safe/jmp/linux-2.6
17 years ago[PATCH] i386: PARAVIRT: Use enums for paravirt lazy flush modi
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:15 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Use enums for paravirt lazy flush modi

Remove #defines, add enum for PARAVIRT_LAZY_FLUSH.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: PARAVIRT: flush lazy mmu updates on kunmap_atomic
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:15 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: flush lazy mmu updates on kunmap_atomic

kunmap_atomic should flush any pending lazy mmu updates, mainly to be
consistent with kmap_atomic, and to preserve its normal behaviour.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: PARAVIRT: add kmap_atomic_pte for mapping highpte pages
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:15 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: add kmap_atomic_pte for mapping highpte pages

Xen and VMI both have special requirements when mapping a highmem pte
page into the kernel address space.  These can be dealt with by adding
a new kmap_atomic_pte() function for mapping highptes, and hooking it
into the paravirt_ops infrastructure.

Xen specifically wants to map the pte page RO, so this patch exposes a
helper function, kmap_atomic_prot, which maps the page with the
specified page protections.

This also adds a kmap_flush_unused() function to clear out the cached
kmap mappings.  Xen needs this to clear out any potential stray RW
mappings of pages which will become part of a pagetable.

[ Zach - vmi.c will need some attention after this patch.  It wasn't
  immediately obvious to me what needs to be done. ]

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
17 years ago[PATCH] i386: PARAVIRT: revert map_pt_hook.
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:15 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: revert map_pt_hook.

Back out the map_pt_hook to clear the way for kmap_atomic_pte.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
17 years ago[PATCH] i386: PARAVIRT: add flush_tlb_others paravirt_op
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:15 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: add flush_tlb_others paravirt_op

This patch adds a pv_op for flush_tlb_others.  Linux running on native
hardware uses cross-CPU IPIs to flush the TLB on any CPU which may
have a particular mm's pagetable entries cached in its TLB.  This is
inefficient in a paravirtualized environment, since the hypervisor
knows which real CPUs actually contain cached mappings, which may be a
small subset of a guest's VCPUs.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: PARAVIRT: add common patching machinery
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: add common patching machinery

Implement the actual patching machinery.  paravirt_patch_default()
contains the logic to automatically patch a callsite based on a few
simple rules:

 - if the paravirt_op function is paravirt_nop, then patch nops
 - if the paravirt_op function is a jmp target, then jmp to it
 - if the paravirt_op function is callable and doesn't clobber too much
    for the callsite, call it directly

paravirt_patch_default is suitable as a default implementation of
paravirt_ops.patch, will remove most of the expensive indirect calls
in favour of either a direct call or a pile of nops.

Backends may implement their own patcher, however.  There are several
helper functions to help with this:

paravirt_patch_nop nop out a callsite
paravirt_patch_ignore leave the callsite as-is
paravirt_patch_call patch a call if the caller and callee
have compatible clobbers
paravirt_patch_jmp patch in a jmp
paravirt_patch_insns patch some literal instructions over
the callsite, if they fit

This patch also implements more direct patches for the native case, so
that when running on native hardware many common operations are
implemented inline.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Anthony Liguori <anthony@codemonkey.ws>
Acked-by: Ingo Molnar <mingo@elte.hu>
17 years ago[PATCH] i386: PARAVIRT: Document asm-i386/paravirt.h
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Document asm-i386/paravirt.h

Clean things up, and broadly document:
 - the paravirt_ops functions themselves
 - the patching mechanism

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
17 years ago[PATCH] i386: PARAVIRT: Consistently wrap paravirt ops callsites to make them patchable
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Consistently wrap paravirt ops callsites to make them patchable

Wrap a set of interesting paravirt_ops calls in a wrapper which makes
the callsites available for patching.  Unfortunately this is pretty
ugly because there's no way to get gcc to generate a function call,
but also wrap just the callsite itself with the necessary labels.

This patch supports functions with 0-4 arguments, and either void or
returning a value.  64-bit arguments must be split into a pair of
32-bit arguments (lower word first).  Small structures are returned in
registers.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Anthony Liguori <anthony@codemonkey.ws>
17 years ago[PATCH] i386: PARAVIRT: Fix patch site clobbers to include return register
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Fix patch site clobbers to include return register

Fix a few clobbers to include the return register.  The clobbers set
is the set of all registers modified (or may be modified) by the code
snippet, regardless of whether it was deliberate or accidental.

Also, make sure that callsites which are used in contexts which don't
allow clobbers actually save and restore all clobberable registers.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Zachary Amsden <zach@vmware.com>
17 years ago[PATCH] i386: PARAVIRT: Use patch site IDs computed from offset in paravirt_ops structure
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Use patch site IDs computed from offset in paravirt_ops structure

Use patch type identifiers derived from the offset of the operation in
the paravirt_ops structure.  This avoids having to maintain a separate
enum for patch site types.

Also, since the identifier is derived from the offset into
paravirt_ops, the offset can be derived from the identifier.  This is
used to remove replicated information in the various callsite macros,
which has been a source of bugs in the past.

This patch also drops the fused save_fl+cli operation, which doesn't
really add much and makes things more complex - specifically because
it breaks the 1:1 relationship between identifiers and offsets.  If
this operation turns out to be particularly beneficial, then the right
answer is to define a new entrypoint for it.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Zachary Amsden <zach@vmware.com>
17 years ago[PATCH] i386: PARAVIRT: rename struct paravirt_patch to paravirt_patch_site for clarity
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: rename struct paravirt_patch to paravirt_patch_site for clarity

Rename struct paravirt_patch to paravirt_patch_site, so that it
clearly refers to a callsite, and not the patch which may be applied
to that callsite.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Zachary Amsden <zach@vmware.com>
17 years ago[PATCH] x86: PARAVIRT: add hooks to intercept mm creation and destruction
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:14 +0000 (19:27 +0200)]
[PATCH] x86: PARAVIRT: add hooks to intercept mm creation and destruction

Add hooks to allow a paravirt implementation to track the lifetime of
an mm.  Paravirtualization requires three hooks, but only two are
needed in common code.  They are:

arch_dup_mmap, which is called when a new mmap is created at fork

arch_exit_mmap, which is called when the last process reference to an
  mm is dropped, which typically happens on exit and exec.

The third hook is activate_mm, which is called from the arch-specific
activate_mm() macro/function, and so doesn't need stub versions for
other architectures.  It's called when an mm is first used.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: linux-arch@vger.kernel.org
Cc: James Bottomley <James.Bottomley@SteelEye.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
17 years ago[PATCH] i386: PARAVIRT: Allow paravirt backend to choose kernel PMD sharing
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Allow paravirt backend to choose kernel PMD sharing

Normally when running in PAE mode, the 4th PMD maps the kernel address space,
which can be shared among all processes (since they all need the same kernel
mappings).

Xen, however, does not allow guests to have the kernel pmd shared between page
tables, so parameterize pgtable.c to allow both modes of operation.

There are several side-effects of this.  One is that vmalloc will update the
kernel address space mappings, and those updates need to be propagated into
all processes if the kernel mappings are not intrinsically shared.  In the
non-PAE case, this is done by maintaining a pgd_list of all processes; this
list is used when all process pagetables must be updated.  pgd_list is
threaded via otherwise unused entries in the page structure for the pgd, which
means that the pgd must be page-sized for this to work.

Normally the PAE pgd is only 4x64 byte entries large, but Xen requires the PAE
pgd to page aligned anyway, so this patch forces the pgd to be page
aligned+sized when the kernel pmd is unshared, to accomodate both these
requirements.

Also, since there may be several distinct kernel pmds (if the user/kernel
split is below 3G), there's no point in allocating them from a slab cache;
they're just allocated with get_free_page and initialized appropriately.  (Of
course the could be cached if there is just a single kernel pmd - which is the
default with a 3G user/kernel split - but it doesn't seem worthwhile to add
yet another case into this code).

[ Many thanks to wli for review comments. ]

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Christoph Lameter <clameter@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: PARAVIRT: Allocate a fixmap slot
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Allocate a fixmap slot

Allocate a fixmap slot for use by a paravirt_ops implementation.  This
is intended for early-boot bootstrap mappings.  Once the zones and
allocator have been set up, it would be better to use get_vm_area() to
allocate some virtual space.

Xen uses this to map the hypervisor's shared info page, which doesn't
have a pseudo-physical page number, and therefore can't be mapped
ordinarily.  It is needed early because it contains the vcpu state,
including the interrupt mask.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
17 years ago[PATCH] i386: PARAVIRT: Hooks to set up initial pagetable
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Hooks to set up initial pagetable

This patch introduces paravirt_ops hooks to control how the kernel's
initial pagetable is set up.

In the case of a native boot, the very early bootstrap code creates a
simple non-PAE pagetable to map the kernel and physical memory.  When
the VM subsystem is initialized, it creates a proper pagetable which
respects the PAE mode, large pages, etc.

When booting under a hypervisor, there are many possibilities for what
paging environment the hypervisor establishes for the guest kernel, so
the constructon of the kernel's pagetable depends on the hypervisor.

In the case of Xen, the hypervisor boots the kernel with a fully
constructed pagetable, which is already using PAE if necessary.  Also,
Xen requires particular care when constructing pagetables to make sure
all pagetables are always mapped read-only.

In order to make this easier, kernel's initial pagetable construction
has been changed to only allocate and initialize a pagetable page if
there's no page already present in the pagetable.  This allows the Xen
paravirt backend to make a copy of the hypervisor-provided pagetable,
allowing the kernel to establish any more mappings it needs while
keeping the existing ones.

A slightly subtle point which is worth highlighting here is that Xen
requires all kernel mappings to share the same pte_t pages between all
pagetables, so that updating a kernel page's mapping in one pagetable
is reflected in all other pagetables.  This makes it possible to
allocate a page and attach it to a pagetable without having to
explicitly enumerate that page's mapping in all pagetables.

And:

+From: "Eric W. Biederman" <ebiederm@xmission.com>

If we don't set the leaf page table entries it is quite possible that
will inherit and incorrect page table entry from the initial boot
page table setup in head.S.  So we need to redo the effort here,
so we pick up PSE, PGE and the like.

Hypervisors like Xen require that their page tables be read-only,
which is slightly incompatible with our low identity mappings, however
I discussed this with Jeremy he has modified the Xen early set_pte
function to avoid problems in this area.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
17 years ago[PATCH] i386: PARAVIRT: Add pagetable accessors to pack and unpack pagetable entries
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Add pagetable accessors to pack and unpack pagetable entries

Add a set of accessors to pack, unpack and modify page table entries
(at all levels).  This allows a paravirt implementation to control the
contents of pgd/pmd/pte entries.  For example, Xen uses this to
convert the (pseudo-)physical address into a machine address when
populating a pagetable entry, and converting back to pphys address
when an entry is read.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
17 years ago[PATCH] i386: PARAVIRT: use paravirt_nop to consistently mark no-op operations
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: use paravirt_nop to consistently mark no-op operations

Add a _paravirt_nop function for use as a stub for no-op operations,
and paravirt_nop #defined void * version to make using it easier
(since all its uses are as a void *).

This is useful to allow the patcher to automatically identify noop
operations so it can simply nop out the callsite.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
[mingo] but only as a cleanup of the current open-coded (void *) casts.
My problem with this is that it loses the types. Not that there is much
to check for, but still, this adds some assumptions about how function
calls look like

17 years ago[PATCH] i386: PARAVIRT: Remove CONFIG_DEBUG_PARAVIRT
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: PARAVIRT: Remove CONFIG_DEBUG_PARAVIRT

Remove CONFIG_DEBUG_PARAVIRT.  When inlining code, this option
attempts to trash registers in the patch-site's "clobber" field, on
the grounds that this should find bugs with incorrect clobbers.
Unfortunately, the clobber field really means "registers modified by
this patch site", which includes return values.

Because of this, this option has outlived its usefulness, so remove
it.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
17 years ago[PATCH] x86-64: update MAINTAINERS
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] x86-64: update MAINTAINERS

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
17 years ago[PATCH] i386: i386 separate hardware-defined TSS from Linux additions
Rusty Russell [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: i386 separate hardware-defined TSS from Linux additions

On Thu, 2007-03-29 at 13:16 +0200, Andi Kleen wrote:
> Please clean it up properly with two structs.

Not sure about this, now I've done it.  Running it here.

If you like it, I can do x86-64 as well.

==
lguest defines its own TSS struct because the "struct tss_struct"
contains linux-specific additions.  Andi asked me to split the struct
in processor.h.

Unfortunately it makes usage a little awkward.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: x86-64 system crashes when no memory populating Node 0
James Puthukattukaran [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] x86-64: x86-64 system crashes when no memory populating Node 0

I have a 4 socket AMD Operton system. The 2.6.18 kernel I have crashes
when there is no memory in node0.

AK: changed call to _nopanic

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Fix x86_64 compilation with DEBUG_SIG on
Glauber de Oliveira Costa [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] x86-64: Fix x86_64 compilation with DEBUG_SIG on

Setting the DEBUG_SIG flag breaks compilation due to a wrong
struct access. Aditionally, it raises two warnings. This is one
patch to fix them all.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Allow boot-time disable of SMP altinstructions
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: Allow boot-time disable of SMP altinstructions

Add "noreplace-smp" to disable SMP instruction replacement.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Remove smp_alt_instructions
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:13 +0000 (19:27 +0200)]
[PATCH] i386: Remove smp_alt_instructions

The .smp_altinstructions section and its corresponding symbols are
completely unused, so remove them.

Also, remove stray #ifdef __KENREL__ in asm-i386/alternative.h

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: Clean up x86 control register and MSR macros (corrected)
H. Peter Anvin [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] x86: Clean up x86 control register and MSR macros (corrected)

This patch is based on Rusty's recent cleanup of the EFLAGS-related
macros; it extends the same kind of cleanup to control registers and
MSRs.

It also unifies these between i386 and x86-64; at least with regards
to MSRs, the two had definitely gotten out of sync.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: Allow percpu variables to be page-aligned
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] x86: Allow percpu variables to be page-aligned

Let's allow page-alignment in general for per-cpu data (wanted by Xen, and
Ingo suggested KVM as well).

Because larger alignments can use more room, we increase the max per-cpu
memory to 64k rather than 32k: it's getting a little tight.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: Enable bank 0 on non K7 Athlon
Andi Kleen [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: Enable bank 0 on non K7 Athlon

As a bug workaround bank 0 on K7s is normally disabled, but no need
to do that on other AMD CPUs.

Cc: davej@redhat.com
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Update smp_call_function* comments
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: Update smp_call_function* comments

Update documentation for i386 smp_call_function* functions.

As reported by Randy Dunlap <rdunlap@xenotime.net>

[ I've posted this before but it seems to have been lost along the way. ]

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Randy Dunlap <rdunlap@xenotime.net>
17 years ago[PATCH] i386: Use menuconfig objects - APM
Jan Engelhardt [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: Use menuconfig objects - APM

(I hope Andi is the right one to Cc, otherwise please add, thanks!)

Use menuconfigs instead of menus, so the whole menu can be disabled at
once instead of going through all options.

Signed-off-by: Jan Engelhardt <jengelh@gmx.de>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: Don't use MWAIT on AMD Family 10
Andi Kleen [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] x86: Don't use MWAIT on AMD Family 10

It doesn't put the CPU into deeper sleep states, so it's better to use the standard
idle loop to save power. But allow to reenable it anyways for benchmarking.

I also removed the obsolete idle=halt on i386

Cc: andreas.herrmann@amd.com
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Clean up asm-x86_64/bugs.h
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] x86-64: Clean up asm-x86_64/bugs.h

Most of asm-x86_64/bugs.h is code which should be in a C file, so put it there.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[PATCH] i386: Make COMPAT_VDSO runtime selectable.
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: Make COMPAT_VDSO runtime selectable.

Now that relocation of the VDSO for COMPAT_VDSO users is done at
runtime rather than compile time, it is possible to enable/disable
compat mode at runtime.

This patch allows you to enable COMPAT_VDSO mode with "vdso=2" on the
kernel command line, or via sysctl.  (Switching on a running system
shouldn't be done lightly; any process which was relying on the compat
VDSO will be upset if it goes away.)

The COMPAT_VDSO config option still exists, but if enabled it just
makes vdso_enabled default to VDSO_COMPAT.

+From: Hugh Dickins <hugh@veritas.com>

Fix oops from i386-make-compat_vdso-runtime-selectable.patch.

Even mingetty at system startup finds it easy to trigger an oops
while reading /proc/PID/maps: though it has a good hold on the mm
itself, that cannot stop exit_mm() from resetting tsk->mm to NULL.

(It is usually show_map()'s call to get_gate_vma() which oopses,
and I expect we could change that to check priv->tail_vma instead;
but no matter, even m_start()'s call just after get_task_mm() is racy.)

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
Cc: "Jan Beulich" <JBeulich@novell.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
17 years ago[PATCH] i386: Relocate VDSO ELF headers to match mapped location with COMPAT_VDSO
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: Relocate VDSO ELF headers to match mapped location with COMPAT_VDSO

Some versions of libc can't deal with a VDSO which doesn't have its
ELF headers matching its mapped address.  COMPAT_VDSO maps the VDSO at
a specific system-wide fixed address.  Previously this was all done at
build time, on the grounds that the fixed VDSO address is always at
the top of the address space.  However, a hypervisor may reserve some
of that address space, pushing the fixmap address down.

This patch does the adjustment dynamically at runtime, depending on
the runtime location of the VDSO fixmap.

[ Patch has been through several hands: Jan Beulich wrote the orignal
  version; Zach reworked it, and Jeremy converted it to relocate phdrs
  as well as sections. ]

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
Cc: "Jan Beulich" <JBeulich@novell.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
17 years ago[PATCH] i386: clean up identify_cpu
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: clean up identify_cpu

identify_cpu() is used to identify both the boot CPU and secondary
CPUs, but it performs some actions which only apply to the boot CPU.
Those functions are therefore really __init functions, but because
they're called by identify_cpu(), they must be marked __cpuinit.

This patch splits identify_cpu() into identify_boot_cpu() and
identify_secondary_cpu(), and calls the appropriate init functions
from each.  Also, identify_boot_cpu() and all the functions it
dominates are marked __init.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Clean up asm-i386/bugs.h
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] i386: Clean up asm-i386/bugs.h

Most of asm-i386/bugs.h is code which should be in a C file, so put it there.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[PATCH] x86-64: Fix vmalloc_32 to really allocate <4GB on 64bit platforms
Andi Kleen [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] x86-64: Fix vmalloc_32 to really allocate <4GB on 64bit platforms

Ugly ifdef, but should handle all 64bit platforms that have suitable
zones. On some like Altix it's probably impossible without IOMMU
use to get memory <4GB this way,  but they have to live with that.
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: fix arithmetic in comment
Avi Kivity [Wed, 2 May 2007 17:27:12 +0000 (19:27 +0200)]
[PATCH] x86-64: fix arithmetic in comment

The xmm space on x86_64 is 256 bytes.

Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Use X86_EFLAGS_IF in x86-64/irqflags.h.
Andi Kleen [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: Use X86_EFLAGS_IF in x86-64/irqflags.h.

As per i386 patch: move X86_EFLAGS_IF et al out to a new header:
processor-flags.h, so we can include it from irqflags.h and use it in
raw_irqs_disabled_flags().

As a side-effect, we could now use these flags in .S files.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: fix amd64-agp aperture validation
Jan Beulich [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86: fix amd64-agp aperture validation

Under CONFIG_DISCONTIGMEM, assuming that a !pfn_valid() implies all
subsequent pfn-s are also invalid is wrong. Thus replace this by
explicitly checking against the E820 map.

AK: make e820 on x86-64 not initdata

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Mark Langsdorf <mark.langsdorf@amd.com>
17 years ago[PATCH] x86-64: Account for module percpu space separately from kernel percpu
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: Account for module percpu space separately from kernel percpu

Rather than using a single constant PERCPU_ENOUGH_ROOM, compute it as
the sum of kernel_percpu + PERCPU_MODULE_RESERVE.  This is now common
to all architectures; if an architecture wants to set
PERCPU_ENOUGH_ROOM to something special, then it may do so (ia64 is
the only one which does).

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Remove unneeded externs in nmi.c
Andi Kleen [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] i386: Remove unneeded externs in nmi.c

All were already in some header
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Change my email address
Andi Kleen [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: Change my email address

Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Add machine_ops interface to abstract halting and rebooting
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] i386: Add machine_ops interface to abstract halting and rebooting

machine_ops is an interface for the machine_* functions defined in
<linux/reboot.h>.  This is intended to allow hypervisors to intercept
the reboot process, but it could be used to implement other x86
subarchtecture reboots.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Add smp_ops interface
Jeremy Fitzhardinge [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] i386: Add smp_ops interface

Add a smp_ops interface.  This abstracts the API defined by
<linux/smp.h> for use within arch/i386.  The primary intent is that it
be used by a paravirtualizing hypervisor to implement SMP, but it
could also be used by non-APIC-using sub-architectures.

This is related to CONFIG_PARAVIRT, but is implemented unconditionally
since it is simpler that way and not a highly performance-sensitive
interface.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
17 years ago[PATCH] i386: cleanup GDT Access
Rusty Russell [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] i386: cleanup GDT Access

Now we have an explicit per-cpu GDT variable, we don't need to keep the
descriptors around to use them to find the GDT: expose cpu_gdt directly.

We could go further and make load_gdt() pack the descriptor for us, or even
assume it means "load the current cpu's GDT" which is what it always does.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: Fix "Section mismatch" compile warning
Bernhard Walle [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: Fix "Section mismatch" compile warning

Fix "Section mismatch" warnings in arch/x86_64/kernel/time.c

Signed-off-by: Bernhard Walle <bwalle@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: adjust EDID retrieval
Jan Beulich [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: adjust EDID retrieval

commit 5e518d7672dea4cd7c60871e40d0490c52f01d13 did the same change to
i386's variant.

With this change, i386's and x86-64's versions are identical, raising
the question whether the x86-64 one should go (just like there's only
one instance of edd.S).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Inhibit machine from asserting an NMI when doing Alt-SysRq-M operation.
Konrad Rzeszutek [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: Inhibit machine from asserting an NMI when doing Alt-SysRq-M operation.

This patch touches the NMI watchdog every MAX_ORDER_NR_PAGES
to inhibit the machine from triggering an NMI while the CPUs
are locked. This situation is happening on boxes with more
than 64CPUs and 128GB of RAM when Alt-SysRq-m is performed.

It has been succesfully tested for regression on uni, 2, 4, 8
32, and 64 CPU boxes with various memory configuration.

Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: vsyscall_gtod_data diet and vgettimeofday() fix
Eric Dumazet [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: vsyscall_gtod_data diet and vgettimeofday() fix

Current vsyscall_gtod_data is large (3 or 4 cache lines dirtied at timer
interrupt). We can shrink it to exactly 64 bytes (1 cache line on AMD64)

Instead of copying a whole struct clocksource, we copy only needed fields.

I deleted an unused field : offset_base

This patch fixes one oddity in vgettimeofday(): It can returns a timeval with
tv_usec = 1000000. Maybe not a bug, but why not doing the right thing ?

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: fix vtime() vsyscall
Eric Dumazet [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86-64: fix vtime() vsyscall

There is a tiny probability that the return value from vtime(time_t *t) is
Signed-off-by: Andi Kleen <ak@suse.de>
different than the value stored in *t

Using a temporary variable solves the problem and gives a faster code.

   17:   48 85 ff                test   %rdi,%rdi
   1a:   48 8b 05 00 00 00 00    mov    0(%rip),%rax        #
__vsyscall_gtod_data.wall_time_tv.tv_sec
   21:   74 03                   je     26
   23:   48 89 07                mov    %rax,(%rdi)
   26:   c9                      leaveq
   27:   c3                      retq

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
17 years ago[PATCH] x86: remove UNEXPECTED_IO_APIC()
Adrian Bunk [Wed, 2 May 2007 17:27:11 +0000 (19:27 +0200)]
[PATCH] x86: remove UNEXPECTED_IO_APIC()

Many years ago, UNEXPECTED_IO_APIC() contained printk()'s (but nothing more).

Now that it's completely empty for years, we can as well remove it.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: sys_ioperm() prototype cleanup
Adrian Bunk [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] x86: sys_ioperm() prototype cleanup

- there's no reason for duplicating the prototype from
  include/linux/syscalls.h in include/asm-x86_64/unistd.h
- every file should #include the headers containing the prototypes for
  it's global functions

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: use lru instead of page->index and page->private for pgd lists manage...
Christoph Lameter [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] x86-64: use lru instead of page->index and page->private for pgd lists management.

x86_64 currently simulates a list using the index and private fields of the
page struct.  Seems that the code was inherited from i386.  But x86_64 does
not use the slab to allocate pgds and pmds etc.  So the lru field is not
used by the slab and therefore available.

This patch uses standard list operations on page->lru to realize pgd
tracking.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: remove the APM_RTC_IS_GMT config option.
Parag Warudkar [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: remove the APM_RTC_IS_GMT config option.

Signed-off-by: Parag Warudkar <parag.warudkar@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Remove unused stext symbol
Andi Kleen [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] x86-64: Remove unused stext symbol

suggested by Jan Beulich

Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: Use X86_EFLAGS_IF in irqflags.h.
Andi Kleen [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: Use X86_EFLAGS_IF in irqflags.h.

Move X86_EFLAGS_IF et al out to a new header: processor-flags.h, so we
can include it from irqflags.h and use it in raw_irqs_disabled_flags().

As a side-effect, we could now use these flags in .S files.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: get rid of unused variables
Parag Warudkar [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: get rid of unused variables

Signed-off-by: Parag Warudkar <parag.warudkar@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: tighten kernel image page access rights
Jan Beulich [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] x86: tighten kernel image page access rights

On x86-64, kernel memory freed after init can be entirely unmapped instead
of just getting 'poisoned' by overwriting with a debug pattern.

On i386 and x86-64 (under CONFIG_DEBUG_RODATA), kernel text and bug table
can also be write-protected.

Compared to the first version, this one prevents re-creating deleted
mappings in the kernel image range on x86-64, if those got removed
previously. This, together with the original changes, prevents temporarily
having inconsistent mappings when cacheability attributes are being
changed on such pages (e.g. from AGP code). While on i386 such duplicate
mappings don't exist, the same change is done there, too, both for
consistency and because checking pte_present() before using various other
pte_XXX functions is a requirement anyway. At once, i386 code gets
adjusted to use pte_huge() instead of open coding this.

AK: split out cpa() changes

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: Improve handling of kernel mappings in change_page_attr
Jan Beulich [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] x86: Improve handling of kernel mappings in change_page_attr

Fix various broken corner cases in i386 and x86-64 change_page_attr.

AK: split off from tighten kernel image access rights

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: rationalize paravirt wrappers
Rusty Russell [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: rationalize paravirt wrappers

paravirt.c used to implement native versions of all low-level
functions.  Far cleaner is to have the native versions exposed in the
headers and as inline native_XXX, and if !CONFIG_PARAVIRT, then simply
#define XXX native_XXX.

There are several nice side effects:

1) write_dt_entry() now takes the correct "struct Xgt_desc_struct *"
   not "void *".

2) load_TLS is reintroduced to the for loop, not manually unrolled
   with a #error in case the bounds ever change.

3) Macros become inlines, with type checking.

4) Access to the native versions is trivial for KVM, lguest, Xen and
   others who might want it.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@muc.de>
Cc: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: Rename boot_gdt_table to boot_gdt
Sebastien Dugue [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: Rename boot_gdt_table to boot_gdt

Rename boot_gdt_table to boot_gdt to avoid the duplicate T(able).

Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: clean up cpu_init()
Rusty Russell [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: clean up cpu_init()

We now have cpu_init() and secondary_cpu_init() doing nothing but calling
_cpu_init() with the same arguments.  Rename _cpu_init() to cpu_init() and use
it as a replcement for secondary_cpu_init().

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: Use per-cpu GDT immediately upon boot
Rusty Russell [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: Use per-cpu GDT immediately upon boot

Now we are no longer dynamically allocating the GDT, we don't need the
"cpu_gdt_table" at all: we can switch straight from "boot_gdt_table" to the
per-cpu GDT.  This means initializing the cpu_gdt array in C.

The boot CPU uses the per-cpu var directly, then in smp_prepare_cpus() it
switches to the per-cpu copy just allocated.  For secondary CPUs, the
early_gdt_descr is set to point directly to their per-cpu copy.

For UP the code is very simple: it keeps using the "per-cpu" GDT as per SMP,
but we never have to move.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: Use per-cpu variables for GDT, PDA
Rusty Russell [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] i386: Use per-cpu variables for GDT, PDA

Allocating PDA and GDT at boot is a pain.  Using simple per-cpu variables adds
happiness (although we need the GDT page-aligned for Xen, which we do in a
followup patch).

[akpm@linux-foundation.org: build fix]
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86: add command line length to boot protocol
Bernhard Walle [Wed, 2 May 2007 17:27:10 +0000 (19:27 +0200)]
[PATCH] x86: add command line length to boot protocol

Because the command line is increased to 2048 characters after 2.6.21, it's
not possible for boot loaders and userspace tools to determine the length
of the command line the kernel can understand.  The benefit of knowing the
length is that users can be warned if the command line size is too long
which prevents surprise if things don't work after bootup.

This patch updates the boot protocol to contain a field called
"cmdline_size" that contain the length of the command line (excluding the
terminating zero).

The patch also adds missing fields (of protocol version 2.05) to the x86_64
setup code.

Signed-off-by: Bernhard Walle <bwalle@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Alon Bar-Lev <alon.barlev@gmail.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: Allow i386 crash kernels to handle x86_64 dumps
Ian Campbell [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] i386: Allow i386 crash kernels to handle x86_64 dumps

The specific case I am encountering is kdump under Xen with a 64 bit
hypervisor and 32 bit kernel/userspace.  The dump created is 64 bit due to
the hypervisor but the dump kernel is 32 bit for maximum compatibility.

It's possibly less likely to be useful in a purely native scenario but I
see no reason to disallow it.

[akpm@linux-foundation.org: build fix]
Signed-off-by: Ian Campbell <ian.campbell@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Vivek Goyal <vgoyal@in.ibm.com>
Cc: Horms <horms@verge.net.au>
Cc: Magnus Damm <magnus.damm@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: Introduce load_TLS to the "for" loop.
Rusty Russell [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86-64: Introduce load_TLS to the "for" loop.

GCC (4.1 at least) unrolls it anyway, but I can't believe this code
was ever justifiable.  (I've also submitted a patch which cleans up
i386, which is even uglier).

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: Initialize esp0 properly all the time
Rusty Russell [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] i386: Initialize esp0 properly all the time

Whenever we schedule, __switch_to calls load_esp0 which does:

tss->esp0 = thread->esp0;

This is never initialized for the initial thread (ie "swapper"), so when we're
scheduling that, we end up setting esp0 to 0.  This is fine: the swapper never
leaves ring 0, so this field is never used.

lguest, however, gets upset that we're trying to used an unmapped page as our
kernel stack.  Rather than work around it there, let's initialize it.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: VDSO_PRELINK warning fix
Andrew Morton [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] i386: VDSO_PRELINK warning fix

The lguest patches somehow managed to trigger this:

In file included from arch/i386/lguest/lguest.c:38:
include/asm/asm-offsets.h:67:1: warning: "VDSO_PRELINK" redefined
In file included from include/linux/elf.h:7,
                 from include/linux/module.h:15,
                 from include/linux/device.h:21,
                 from include/linux/interrupt.h:15,
                 from arch/i386/lguest/lguest.c:27:
include/asm/elf.h:140:1: warning: this is the location of the previous definition

I assume that using the same identifier twice was a bad idea..

Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: fake numa for cpusets document
David Rientjes [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86-64: fake numa for cpusets document

Create a document to explain how to use numa=fake in conjunction with cpusets
for coarse memory resource management.

An attempt to get more awareness and testing for this feature.

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86: remove constant_tsc reporting from /proc/cpuinfo' power flags
Joerg Roedel [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86: remove constant_tsc reporting from /proc/cpuinfo' power flags

remove the reporting of the constant_tsc flag from the "power management"
field in /proc/cpuinfo.  The NULL value there was replaced by "" because
the former would result in a printout of [8] if the flag is set.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: fixed size remaining fake nodes
David Rientjes [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86-64: fixed size remaining fake nodes

Extends the numa=fake x86_64 command-line option to split the remaining system
memory into nodes of fixed size.  Any leftover memory is allocated to a final
node unless the command-line ends with a comma.

For example:
  numa=fake=2*512,*128 gives two 512M nodes and the remaining system
memory is split into nodes of 128M each.

This is beneficial for systems where the exact size of RAM is unknown or not
necessarily relevant, but the size of the remaining nodes to be allocated is
known based on their capacity for resource management.

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: split remaining fake nodes equally
David Rientjes [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86-64: split remaining fake nodes equally

Extends the numa=fake x86_64 command-line option to split the remaining
system memory into equal-sized nodes.

For example:
numa=fake=2*512,4* gives two 512M nodes and the remaining system
memory is split into four approximately equal
chunks.

This is beneficial for systems where the exact size of RAM is unknown or not
necessarily relevant, but the granularity with which nodes shall be allocated
is known.

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: configurable fake numa node sizes
David Rientjes [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86-64: configurable fake numa node sizes

Extends the numa=fake x86_64 command-line option to allow for configurable
node sizes.  These nodes can be used in conjunction with cpusets for coarse
memory resource management.

The old command-line option is still supported:
  numa=fake=32 gives 32 fake NUMA nodes, ignoring the NUMA setup of the
actual machine.

But now you may configure your system for the node sizes of your choice:
  numa=fake=2*512,1024,2*256
gives two 512M nodes, one 1024M node, two 256M nodes, and
the rest of system memory to a sixth node.

The existing hash function is maintained to support the various node sizes
that are possible with this implementation.

Each node of the same size receives roughly the same amount of available
pages, regardless of any reserved memory with its address range.  The total
available pages on the system is calculated and divided by the number of equal
nodes to allocate.  These nodes are then dynamically allocated and their
borders extended until such time as their number of available pages reaches
the required size.

Configurable node sizes are recommended when used in conjunction with cpusets
for memory control because it eliminates the overhead associated with scanning
the zonelists of many smaller full nodes on page_alloc().

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: fix GDT's number of quadwords in comment
Ahmed S. Darwish [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] i386: fix GDT's number of quadwords in comment

Fix comments to represent the true number of quadwords in GDT.

Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: vmi_pmd_clear() static
Adrian Bunk [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] i386: vmi_pmd_clear() static

This patch makes the needlessly global vmi_pmd_clear() static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Zachary Amsden <zach@vmware.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: make simnow_init() static
Adrian Bunk [Wed, 2 May 2007 17:27:09 +0000 (19:27 +0200)]
[PATCH] x86-64: make simnow_init() static

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: remove extra smp_processor_id calling
Yinghai Lu [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: remove extra smp_processor_id calling

Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: fix ia32_binfmt.c build error
Ralf Baechle [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: fix ia32_binfmt.c build error

Reorder code to avoid multiple inclusion of elf.h.

#undef several symbols to avoid build errors over redefinitions.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86: Log reason why TSC was marked unstable
john stultz [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86: Log reason why TSC was marked unstable

Change mark_tsc_unstable() so it takes a string argument, which holds the
reason the TSC was marked unstable.

This is then displayed the first time mark_tsc_unstable is called.

This should help us better debug why the TSC was marked unstable on certain
systems and allow us to make sure we're not being overly paranoid when
throwing out this troublesome clocksource.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: workaround for a -Wmissing-prototypes warning
Adrian Bunk [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] i386: workaround for a -Wmissing-prototypes warning

Work around a warning with -Wmissing-prototypes in
arch/i386/kernel/asm-offsets.c

The warning isn't gcc's fault - asm-offsets.c is simply a special file.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: type cast clean up for find_next_zero_bit
Ken Chen [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] i386: type cast clean up for find_next_zero_bit

clean up unneeded type cast by properly declare data type.

Signed-off-by: Ken Chen <kenchen@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] i386: make struct vmi_ops static
Adrian Bunk [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] i386: make struct vmi_ops static

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] i386: modpost apic related warning fixes
Vivek Goyal [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] i386: modpost apic related warning fixes

o Modpost generates warnings for i386 if compiled with CONFIG_RELOCATABLE=y

WARNING: vmlinux - Section mismatch: reference to .init.text:find_unisys_acpi_oem_table from .text between 'acpi_madt_oem_check' (at offset 0xc0101eda) and 'enable_apic_mode'
WARNING: vmlinux - Section mismatch: reference to .init.text:acpi_get_table_header_early from .text between 'acpi_madt_oem_check' (at offset 0xc0101ef0) and 'enable_apic_mode'
WARNING: vmlinux - Section mismatch: reference to .init.text:parse_unisys_oem from .text between 'acpi_madt_oem_check' (at offset 0xc0101f2e) and 'enable_apic_mode'
WARNING: vmlinux - Section mismatch: reference to .init.text:setup_unisys from .text between 'acpi_madt_oem_check' (at offset 0xc0101f37) and 'enable_apic_mode'WARNING: vmlinux - Section mismatch: reference to .init.text:parse_unisys_oem from .text between 'mps_oem_check' (at offset 0xc0101ec7) and 'acpi_madt_oem_check'
WARNING: vmlinux - Section mismatch: reference to .init.text:es7000_sw_apic from .text between 'enable_apic_mode' (at offset 0xc0101f48) and 'check_apicid_present'

o Some functions which are inline (acpi_madt_oem_check) are not inlined by
  compiler as these functions are accessed using function pointer. These
  functions are put in .text section and they in-turn access __init type
  functions hence modpost generates warnings.

o Do not iniline acpi_madt_oem_check, instead make it __init.

Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Len Brown <lenb@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: Set HASHDIST_DEFAULT to 1 for x86_64 NUMA
Ravikiran G Thirumalai [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: Set HASHDIST_DEFAULT to 1 for x86_64 NUMA

Enable system hashtable memory to be distributed among nodes on x86_64 NUMA

Forcing the kernel to use node interleaved vmalloc instead of bootmem for
the system hashtable memory (alloc_large_system_hash) reduces the memory
imbalance on node 0 by around 40MB on a 8 node x86_64 NUMA box:

Before the following patch, on bootup of a 8 node box:

Node 0 MemTotal:      3407488 kB
Node 0 MemFree:       3206296 kB
Node 0 MemUsed:        201192 kB
Node 0 Active:           7012 kB
Node 0 Inactive:          512 kB
Node 0 Dirty:               0 kB
Node 0 Writeback:           0 kB
Node 0 FilePages:        1912 kB
Node 0 Mapped:            420 kB
Node 0 AnonPages:        5612 kB
Node 0 PageTables:        468 kB
Node 0 NFS_Unstable:        0 kB
Node 0 Bounce:              0 kB
Node 0 Slab:             5408 kB
Node 0 SReclaimable:      644 kB
Node 0 SUnreclaim:       4764 kB

After the patch (or using hashdist=1 on the kernel command line):

Node 0 MemTotal:      3407488 kB
Node 0 MemFree:       3247608 kB
Node 0 MemUsed:        159880 kB
Node 0 Active:           3012 kB
Node 0 Inactive:          616 kB
Node 0 Dirty:               0 kB
Node 0 Writeback:           0 kB
Node 0 FilePages:        2424 kB
Node 0 Mapped:            380 kB
Node 0 AnonPages:        1200 kB
Node 0 PageTables:        396 kB
Node 0 NFS_Unstable:        0 kB
Node 0 Bounce:              0 kB
Node 0 Slab:             6304 kB
Node 0 SReclaimable:     1596 kB
Node 0 SUnreclaim:       4708 kB

I guess it is a good idea to keep HASHDIST_DEFAULT "on" for x86_64 NUMA
since x86_64 has no dearth of vmalloc space?  Or maybe enable hash
distribution for all 64bit NUMA arches?  The following patch does it only
for x86_64.

I ran a HPC MPI benchmark -- 'Ansys wingsolid', which takes up quite a bit of
memory and uses up tlb entries.  This was on a 4 way, 2 socket
Tyan AMD box (non vsmp), with 8G total memory (4G pernode).

The results with and without hash distribution are:

1. Vanilla - runtime of 1188.000s
2. With hashdist=1 runtime of 1154.000s

Oprofile output for the duration of run is:

1. Vanilla:
PU: AMD64 processors, speed 2411.16 MHz (estimated)
Counted L1_AND_L2_DTLB_MISSES events (L1 and L2 DTLB misses) with a unit
mask of 0x00 (No unit mask) count 500
samples  %        app name                 symbol name
163054    6.5513  libansys1.so             MultiFront::decompose(int, int,
Elemset *, int *, int, int, int)
162061    6.5114  libansys3.so             blockSaxpy6L_fd
162042    6.5107  libansys3.so             blockInnerProduct6L_fd
156286    6.2794  libansys3.so             maxb33_
87879     3.5309  libansys1.so             elmatrixmultpcg_
84857     3.4095  libansys4.so             saxpy_pcg
58637     2.3560  libansys4.so             .st4560
46612     1.8728  libansys4.so             .st4282
43043     1.7294  vmlinux-t                copy_user_generic_string
41326     1.6604  libansys3.so             blockSaxpyBackSolve6L_fd
41288     1.6589  libansys3.so             blockInnerProductBackSolve6L_fd

2. With hashdist=1
CPU: AMD64 processors, speed 2411.13 MHz (estimated)
Counted L1_AND_L2_DTLB_MISSES events (L1 and L2 DTLB misses) with a unit
mask of 0x00 (No unit mask) count 500
samples  %        app name                 symbol name
162993    6.9814  libansys1.so             MultiFront::decompose(int, int,
Elemset *, int *, int, int, int)
160799    6.8874  libansys3.so             blockInnerProduct6L_fd
160459    6.8729  libansys3.so             blockSaxpy6L_fd
156018    6.6826  libansys3.so             maxb33_
84700     3.6279  libansys4.so             saxpy_pcg
83434     3.5737  libansys1.so             elmatrixmultpcg_
58074     2.4875  libansys4.so             .st4560
46000     1.9703  libansys4.so             .st4282
41166     1.7632  libansys3.so             blockSaxpyBackSolve6L_fd
41033     1.7575  libansys3.so             blockInnerProductBackSolve6L_fd
35762     1.5318  libansys1.so             inner_product_sub
35591     1.5245  libansys1.so             inner_product_sub2
28259     1.2104  libansys4.so             addVectors

Signed-off-by: Pravin B. Shelar <pravin.shelar@calsoftinc.com>
Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Christoph Lameter <clameter@engr.sgi.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: Minor white space cleanup in traps.c
Andi Kleen [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: Minor white space cleanup in traps.c

Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Allow sys_uselib unconditionally
Andi Kleen [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: Allow sys_uselib unconditionally

Previously it wasn't enabled in the binfmt_aout is a module case.

Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Don't disable basic block reordering
Andi Kleen [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: Don't disable basic block reordering

When compiling with -Os (which is default) the compiler defaults to it
anyways. And with -O2 it probably generates somewhat better (although
also larger) code.

Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: fix x86_64-mm-sched-clock-share
Andrew Morton [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: fix x86_64-mm-sched-clock-share

Fix for the following patch. Provide dummy cpufreq functions when
CPUFREQ is not compiled in.

Cc: Andi Kleen <ak@suse.de>
Cc: Dave Jones <davej@codemonkey.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Move cpu verification code to common file
Vivek Goyal [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: Move cpu verification code to common file

o This patch moves the code to verify long mode and SSE to a common file.
  This code is now shared by trampoline.S, wakeup.S, boot/setup.S and
  boot/compressed/head.S

o So far we used to do very limited check in trampoline.S, wakeup.S and
  in 32bit entry point. Now all the entry paths are forced to do the
  exhaustive check, including SSE because verify_cpu is shared.

o I am keeping this patch as last in the x86 relocatable series because
  previous patches have got quite some amount of testing done and don't want
  to distrub that. So that if there is problem introduced by this patch, at
  least it can be easily isolated.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Extend bzImage protocol for relocatable bzImage
Vivek Goyal [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: Extend bzImage protocol for relocatable bzImage

o Extend the bzImage protocol (same as i386) to allow bzImage loaders to
  load the protected mode kernel at non-1MB address. Now protected mode
  component is relocatable and can be loaded at non-1MB addresses.

o As of today kdump uses it to run a second kernel from a reserved memory
  area.

Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: build-time checking
Vivek Goyal [Wed, 2 May 2007 17:27:08 +0000 (19:27 +0200)]
[PATCH] x86-64: build-time checking

o X86_64 kernel should run from 2MB aligned address for two reasons.
- Performance.
- For relocatable kernels, page tables are updated based on difference
  between compile time address and load time physical address.
  This difference should be multiple of 2MB as kernel text and data
  is mapped using 2MB pages and PMD should be pointing to a 2MB
  aligned address. Life is simpler if both compile time and load time
  kernel addresses are 2MB aligned.

o Flag the error at compile time if one is trying to build a kernel which
  does not meet alignment restrictions.

Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[PATCH] x86-64: Relocatable Kernel Support
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86-64: Relocatable Kernel Support

This patch modifies the x86_64 kernel so that it can be loaded and run
at any 2M aligned address, below 512G.  The technique used is to
compile the decompressor with -fPIC and modify it so the decompressor
is fully relocatable.  For the main kernel the page tables are
modified so the kernel remains at the same virtual address.  In
addition a variable phys_base is kept that holds the physical address
the kernel is loaded at.  __pa_symbol is modified to add that when
we take the address of a kernel symbol.

When loaded with a normal bootloader the decompressor will decompress
the kernel to 2M and it will run there.  This both ensures the
relocation code is always working, and makes it easier to use 2M
pages for the kernel and the cpu.

AK: changed to not make RELOCATABLE default in Kconfig

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: __pa and __pa_symbol address space separation
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86: __pa and __pa_symbol address space separation

Currently __pa_symbol is for use with symbols in the kernel address
map and __pa is for use with pointers into the physical memory map.
But the code is implemented so you can usually interchange the two.

__pa which is much more common can be implemented much more cheaply
if it is it doesn't have to worry about any other kernel address
spaces.  This is especially true with a relocatable kernel as
__pa_symbol needs to peform an extra variable read to resolve
the address.

There is a third macro that is added for the vsyscall data
__pa_vsymbol for finding the physical addesses of vsyscall pages.

Most of this patch is simply sorting through the references to
__pa or __pa_symbol and using the proper one.  A little of
it is continuing to use a physical address when we have it
instead of recalculating it several times.

swapper_pgd is now NULL.  leave_mm now uses init_mm.pgd
and init_mm.pgd is initialized at boot (instead of compile time)
to the physmem virtual mapping of init_level4_pgd.  The
physical address changed.

Except for the for EMPTY_ZERO page all of the remaining references
to __pa_symbol appear to be during kernel initialization.  So this
should reduce the cost of __pa in the common case, even on a relocated
kernel.

As this is technically a semantic change we need to be on the lookout
for anything I missed.  But it works for me (tm).

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: do not use virt_to_page on kernel data address
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86-64: do not use virt_to_page on kernel data address

o virt_to_page() call should be used on kernel linear addresses and not
  on kernel text and data addresses. Swsusp code uses it on kernel data
  (statically allocated swsusp_header).

o Allocate swsusp_header dynamically so that virt_to_page() can be used
  safely.

o I am changing this because in next few patches, __pa() on x86_64 will
  no longer support kernel text and data addresses and hibernation breaks.

Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86: Move swsusp __pa() dependent code to arch portion
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86: Move swsusp __pa() dependent code to arch portion

o __pa() should be used only on kernel linearly mapped virtual addresses
  and not on kernel text and data addresses.

o Hibernation code needs to determine the physical address associated
  with kernel symbol to mark a section boundary which contains pages which
  don't have to be saved and restored during hibernate/resume operation.

o Move this piece of code in arch dependent section. So that architectures
  which don't have kernel text/data mapped into kernel linearly mapped
  region can come up with their own ways of determining physical addresses
  associated with a kernel text.

Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Remove the identity mapping as early as possible
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86-64: Remove the identity mapping as early as possible

With the rewrite of the SMP trampoline and the early page
allocator there is nothing that needs identity mapped pages,
once we start executing C code.

So add zap_identity_mappings into head64.c and remove
zap_low_mappings() from much later in the code.  The functions
 are subtly different thus the name change.

This also kills boot_level4_pgt which was from an earlier
attempt to move the identity mappings as early as possible,
and is now no longer needed.  Essentially I have replaced
boot_level4_pgt with trampoline_level4_pgt in trampoline.S

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: Modify discover_ebda to use virtual addresses
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86-64: Modify discover_ebda to use virtual addresses

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: 64bit ACPI wakeup trampoline
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86-64: 64bit ACPI wakeup trampoline

o Moved wakeup_level4_pgt into the wakeup routine so we can
  run the kernel above 4G.

o Now we first go to 64bit mode and continue to run from trampoline and
  then then start accessing kernel symbols and restore processor context.
  This enables us to resume even in relocatable kernel context when
  kernel might not be loaded at physical addr it has been compiled for.

o Removed the need for modifying any existing kernel page table.

o Increased the size of the wakeup routine to 8K. This is required as
  wake page tables are on trampoline itself and they got to be at 4K
  boundary, hence one page is not sufficient.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
17 years ago[PATCH] x86-64: wakeup.S misc cleanups
Vivek Goyal [Wed, 2 May 2007 17:27:07 +0000 (19:27 +0200)]
[PATCH] x86-64: wakeup.S misc cleanups

o Various cleanups. One of the main purpose of cleanups is that make
  wakeup.S as close as possible to trampoline.S.

o Following are the changes
- Indentations for comments.
- Changed the gdt table to compact form and to resemble the
  one in trampoline.S
- Take the jump to 32bit from real mode using ljmpl. Makes code
  more readable.
- After enabling long mode, directly take a long jump for 64bit
  mode. No need to take an extra jump to "reach_comaptibility_mode"
- Stack is not used after real mode. So don't load stack in
    32 bit mode.
- No need to enable PGE here.
- No need to do extra EFER read, anyway we trash the read contents.
- No need to enable system call (EFER_SCE). Anyway it will be
  enabled when original EFER is restored.
- No need to set MP, ET, NE, WP, AM bits in cr0. Very soon we will
     reload the original cr0 while restroing the processor state.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>