Commit Graph

1250 Commits

Author SHA1 Message Date
Jack Steiner
34d0559178 x86: UV startup of slave cpus
This patch changes smpboot.c so that it can start slave cpus running
in UV non-unique apicid mode. The SIPI must be sent using a UV-specific
mechanism.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-04-19 19:19:58 +02:00
Glauber Costa
098cb7f27e x86: integrate pci-dma.c
The code in pci-dma_{32,64}.c are now sufficiently
close to each other. We merge them in pci-dma.c.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
bb8ada95a7 x86: don't do dma if mask is NULL.
if the device hasn't provided a mask, abort allocation.
Note that we're using a fallback device now, so it does not cover
the case of a NULL device: just drivers passing NULL masks around.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
da60cab4dd x86: return conditional to mmu
Just return our allocation if we don't have an mmu. For i386, where this patch
is being applied, we never have. So our goal is just to have the code to look like
x86_64's.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
aa99b16faa x86: remove kludge from x86_64
The claim is that i386 does it. Just it does not.
So remove it.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
8f19ca1341 x86: unify gfp masks
Use the same gfp masks for x86_64 and i386.
It involves using HIGHMEM or DMA32 where necessary, for the sake
of code compatibility, (no real effect), and using the NORETRY
mask for i386.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
5fa78ca75d x86: retry allocation if failed
This patch puts in the code to retry allocation in case it fails. By its
own, it does not make much sense but making the code look like x86_64.
But later patches in this series will make we try to allocate from
zones other than DMA first, which will possibly fail.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
8779f2fc3b x86: don't try to allocate from DMA zone at first
If we fail, we'll loop into the allocation again,
and then allocate in the DMA zone.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
45a07e7749 x86: use a fallback dev for i386
We can use a fallback dev for cases of a NULL device being passed (mostly ISA)
This comes from x86_64 implementation.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
d1a0790290 x86: use numa allocation function in i386
We can do it here to, in the same way x86_64 does.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
71848d687e x86: remove virt_to_bus in pci-dma_64.c
virt_to_bus() is deprecated according to the docs, and moreover,
won't return the right thing in i386 if we're dealing with high memory mappings.
So we make our allocation function return a page, and then use page_address() (for
virtual addr) and page_to_phys() (for physical addr) instead.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
2e33e36118 x86: adjust dma_free_coherent for i386
We call unmap_single, if available.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
cac67877d2 x86: move bad_dma_address
It goes to pci-dma.c, and is removed from the arch-specific files.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
d09d815c1b x86: isolate coherent mapping functions
i386 implements the declare coherent memory API, and x86_64 does not
it is reflected in pieces of dma_alloc_coherent and dma_free_coherent.
Those pieces are isolated in separate functions, that are declared
as empty macros in x86_64. This way we can make the code the same.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:58 +02:00
Glauber Costa
8e8edc6401 x86: move dma_coherent functions to pci-dma.c
They are placed in an ifdef, since they are i386 specific
the structure definition goes to dma-mapping.h.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
fae9a0d8ca x86: merge iommu initialization parameters
we merge the iommu initialization parameters in pci-dma.c
Nice thing, that both architectures at least recognize the same
parameters.

usedac i386 parameter is marked for deprecation

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
8e0c379718 x86: merge dma_supported
The code for both arches are very similar, so this patch merge them.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
bca5c09663 x86: move pci fixup to pci-dma.c
via_no_dac provides a fixup that is the same for both
architectures. Move it to pci-dma.c.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
116890d556 x86: move x86_64-specific to common code.
This patch moves the bootmem functions, that are largely
x86_64-specific into pci-dma.c. The code goes inside an ifdef.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
cb5867a5d8 x86: move initialization functions to pci-dma.c
initcalls that triggers the various possibiities for
dma subsys are moved to pci-dma.c.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
f9c258de34 x86: unify pci-nommu
merge pci-base_32.c and pci-nommu_64.c into pci-nommu.c
Their code were made the same, so now they can be merged.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
85c246ee16 x86: move definition to pci-dma.c
Move dma_ops structure definition to pci-dma.c, where it
belongs.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
d741bde26d x86: use dma_length in i386
This is done to get the code closer to x86_64.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
5b3e5b7273 x86: use WARN_ON in mapping functions
In the very same way i386 do, we use WARN_ON functions
in map_simple and map_sg.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
30db2cbf38 x86: use sg_phys in x86_64
To make the code usable in i386, where we have high memory mappings,
we drop te virt_to_bus(sg_virt()) construction in favour of sg_phys.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
e4dcdd6b4f x86: Add flush_write_buffers in nommu functions
This patch adds flush_write_buffers() in some functions of pci-nommu_64.c
They are added anywhere i386 would also have it. This is not a problem
for x86_64, since flush_rite_buffers() an nop for it.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
9f9ab46d55 x86: implement mapping_error in pci-nommu_64.c
This patch implements mapping_error for pci-nommu_64.c.
It takes care to keep the same compatible behaviour it already
had. Although this file is not (yet) used for i386, we introduce
the i386 version here. Again, care is taken, even at the expense of
an ifdef, to keep the same behaviour inconditionally.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
d5df63f48a x86: delete empty functions from pci-nommu_64.c
This functions are now called conditionally on their
existence in the struct. So just delete them, instead
of keeping an empty implementation.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
459121c9ec x86: introduce pci-dma.c
This patch introduces pci-dma.c, a common file for pci dma
between i386 and x86_64. As a start, dma_set_mask() is the same
between architectures, and is placed there.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Mark McLoughlin
19e395afb4 x86: move dma_supported and dma_set_mask to pci-dma_32.c, fix
ERROR: "dma_supported" [drivers/ssb/ssb.ko] undefined!
ERROR: "dma_set_mask" [drivers/scsi/qla2xxx/qla2xxx.ko] undefined!
ERROR: "dma_set_mask" [drivers/scsi/aic7xxx/aic7xxx.ko] undefined!
ERROR: "dma_set_mask" [drivers/scsi/aic7xxx/aic79xx.ko] undefined!
ERROR: "dma_supported" [drivers/net/pcnet32.ko] undefined!
ERROR: "dma_supported" [drivers/media/video/saa7134/saa7134.ko] undefined!
ERROR: "dma_set_mask" [drivers/media/video/meye.ko] undefined!
ERROR: "dma_supported" [drivers/media/video/cx88/cx8802.ko] undefined!
ERROR: "dma_supported" [drivers/media/video/cx88/cx8800.ko] undefined!
ERROR: "dma_supported" [drivers/media/video/cx88/cx88-alsa.ko] undefined!
ERROR: "dma_supported" [drivers/media/video/cx23885/cx23885.ko] undefined!

They just need to be exported like on x86_64.

dma_supported() and dma_set_mask() were previously inlined,
but are now moved to pci-dma_32.c.

Since they're used by various drivers, they need to be
exported.

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
c786df08f6 x86: unify dma_mapping_error
We provide a map_error function in pci-base_32.c to make
sure i386 keeps with the same behaviour it used to.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:57 +02:00
Glauber Costa
7c18341665 x86: provide a bad_dma_address symbol for i386
It's initially 0, since we don't expect any DMA there.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
802c1f6648 x86: move dma_supported and dma_set_mask to pci-dma_32.c
This is the way x86_64 does, so this make them equal. They have
to be extern now in the header, and the extern definition is moved to
the common dma-mapping.h header.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Ingo Molnar
2be621498d x86: dma-ops on highmem fix
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
e7f3a913f9 x86: move dma_sync_sg_for_device to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
ed435dee9c x86: move dma_sync_sg_for_cpu to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
713623326c x86: move dma_sync_single_range_for_device to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
627610fcb7 x86: move dma_sync_single_range_for_cpu to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
9231b269e0 x86: move dma_sync_single_for_device to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
c01dd8cf7d x86: move dma_sync_single_for_cpu to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
72c784f82c x86: move dma_unmap_sg to common header
i386 gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
16a3ce9bae x86: move dma_map_sg to common header
the old i386 implementation is moved to pci-base_32.c

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
0cb0ae6832 x86: move dma_unmap_single to common header
i386 base does not need it, so it gets an empty function.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Glauber Costa
22456b9714 x86: implement dma_map_single through dma_ops
That's already the name of the game for x86_64. For i386,
we add a pci-base_32.c, that will hold the default operations.
The function call itself goes through dma-mapping.h , the common
header

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:56 +02:00
Yinghai Lu
752bea4abb x86: reserve dma32 early for gart
a system with 256 GB of RAM, when NUMA is disabled crashes the
following way:

Your BIOS doesn't leave a aperture memory hole
Please enable the IOMMU option in the BIOS setup
This costs you 64 MB of RAM
Cannot allocate aperture memory hole (ffff8101c0000000,65536K)
Kernel panic - not syncing: Not enough memory for aperture
Pid: 0, comm: swapper Not tainted 2.6.25-rc4-x86-latest.git #33

Call Trace:
 [<ffffffff84037c62>] panic+0xb2/0x190
 [<ffffffff840381fc>] ? release_console_sem+0x7c/0x250
 [<ffffffff847b1628>] ? __alloc_bootmem_nopanic+0x48/0x90
 [<ffffffff847b0ac9>] ? free_bootmem+0x29/0x50
 [<ffffffff847ac1f7>] gart_iommu_hole_init+0x5e7/0x680
 [<ffffffff847b255b>] ? alloc_large_system_hash+0x16b/0x310
 [<ffffffff84506a2f>] ? _etext+0x0/0x1
 [<ffffffff847a2e8c>] pci_iommu_alloc+0x1c/0x40
 [<ffffffff847ac795>] mem_init+0x45/0x1a0
 [<ffffffff8479ff35>] start_kernel+0x295/0x380
 [<ffffffff8479f1c2>] _sinittext+0x1c2/0x230

the root cause is : memmap PMD is too big,
[ffffe200e0600000-ffffe200e07fffff] PMD ->ffff81383c000000 on node 0
almost near 4G..., and vmemmap_alloc_block will use up the ram under 4G.

solution will be:
1. make memmap allocation get memory above 4G...
2. reserve some dma32 range early before we try to set up memmap for all.
and release that before pci_iommu_alloc, so gart or swiotlb could get some
range under 4g limit for sure.

the patch is using method 2.
because method1 may need more code to handle SPARSEMEM and SPASEMEM_VMEMMAP

will get
Your BIOS doesn't leave a aperture memory hole
Please enable the IOMMU option in the BIOS setup
This costs you 64 MB of RAM
Mapping aperture over 65536 KB of RAM @ 4000000
Memory: 264245736k/268959744k available (8484k kernel code, 4187464k reserved, 4004k data, 724k init)

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:55 +02:00
Suresh Siddha
1679f2710a x86: fpu xstate split cleanup
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:55 +02:00
Suresh Siddha
aa283f4927 x86, fpu: lazy allocation of FPU area - v5
Only allocate the FPU area when the application actually uses FPU, i.e., in the
first lazy FPU trap. This could save memory for non-fpu using apps.

for example: on my system after boot, there are around 300 processes, with
only 17 using FPU.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:55 +02:00
Suresh Siddha
61c4628b53 x86, fpu: split FPU state from task struct - v5
Split the FPU save area from the task struct. This allows easy migration
of FPU context, and it's generally cleaner. It also allows the following
two optimizations:

1) only allocate when the application actually uses FPU, so in the first
lazy FPU trap. This could save memory for non-fpu using apps. Next patch
does this lazy allocation.

2) allocate the right size for the actual cpu rather than 512 bytes always.
Patches enabling xsave/xrstor support (coming shortly) will take advantage
of this.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:55 +02:00
Ingo Molnar
fa5c463941 x86: rename find_max_pfn() to propagate_e820_map()
this function doesnt just 'find' the max_pfn - it also has
other side-effects such as registering sparse memory maps.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-04-19 19:19:55 +02:00
Thomas Gleixner
d8bb6f4c16 x86: tsc prevent time going backwards
We already catch most of the TSC problems by sanity checks, but there
is a subtle bug which has been in the code forever. This can cause
time jumps in the range of hours.

This was reported in:
     http://lkml.org/lkml/2007/8/23/96
and
     http://lkml.org/lkml/2008/3/31/23

I was able to reproduce the problem with a gettimeofday loop test on a
dual core and a quad core machine which both have sychronized
TSCs. The TSCs seems not to be perfectly in sync though, but the
kernel is not able to detect the slight delta in the sync check. Still
there exists an extremly small window where this delta can be observed
with a real big time jump. So far I was only able to reproduce this
with the vsyscall gettimeofday implementation, but in theory this
might be observable with the syscall based version as well.

CPU 0 updates the clock source variables under xtime/vyscall lock and
CPU1, where the TSC is slighty behind CPU0, is reading the time right
after the seqlock was unlocked.

The clocksource reference data was updated with the TSC from CPU0 and
the value which is read from TSC on CPU1 is less than the reference
data. This results in a huge delta value due to the unsigned
subtraction of the TSC value and the reference value. This algorithm
can not be changed due to the support of wrapping clock sources like
pm timer.

The huge delta is converted to nanoseconds and added to xtime, which
is then observable by the caller. The next gettimeofday call on CPU1
will show the correct time again as now the TSC has advanced above the
reference value.

To prevent this TSC specific wreckage we need to compare the TSC value
against the reference value and return the latter when it is larger
than the actual TSC value.

I pondered to mark the TSC unstable when the readout is smaller than
the reference value, but this would render an otherwise good and fast
clocksource unusable without a real good reason.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-04-19 19:19:55 +02:00