We noticed very erratic behavior [throughput] with the AIM7 shared
workload running on recent distro [SLES11] and mainline kernels on an
8-socket, 32-core, 256GB x86_64 platform. On the SLES11 kernel
[2.6.27.19+] with Barcelona processors, as we increased the load [10s of
thousands of tasks], the throughput would vary between two "plateaus"--one
at ~65K jobs per minute and one at ~130K jpm. The simple patch below
causes the results to smooth out at the ~130k plateau.
But wait, there's more:
We do not see this behavior on smaller platforms--e.g., 4 socket/8 core.
This could be the result of the larger number of cpus on the larger
platform--a scalability issue--or it could be the result of the larger
number of interconnect "hops" between some nodes in this platform and how
the tasks for a given load end up distributed over the nodes' cpus and
memories--a stochastic NUMA effect.
The variability in the results are less pronounced [on the same platform]
with Shanghai processors and with mainline kernels. With 31-rc6 on
Shanghai processors and 288 file systems on 288 fibre attached storage
volumes, the curves [jpm vs load] are both quite flat with the patched
kernel consistently producing ~3.9% better throughput [~80K jpm vs ~77K
jpm] than the unpatched kernel.
Profiling indicated that the "slow" runs were incurring high[er]
contention on an anon_vma lock in vma_adjust(), apparently called from the
sbrk() system call.
The patch:
A comment in mm/mmap.c:vma_adjust() suggests that we don't really need the
anon_vma lock when we're only adjusting the end of a vma, as is the case
for brk(). The comment questions whether it's worth while to optimize for
this case. Apparently, on the newer, larger x86_64 platforms, with
interesting NUMA topologies, it is worth while--especially considering
that the patch [if correct!] is quite simple.
We can detect this condition--no overlap with next vma--by noting a NULL
"importer". The anon_vma pointer will also be NULL in this case, so
simply avoid loading vma->anon_vma to avoid the lock.
However, we DO need to take the anon_vma lock when we're inserting a vma
['insert' non-NULL] even when we have no overlap [NULL "importer"], so we
need to check for 'insert', as well. And Hugh points out that we should
also take it when adjusting vm_start (so that rmap.c can rely upon
vma_address() while it holds the anon_vma lock).
akpm: Zhang Yanmin reprts a 150% throughput improvement with aim7, so it
might be -stable material even though thiss isn't a regression: "this
issue is not clear on dual socket Nehalem machine (2*4*2 cpu), but is
severe on large machine (4*8*2 cpu)"
[hugh.dickins@tiscali.co.uk: test vma start too] Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: Nick Piggin <npiggin@suse.de> Cc: Eric Whitney <eric.whitney@hp.com> Tested-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
do_anonymous_page() has been wrong to dirty the pte regardless.
If it's not going to mark the pte writable, then it won't help
to mark it dirty here, and clogs up memory with pages which will
need swap instead of being thrown away. Especially wrong if no
overcommit is chosen, and this vma is not yet VM_ACCOUNTed -
we could exceed the limit and OOM despite no overcommit.
thinkpad-acpi: fix incorrect use of TPACPI_BRGHT_MODE_ECNVRAM
HBRV-based default selection of backlight control strategy didn't work
well, at least the X41 defines it but doesn't use it and I don't think
it will stop there. Switch to a blacklist, and make sure only Radeon-
based models get ECNVRAM.
Symptoms of incorrect backlight mode selection are:
1. Non-working backlight control through sysfs;
2. Backlight gets reset to the lowest level at every shutdown, reboot
and when thinkpad-acpi gets unloaded;
This fixes a regression in 2.6.30, bugzilla #13826. This fix is
already present on 2.6.31.
Since 2.6.29 the PCI PM core have been restoring the standard
configuration registers of PCI devices in the early phase of
resume. In particular, PCI devices without drivers have been handled
this way since commit 355a72d75b3b4f4877db4c9070c798238028ecb5
(PCI: Rework default handling of suspend and resume). Unfortunately,
this leads to post-resume problems with CardBus devices which cannot
be accessed in the early phase of resume, because the sockets they
are on have not been woken up yet at that point.
To solve this problem, move the yenta socket resume to the early
phase of resume and, analogously, move the suspend of it to the late
phase of suspend. Additionally, remove some unnecessary PCI code
from the yenta socket's resume routine.
Fixes http://bugzilla.kernel.org/show_bug.cgi?id=13092, which is a
post-2.6.28 regression.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl> Reported-by: Florian <fs-kernelbugzilla@spline.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
powerpc: Fix incorrect setting of __HAVE_ARCH_PTE_SPECIAL
[I'm going to fix upstream differently, by having all CPU types
actually support _PAGE_SPECIAL, but I prefer the simple and obvious
fix for -stable. -- Ben]
The test that decides whether to define __HAVE_ARCH_PTE_SPECIAL on
powerpc is bogus and will end up always defining it, even when
_PAGE_SPECIAL is not supported (in which case it's 0) such as on
8xx or 40x processors.
Signed-off-by: Bernhard Weirich <bernhard.weirich@riedel.net> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
After upgrading to the latest kernel on my mpc875 userspace started
running incredibly slow (hours to get to a shell, even!).
I tracked it down to commit 8d30c14cab30d405a05f2aaceda1e9ad57800f36,
that patch removed a work-around for the 8xx. Adding it
back makes my problem go away.
Signed-off-by: Rex Feany <rfeany@mrv.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Lee Schermerhorn [Tue, 22 Sep 2009 00:01:04 +0000 (17:01 -0700)]
hugetlb: restore interleaving of bootmem huge pages (2.6.31)
Not upstream as it is fixed differently in .32
I noticed that alloc_bootmem_huge_page() will only advance to the next
node on failure to allocate a huge page. I asked about this on linux-mm
and linux-numa, cc'ing the usual huge page suspects. Mel Gorman
responded:
I strongly suspect that the same node being used until allocation
failure instead of round-robin is an oversight and not deliberate
at all. It appears to be a side-effect of a fix made way back in
commit 63b4613c3f0d4b724ba259dc6c201bb68b884e1a ["hugetlb: fix
hugepage allocation with memoryless nodes"]. Prior to that patch
it looked like allocations would always round-robin even when
allocation was successful.
Andy Whitcroft countered that the existing behavior looked like Andi
Kleen's original implementation and suggested that we ask him. We did and
Andy replied that his intention was to interleave the allocations. So,
...
This patch moves the advance of the hstate next node from which to
allocate up before the test for success of the attempted allocation. This
will unconditionally advance the next node from which to alloc,
interleaving successful allocations over the nodes with sufficient
contiguous memory, and skipping over nodes that fail the huge page
allocation attempt.
Note that alloc_bootmem_huge_page() will only be called for huge pages of
order > MAX_ORDER.
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Reviewed-by: Andi Kleen <ak@linux.intel.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: David Rientjes <rientjes@google.com> Cc: Adam Litke <agl@us.ibm.com> Cc: Andy Whitcroft <apw@canonical.com> Cc: Eric Whitney <eric.whitney@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Git commit 79741dd changes idle cputime accounting, but unfortunately
the /proc/uptime file hasn't caught up. Here the idle time calculation
from /proc/stat is copied over.
Signed-off-by: Michael Abbott <michael.abbott@diamond.ac.uk> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Kernel 2.6.30 introduced a patch [1] for the persistent option in the
netfilter SNAT target. This is exactly what we need here so I had a quick look
at the code and noticed that the patch is wrong. The logic is simply inverted.
The patch below fixes this.
Also note that because of this the default behavior of the SNAT target has
changed since kernel 2.6.30 as it now ignores the destination IP in choosing
the source IP for nating (which should only be the case if the persistent
option is set).
Commit 19eda87 (netfilter: change return types of check functions for
Ebtables extensions) broke the ebtables ulog module by missing a return
value conversion.
Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The length of the to-copy data structure is currently stored in
a signed integer. However many comparisons are done with sizeof(..)
which is unsigned. It's more suitable for this variable to be unsigned
to make these comparisons more naturally right.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: David S. Miller <davem@davemloft.net> Cc: Ingo Molnar <mingo@elte.hu> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
as a security check against optlen being negative (or zero) in the
set socket option.
Unfortunately, "sizeof(int)" is an unsigned property, with the
result that the whole comparison is done in unsigned, letting
negative values slip through.
This patch changes this to
if (optlen < (int)sizeof(int))
return -EINVAL;
so that the comparison is done as signed, and negative values
get properly caught.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: David S. Miller <davem@davemloft.net> Cc: Ingo Molnar <mingo@elte.hu> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Commit a5bfc4714b3f01365aef89a92673f2ceb1ccf246 dropped explicit
pci_intx() manipulation from ahci because it seemed unnecessary and
ahci doesn't seem to be the right place to be tweaking it if it were.
This was largely okay but there are exceptions. There was one on an
embedded platform which was fixed via firmware and now bko#14124
reports it on a HP DL320.
http://bugzilla.kernel.org/show_bug.cgi?id=14124
I still think this isn't something libata drivers should be caring
about (the only ones which are calling pci_intx() explicitly are
libata ones and one other driver) but for now reverting the change
seems to be the right thing to do.
Signed-off-by: Tejun Heo <tj@kernel.org> Reported-by: Thomas Jarosch <thomas.jarosch@intra2net.com> Signed-off-by: Jeff Garzik <jgarzik@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
To my understanding, it became obsolete with the advent of the more
robust check in mmu_alloc_roots (89da4ff17f). Moreover, it prevents
the conceptually safe pattern
1. set sregs
2. register mem-slots
3. run vcpu
by setting a sticky triple fault during step 1.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Signed-off-by: Avi Kivity <avi@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The use of __pa() to calculate the address of a C-visible symbol
is wrong, and can lead to unpredictable results. See arch/x86/include/asm/page.h
for details.
It should be replaced with __pa_symbol(), that does the correct math here,
by taking relocations into account. This ensures the correct wallclock data
structure physical address is passed to the hypervisor.
Signed-off-by: Glauber Costa <glommer@redhat.com> Signed-off-by: Avi Kivity <avi@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
n_requested_mmu_pages/n_free_mmu_pages are used by
kvm_mmu_change_mmu_pages to calculate the number of pages to zap.
alloc_mmu_pages, called from the vcpu initialization path, modifies this
variables without proper locking, which can result in a negative value
in kvm_mmu_change_mmu_pages (say, with cpu hotplug).
So far unprivileged guest callers running in ring 3 can issue, e.g., MMU
hypercalls. Normally, such callers cannot provide any hand-crafted MMU
command structure as it has to be passed by its physical address, but
they can still crash the guest kernel by passing random addresses.
To close the hole, this patch considers hypercalls valid only if issued
from guest ring 0. This may still be relaxed on a per-hypercall base in
the future once required.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Signed-off-by: Avi Kivity <avi@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
KVM: VMX: Fix cr8 exiting control clobbering by EPT
(cherry picked from commit 5fff7d270bd6a4759b6d663741b729cdee370257)
Don't call adjust_vmx_controls() two times for the same control.
It restores options that were dropped earlier. This loses us the cr8
exit control, which causes a massive performance regression Windows x64.
Debug registers may only be accessed from cpl 0. Unfortunately, vmx will
code to emulate the instruction even though it was issued from guest
userspace, possibly leading to an unexpected trap later.
commit ac68392460ffefed13020967bae04edc4d3add06 ("[CIFS] Allow raw
ntlmssp code to be enabled with sec=ntlmssp") added a new bit to the
allowed security flags mask but seems to have inadvertently removed
Lanman security from the allowed flags. Add it back.
Signed-off-by: Chuck Ebbert <cebbert@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch adds a new usbid for Zcomax XG-705A to the device table.
Reported-by: Jari Jaakola <jari.jaakola@gmail.com> Signed-off-by: Christian Lamparter <chunkeey@googlemail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This will fix file system corruption which infrequently happens after
mount. The problem was reported from users with the title "[NILFS
users] Fail to mount NILFS." (Message-ID:
<200908211918.34720.yuri@itinteg.net>), and so forth. I've also
experienced the corruption multiple times on kernel 2.6.30 and 2.6.31.
The problem turned out to be caused due to discordance between
mapping->nrpages of a btree node cache and the actual number of pages
hung on the cache; if the mapping->nrpages becomes zero even as it has
pages, truncate_inode_pages() returns without doing anything. Usually
this is harmless except it may cause page leak, but garbage collection
fairly infrequently sees a stale page remained in the btree node cache
of DAT (i.e. disk address translation file of nilfs), and induces the
corruption.
I identified a missing initialization in btree node caches was the
root cause. This corrects the bug.
Commit b478b782e110fdb4135caa3062b6d687e989d994 "kallsyms, tracing: output
more proper symbol name" introduces a "bugfix" that introduces a segfault
in kallsyms in my configurations.
The cause is the introduction of prefix_underscores_count() which attempts
to count underscores, even in symbols that do not have them. As a result,
it just uselessly runs past the end of the buffer until it crashes:
In theory it could happen that on one CPU we initialize a new inode but
clearing of I_NEW | I_LOCK gets reordered before some of the
initialization. Thus on another CPU we return not fully uptodate inode
from iget_locked().
This seems to fix a corruption issue on ext3 mounted over NFS.
[akpm@linux-foundation.org: add some commentary] Signed-off-by: Jan Kara <jack@suse.cz> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Similar to commit b6adc195 (PCI hotplug: acpiphp wants a 64-bit
_SUN), pci_slot.ko reads and creates sysfs directories based on
the _SUN method.
Certain HP platforms return 64 bits in _SUN. This change to
pci_slot.ko allows us to see the correct sysfs directories.
Reported-by: Chad Smith <chad.smith@hp.com> Signed-off-by: Alex Chiang <achiang@hp.com> Signed-off-by: Len Brown <len.brown@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The SLB can change sizes across a live migration, which was not
being handled, resulting in possible machine crashes during
migration if migrating to a machine which has a smaller max SLB
size than the source machine. Fix this by first reducing the
SLB size to the minimum possible value, which is 32, prior to
migration. Then during the device tree update which occurs after
migration, we make the call to ensure the SLB gets updated. Also
add the slb_size to the lparcfg output so that the migration
tools can check to make sure the kernel has this capability
before allowing migration in scenarios where the SLB size will change.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
One more form factor for Compaq Evo D510, which needs the same quirk
as the other form factors. Apparently there's no hardware monitoring
chip on that one, but SPD EEPROMs, so it's still worth unhiding the
SMBus.
ata_tf_read_block() has off-by-one error when converting CHS address
to LBA. The bug isn't very visible because ata_tf_read_block() is
used only when generating sense data for a failed RW command and CHS
addressing isn't used too often these days.
By default a block driver bounces highmem requests, but virtio-blk is
perfectly fine with any request that fit into it's 64 bit addressing scheme,
mapped in the kernel virtual space or not.
Besides improving performance on highmem systems this also makes the
reproducible oops in __bounce_end_io go away (but hiding the real cause).
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> Cc: Chuck Ebbert <cebbert@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Fix address passed to cpa_flush_range() when changing page
attributes from WB to UC. The address (*addr) is
modified by __change_page_attr_set_clr(). The result is that
the pages being flushed start at the _end_ of the changed range
instead of the beginning.
This should be considered for 2.6.30-stable and 2.6.31-stable.
Signed-off-by: Jack Steiner <steiner@sgi.com> Acked-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: H. Peter Anvin <hpa@zytor.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
In Intel Atom microarchitecture, the address generation unit
assumes that the segment base will be 0 by default. Non-zero
segment base will cause load and store operations to experience
a delay.
- If the segment base isn't aligned to a cache line
boundary, the max throughput of memory operations is
reduced to one [e]very 9 cycles.
[...]
Assembly/Compiler Coding Rule 15. (H impact, ML generality)
For Intel Atom processors, use segments with base set to 0
whenever possible; avoid non-zero segment base address that is
not aligned to cache line boundary at all cost.
We can't avoid having a non-zero base for the stack-protector
segment, but we can make it cache-aligned.
When the volume is changed continuously (e.g., when the user drags a
volume slider with the mouse), the driver does lots of I2C writes.
Apparently, the sound chip can get confused when we poll the I2C status
register too much, and fails to complete a read from it. On the PCI-E
models, the PCI-E/PCI bridge gets upset by this and generates a machine
check exception.
To avoid this, this patch replaces the polling with an unconditional
wait that is guaranteed to be long enough.
Signed-off-by: Clemens Ladisch <clemens@ladisch.de> Tested-by: Johann Messner <johann.messner at jku.at> Signed-off-by: Takashi Iwai <tiwai@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
http://bugzilla.kernel.org/show_bug.cgi?id=12542 reports that with the
quirk not applied on resume, msi stops working after resuming and mcp78s
ahci fails due to IRQ mis-delivery. Apply it on resume too.
The current implementation allocates a single host page for EQ context
memory, which was OK when we only allocated a few EQs. However, since
we now allocate an EQ for each CPU core, this patch removes the
hard-coded limit (which we exceed with 4 KB pages and 128 byte EQ
context entries with 32 CPUs) and uses the same ICM table code as all
other context tables, which ends up simplifying the code quite a bit
while fixing the problem.
This problem was actually hit in practice on a dual-socket Nehalem box
with 16 real hardware threads and sufficiently odd ACPI tables that it
shows on boot
SMP: Allowing 32 CPUs, 16 hotplug CPUs
so num_possible_cpus() ends up 32, and mlx4 ends up creating 33 MSI-X
interrupts and 33 EQs. This mlx4 bug means that mlx4 can't even
initialize at all on this quite mainstream system.
Reported-by: Eli Cohen <eli@mellanox.co.il> Tested-by: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: Roland Dreier <rolandd@cisco.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Let's suppose a highmem page is kmap'd with kmap(). A pkmap entry is
used, the page mapped to it, and the virtual cache is dirtied. Then
kunmap() is used which does virtually nothing except for decrementing a
usage count.
Then, let's suppose the _same_ page gets mapped using kmap_atomic().
It is therefore mapped onto a fixmap entry instead, which has a
different virtual address unaware of the dirty cache data for that page
sitting in the pkmap mapping.
Fortunately it is easy to know if a pkmap mapping still exists for that
page and use it directly with kmap_atomic(), thanks to kmap_high_get().
And actual testing with a printk in the added code path shows that this
condition is actually met *extremely* frequently. Seems that we've been
quite lucky that things have worked so well with highmem so far.
Signed-off-by: Nicolas Pitre <nico@marvell.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Fix minimum period size for cs46xx cards. This fixes a problem in the
case where neither a period size nor a buffer size is passed to ALSA;
this is the case in Audacious, OpenAL, and others.
Signed-off-by: Sophie Hamilton <kernel@theblob.org> Signed-off-by: Takashi Iwai <tiwai@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Some drives report 0 as the number of written blocks when there are some blocks
recorded. Use device size in such case so that we can automagically mount such
media.
Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
When probing the device in tpm_tis_init the call request_locality
uses timeout_a, which wasn't being initalized until after
request_locality. This results in request_locality falsely timing
out if the chip is still starting. Move the initialization to before
request_locality.
This probably only matters for embedded cases (ie mine), a BIOS likely
gets the TPM into a state where this code path isn't necessary.
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> Acked-by: Rajiv Andrade <srajiv@linux.vnet.ibm.com> Signed-off-by: James Morris <jmorris@namei.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
My 353d5c30c666580347515da609dd74a2b8e9b828 "mm: fix hugetlb bug due to
user_shm_unlock call" broke the CONFIG_SYSVIPC !CONFIG_MMU build of both
2.6.31 and 2.6.30.6: "undefined reference to `user_shm_unlock'".
gcc didn't understand my comment! so couldn't figure out to optimize
away user_shm_unlock() from the error path in the hugetlb-less case, as
it does elsewhere. Help it to do so, in a language it understands.
The find_ie() function uses a size_t for the len parameter, and
directly uses len as a loop variable. If any received packets
are malformed, it is possible for the decrease of len to overflow,
and since the result is unsigned, the loop will not terminate.
Change it to a signed int so the loop conditional works for
negative values.
Signed-off-by: Bob Copeland <me@bobcopeland.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
After applying the patch:
$ ./hello
Trace trap # user-mode execution after execve() finishes
If the ELF headers are actually self-inconsistent, then dying is fine.
But having no PROT_WRITE segment is perfectly normal and correct if
there is no segment with p_memsz > p_filesz (i.e. bss). John Reiser
suggested checking for PROT_WRITE in the bss logic. I think it makes
most sense to simply apply the bss logic only when there is bss.
This patch looks less trivial than it is due to some reindentation.
It just moves the "if (last_bss > elf_bss) {" test up to include the
partial-page bss logic as well as the more-pages bss logic.
Reported-by: John Reiser <jreiser@bitwagon.com> Signed-off-by: Roland McGrath <roland@redhat.com> Signed-off-by: James Morris <jmorris@namei.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
"Ath5k: unify resets"
introduced a regression into 2.6.28 where the PCU registers are never
initialized, due to ath5k_reset() always passing true for change_channel.
We subsequently program a lot of these registers but several may start
in an unknown state.
Reported-by: Forrest Zhang <forrest@hifulltech.com> Signed-off-by: Bob Copeland <me@bobcopeland.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
As early pci resume has already restored config for host
bridge and graphics device, don't need to restore it again,
This removes an original order hack for graphics device restore.
This fixed the resume hang issue found by Alan Stern on 845G,
caused by extra config restore on graphics device.
Cc: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com> Signed-off-by: Dave Airlie <airlied@linux.ie> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
When the allocation fails in sg_build_indirect(), an oops happens in
the error path. It's caused by an obvious typo.
Signed-off-by: Michal Schmidt <mschmidt@redhat.com> Reported-by: Bob Tracy <rct@gherkin.frus.com> Acked-by: Douglas Gilbert <dgilbert@interlog.com> Signed-off-by: James Bottomley <James.Bottomley@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The KEY_MAX change in 2.6.28 changed the values of the JSIOCSBTNMAP and
JSIOCGBTNMAP constants; software compiled with the old values no longer
works with kernels following 2.6.28, because the ioctl switch statement
no longer matches the values given by the software. This patch handles
these ioctls independently of the length of data specified, and applies the
same treatment to JSIOCSAXMAP and JSIOCGAXMAP which currently depend on
ABS_MAX.
Fix some problems seen in the chunk size processing when activating a
pre-existing snapshot.
For a new snapshot, the chunk size can either be supplied by the creator
or a default value can be used. For an existing snapshot, the
chunk size in the snapshot header on disk should always be used.
If someone attempts to load an existing snapshot and has the 'default
chunk size' option set, the kernel uses its default value even when it
is incorrect for the snapshot being loaded. This patch ensures the
correct on-disk value is always used.
Secondly, when the code does use the chunk size stored on the disk it is
prudent to revalidate it, so the code can exit cleanly if it got
corrupted as happened in
https://bugzilla.redhat.com/show_bug.cgi?id=461506 .
If a persistent snapshot fills up, a race can corrupt the on-disk header
which causes a crash on any future attempt to activate the snapshot
(typically while booting). This patch fixes the race.
When the snapshot overflows, __invalidate_snapshot is called, which calls
snapshot store method drop_snapshot. It goes to persistent_drop_snapshot that
calls write_header. write_header constructs the new header in the "area"
location.
Concurrently, an existing kcopyd job may finish, call copy_callback
and commit_exception method, that goes to persistent_commit_exception.
persistent_commit_exception doesn't do locking, relying on the fact that
callbacks are single-threaded, but it can race with snapshot invalidation and
overwrite the header that is just being written while the snapshot is being
invalidated.
The result of this race is a corrupted header being written that can
lead to a crash on further reactivation (if chunk_size is zero in the
corrupted header).
The fix is to use separate memory areas for each.
See the bug: https://bugzilla.redhat.com/show_bug.cgi?id=461506
This patch fixes a bug which was triggering a case where the primary leg
could not be changed on failure even when the mirror was in-sync.
The case involves the failure of the primary device along with
the transient failure of the log device. The problem is that
bios can be put on the 'failures' list (due to log failure)
before 'fail_mirror' is called due to the primary device failure.
Normally, this is fine, but if the log device failure is transient,
a subsequent iteration of the work thread, 'do_mirror', will
reset 'log_failure'. The 'do_failures' function then resets
the 'in_sync' variable when processing bios on the failures list.
The 'in_sync' variable is what is used to determine if the
primary device can be switched in the event of a failure. Since
this has been reset, the primary device is incorrectly assumed
to be not switchable.
The case has been seen in the cluster mirror context, where one
machine realizes the log device is dead before the other machines.
As the responsibilities of the server migrate from one node to
another (because the mirror is being reconfigured due to the failure),
the new server may think for a moment that the log device is fine -
thus resetting the 'log_failure' variable.
In any case, it is inappropiate for us to reset the 'log_failure'
variable. The above bug simply illustrates that it can actually
hurt us.
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Do not forget to program the MCLK ratio for the I2S output.
Otherwise, the master clock frequency can be too high for
the DACs at sample frequencies above 96 kHz.
The card model detection code introduced in 2.6.30 that tries to work
around partially broken EEPROM contents by reading the EEPROM directly
does not handle cards where the EEPROM has been omitted. In this case,
we have to use the default ID to allow the driver to load.
The async split up of probing in sd.c created a potential failure case where
something goes wrong with device_add(), but which we don't recover properly.
Since, in general, asynchronous error handling is hard, move the device_add()
into the asynchronous path (it should be fast) and make sure all the deferred
processing cannot fail.
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
An SR-IOV capable device includes an SR-IOV PCIe capability which
describes the Virtual Function (VF) BAR requirements. A typical SR-IOV
device can support multiple VFs whose BARs must be in a contiguous region,
effectively an array of VF BARs. The BAR reports the size requirement
for a single VF. We calculate the full range needed by simply multiplying
the VF BAR size with the number of possible VFs and create a resource
spanning the full range.
This all seems sane enough except it artificially inflates the alignment
requirement for the VF BAR. The VF BAR need only be aligned to the size
of a single BAR not the contiguous range of VF BARs. This can cause us
to fail to allocate resources for the BAR despite the fact that we
actually have enough space.
This patch adds a thin PCI specific layer over the generic
resource_alignment() function which is aware of the special nature of
VF BARs and does sorting and allocation based on the smaller alignment
requirement.
I recognize that while resource_alignment is generic, it's basically a
PCI helper. An alternative to this patch is to add PCI VF BAR specific
information to struct resource. I opted for the extra layer rather than
adding such PCI specific information to struct resource. This does
have the slight downside that we don't cache the BAR size and re-read
for each alignment query (happens a small handful of times during boot
for each VF BAR).
Signed-off-by: Chris Wright <chrisw@sous-sol.org> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matthew Wilcox <matthew@wil.cx> Cc: Yu Zhao <yu.zhao@intel.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This will fix the following preempt count underflow reported from
users with the title "[NILFS users] segctord problem" (Message-ID:
<949415.6494.qm@web58808.mail.re1.yahoo.com> and Message-ID:
<debc30fc0908270825v747c1734xa59126623cfd5b05@mail.gmail.com>):
kmem_cache_destroy() should call rcu_barrier() *after* kmem_cache_close() and
*before* sysfs_slab_remove() or risk rcu_free_slab() being called after
kmem_cache is deleted (kfreed).
rmmod nf_conntrack can crash the machine because it has to kmem_cache_destroy()
a SLAB_DESTROY_BY_RCU enabled cache.
Reported-by: Zdenek Kabelac <zdenek.kabelac@gmail.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The function jffs2_nor_wbuf_flash_setup() doesn't allocate the verify buffer
if CONFIG_JFFS2_FS_WBUF_VERIFY is defined, so causing a kernel panic when
that macro is enabled and the verify function is called. Similarly the
jffs2_nor_wbuf_flash_cleanup() must free the buffer if
CONFIG_JFFS2_FS_WBUF_VERIFY is enabled.
The following patch fixes the problem.
The following patch applies to 2.6.30 kernel.
I think arch/sparc/kernel/sys32.S has an incorrect splice definition:
SIGN2(sys32_splice, sys_splice, %o0, %o1)
The splice() prototype looks like :
long splice(int fd_in, loff_t *off_in, int fd_out,
loff_t *off_out, size_t len, unsigned int flags);
So I think we should have :
SIGN2(sys32_splice, sys_splice, %o0, %o2)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Functions invoked early when booting up a cpu can't use
tracing because mcount requires a valid 'current_thread_info()'
and TLB mappings to be setup.
The code path of sun4v_register_mondo_queues --> register_one_mondo
is one such case. sun4v_register_mondo_queues already has the
necessary 'notrace' annotation, but register_one_mondo does not.
Normally register_one_mondo is inlined so the bug doesn't trigger,
but with some config/compiler combinations, it won't be so we
must properly mark it notrace.
While we're here, add 'notrace' annoations to prom_printf and
prom_halt so that early error handling won't have the same problem.
Reported-by: Alexander Beregalov <a.beregalov@gmail.com> Reported-by: Leif Sawyer <lsawyer@gci.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
When page alloc debugging is not enabled, we essentially accept any
virtual address for linear kernel TLB misses. But with kgdb, kernel
address probing, and other facilities we can try to access arbitrary
crap.
So, make sure the address we miss on will translate to physical memory
that actually exists.
In order to make this work we have to embed the valid address bitmap
into the kernel image. And in order to make that less expensive we
make an adjustment, in that the max physical memory address is
decreased to "1 << 41", even on the chips that support a 42-bit
physical address space. We can do this because bit 41 indicates
"I/O space" and thus covers non-memory ranges.
The result of this is that:
1) kpte_linear_bitmap shrinks from 2K to 1K in size
2) we need 64K more for the valid address bitmap
We can't let the valid address bitmap be dynamically allocated
once we start using it to validate TLB misses, otherwise we have
crazy issues to deal with wrt. recursive TLB misses and such.
If we're in a TLB miss it could be the deepest trap level that's legal
inside of the cpu. So if we TLB miss referencing the bitmap, the cpu
will be out of trap levels and enter RED state.
To guard against out-of-range accesses to the bitmap, we have to check
to make sure no bits in the physical address above bit 40 are set. We
could export and use last_valid_pfn for this check, but that's just an
unnecessary extra memory reference.
On the plus side of all this, since we load all of these translations
into the special 4MB mapping TSB, and we check the TSB first for TLB
misses, there should be absolutely no real cost for these new checks
in the TLB miss path.
Reported-by: heyongli@gmail.com Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This is a compromise and a temporary workaround for bootup NMI
watchdog triggers some people see with qla2xxx devices present.
This happens when, for example:
CPU 0 is in the driver init and looping submitting mailbox commands to
load the firmware, then waiting for completion.
CPU 1 is receiving the device interrupts. CPU 1 is where the NMI
watchdog triggers.
CPU 0 is submitting mailbox commands fast enough that by the time CPU
1 returns from the device interrupt handler, a new one is pending.
This sequence runs for more than 5 seconds.
The problematic case is CPU 1's timer interrupt running when the
barrage of device interrupts begin. Then we have:
timer interrupt
return for softirq checking
pending, thus enable interrupts
qla2xxx interrupt
return
qla2xxx interrupt
return
... 5+ seconds pass
final qla2xxx interrupt for fw load
return
run timer softirq
return
At some point in the multi-second qla2xxx interrupt storm we trigger
the NMI watchdog on CPU 1 from the NMI interrupt handler.
The timer softirq, once we get back to running it, is smart enough to
run the timer work enough times to make up for the missed timer
interrupts.
However, the NMI watchdogs (both x86 and sparc) use the timer
interrupt count to notice the cpu is wedged. But in the above
scenerio we'll receive only one such timer interrupt even if we last
all the way back to running the timer softirq.
The default watchdog trigger point is only 5 seconds, which is pretty
low (the softwatchdog triggers at 60 seconds). So increase it to 30
seconds for now.
Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
memcpy() should take into account size of pointers,
not only number of pointers to copy.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Acked-by: Pavel Emelyanov <xemul@openvz.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Failure to call unregister_pernet_gen_device() can exhaust memory
if module is loaded/unloaded many times.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Acked-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch fixes the corner cases where the sum of MTU of the free
channels (adjusted for fragmentation overheads) is less than the MTU
of PPP link. There are at least 3 situations where this case might
arise:
- some of the channels are busy
- the multilink session is running in a degraded state (i.e. with less
than its full complement of active channels)
- by design, where multilink protocol is being used to artificially
increase the effective link MTU of a single link.
Without this patch, at most 1 fragment is ever sent per free channel
for a given PPP frame and any remaining part of the PPP frame that
does not fit into those fragments is silently discarded.
This patch restores the original behaviour which was broken by commit 9c705260feea6ae329bc6b6d5f6d2ef0227eda0a 'ppp:ppp_mp_explode()
redesign'. Once all 'free' channels have been given a fragment, an
additional fragment is queued to each available channel in turn, as many
times as necessary, until the entire PPP frame has been consumed.
Signed-off-by: Ben McKeegan <ben@netservers.co.uk> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The GRE header length should be subtracted when the tunnel MTU is
calculated. This just corrects for the associativity change
introduced by commit 42aa916265d740d66ac1f17290366e9494c884c2
("gre: Move MTU setting out of ipgre_tunnel_bind_dev").
Signed-off-by: Tom Goff <thomas.goff@boeing.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
E100 places it's RX packet descriptors inside skb->data and uses them
with bidirectional streaming DMA mapping. Data in descriptors is
accessed simultaneously by the chip (writing status and size when
a packet is received) and CPU (reading to check if the packet was
received). This isn't a valid usage of PCI DMA API, which requires use
of the coherent (consistent) memory for such purpose. Unfortunately e100
chips working in "simplified" RX mode have to store received data
directly after the descriptor. Fixing the driver to conform to the API
would require using unsupported "flexible" RX mode or receiving data
into a coherent memory and using CPU to copy it to network buffers.
This patch, while not yet making the driver conform to the PCI DMA API,
allows it to work correctly on X86 with swiotlb (while not breaking
other architectures).
Signed-off-by: Krzysztof Hałasa <khc@pm.waw.pl> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
percpu counter dccp_orphan_count is init in dccp_init() by
percpu_counter_init() while dccp module is loaded, but the
destroy of it is missing while dccp module is unloaded. We
can get the kernel WARNING about this. Reproduct by the
following commands:
Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com> Acked-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sllc_arphrd member of sockaddr_llc might not be changed. Zero sllc
before copying to the above layer's structure.
Signed-off-by: Jiri Slaby <jirislaby@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
atalk_getname() can leak 8 bytes of kernel memory to user
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
raw_getname() can leak 10 bytes of kernel memory to user
(two bytes hole between can_family and can_ifindex,
8 bytes at the end of sockaddr_can structure)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Acked-by: Oliver Hartkopp <oliver@hartkopp.net> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
As soon as the framebuffer is registered, our methods may be called by the
kernel. This leads to a crash as xenfb_refresh() gets called before we have
the irq.
Connect to the backend before registering our framebuffer with the kernel.
Backport done by Christian Lamparter <chunkeey@googlemail.com>
queue == __AR9170_NUM_TXQ would cause a bug on the next line.
found by Smatch ( http://repo.or.cz/w/smatch.git ).
Reported-by: Dan Carpenter <error27@gmail.com> Signed-off-by: Dan Carpenter <error27@gmail.com> Signed-off-by: Christian Lamparter <chunkeey@web.de> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch (as1273) fixes two(!) bugs introduced by the new
Clear-TT-Buffer implementation in ehci-hcd.
It is now possible for an idle QH to have some URBs on its
queue -- this will happen if a Clear-TT-Buffer is pending for
the QH's endpoint. Consequently we should not issue a warning
when someone tries to unlink an URB from an idle QH; instead
we should process the request immediately.
The refcounts for QHs could get messed up, because
submit_async() would increment the refcount when calling
qh_link_async() and qh_link_async() would then refuse to link
the QH into the schedule if a Clear-TT-Buffer was pending.
Instead we should increment the refcount only when the QH
actually is added to the schedule. The current code tries to
be clever by leaving the refcount alone if an unlink is
immediately followed by a relink; the patch changes this to an
unconditional decrement and increment (although they occur in
the opposite order).
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: David Brownell <david-b@pacbell.net> Tested-by: Manuel Lauss <manuel.lauss@gmail.com> Tested-by: Matthijs Kooijman <matthijs@stdin.nl> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch (as1256) changes ehci-hcd and all the other drivers in the
EHCI family to make use of the new clear_tt_buffer callbacks. When a
Clear-TT-Buffer request is in progress for a QH, the QH is not allowed
to be linked into the async schedule until the request is finished.
At that time, if there are any URBs queued for the QH, it is linked
into the async schedule.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch (as1255) updates the interface for calling
usb_hub_clear_tt_buffer(). Even the name of the function is changed!
When an async URB (i.e., Control or Bulk) going through a high-speed
hub to a non-high-speed device is cancelled or fails, the hub's
Transaction Translator buffer may be left busy still trying to
complete the transaction. The buffer has to be cleared; that's what
usb_hub_clear_tt_buffer() does.
It isn't safe to send any more URBs to the same endpoint until the TT
buffer is fully clear. Therefore the HCD needs to be told when the
Clear-TT-Buffer request has finished. This patch adds a callback
method to struct hc_driver for that purpose, and makes the hub driver
invoke the callback at the proper time.
The patch also changes a couple of names; "hub_tt_kevent" and
"tt.kevent" now look rather antiquated.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Commit 63d9950b08184e6531adceb65f64b429909cc101
(ipv6: Make v4-mapped bindings consistent with IPv4)
changes behavior of inet6_bind() for v4-mapped addresses so it should
behave the same way as inet_bind().
During this change setting of err to -EADDRNOTAVAIL got lost:
Oleg Nesterov [Mon, 24 Aug 2009 10:45:29 +0000 (12:45 +0200)]
kthreads: fix kthread_create() vs kthread_stop() race
The bug should be "accidently" fixed by recent changes in 2.6.31,
all kernels <= 2.6.30 need the fix. The problem was never noticed before,
it was found because it causes mysterious failures with GFS mount/umount.
Credits to Robert Peterson. He blaimed kthread.c from the very beginning.
But, despite my promise, I forgot to inspect the old implementation until
he did a lot of testing and reminded me. This led to huge delay in fixing
this bug.
kthread_stop() does put_task_struct(k) before it clears kthread_stop_info.k.
This means another kthread_create() can re-use this task_struct, but the
new kthread can still see kthread_should_stop() == T and exit even without
calling threadfn().
Reported-by: Robert Peterson <rpeterso@redhat.com> Tested-by: Robert Peterson <rpeterso@redhat.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Acked-by: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>