obsolete.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace arch merge
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: x86 ptrace arch merge

This adds 64-bit support to arch_ptrace in arch/x86/kernel/ptrace.c,
so this function can be used for native ptrace on both 32 and 64.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace getreg/putreg merge
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: x86 ptrace getreg/putreg merge

This merges 64-bit support into the low-level register access
functions in arch/x86/kernel/ptrace.c, paving the way to share
this file between 32-bit and 64-bit builds.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace getreg/putreg cleanup
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: x86 ptrace getreg/putreg cleanup

This cleans up the getreg/putreg functions to move the special cases
(segment registers and eflags) out into their own subroutines.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: ptrace FLAG_MASK cleanup
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: ptrace FLAG_MASK cleanup

This cleans up the FLAG_MASK macro to use symbolic constants instead of a
magic number.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: ptrace_32 renamed
Roland McGrath [Wed, 30 Jan 2008 12:31:01 +0000 (13:31 +0100)]
x86: ptrace_32 renamed

This renames ptrace_32.c back to ptrace.c, in preparation
for merging the 32/64 versions of these files.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-32 thread_struct.debugreg
Roland McGrath [Wed, 30 Jan 2008 12:30:59 +0000 (13:30 +0100)]
x86: x86-32 thread_struct.debugreg

This replaces the debugreg[7] member of thread_struct with individual
members debugreg0, etc.  This saves two words for the dummies 4 and 5,
and harmonizes the code between 32 and 64.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 ia32 ptrace get/putreg32 current task
Roland McGrath [Wed, 30 Jan 2008 12:30:58 +0000 (13:30 +0100)]
x86: x86-64 ia32 ptrace get/putreg32 current task

This generalizes the getreg32 and putreg32 functions so they can be used on
the current task, as well as on a task stopped in TASK_TRACED and switched
off.  This lays the groundwork to share this code for all kinds of
user-mode machine state access, not just ptrace.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-32 ptrace get/putreg current task
Roland McGrath [Wed, 30 Jan 2008 12:30:58 +0000 (13:30 +0100)]
x86: x86-32 ptrace get/putreg current task

This generalizes the getreg and putreg functions so they can be used on the
current task, as well as on a task stopped in TASK_TRACED and switched off.
This lays the groundwork to share this code for all kinds of user-mode
machine state access, not just ptrace.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 ptrace get/putreg current task
Roland McGrath [Wed, 30 Jan 2008 12:30:58 +0000 (13:30 +0100)]
x86: x86-64 ptrace get/putreg current task

This generalizes the getreg and putreg functions so they can be used on the
current task, as well as on a task stopped in TASK_TRACED and switched off.
This lays the groundwork to share this code for all kinds of user-mode
machine state access, not just ptrace.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-32 ptrace whitespace
Roland McGrath [Wed, 30 Jan 2008 12:30:58 +0000 (13:30 +0100)]
x86: x86-32 ptrace whitespace

This canonicalizes the indentation in the getreg and putreg functions.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 ptrace whitespace
Roland McGrath [Wed, 30 Jan 2008 12:30:58 +0000 (13:30 +0100)]
x86: x86-64 ptrace whitespace

This canonicalizes the indentation in the getreg and putreg functions.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 ia32 ptrace pt_regs cleanup
Roland McGrath [Wed, 30 Jan 2008 12:30:57 +0000 (13:30 +0100)]
x86: x86-64 ia32 ptrace pt_regs cleanup

This cleans up the getreg32/putreg32 functions to use struct pt_regs in a
straightforward fashion, instead of equivalent ugly pointer arithmetic.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: eflags enum
Roland McGrath [Wed, 30 Jan 2008 12:30:57 +0000 (13:30 +0100)]
x86: eflags enum

This removes the EF_* enum from <asm/ptrace.h>.  It is no longer used,
and duplicates the X86_EFLAGS_* constants from <asm/processor-flags.h>.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: setup64 eflags constants
Roland McGrath [Wed, 30 Jan 2008 12:30:57 +0000 (13:30 +0100)]
x86: setup64 eflags constants

This cleans up arch/x86/kernel/setup64.c to use the X86_EFLAGS_* constants
from <asm/processor-flags.h> instead of the EF_* enum in <asm/ptrace.h>.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use generic register names in struct sigcontext
H. Peter Anvin [Wed, 30 Jan 2008 12:30:56 +0000 (13:30 +0100)]
x86: use generic register names in struct sigcontext

Switch struct sigcontext (defined in <asm/sigcontext*.h>) to using
register names withut e- or r-prefixes for both 32- and 64-bit x86.
This is intended as a preliminary step in unifying this code between
architectures.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use generic register names in struct user_regs_struct
H. Peter Anvin [Wed, 30 Jan 2008 12:30:56 +0000 (13:30 +0100)]
x86: use generic register names in struct user_regs_struct

Switch struct user_regs_struct (defined in <asm/user.h>, which is no
longer exported to userspace) to using register names without e- or
r-prefixes for both 32 and 64 bit x86.  This is intended as a
preliminary step in unifying this code between architectures.

Also, be a bit more strict in truncating 32-bit "extended" segment
register values to 16 bits.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: rename the struct pt_regs members for 32/64-bit consistency
H. Peter Anvin [Wed, 30 Jan 2008 12:30:56 +0000 (13:30 +0100)]
x86: rename the struct pt_regs members for 32/64-bit consistency

We have a lot of code which differs only by the naming of specific
members of structures that contain registers.  In order to enable
additional unifications, this patch drops the e- or r- size prefix
from the register names in struct pt_regs, and drops the x- prefixes
for segment registers on the 32-bit side.

This patch also performs the equivalent renames in some additional
places that might be candidates for unification in the future.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add set/clear_cpu_cap operations
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:30:55 +0000 (13:30 +0100)]
x86: add set/clear_cpu_cap operations

The patch to suppress bitops-related warnings added a pile of ugly
casts.  Many of these were related to the management of x86 CPU
capabilities.  Clean these up by adding specific set/clear_cpu_cap
macros, and use them consistently.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: clean up bitops-related warnings
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:30:55 +0000 (13:30 +0100)]
x86: clean up bitops-related warnings

Add casts to appropriate places to silence spurious bitops warnings.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: partial unification of asm-x86/bitops.h
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:30:55 +0000 (13:30 +0100)]
x86: partial unification of asm-x86/bitops.h

This unifies the set/clear/test bit functions of asm/bitops.h.

I have not attempted to merge the bit-finding functions, since they
rely on the machine word size and can't be easily restructured to work
generically without a lot of #ifdefs.  In particular, the 64-bit code
can assume the presence of conditional move instructions, whereas
32-bit needs to be more careful.

The inline assembly for the bit operations has been changed to remove
explicit sizing hints on the instructions, so the assembler will pick
the appropriate instruction forms depending on the architecture and
the context.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: prevent dereferencing non-allocated per_cpu variables
Mike Travis [Wed, 30 Jan 2008 12:30:55 +0000 (13:30 +0100)]
x86: prevent dereferencing non-allocated per_cpu variables

'for_each_possible_cpu(i)' when there's a _remote possibility_ of
dereferencing a non-allocated per_cpu variable involved.

All files except mm/vmstat.c are x86 arch.

Thanks to pageexec@freemail.hu for pointing this out.

Signed-off-by: Mike Travis <travis@sgi.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: <pageexec@freemail.hu>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: PTRACE_SINGLEBLOCK
Roland McGrath [Wed, 30 Jan 2008 12:30:55 +0000 (13:30 +0100)]
x86: PTRACE_SINGLEBLOCK

This adds the PTRACE_SINGLEBLOCK request on x86, matching the ia64 feature.
The implementation comes from the generic ptrace code and relies on the
low-level machine support provided by arch_has_block_step() et al.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: debugctlmsr kprobes
Roland McGrath [Wed, 30 Jan 2008 12:30:54 +0000 (13:30 +0100)]
x86: debugctlmsr kprobes

This adjusts the x86 kprobes implementation to cope with per-thread
MSR_IA32_DEBUGCTLMSR being set for user mode.  I haven't delved deep
enough into the kprobes code to be really sure this covers all the
cases where the user-mode BTF setting needs to be cleared or restored.
It looks about right to me.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: debugctlmsr arch_has_block_step
Roland McGrath [Wed, 30 Jan 2008 12:30:54 +0000 (13:30 +0100)]
x86: debugctlmsr arch_has_block_step

This implements user-mode step-until-branch on x86 using the BTF bit
in MSR_IA32_DEBUGCTLMSR.  It's just like single-step, only less so.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: debugctlmsr context switch
Roland McGrath [Wed, 30 Jan 2008 12:30:54 +0000 (13:30 +0100)]
x86: debugctlmsr context switch

This adds low-level support for a per-thread value of MSR_IA32_DEBUGCTLMSR.
The per-thread value is switched in when TIF_DEBUGCTLMSR is set.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: debugctlmsr kconfig
Roland McGrath [Wed, 30 Jan 2008 12:30:54 +0000 (13:30 +0100)]
x86: debugctlmsr kconfig

This adds the (internal) Kconfig macro CONFIG_X86_DEBUGCTLMSR,
to be defined when configuring to support only hardware that
definitely supports MSR_IA32_DEBUGCTLMSR with the BTF flag.

The Intel documentation says "P6 family" and later processors all have it.
I think the Kconfig dependencies are right to have it set for those and
unset for others (i.e., when 586 and earlier are supported).

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: debugctlmsr constants
Roland McGrath [Wed, 30 Jan 2008 12:30:54 +0000 (13:30 +0100)]
x86: debugctlmsr constants

This adds constant macros for a few of the bits in MSR_IA32_DEBUGCTLMSR.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoptrace: generic PTRACE_SINGLEBLOCK
Roland McGrath [Wed, 30 Jan 2008 12:30:53 +0000 (13:30 +0100)]
ptrace: generic PTRACE_SINGLEBLOCK

This makes ptrace_request handle PTRACE_SINGLEBLOCK along with
PTRACE_CONT et al.  The new generic code makes use of the
arch_has_block_step macro and generic entry points on machines
that define them.

[ mingo@elte.hu: bugfix ]

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoptrace: arch_has_block_step
Roland McGrath [Wed, 30 Jan 2008 12:30:53 +0000 (13:30 +0100)]
ptrace: arch_has_block_step

This defines the new macro arch_has_block_step() in linux/ptrace.h, a
default for when asm/ptrace.h does not define it.  This is the analog
of arch_has_single_step() for step-until-branch on machines that have
it.  It declares the new user_enable_block_step function, which goes
with the existing user_enable_single_step and user_disable_single_step.
This is not used yet, but paves the way to harmonize on this interface
for the arch-specific calls on all machines.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-32 ptrace debugreg cleanup
Roland McGrath [Wed, 30 Jan 2008 12:30:52 +0000 (13:30 +0100)]
x86: x86-32 ptrace debugreg cleanup

This cleans up the 32-bit ptrace code to separate the guts of the
debug register access from the implementation of PTRACE_PEEKUSR and
PTRACE_POKEUSR.  The new functions ptrace_[gs]et_debugreg match the
new 64-bit entry points for parity, but they don't need to be global.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 ia32 ptrace debugreg cleanup
Roland McGrath [Wed, 30 Jan 2008 12:30:52 +0000 (13:30 +0100)]
x86: x86-64 ia32 ptrace debugreg cleanup

This cleans up the ia32 compat ptrace code to use shared code from
native ptrace for the implementation guts of debug register access.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 ptrace debugreg cleanup
Roland McGrath [Wed, 30 Jan 2008 12:30:52 +0000 (13:30 +0100)]
x86: x86-64 ptrace debugreg cleanup

This cleans up the 64-bit ptrace code to separate the guts of the
debug register access from the implementation of PTRACE_PEEKUSR and
PTRACE_POKEUSR.  The new functions ptrace_[gs]et_debugreg are made
global so that the ia32 code can later be changed to call them too.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86-64 ptrace: use task_pt_regs
Roland McGrath [Wed, 30 Jan 2008 12:30:52 +0000 (13:30 +0100)]
x86-64 ptrace: use task_pt_regs

This cleans up the 64-bit ptrace code to use task_pt_regs instead of its
own redundant code that does the same thing a different way.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86-32 ptrace: use task_pt_regs
Roland McGrath [Wed, 30 Jan 2008 12:30:52 +0000 (13:30 +0100)]
x86-32 ptrace: use task_pt_regs

This cleans up the 32-bit ptrace code to use task_pt_regs instead of its
own redundant code that does the same thing a different way.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agopowerpc: ptrace generic resume
Roland McGrath [Wed, 30 Jan 2008 12:30:51 +0000 (13:30 +0100)]
powerpc: ptrace generic resume

This removes the handling for PTRACE_CONT et al from the powerpc
ptrace code, so it uses the new generic code via ptrace_request.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agopowerpc: arch_has_single_step
Roland McGrath [Wed, 30 Jan 2008 12:30:51 +0000 (13:30 +0100)]
powerpc: arch_has_single_step

This defines the new standard arch_has_single_step macro.  It makes the
existing set_single_step and clear_single_step entry points global, and
renames them to the new standard names user_enable_single_step and
user_disable_single_step, respectively.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86-32: ptrace generic resume
Roland McGrath [Wed, 30 Jan 2008 12:30:51 +0000 (13:30 +0100)]
x86-32: ptrace generic resume

This removes the handling for PTRACE_CONT et al from the 32-bit
ptrace code, so it uses the new generic code via ptrace_request.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86-64: ptrace generic resume
Roland McGrath [Wed, 30 Jan 2008 12:30:51 +0000 (13:30 +0100)]
x86-64: ptrace generic resume

This removes the handling for PTRACE_CONT et al from the 64-bit
ptrace code, so it uses the new generic code via ptrace_request.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoptrace: generic resume
Roland McGrath [Wed, 30 Jan 2008 12:30:51 +0000 (13:30 +0100)]
ptrace: generic resume

This makes ptrace_request handle all the ptrace requests that wake
up the traced task.  These do low-level ptrace implementation magic
that is not arch-specific and should be kept out of arch code.  The
implementations on each arch usually do the same thing.  The new
generic code makes use of the arch_has_single_step macro and generic
entry points to handle PTRACE_SINGLESTEP.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86 single_step: TIF_FORCED_TF
Roland McGrath [Wed, 30 Jan 2008 12:30:50 +0000 (13:30 +0100)]
x86 single_step: TIF_FORCED_TF

This changes the single-step support to use a new thread_info flag
TIF_FORCED_TF instead of the PT_DTRACE flag in task_struct.ptrace.
This keeps arch implementation uses out of this non-arch field.

This changes the ptrace access to eflags to mask TF and maintain
the TIF_FORCED_TF flag directly if userland sets TF, instead of
relying on ptrace_signal_deliver.  The 64-bit and 32-bit kernels
are harmonized on this same behavior.  The ptrace_signal_deliver
approach works now, but this change makes the low-level register
access code reliable when called from different contexts than a
ptrace stop, which will be possible in the future.

The 64-bit do_debug exception handler is also changed not to clear TF
from user-mode registers.  This matches the 32-bit kernel's behavior.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: single_step: share code
Roland McGrath [Wed, 30 Jan 2008 12:30:50 +0000 (13:30 +0100)]
x86: single_step: share code

This removes the single-step code from ptrace_32.c and uses the step.c code
shared with the 64-bit kernel.  The two versions of the code were nearly
identical already, so the shared code has only a couple of simple #ifdef's.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: single_step 0xf0
Roland McGrath [Wed, 30 Jan 2008 12:30:50 +0000 (13:30 +0100)]
x86: single_step 0xf0

This fixes the 64-bit single-step handling code's instruction
decoder to grok the 0xf0 (lock) prefix, which the 32-bit code
already does correctly.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: single_step segment macros
Roland McGrath [Wed, 30 Jan 2008 12:30:50 +0000 (13:30 +0100)]
x86: single_step segment macros

This cleans up the single-step code to use the asm/segment.h macros
for segment selector magic bits, rather than its own constant.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: single_step moved
Roland McGrath [Wed, 30 Jan 2008 12:30:50 +0000 (13:30 +0100)]
x86: single_step moved

This moves the single-step support code from ptrace_64.c into a new file
step.c, verbatim.  This paves the way for consolidating this code between
64-bit and 32-bit versions.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: arch_has_single_step
Roland McGrath [Wed, 30 Jan 2008 12:30:48 +0000 (13:30 +0100)]
x86: arch_has_single_step

This defines the new standard arch_has_single_step macro.  It makes the
existing set_singlestep and clear_singlestep entry points global, and
renames them to the new standard names user_enable_single_step and
user_disable_single_step, respectively.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove TRAP_FLAG
Roland McGrath [Wed, 30 Jan 2008 12:30:48 +0000 (13:30 +0100)]
x86: remove TRAP_FLAG

This gets rid of the local constant macro TRAP_FLAG.
It's redundant with the public constant macro X86_EFLAGS_TF.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: segment selector macros
Roland McGrath [Wed, 30 Jan 2008 12:30:48 +0000 (13:30 +0100)]
x86: segment selector macros

This copies into asm-x86/segment_64.h some macros from asm-x86/segment_32.h
for dissecting segment selectors.  This lets other code use these macros
uniformly on 32/64-bit rather than duplicating the constants elsewhere.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoptrace: arch_has_single_step
Roland McGrath [Wed, 30 Jan 2008 12:30:47 +0000 (13:30 +0100)]
ptrace: arch_has_single_step

This defines the new macro arch_has_single_step() in linux/ptrace.h, a
default for when asm/ptrace.h does not define it.  It declares the new
user_enable_single_step and user_disable_single_step functions.
This is not used yet, but paves the way to harmonize on this interface
for the arch-specific calls on all machines.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: kmap_atomic() debugging
Andrew Morton [Wed, 30 Jan 2008 12:30:47 +0000 (13:30 +0100)]
x86: kmap_atomic() debugging

[ mingo@elte.hu: cleanups and made dependent on CONFIG_DEBUG_HIGHMEM.

  this caught a handful of bugs already, so lets apply it. If it gets
  things wrong we'll disable it. ]

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fall back on interrupt disable in cmpxchg8b on 80386 and 80486
Mathieu Desnoyers [Wed, 30 Jan 2008 12:30:47 +0000 (13:30 +0100)]
x86: fall back on interrupt disable in cmpxchg8b on 80386 and 80486

Actually, on 386, cmpxchg and cmpxchg_local fall back on
cmpxchg_386_u8/16/32: it disables interruptions around non atomic
updates to mimic the cmpxchg behavior.

The comment:
/* Poor man's cmpxchg for 386. Unsuitable for SMP */

already present in cmpxchg_386_u32 tells much about how this cmpxchg
implementation should not be used in a SMP context. However, the cmpxchg_local
can perfectly use this fallback, since it only needs to be atomic wrt the local
cpu.

This patch adds a cmpxchg_486_u64 and uses it as a fallback for cmpxchg64
and cmpxchg64_local on 80386 and 80486.

Q:
but why is it called cmpxchg_486 when the other functions are called

A:
Because the standard cmpxchg is missing only on 386, but cmpxchg8b is
missing both on 386 and 486.

Citing Intel's Instruction set reference:

cmpxchg:
This instruction is not supported on Intel processors earlier than the
Intel486 processors.

cmpxchg8b:
This instruction encoding is not supported on Intel processors earlier
than the Pentium processors.

Q:
What's the reason to have cmpxchg64_local on 32 bit architectures?
Without that need all this would just be a few simple defines.

A:
cmpxchg64_local on 32 bits architectures takes unsigned long long
parameters, but cmpxchg_local only takes longs. Since we have cmpxchg8b
to execute a 8 byte cmpxchg atomically on pentium and +, it makes sense
to provide a flavor of cmpxchg and cmpxchg_local using this instruction.

Also, for 32 bits architectures lacking the 64 bits atomic cmpxchg, it
makes sense _not_ to define cmpxchg64 while cmpxchg could still be
available.

Moreover, the fallback for cmpxchg8b on i386 for 386 and 486 is a

However, cmpxchg64_local will be emulated by disabling interrupts on all
architectures where it is not supported atomically.

Therefore, we *could* turn cmpxchg64_local into a cmpxchg_local, but it
would make the 386/486 fallbacks ugly, make its design different from
cmpxchg/cmpxchg64 (which really depends on atomic operations and cannot
be emulated) and require the __cmpxchg_local to be expressed as a macro
rather than an inline function so the parameters would not be fixed to
unsigned long long in every case.

So I think cmpxchg64_local makes sense there, but I am open to
suggestions.

Q:
Are there any callers?

A:
I am actually using it in LTTng in my timestamping code. I use it to
work around CPUs with asynchronous TSCs. I need to update 64 bits
values atomically on this 32 bits architecture.

Changelog:
- Ran though checkpatch.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agomips, x86: optimize the i8259 code a bit
Ralf Baechle [Wed, 30 Jan 2008 12:30:47 +0000 (13:30 +0100)]
mips, x86: optimize the i8259 code a bit

The timer code always calls the clock_event_device set_net_event and
set_mode methods with interrupts disabled, so no need to use
spin_lock_irqsave / spin_unlock_irqrestore for those.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Acked-by:Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: 64-bit, make sparsemem vmemmap the only memory model
Christoph Lameter [Wed, 30 Jan 2008 12:30:47 +0000 (13:30 +0100)]
x86: 64-bit, make sparsemem vmemmap the only memory model

Use sparsemem as the only memory model for UP, SMP and NUMA.  Measurements
indicate that DISCONTIGMEM has a higher overhead than sparsemem.  And
FLATMEMs benefits are minimal.  So I think its best to simply standardize
on sparsemem.

Results of page allocator tests (test can be had via git from slab git
tree branch tests)

Measurements in cycle counts. 1000 allocations were performed and then the
average cycle count was calculated.

Order FlatMem Discontig SparseMem
0   639   665   641
1   567   647   593
2   679   774   692
3   763   967   781
4   961  1501   962
5  1356  2344  1392
6  2224  3982  2336
7  4869  7225  5074
8 12500 14048 12732
9 27926 28223 28165
10 58578 58714 58682

(Note that FlatMem is an SMP config and the rest NUMA configurations)

Memory use:

SMP Sparsemem
-------------

Kernel size:

   text    data     bss     dec     hex filename
3849268  397739 1264856 5511863  541ab7 vmlinux

             total       used       free     shared    buffers     cached
Mem:       8242252      41164    8201088          0        352      11512
-/+ buffers/cache:      29300    8212952
Swap:      9775512          0    9775512

SMP Flatmem
-----------

Kernel size:

   text    data     bss     dec     hex filename
3844612  397739 1264536 5506887  540747 vmlinux

So 4.5k growth in text size vs. FLATMEM.

             total       used       free     shared    buffers     cached
Mem:       8244052      40544    8203508          0        352      11484
-/+ buffers/cache:      28708    8215344

2k growth in overall memory use after boot.

NUMA discontig:

   text    data     bss     dec     hex filename
3888124  470659 1276504 5635287  55fcd7 vmlinux

             total       used       free     shared    buffers     cached
Mem:       8256256      56908    8199348          0        352      11496
-/+ buffers/cache:      45060    8211196
Swap:      9775512          0    9775512

NUMA sparse:

   text    data     bss     dec     hex filename
3896428  470659 1276824 5643911  561e87 vmlinux

8k text growth. Given that we fully inline virt_to_page and friends now
that is rather good.

             total       used       free     shared    buffers     cached
Mem:       8264720      57240    8207480          0        352      11516
-/+ buffers/cache:      45372    8219348
Swap:      9775512          0    9775512

The total available memory is increased by 8k.

This patch makes sparsemem the default and removes discontig and
flatmem support from x86.

[ akpm@linux-foundation.org: allnoconfig build fix ]

Acked-by: Andi Kleen <ak@suse.de>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: vmlinux_32.lds.S: remove repeated comment from the x86-32 linker script
Borislav Petkov [Wed, 30 Jan 2008 12:30:46 +0000 (13:30 +0100)]
x86: vmlinux_32.lds.S: remove repeated comment from the x86-32 linker script

Remove repeated comment from the linker script for the x86-32 target.

Signed-off-by: Borislav Petkov <bbpetkov@yahoo.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: do not set boot cpu in cpu_online_map at x86_64_start_kernel()
Yinghai Lu [Wed, 30 Jan 2008 12:30:46 +0000 (13:30 +0100)]
x86: do not set boot cpu in cpu_online_map at x86_64_start_kernel()

In init/main.c boot_cpu_init() does that later.

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: set cpu_index to nr_cpus instead of 0
Yinghai Lu [Wed, 30 Jan 2008 12:30:46 +0000 (13:30 +0100)]
x86: set cpu_index to nr_cpus instead of 0

Some BIOSes that support two/four dualcore/quadcore systems, will get:

ACPI: LAPIC (acpi_id[0x01] lapic_id[0x00] enabled)
Processor #0 15:1 APIC version 16
ACPI: LAPIC (acpi_id[0x02] lapic_id[0x01] enabled)
Processor #1 15:1 APIC version 16
ACPI: LAPIC (acpi_id[0x03] lapic_id[0x02] enabled)
Processor #2 15:1 APIC version 16
ACPI: LAPIC (acpi_id[0x04] lapic_id[0x03] enabled)
Processor #3 15:1 APIC version 16
ACPI: LAPIC (acpi_id[0x05] lapic_id[0x84] disabled)
ACPI: LAPIC (acpi_id[0x06] lapic_id[0x85] disabled)
ACPI: LAPIC (acpi_id[0x07] lapic_id[0x86] disabled)
ACPI: LAPIC (acpi_id[0x08] lapic_id[0x87] disabled)
ACPI: LAPIC (acpi_id[0x09] lapic_id[0x88] disabled)
ACPI: LAPIC (acpi_id[0x0a] lapic_id[0x89] disabled)
ACPI: LAPIC (acpi_id[0x0b] lapic_id[0x8a] disabled)
ACPI: LAPIC (acpi_id[0x0c] lapic_id[0x8b] disabled)
ACPI: LAPIC (acpi_id[0x0d] lapic_id[0x8c] disabled)
ACPI: LAPIC (acpi_id[0x0e] lapic_id[0x8d] disabled)
ACPI: LAPIC (acpi_id[0x0f] lapic_id[0x8e] disabled)
ACPI: LAPIC (acpi_id[0x10] lapic_id[0x8f] disabled)

SMP: Allowing 16 CPUs, 12 hotplug CPUs

the /proc/cpuinfo will show a bunch of NULL cpus with cpu_index=0

so assign impossible cpu_index value at first instead of 0.

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: sanitize user specified e820 memmap values
Vladimir Berezniker [Wed, 30 Jan 2008 12:30:46 +0000 (13:30 +0100)]
x86: sanitize user specified e820 memmap values

Sanitize user specified e820 memory ranges, using the same logic that is
applied to the values returned by the BIOS.  This ensures consistent
handling regardless of the source of the memory mappings.

Allows overriding portions of the memory map without specifying one in
it's entirety (memmap=exactmap).

E.g. marking a range of bad RAM as reserved with memmap=48M$528M

BIOS supplied range

BIOS-e820: 0000000000100000 - 000000007fe80000 (usable)

becomes

user: 0000000000100000 - 0000000021000000 (usable)
user: 0000000021000000 - 0000000024000000 (reserved)
user: 0000000024000000 - 000000007fe80000 (usable)

Previously this did not work, as the original BIOS range was left
untouched while the user defined range was appended to the end of the
memory map.

[ tglx: arch/x86 adaptation ]

Signed-off-by: Vladimir Berezniker <vmpn@hitechman.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: TLS cleanup
Roland McGrath [Wed, 30 Jan 2008 12:30:46 +0000 (13:30 +0100)]
x86: TLS cleanup

This consolidates the four different places that implemented the same
encoding magic for the GDT-slot 32-bit TLS support.  The old tls32.c was
renamed and is now only slightly modified to be the shared implementation.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Zachary Amsden <zach@vmware.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: tls32 moved
Roland McGrath [Wed, 30 Jan 2008 12:30:45 +0000 (13:30 +0100)]
x86: tls32 moved

This renames arch/x86/ia32/tls32.c to arch/x86/kernel/tls.c, which does
nothing now but paves the way to consolidate this code for 32-bit too.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Zachary Amsden <zach@vmware.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: desc_empty
Roland McGrath [Wed, 30 Jan 2008 12:30:45 +0000 (13:30 +0100)]
x86: desc_empty

This replaces the desc_empty macro with an inline.  It now handles
easily any of the four different types used between 32/64 code to
refer to these 8 bytes.  It's identical in both asm-x86/processor_64.h
and asm-x86/processor_32.h, so if these files ever get merged this
function can be in the common code.

This also removes the desc_equal macro because nothing uses it.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: ptrace fs/gs_base
Roland McGrath [Wed, 30 Jan 2008 12:30:45 +0000 (13:30 +0100)]
x86: ptrace fs/gs_base

The fs_base and gs_base fields are available in user_regs_struct.
But reading these via ptrace (PTRACE_GETREGS or PTRACE_PEEKUSR) does
not give a reliably useful value.  The thread_struct fields are 0
when do_arch_prctl decided to use a GDT slot instead of MSR_FS_BASE,
which it does for a value under 1<<32.

This changes ptrace access to fs_base and gs_base to work like
PTRACE_ARCH_PRCTL does.  That is, it reads the base address that
user-mode memory access using the fs/gs instruction prefixes will
use, regardless of how it's being implemented in the kernel.  The
MSR vs GDT is an implementation detail that is pretty much hidden
from userland in the actual using, and there is no reason that
ptrace should give the internal implementation picture rather than
the user-mode semantic picture.  In the case of setting the value,
this can implicitly change the fsindex/gsindex value (also
separately in user_regs_struct), which is what happens when the
thread calls arch_prctl itself.  In a PTRACE_SETREGS, the fs_base
change will come after the fsindex change due to the order of the
struct, and so a change the debugger made to fs_base will have the
effect intended, another part of the user_regs_struct will now
differ when read back from what the debugger wrote.

This makes PTRACE_ARCH_PRCTL obsolete.  We could consider declaring
it deprecated and removing it one day, though there is no hurry.
For the foreseeable future, debuggers have to assume an old kernel
that does not report reliable fs_base/gs_base values in user_regs_struct
and stick to PTRACE_ARCH_PRCTL anyway.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>