SYSENTER can cause a NT to be set which might cause crashes on the IRET
in the next task.

Following similar i386 patch from Linus.
Signed-off-by: NAndi Kleen <ak@suse.de>

Current gcc generates calls not jumps to noreturn functions. When that happens the
return address can point to the next function, which confuses the unwinder.

This patch works around it by marking asynchronous exception
frames in contrast normal call frames in the unwind information.  Then teach
the unwinder to decode this.

For normal call frames the unwinder now subtracts one from the address which avoids
this problem.  The standard libgcc unwinder uses the same trick.

It doesn't include adjustment of the printed address (i.e. for the original
example, it'd still be kernel_math_error+0 that gets displayed, but the
unwinder wouldn't get confused anymore.

This only works with binutils 2.6.17+ and some versions of H.J.Lu's 2.6.16
unfortunately because earlier binutils don't support .cfi_signal_frame

[AK: added automatic detection of the new binutils and wrote description]
Signed-off-by: NJan Beulich <jbeulich@novell.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

This was old code that was needed for iBCS and x86-64 never supported that.

Pointed out by Albert Cahalan
Signed-off-by: NAndi Kleen <ak@suse.de>

We do some additional CPU synchronization in gettimeofday et.al. to make
sure the time stamps are always monotonic over multiple CPUs. But on
single core systems that is not needed. So don't do it.
Signed-off-by: NAndi Kleen <ak@suse.de>

It is faster than using a unrolled loop for the use cases the kernel
cares about (cached, sizes typically < 4K)
Signed-off-by: NAndi Kleen <ak@suse.de>

Got it. i8259A_resume calls init_8259A(0) unconditionally, even if
auto_eoi has been set. Keep track of the current status and restore that
on resume. This fixes it for AMD64 and i386.
Signed-off-by: NMatthew Garrett <mjg59@srcf.ucam.org>
Signed-off-by: NAndi Kleen <ak@suse.de>

Patch inserts the GART region into the iomem resource map. The GART will then
be visible within /proc/iomem. It will also allow for other users
utilizing the GART to subreserve the region (agp or IOMMU).
Signed-off-by: NAaron Durbin <adurbin@google.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

Previously exit_idle would be called more often than enter_idle

Now instead of using complicated tests just keep track of it
using the per CPU variable as a flip flop.  I moved the idle state into the
PDA to make the access more efficient.

Original bug report and an initial patch from Stephane Eranian,
but redone by AK.

Cc: Stephane Eranian <eranian@hpl.hp.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

Before 2.6.16 this was changed to work around code that accessed
CPUs not in the possible map. But that code should be all fixed now,
so mark it __initdata again.
Signed-off-by: NAndi Kleen <ak@suse.de>

- Don't zero for __copy_from_user_inatomic following i386.
This will prevent spurious zeros for parallel file system writers when
one does a exception
- The string instruction version didn't zero the output on
exception. Oops.

Also I cleaned up the code a bit while I was at it and added a minor
optimization to the string instruction path.
Signed-off-by: NAndi Kleen <ak@suse.de>

This patch places the IOAPIC(s) and the Local APIC specified by ACPI
tables into the resource map. The APICs will then be visible within
/proc/iomem
Signed-off-by: NAaron Durbin <adurbin@google.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

Fix
linux/arch/x86_64/kernel/process.c: In function __switch_to:
linux/arch/x86_64/kernel/process.c:626: warning: assignment makes integer from pointer without a cast
Signed-off-by: NAndi Kleen <ak@suse.de>

Signed-off-by: NAndi Kleen <ak@suse.de>

This patch adds the per thread cookie field to the task struct and the PDA.
Also it makes sure that the PDA value gets the new cookie value at context
switch, and that a new task gets a new cookie at task creation time.
Signed-off-by: NArjan van Ven <arjan@linux.intel.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NAndi Kleen <ak@suse.de>
CC: Andi Kleen <ak@suse.de>

Because it can take spinlocks.

Suggested by Mathieu Desnoyers

Cc: Mathieu Desnoyers <compudj@krystal.dyndns.org>
Signed-off-by: NAndi Kleen <ak@suse.de>

kexec: Avoid overwriting the current pgd (V4, x86_64)

This patch upgrades the x86_64-specific kexec code to avoid overwriting the
current pgd. Overwriting the current pgd is bad when CONFIG_CRASH_DUMP is used
to start a secondary kernel that dumps the memory of the previous kernel.

The code introduces a new set of page tables. These tables are used to provide
an executable identity mapping without overwriting the current pgd.
Signed-off-by: NMagnus Damm <magnus@valinux.co.jp>
Signed-off-by: NAndi Kleen <ak@suse.de>

Remove most of the special cases for the debug IST stack.  This is a
follow on clean up patch, it requires the bug fix patch that adds
orig_ist.
Signed-off-by: NKeith Owens <kaos@ocs.com.au>
Signed-off-by: NAndi Kleen <ak@suse.de>

Based on a idea by Jeremy Fitzhardinge:

Replace the volatiles and memory clobbers in the PDA access with
telling gcc about access to a proxy PDA structure that doesn't
actually exist. But the dummy accesses give a defined ordering for
read/write accesses.

Also add some memory barriers to the early GS initialization to
make sure no PDA access is moved before it.

Advantage is some .text savings (probably most from better
code for accessing "current"):

   text    data     bss     dec     hex filename
4845647 1223688  615864 6685199  66020f vmlinux
4837780 1223688  615864 6677332  65e354 vmlinux-pda

1.2% smaller code

Cc:  Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: NAndi Kleen <ak@suse.de>

This patch updates x86_64 linker script to pack any .note.* sections
into a PT_NOTE segment in the output file.

To do this, we tell ld that we need a PT_NOTE segment.  This requires
us to start explicitly mapping sections to segments, so we also need
to explicitly create PT_LOAD segments for text and data, and map the
sections to them appropriately.  Fortunately, each section will
default to its previous section's segment, so it doesn't take many
changes to vmlinux.lds.S.

The corresponding change is already made for i386 in -mm and I'd like
this patch to join it. The section to segment mappings do change as do
the segment flags so some time in -mm would be good for that reason as
well, just in case.

In particular .data and .bss move from the text segment to the data
segment and .data.cacheline_aligned .data.read_mostly are put in the
data segment instead of a separate one.

I think that it would be possible to exactly match the existing section
to segment mapping and flags but it would be a more intrusive change and
I'm not sure there is a reason for the existing layout other than it is
what you get by default if you don't explicitly specify something else.
If there is a reason for the existing layout then I will of course make
the more intrusive change. If there is no reason we could probably drop
the executable or writable flags from some segments but I don't know how
much attention is paid to them anyway so it might not be worth the
effort.

The vsyscall related sections need to go in a different segment to the
normal data segment and so I invented a "user" segment to contain them.
I believe this should appear to be another data segment as far as the
kernel is concerned so the flags are setup accordingly.

The notes will be used in the Xen paravirt_ops backend to provide
additional information to the domain builder. I am in the process of
converting the xen-unstable kernels and tools over to this scheme at the
moment to support this in the future.

It has been suggested to me that the notes segment should have flags 0
(i.e. not readable) since it is only used by the loader and is not used
at runtime. For now I went with a readable segment since that is what
the i386 patch uses.

AK: dropped NOTES addition right now because the needed infrastructure
for that is not merged yet
Signed-off-by: NIan Campbell <ian.campbell@xensource.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

In long mode the %cs is largely a relic.  However there are a few cases
like iret where it matters that we have a valid value.  Without this
patch it is possible to enter the kernel in startup_64 without setting
%cs to a valid value.  With this patch we don't care what %cs value
we enter the kernel with, so long as the cs shadow register indicates
it is a privileged code segment.

Thanks to Magnus Damm for finding this problem and posting the
first workable patch.  I have moved the jump to set %cs down a
few instructions so we don't need to take an extra jump.  Which
keeps the code simpler.
Signed-of-by: NEric W. Biederman <ebiederm@xmission.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

Drop support for non e820 BIOS calls to get the memory map.

The boot assembler code still has some support, but not the C code now.
Signed-off-by: NAndi Kleen <ak@suse.de>

NMIs are not supposed to track the irq flags, but TRACE_IRQS_IRETQ
did it anyways. Add a check.

Cc: mingo@elte.hu
Signed-off-by: NAndi Kleen <ak@suse.de>

Give the printks a consistent prefix.
Add some missing white space.

Cc: len.brown@intel.com
Signed-off-by: NAndi Kleen <ak@suse.de>

- Remove a define that was used only once
- Remove the too large APIC ID check because we always support
the full 8bit range of APICs.
- Restructure code a bit to be simpler.

Cc: len.brown@intel.com
Signed-off-by: NAndi Kleen <ak@suse.de>

ACPI went to great trouble to get the APIC version and CPU capabilities
of different CPUs before passing them to the mpparser. But all
that data was used was to print it out.  Actually it even faked some data
based on the boot cpu, not on the actual CPU being booted.

Remove all this code because it's not needed.

Cc: len.brown@intel.com
Signed-off-by: NAndi Kleen <ak@suse.de>

And replace all users with ordinary smp_processor_id.  The function
was originally added to get some basic oops information out even
if the GS register was corrupted. However that didn't
work for some anymore because printk is needed to print the oops
and it uses smp_processor_id() already. Also GS register corruptions
are not particularly common anymore.

This also helps the Xen port which would otherwise need to
do this in a special way because it can't access the local APIC.

Cc: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: NAndi Kleen <ak@suse.de>

Detect the situations in which the time after a resume from disk would
be earlier than the time before the suspend and prevent them from
happening on x86_64.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NAndi Kleen <ak@suse.de>

From i386 x86-64 inherited code to force reserve the 640k-1MB area.
That was needed on some old systems.

But we generally trust the e820 map to be correct on 64bit systems
and mark all areas that are not memory correctly.

This patch will allow to use the real memory in there.

Or rather the only way to find out if it's still needed is to
try. So far I'm optimistic.
Signed-off-by: NAndi Kleen <ak@suse.de>

The init_amd() function is only called from identify_cpu() which is already
marked as __cpuinit. So let's mark it as __cpuinit.
Signed-off-by: NMagnus Damm <magnus@valinux.co.jp>
Signed-off-by: NAndi Kleen <ak@suse.de>

Implement pause_on_oops() on x86_64.

AK: I redid the patch to do the oops_enter/exit in the existing
oops_begin()/end(). This makes it much shorter.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NAndi Kleen <ak@suse.de>

Right now the kernel on x86-64 has a 100% lazy fpu behavior: after *every*
context switch a trap is taken for the first FPU use to restore the FPU
context lazily.  This is of course great for applications that have very
sporadic or no FPU use (since then you avoid doing the expensive
save/restore all the time).  However for very frequent FPU users...  you
take an extra trap every context switch.

The patch below adds a simple heuristic to this code: After 5 consecutive
context switches of FPU use, the lazy behavior is disabled and the context
gets restored every context switch.  If the app indeed uses the FPU, the
trap is avoided.  (the chance of the 6th time slice using FPU after the
previous 5 having done so are quite high obviously).

After 256 switches, this is reset and lazy behavior is returned (until
there are 5 consecutive ones again).  The reason for this is to give apps
that do longer bursts of FPU use still the lazy behavior back after some
time.

[akpm@osdl.org: place new task_struct field next to jit_keyring to save space]
Signed-off-by: NArjan van de Ven <arjan@linux.intel.com>
Signed-off-by: NAndi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>

Now for a completely different but trivial approach.
I just boot tested it with 255 CPUS and everything worked.

Currently everything (except module data) we place in
the per cpu area we know about at compile time.  So
instead of allocating a fixed size for the per_cpu area
allocate the number of bytes we need plus a fixed constant
for to be used for modules.

It isn't perfect but it is much less of a pain to
work with than what we are doing now.

AK: fixed warning
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

Since it's all zero.

Actually I think gcc 4+ will do that automatically, but earlier compilers won't
Signed-off-by: NAndi Kleen <ak@suse.de>

I've noticed some erratic behavior while testing the X86_64 version
of monotonic_clock().

While spinning in a loop reading monotonic clock values (pinned to a
single cpu) I noticed that the difference between subsequent values
occasionally went negative (time going backwards).

I found that in the following code:
                this_offset = get_cycles_sync();
                /* FIXME: 1000 or 1000000? */
-->             offset = (this_offset - last_offset)*1000 / cpu_khz;
        }
        return base + offset;

the offset sometimes turns out to be 0, even though
this_offset > last_offset.

+Added fix From: Toyo Abe <toyoa@mvista.com>

The x86_64-mm-monotonic-clock.patch in 2.6.18-rc4-mm2 made a change to
the updating of monotonic_base. It now uses cycles_2_ns().

I suggest that a set_cyc2ns_scale() should be done prior to the setup_irq().
Because cycles_2_ns() can be called from the timer ISR right after the irq0
is enabled.
Signed-off-by: NToyo Abe <toyoa@mvista.com>
Signed-off-by: NDimitri Sivanich <sivanich@sgi.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

This patch moves the entry.S:error_entry to .kprobes.text section,
since code marked unsafe for kprobes jumps directly to entry.S::error_entry,
that must be marked unsafe as well.
This patch also moves all the ".previous.text" asm directives to ".previous"
for kprobes section.

AK: Following a similar i386 patch from Chuck Ebbert
AK: Also merged Jeremy's fix in.

+From: Jeremy Fitzhardinge <jeremy@goop.org>

KPROBE_ENTRY does a .section .kprobes.text, and expects its users to
do a .previous at the end of the function.

Unfortunately, if any code within the function switches sections, for
example .fixup, then the .previous ends up putting all subsequent code
into .fixup.  Worse, any subsequent .fixup code gets intermingled with
the code its supposed to be fixing (which is also in .fixup).  It's
surprising this didn't cause more havok.

The fix is to use .pushsection/.popsection, so this stuff nests
properly.  A further cleanup would be to get rid of all
.section/.previous pairs, since they're inherently fragile.

+From: Chuck Ebbert <76306.1226@compuserve.com>

Because code marked unsafe for kprobes jumps directly to
entry.S::error_code, that must be marked unsafe as well.
The easiest way to do that is to move the page fault entry
point to just before error_code and let it inherit the same
section.

Also moved all the ".previous" asm directives for kprobes
sections to column 1 and removed ".text" from them.
Signed-off-by: NChuck Ebbert <76306.1226@compuserve.com>
Signed-off-by: NAndi Kleen <ak@suse.de>

Signed-off-by: NAndi Kleen <ak@suse.de>

This unifies the standard backtracer and the new stacktrace
in memory backtracer. The standard one is converted to use callbacks
and then reimplement stacktrace using new callbacks.

The main advantage is that stacktrace can now use the new dwarf2 unwinder
and avoid false positives in many cases.

I kept it simple to make sure the standard backtracer stays reliable.

Cc: mingo@elte.hu
Signed-off-by: NAndi Kleen <ak@suse.de>

Lockdep can call the dwarf2 unwinder early, and the dwarf2 code
uses safe_smp_processor_id which tries to access the local APIC page.
But that doesn't work before the APIC code has set up its fixmap.

Check for this case and always return boot cpu then.

Cc: jbeulich@novell.com
Cc: mingo@elte.hu
Signed-off-by: NAndi Kleen <ak@suse.de>

- Remove unused all_contexts parameter
No caller used it
- Move skip argument into the structure (needed for
followon patches)

Cc: mingo@elte.hu
Signed-off-by: NAndi Kleen <ak@suse.de>

Signed-off-by: NMuli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: NJon Mason <jdmason@us.ibm.com>
Signed-off-by: NAndi Kleen <ak@suse.de>