This is a new slab allocator which was motivated by the complexity of the
existing code in mm/slab.c. It attempts to address a variety of concerns
with the existing implementation.
A. Management of object queues
A particular concern was the complex management of the numerous object
queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for
each allocating CPU and use objects from a slab directly instead of
queueing them up.
B. Storage overhead of object queues
SLAB Object queues exist per node, per CPU. The alien cache queue even
has a queue array that contain a queue for each processor on each
node. For very large systems the number of queues and the number of
objects that may be caught in those queues grows exponentially. On our
systems with 1k nodes / processors we have several gigabytes just tied up
for storing references to objects for those queues This does not include
the objects that could be on those queues. One fears that the whole
memory of the machine could one day be consumed by those queues.
C. SLAB meta data overhead
SLAB has overhead at the beginning of each slab. This means that data
cannot be naturally aligned at the beginning of a slab block. SLUB keeps
all meta data in the corresponding page_struct. Objects can be naturally
aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte
boundaries and can fit tightly into a 4k page with no bytes left over.
SLAB cannot do this.
D. SLAB has a complex cache reaper
SLUB does not need a cache reaper for UP systems. On SMP systems
the per CPU slab may be pushed back into partial list but that
operation is simple and does not require an iteration over a list
of objects. SLAB expires per CPU, shared and alien object queues
during cache reaping which may cause strange hold offs.
E. SLAB has complex NUMA policy layer support
SLUB pushes NUMA policy handling into the page allocator. This means that
allocation is coarser (SLUB does interleave on a page level) but that
situation was also present before 2.6.13. SLABs application of
policies to individual slab objects allocated in SLAB is
certainly a performance concern due to the frequent references to
memory policies which may lead a sequence of objects to come from
one node after another. SLUB will get a slab full of objects
from one node and then will switch to the next.
F. Reduction of the size of partial slab lists
SLAB has per node partial lists. This means that over time a large
number of partial slabs may accumulate on those lists. These can
only be reused if allocator occur on specific nodes. SLUB has a global
pool of partial slabs and will consume slabs from that pool to
decrease fragmentation.
G. Tunables
SLAB has sophisticated tuning abilities for each slab cache. One can
manipulate the queue sizes in detail. However, filling the queues still
requires the uses of the spin lock to check out slabs. SLUB has a global
parameter (min_slab_order) for tuning. Increasing the minimum slab
order can decrease the locking overhead. The bigger the slab order the
less motions of pages between per CPU and partial lists occur and the
better SLUB will be scaling.
G. Slab merging
We often have slab caches with similar parameters. SLUB detects those
on boot up and merges them into the corresponding general caches. This
leads to more effective memory use. About 50% of all caches can
be eliminated through slab merging. This will also decrease
slab fragmentation because partial allocated slabs can be filled
up again. Slab merging can be switched off by specifying
slub_nomerge on boot up.
Note that merging can expose heretofore unknown bugs in the kernel
because corrupted objects may now be placed differently and corrupt
differing neighboring objects. Enable sanity checks to find those.
H. Diagnostics
The current slab diagnostics are difficult to use and require a
recompilation of the kernel. SLUB contains debugging code that
is always available (but is kept out of the hot code paths).
SLUB diagnostics can be enabled via the "slab_debug" option.
Parameters can be specified to select a single or a group of
slab caches for diagnostics. This means that the system is running
with the usual performance and it is much more likely that
race conditions can be reproduced.
I. Resiliency
If basic sanity checks are on then SLUB is capable of detecting
common error conditions and recover as best as possible to allow the
system to continue.
J. Tracing
Tracing can be enabled via the slab_debug=T,<slabcache> option
during boot. SLUB will then protocol all actions on that slabcache
and dump the object contents on free.
K. On demand DMA cache creation.
Generally DMA caches are not needed. If a kmalloc is used with
__GFP_DMA then just create this single slabcache that is needed.
For systems that have no ZONE_DMA requirement the support is
completely eliminated.
L. Performance increase
Some benchmarks have shown speed improvements on kernbench in the
range of 5-10%. The locking overhead of slub is based on the
underlying base allocation size. If we can reliably allocate
larger order pages then it is possible to increase slub
performance much further. The anti-fragmentation patches may
enable further performance increases.
Tested on:
i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator
SLUB Boot options
slub_nomerge Disable merging of slabs
slub_min_order=x Require a minimum order for slab caches. This
increases the managed chunk size and therefore
reduces meta data and locking overhead.
slub_min_objects=x Mininum objects per slab. Default is 8.
slub_max_order=x Avoid generating slabs larger than order specified.
slub_debug Enable all diagnostics for all caches
slub_debug=<options> Enable selective options for all caches
slub_debug=<o>,<cache> Enable selective options for a certain set of
caches
Available Debug options
F Double Free checking, sanity and resiliency
R Red zoning
P Object / padding poisoning
U Track last free / alloc
T Trace all allocs / frees (only use for individual slabs).
To use SLUB: Apply this patch and then select SLUB as the default slab
allocator.
[hugh@veritas.com: fix an oops-causing locking error]
[akpm@linux-foundation.org: various stupid cleanups and small fixes]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was broken. It adds complexity, for no good reason. Rather than
separate __pa() and __pa_symbol(), we should deprecate __pa_symbol(),
and preferably __pa() too - and just use "virt_to_phys()" instead, which
is more readable and has nicer semantics.
However, right now, just undo the separation, and make __pa_symbol() be
the exact same as __pa(). That fixes the bugs this patch introduced,
and we can do the fairly obvious cleanups later.
Do the new __phys_addr() function (which is now the actual workhorse for
the unified __pa()/__pa_symbol()) as a real external function, that way
all the potential issues with compile/link-time optimizations of
constant symbol addresses go away, and we can also, if we choose to, add
more sanity-checking of the argument.
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Vivek Goyal <vgoyal@in.ibm.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'for-linus' of git://one.firstfloor.org/home/andi/git/linux-2.6: (231 commits)
[PATCH] i386: Don't delete cpu_devs data to identify different x86 types in late_initcall
[PATCH] i386: type may be unused
[PATCH] i386: Some additional chipset register values validation.
[PATCH] i386: Add missing !X86_PAE dependincy to the 2G/2G split.
[PATCH] x86-64: Don't exclude asm-offsets.c in Documentation/dontdiff
[PATCH] i386: avoid redundant preempt_disable in __unlazy_fpu
[PATCH] i386: white space fixes in i387.h
[PATCH] i386: Drop noisy e820 debugging printks
[PATCH] x86-64: Fix allnoconfig error in genapic_flat.c
[PATCH] x86-64: Shut up warnings for vfat compat ioctls on other file systems
[PATCH] x86-64: Share identical video.S between i386 and x86-64
[PATCH] x86-64: Remove CONFIG_REORDER
[PATCH] x86-64: Print type and size correctly for unknown compat ioctls
[PATCH] i386: Remove copy_*_user BUG_ONs for (size < 0)
[PATCH] i386: Little cleanups in smpboot.c
[PATCH] x86-64: Don't enable NUMA for a single node in K8 NUMA scanning
[PATCH] x86: Use RDTSCP for synchronous get_cycles if possible
[PATCH] i386: Add X86_FEATURE_RDTSCP
[PATCH] i386: Implement X86_FEATURE_SYNC_RDTSC on i386
[PATCH] i386: Implement alternative_io for i386
...
Fix up trivial conflict in include/linux/highmem.h manually.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* master.kernel.org:/pub/scm/linux/kernel/git/jejb/voyager-2.6:
[VOYAGER] add smp alternatives
[VOYAGER] Use modern techniques to setup and teardown low identiy mappings.
[VOYAGER] Convert the monitor thread to use the kthread API
[VOYAGER] clockevents driver: bring voyager in to line
[VOYAGER] clockevents: correct boot cpu is zero assumption
[VOYAGER] add smp_call_function_single
* master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6: (59 commits)
PCI: Free resource files in error path of pci_create_sysfs_dev_files()
pci-quirks: disable MSI on RS400-200 and RS480
PCI hotplug: Use menuconfig objects
PCI: ZT5550 CPCI Hotplug driver fix
PCI: rpaphp: Remove semaphores
PCI: rpaphp: Ensure more pcibios_add/pcibios_remove symmetry
PCI: rpaphp: Use pcibios_remove_pci_devices() symmetrically
PCI: rpaphp: Document is_php_dn()
PCI: rpaphp: Document find_php_slot()
PCI: rpaphp: Rename rpaphp_register_pci_slot() to rpaphp_enable_slot()
PCI: rpaphp: refactor tail call to rpaphp_register_slot()
PCI: rpaphp: remove rpaphp_set_attention_status()
PCI: rpaphp: remove print_slot_pci_funcs()
PCI: rpaphp: Remove setup_pci_slot()
PCI: rpaphp: remove a call that does nothing but a pointer lookup
PCI: rpaphp: Remove another wrappered function
PCI: rpaphp: Remve another call that is a wrapper
PCI: rpaphp: remove a function that does nothing but wrap debug printks
PCI: rpaphp: Remove un-needed goto
PCI: rpaphp: Fix a memleak; slot->location string was never freed
...
Add more information to PCI resource collision message
to help with debugging.
Signed-off-by: Chuck Ebbert <cebbert@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
set_irq_msi() currently connects an irq_desc to an msi_desc. The archs call
it at some point in their setup routine, and then the generic code sets up the
reverse mapping from the msi_desc back to the irq.
set_irq_msi() should do both connections, making it the one and only call
required to connect an irq with it's MSI desc and vice versa.
The arch code MUST call set_irq_msi(), and it must do so only once it's sure
it's not going to fail the irq allocation.
Given that there's no need for the arch to return the irq anymore, the return
value from the arch setup routine just becomes 0 for success and anything else
for failure.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Allows architectures to advertise that they support MSI rather than listing
each architecture as a PCI_MSI dependency.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
At one time, if a BIOS ROM shadow was detected for the boot video
device (stored at offset 0xc0000), we'd set a special resource flag,
IORESOURCE_ROM_SHADOW, so that the sysfs ROM file code could handle
it properly. That broke along the way somewhere though, so current
kernels will be missing 'rom' files in sysfs if the video device
doesn't have an explicit ROM BAR.
This patch fixes the regression by moving the video fixup quirk to a
little later in the boot cycle (to avoid having its work undone by
PCI resource allocation) and checking in the PCI sysfs code whether
a rom file should be created due to a shadow resource, which is also
moved to a little later in the boot cycle so it will occur after the
video fixup. Tested and works on my i386 test box.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
I noticed that many source files include <linux/pci.h> while they do
not appear to need it. Here is an attempt to clean it all up.
In order to find all possibly affected files, I searched for all
files including <linux/pci.h> but without any other occurence of "pci"
or "PCI". I removed the include statement from all of these, then I
compiled an allmodconfig kernel on both i386 and x86_64 and fixed the
false positives manually.
My tests covered 66% of the affected files, so there could be false
positives remaining. Untested files are:
arch/alpha/kernel/err_common.c
arch/alpha/kernel/err_ev6.c
arch/alpha/kernel/err_ev7.c
arch/ia64/sn/kernel/huberror.c
arch/ia64/sn/kernel/xpnet.c
arch/m68knommu/kernel/dma.c
arch/mips/lib/iomap.c
arch/powerpc/platforms/pseries/ras.c
arch/ppc/8260_io/enet.c
arch/ppc/8260_io/fcc_enet.c
arch/ppc/8xx_io/enet.c
arch/ppc/syslib/ppc4xx_sgdma.c
arch/sh64/mach-cayman/iomap.c
arch/xtensa/kernel/xtensa_ksyms.c
arch/xtensa/platform-iss/setup.c
drivers/i2c/busses/i2c-at91.c
drivers/i2c/busses/i2c-mpc.c
drivers/media/video/saa711x.c
drivers/misc/hdpuftrs/hdpu_cpustate.c
drivers/misc/hdpuftrs/hdpu_nexus.c
drivers/net/au1000_eth.c
drivers/net/fec_8xx/fec_main.c
drivers/net/fec_8xx/fec_mii.c
drivers/net/fs_enet/fs_enet-main.c
drivers/net/fs_enet/mac-fcc.c
drivers/net/fs_enet/mac-fec.c
drivers/net/fs_enet/mac-scc.c
drivers/net/fs_enet/mii-bitbang.c
drivers/net/fs_enet/mii-fec.c
drivers/net/ibm_emac/ibm_emac_core.c
drivers/net/lasi_82596.c
drivers/parisc/hppb.c
drivers/sbus/sbus.c
drivers/video/g364fb.c
drivers/video/platinumfb.c
drivers/video/stifb.c
drivers/video/valkyriefb.c
include/asm-arm/arch-ixp4xx/dma.h
sound/oss/au1550_ac97.c
I would welcome test reports for these files. I am fine with removing
the untested files from the patch if the general opinion is that these
changes aren't safe. The tested part would still be nice to have.
Note that this patch depends on another header fixup patch I submitted
to LKML yesterday:
[PATCH] scatterlist.h needs types.h
http://lkml.org/lkml/2007/3/01/141
Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
In arch/i386/cpu/common.c there is:
cpu_devs[X86_VENDOR_INTEL]
cpu_devs[X86_VENDOR_CYRIX]
cpu_devs[X86_VENDOR_AMD]
...
They are all filled with data early.
The data (struct) got set to NULL for all, but Intel in different
late_initcall (exit_cpu_vendor) calls.
I don't see what sense this makes at all, maybe something that got
forgotten with the HOTPLUG_CPU extenstions?
Please check/review whether initdata, cpuinitdata is still ok and this
still works with HOTPLUG_CPU and without, it should...
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: davej@redhat.com
In the case of !CONFIG_PCI_DIRECT && !CONFIG_PCI_MMCONFIG, type is
unreferened.
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
On i945, a mmconfig range hitting the f0000000-ffffffff zone conflicts
with the APIC registers and others. Consider it invalid.
On E7520, values 0000 and f000 for the window register are defined
invalid in the documentation.
I haven't seen a bios use these values, but who trusts biosen these
days?
Signed-off-by: Olivier Galibert <galibert@pobox.com>
Signed-off-by: Andi Kleen <ak@suse.de>
arch/i386/pci/mmconfig-shared.c | 25 +++++++++++++++++--------
1 file changed, 17 insertions(+), 8 deletions(-)
Only 1GB-aligned kernel/user splits are now handled for PAE. The
2GB/2GB split attempts to avoid aliasing vmallocspace with the 1:1
mapping for physical memory by using an actual split of 1.875/2.125
to accommodate 128MB of vmallocspace out of what would otherwise
be a full 2GB for userspace. That attempt disturbs the alignment
required by PAE for 2GB/2GB splits, and furthermore does not provide
a 2GB/2GB split as advertised.
This patch resolves the issues here in two manners. The first is
by providing a true 2GB/2GB split in addition to the 1.875/2.125
split. The second is by renaming the 1.875/2.125 split to
CONFIG_VMSPLIT_2G_OPT analogously to CONFIG_VMSPLIT_3G_OPT, which
performs a similar manuever to avoid aliasing vmallocspace with
the 1:1 mapping for physical memory around the 3GB boundary. With
the 1.875/2.125 split properly-named, its config option is then
tagged as depending on !HIGHMEM to express the PAE implementation's
current inability to deal with such unaligned splits.
This patch is essentially a combination of two patches, one written
by Eric Biederman and the other by Eric Dumazet. If they could add
their Signed-off-by: to this, I'd be much obliged.
Signed-off-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Cc: Mark Lord <lkml@rtr.ca>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <ak@suse.de>
Check some CPUID bits that are needed for compiler generated early in boot.
When the system is still in real mode before changing the VESA BIOS mode
it is possible to still display an visible error message on the screen.
Similar to x86-64.
Includes cleanups from Eric Biederman
Signed-off-by: Andi Kleen <ak@suse.de>
- Introduce a wd_ops structure
- Convert the various nmi watchdogs over to it
- This allows to split the perfctr reservation from the watchdog
setup cleanly.
- Do perfctr reservation globally as it should have always been
- Remove dead code referenced only by unused EXPORT_SYMBOLs
Signed-off-by: Andi Kleen <ak@suse.de>
This patch just trivial converts from calling kernel_thread and daemonize
to just calling kthread_run.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The locking of the xtime_lock around the cpu notifier is unessesary now.
At one time the tsc was used after a frequency change for timekeeping, but
the re-write of timekeeping no longer uses the TSC unless the frequency is
constant.
The variables that are changed in this section of code had also once been
used for timekeeping, but not any longer ..
Signed-off-by: Daniel Walker <dwalker@mvista.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: john stultz <johnstul@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Currently the i386 architecture checks the family for mce capability and this
removes that and uses the CPUID information. Tested on a K8 revE and a
family10h processor.
This eliminates checking of a set AMD procesor family if mce is
allowed and relies on the information being in CPUID.
Signed-off-by: Joachim Deguara <joachim.deguara@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
IRQ is already disabled through local_irq_disable(). So
spin_lock_irqsave() can be replaced with spin_lock().
Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Avoid checking for cpu gone in mm hot path when CONFIG_HOTPLUG_CPU is not
defined.
Signed-off-by: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
startup_ipi_hook depends on CONFIG_X86_LOCAL_APIC, so move it to the
right part of the paravirt_ops initialization.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Fix section mismatch warnings in mtrr code.
Fix line length on one source line.
WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data: from .text.get_mtrr_state after 'get_mtrr_state' (at offset 0x103)
WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text: from .text.get_mtrr_state after 'get_mtrr_state' (at offset 0x180)
WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text: from .text.get_mtrr_state after 'get_mtrr_state' (at offset 0x199)
WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text: from .text.get_mtrr_state after 'get_mtrr_state' (at offset 0x1c1)
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2007-05-02 19:27:18 +02:00
Fernando Luis [** ISO-8859-1 charset **] VzquezCao
Use safe_apic_wait_icr_idle to check ICR idle bit if the vector is
NMI_VECTOR to avoid potential hangups in the event of crash when kdump
tries to stop the other CPUs.
Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Andi Kleen <ak@suse.de>
2007-05-02 19:27:18 +02:00
Fernando Luis [** ISO-8859-1 charset **] VzquezCao
Implement __send_IPI_dest_field which can be used to send IPIs when the
"destination shorthand" field of the ICR is set to 00 (destination
field). Use it whenever possible.
Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Andi Kleen <ak@suse.de>
__inquire_remote_apic is used for APIC debugging, so use
safe_apic_wait_icr_idle instead of apic_wait_icr_idle to avoid possible
lockups when APIC delivery fails.
Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Andi Kleen <ak@suse.de>
The functionality provided by the new safe_apic_wait_icr_idle is being
open-coded all over "kernel/smpboot.c". Use safe_apic_wait_icr_idle
instead to consolidate code and ease maintenance.
Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Andi Kleen <ak@suse.de>
apic_wait_icr_idle looks like this:
static __inline__ void apic_wait_icr_idle(void)
{
while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
cpu_relax();
}
The busy loop in this function would not be problematic if the
corresponding status bit in the ICR were always updated, but that does
not seem to be the case under certain crash scenarios. Kdump uses an IPI
to stop the other CPUs in the event of a crash, but when any of the
other CPUs are locked-up inside the NMI handler the CPU that sends the
IPI will end up looping forever in the ICR check, effectively
hard-locking the whole system.
Quoting from Intel's "MultiProcessor Specification" (Version 1.4), B-3:
"A local APIC unit indicates successful dispatch of an IPI by
resetting the Delivery Status bit in the Interrupt Command
Register (ICR). The operating system polls the delivery status
bit after sending an INIT or STARTUP IPI until the command has
been dispatched.
A period of 20 microseconds should be sufficient for IPI dispatch
to complete under normal operating conditions. If the IPI is not
successfully dispatched, the operating system can abort the
command. Alternatively, the operating system can retry the IPI by
writing the lower 32-bit double word of the ICR. This “time-out”
mechanism can be implemented through an external interrupt, if
interrupts are enabled on the processor, or through execution of
an instruction or time-stamp counter spin loop."
Intel's documentation suggests the implementation of a time-out
mechanism, which, by the way, is already being open-coded in some parts
of the kernel that tinker with ICR.
Create a apic_wait_icr_idle replacement that implements the time-out
mechanism and that can be used to solve the aforementioned problem.
AK: moved both functions out of line
AK: added improved loop from Keith Owens
Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Andi Kleen <ak@suse.de>
If our copy of the MTRRs of the BSP has RdMem or WrMem set, and
we are running on an AMD64/K8 system, the boot CPU must have had
MtrrFixDramEn and MtrrFixDramModEn set (otherwise our RDMSR would
have copied these bits cleared), so we set them on this CPU as well.
This allows us to keep the AMD64/K8 RdMem and WrMem bits in sync
across the CPUs of SMP systems in order to fullfill the duty of
system software to "initialize and maintain MTRR consistency
across all processors." as written in the AMD and Intel manuals.
If an WRMSR instruction fails because MtrrFixDramModEn is not
set, I expect that also the Intel-style MTRR bits are not updated.
AK: minor cleanup, moved MSR defines around
Signed-off-by: Bernhard Kaindl <bk@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andi Kleen <ak@suse.de>
Cc: Dave Jones <davej@codemonkey.org.uk>
Note: This patch didn'nt need an update since it's initial post.
Some BIOSes may modify fixed-range MTRRs in SMM, e.g. when they
transition the system into ACPI mode, which is entered thru an SMI,
triggered by Linux in acpi_enable().
SMIs which cause that Linux is interrupted and BIOS code is
executed (which may change e.g. fixed-range MTRRs) in SMM may
be raised by an embedded system controller which is often found
in notebooks also at other occasions.
If we would not update our copy of the fixed-range MTRRs before
suspending to RAM or to disk, restore_processor_state() would
set the fixed-range MTRRs of the BSP using old backup values
which may be outdated and this could cause the system to fail
later during resume.
This patch ensures that our copy of the fixed-range MTRRs
is updated when saving the boot processor state on suspend
to disk and suspend to RAM.
In combination with other patches this allows to fix s2ram
and s2disk on the Acer Ferrari 1000 notebook and at least
s2disk on the Acer Ferrari 5000 notebook.
Signed-off-by: Bernhard Kaindl <bk@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andi Kleen <ak@suse.de>
Cc: Dave Jones <davej@codemonkey.org.uk>
Applied fix by Andew Morton:
http://lkml.org/lkml/2007/4/8/88 - Fix `make headers_check'.
AMD and Intel x86 CPU manuals state that it is the responsibility of
system software to initialize and maintain MTRR consistency across
all processors in Multi-Processing Environments.
Quote from page 188 of the AMD64 System Programming manual (Volume 2):
7.6.5 MTRRs in Multi-Processing Environments
"In multi-processing environments, the MTRRs located in all processors must
characterize memory in the same way. Generally, this means that identical
values are written to the MTRRs used by the processors." (short omission here)
"Failure to do so may result in coherency violations or loss of atomicity.
Processor implementations do not check the MTRR settings in other processors
to ensure consistency. It is the responsibility of system software to
initialize and maintain MTRR consistency across all processors."
Current Linux MTRR code already implements the above in the case that the
BIOS does not properly initialize MTRRs on the secondary processors,
but the case where the fixed-range MTRRs of the boot processor are changed
after Linux started to boot, before the initialsation of a secondary
processor, is not handled yet.
In this case, secondary processors are currently initialized by Linux
with MTRRs which the boot processor had very early, when mtrr_bp_init()
did run, but not with the MTRRs which the boot processor uses at the
time when that secondary processors is actually booted,
causing differing MTRR contents on the secondary processors.
Such situation happens on Acer Ferrari 1000 and 5000 notebooks where the
BIOS enables and sets AMD-specific IORR bits in the fixed-range MTRRs
of the boot processor when it transitions the system into ACPI mode.
The SMI handler of the BIOS does this in SMM, entered while Linux ACPI
code runs acpi_enable().
Other occasions where the SMI handler of the BIOS may change bits in
the MTRRs could occur as well. To initialize newly booted secodary
processors with the fixed-range MTRRs which the boot processor uses
at that time, this patch saves the fixed-range MTRRs of the boot
processor before new secondary processors are started. When the
secondary processors run their Linux initialisation code, their
fixed-range MTRRs will be updated with the saved fixed-range MTRRs.
If CONFIG_MTRR is not set, we define mtrr_save_state
as an empty statement because there is nothing to do.
Possible TODOs:
*) CPU-hotplugging outside of SMP suspend/resume is not yet tested
with this patch.
*) If, even in this case, an AP never runs i386/do_boot_cpu or x86_64/cpu_up,
then the calls to mtrr_save_state() could be replaced by calls to
mtrr_save_fixed_ranges(NULL) and mtrr_save_state() would not be
needed.
That would need either verification of the CPU-hotplug code or
at least a test on a >2 CPU machine.
*) The MTRRs of other running processors are not yet checked at this
time but it might be interesting to syncronize the MTTRs of all
processors before booting. That would be an incremental patch,
but of rather low priority since there is no machine known so
far which would require this.
AK: moved prototypes on x86-64 around to fix warnings
Signed-off-by: Bernhard Kaindl <bk@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Dave Jones <davej@codemonkey.org.uk>
In this current implementation which is used in other patches,
mtrr_save_fixed_ranges() accepts a dummy void pointer because
in the current implementation of one of these patches, this
function may be called from smp_call_function_single() which
requires that this function takes a void pointer argument.
This function calls get_fixed_ranges(), passing mtrr_state.fixed_ranges
which is the element of the static struct which stores our current
backup of the fixed-range MTRR values which all CPUs shall be
using.
Because mtrr_save_fixed_ranges calls get_fixed_ranges after
kernel initialisation time, __init needs to be removed from
the declaration of get_fixed_ranges().
If CONFIG_MTRR is not set, we define mtrr_save_fixed_ranges
as an empty statement because there is nothing to do.
AK: Moved prototypes for x86-64 around to fix warnings
Signed-off-by: Bernhard Kaindl <bk@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andi Kleen <ak@suse.de>
Cc: Dave Jones <davej@codemonkey.org.uk>
Three cleanups:
1: ELF notes are never mapped, so there's no need to have any access
flags in their phdr.
2: When generating them from asm, tell the assembler to use a SHT_NOTE
section type. There doesn't seem to be a way to do this from C.
3: Use ANSI rather than traditional cpp behaviour to stringify the
macro argument.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Otherwise non GPL modules cannot even do basic operations
like disabling interrupts anymore, which would be excessive.
Longer term should split the single structure up into
internal and external symbols and not export the internal
ones at all.
Signed-off-by: Andi Kleen <ak@suse.de>
The other symbols used to delineate the alt-instructions sections have the
form __foo/__foo_end. Rename parainstructions to match.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Convert VMI timer to use clock events, making it properly able to use the NO_HZ
infrastructure. On UP systems, with no local APIC, we just continue to route
these events through the PIT. On systems with a local APIC, or SMP, we provide
a single source interrupt chip which creates the local timer IRQ. It actually
gets delivered by the APIC hardware, but we don't want to use the same local
APIC clocksource processing, so we create our own handler here.
Signed-off-by: Zachary Amsden <zach@vmware.com>
Signed-off-by: Andi Kleen <ak@suse.de>
CC: Dan Hecht <dhecht@vmware.com>
CC: Ingo Molnar <mingo@elte.hu>
CC: Thomas Gleixner <tglx@linutronix.de>
Implement vmi_kmap_atomic_pte in terms of the backend set_linear_mapping
operation. The conversion is rather straighforward; call kmap_atomic
and then inform the hypervisor of the page mapping.
The _flush_tlb damage is due to macros being pulled in from highmem.h.
Signed-off-by: Zachary Amsden <zach@vmware.com>
Signed-off-by: Andi Kleen <ak@suse.de>
No, just no. You do not use goto to skip a code block. You do not
return an obvious variable from a singly-inlined function and give
the function a return value. You don't put unexplained comments
about kmalloc in code which doesn't do dynamic allocation. And
you don't leave stray warnings around for no good reason.
Also, when possible, it is better to use block scoped variables
because gcc can sometime generate better code.
Signed-off-by: Zachary Amsden <zach@vmware.com>
Signed-off-by: Andi Kleen <ak@suse.de>
head.S creates the very initial pagetable for the kernel. This just
maps enough space for the kernel itself, and an allocation bitmap.
The amount of mapped memory is rounded up to 4Mbytes, and so this
typically ends up mapping 8Mbytes of memory.
When booting, pagetable_init() needs to create mappings for all
lowmem, and the pagetables for these mappings are allocated from the
free pages around the kernel in low memory. If the number of
pagetable pages + kernel size exceeds head.S's initial mapping, it
will end up faulting on an unmapped page. This will only happen with
specific combinations of kernel size and memory size.
This patch makes sure that head.S also maps enough space to fit the
kernel pagetables as well as the kernel itself. It ends up using an
additional two pages of unreclaimable memory.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
