GitHub/moto-9609/android_kernel_motorola_exynos9610.git
15 years agox86, pat: don't use rb-tree based lookup in reserve_memtype()
Suresh Siddha [Wed, 16 Sep 2009 21:28:03 +0000 (14:28 -0700)]
x86, pat: don't use rb-tree based lookup in reserve_memtype()

Recent enhancement of rb-tree based lookup exposed a  bug with the lookup
mechanism in the reserve_memtype() which ensures that there are no conflicting
memtype requests for the memory range.

memtype_rb_search() returns an entry which has a start address <= new start
address. And from here we traverse the linear linked list to check if there
any conflicts with the existing mappings. As the rbtree is based on the
start address of the memory range, it is quite possible that we have several
overlapped mappings whose start address is much less than new requested start
but the end is >= new requested end. This results in conflicting memtype
mappings.

Same bug exists with the old code which uses cached_entry from where
we traverse the linear linked list. But the new rb-tree code exposes this
bug fairly easily.

For now, don't use the memtype_rb_search() and always start the search from
the head of linear linked list in reserve_memtype(). Linear linked list
for most of the systems grow's to few 10's of entries(as we track memory type
of RAM pages using struct page). So we should be ok for now.

We still retain the rbtree and use it to speed up free_memtype() which
doesn't have the same bug(as we know what exactly we are searching for
in free_memtype).

Also use list_for_each_entry_from() in free_memtype() so that we start
the search from rb-tree lookup result.

Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
LKML-Reference: <1253136483.4119.12.camel@sbs-t61.sc.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Fix cacheflush address in change_page_attr_set_clr()
Jack Steiner [Thu, 3 Sep 2009 17:56:02 +0000 (12:56 -0500)]
x86, pat: Fix cacheflush address in change_page_attr_set_clr()

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>
Cc: Stable team <stable@kernel.org>
15 years agomm: remove !NUMA condition from PAGEFLAGS_EXTENDED condition set
H. Peter Anvin [Mon, 31 Aug 2009 18:17:44 +0000 (11:17 -0700)]
mm: remove !NUMA condition from PAGEFLAGS_EXTENDED condition set

CONFIG_PAGEFLAGS_EXTENDED disables a trick to conserve pageflags.
This trick is indended to be enabled when the pressure on page flags
is very high.

The previous condition was:

-       depends on 64BIT || SPARSEMEM_VMEMMAP || !NUMA || !SPARSEMEM

... however, the sparsemem code already has a way to crowd out the
node number from the pageflags, which means that !NUMA actually
doesn't contribute to hard pageflags exhaustion.

This is required for the new PG_uncached flag to not cause pageflags
exhaustion on x86_32 + PAE + SPARSEMEM + !NUMA.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
LKML-Reference: <4A9828F4.4040905@zytor.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Suresh Siddha <suresh.siddha@intel.com>
15 years agox86: Fix earlyprintk=dbgp for machines without NX
Jan Beulich [Fri, 28 Aug 2009 13:11:57 +0000 (14:11 +0100)]
x86: Fix earlyprintk=dbgp for machines without NX

Since parse_early_param() may (e.g. for earlyprintk=dbgp)
involve calls to page table manipulation functions (here
set_fixmap_nocache()), NX hardware support must be determined
before calling that function (so that __supported_pte_mask gets
properly set up).

But the call after parse_early_param() can also not go away, as
that will honor eventual command line specified disabling of
the NX functionality.

( This will then just result in whatever mappings got
  established during parse_early_param() having the NX bit set
  despite it being disabled on the command line, but I think
  that's tolerable).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
LKML-Reference: <4A97F3BD02000078000121B9@vpn.id2.novell.com>
[ merged to x86/pat to resolve a conflict. ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoMerge branch 'x86/urgent' into x86/pat
H. Peter Anvin [Thu, 27 Aug 2009 00:17:51 +0000 (17:17 -0700)]
Merge branch 'x86/urgent' into x86/pat

Reason: Change to is_new_memtype_allowed() in x86/urgent

Resolved semantic conflicts in:

 arch/x86/mm/pat.c
 arch/x86/mm/ioremap.c

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Sanity check remap_pfn_range for RAM region
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:41 +0000 (09:57 -0700)]
x86, pat: Sanity check remap_pfn_range for RAM region

Add sanity check for remap_pfn_range of RAM regions using
lookup_memtype(). Previously, we did not have anyway to get the type of
RAM memory regions as they were tracked using a single bit in
page_struct (WB, nonWB). Now we can get the actual type from page struct
(WB, WC, UC_MINUS) and make sure the requester gets that type.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Lookup the protection from memtype list on vm_insert_pfn()
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:40 +0000 (09:57 -0700)]
x86, pat: Lookup the protection from memtype list on vm_insert_pfn()

Lookup the reserved memtype during vm_insert_pfn and use that memtype
for the new mapping. This takes care or handling of vm_insert_pfn()
interface in track_pfn_vma*/untrack_pfn_vma.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Add lookup_memtype to get the current memtype of a paddr
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:39 +0000 (09:57 -0700)]
x86, pat: Add lookup_memtype to get the current memtype of a paddr

Add a new routine lookup_memtype() to get the current memtype based on
the PAT reserves and frees.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Use page flags to track memtypes of RAM pages
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:38 +0000 (09:57 -0700)]
x86, pat: Use page flags to track memtypes of RAM pages

Change reserve_ram_pages_type and free_ram_pages_type to use 2 page
flags to track UC_MINUS, WC, WB and default types. Previous RAM tracking
just tracked WB or NonWB, which was not complete and did not allow
tracking of RAM fully and there was no way to get the actual type
reserved by looking at the page flags.

We use the memtype_lock spinlock for atomicity in dealing with
memtype tracking in struct page.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Generalize the use of page flag PG_uncached
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:37 +0000 (09:57 -0700)]
x86, pat: Generalize the use of page flag PG_uncached

Only IA64 was using PG_uncached as of now. We now intend to use this bit
in x86 as well, to keep track of memory type of those addresses that
have page struct for them. So, generalize the use of that bit across
ia64 and x86.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Add rbtree to do quick lookup in memtype tracking
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:36 +0000 (09:57 -0700)]
x86, pat: Add rbtree to do quick lookup in memtype tracking

PAT memtype tracking uses a linear link list to keep track of IO
(non-RAM) regions and their memtypes. The code used a last_accessed
pointer as a cache to speedup the lookup. As per discussions with
H. Peter Anvin a while back, having a rbtree here will avoid bad
performances in pathological cases where we may end up with huge
linked list. This may not add any noticable performance speedup
in normal case as the number of entires in PAT memtype list tend
to be ~20-30 range. The patch removes the "cached_entry" logic
as with rbtree we have more generic way of speeding up the lookup.

With this patch, we use rbtree to do the quick lookup. We still use
linked list as the memtype range tracked can be of different sizes
and can overlap in different ways. We also keep track of usage counts
with linked list.

Example:
Multiple ioremaps with different sizes
uncached-minus @ 0xfffff00000-0xfffff04000
uncached-minus @ 0xfffff02000-0xfffff03000

And one userlevel mmap and the thread forks a new process
uncached-minus @ 0xbf453000-0xbf454000
uncached-minus @ 0xbf453000-0xbf454000

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Add PAT reserve free to io_mapping* APIs
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:35 +0000 (09:57 -0700)]
x86, pat: Add PAT reserve free to io_mapping* APIs

io_mapping_* interfaces were added, mainly for graphics drivers.
Make this interface go through the PAT reserve/free, instead of
hardcoding WC mapping. This makes sure that there are no
aliases due to unconditional WC setting.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: New i/f for driver to request memtype for IO regions
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:34 +0000 (09:57 -0700)]
x86, pat: New i/f for driver to request memtype for IO regions

Add new routines to request memtype for IO regions. This will currently
be a backend for io_mapping_* routines. But, it can also be made available
to drivers directly in future, in case it is needed.

reserve interface reserves the memory, makes sure we have a compatible
memory type available and keeps the identity map in sync when needed.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: ioremap to follow same PAT restrictions as other PAT users
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:33 +0000 (09:57 -0700)]
x86, pat: ioremap to follow same PAT restrictions as other PAT users

ioremap has this hard-coded check for new type and requested type. That
check differs from other PAT users like /dev/mem mmap, remap_pfn_range
in only one condition where requested type is UC_MINUS and new type
is WC. Under that condition, ioremap fails. But other PAT interfaces succeed
with a WC mapping.

Change to make ioremap be in sync with other PAT APIs and use the same
macro as others. Also changes the error print to KERN_ERR instead of
pr_debug.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, pat: Keep identity maps consistent with mmaps even when pat_disabled
Venkatesh Pallipadi [Fri, 10 Jul 2009 16:57:32 +0000 (09:57 -0700)]
x86, pat: Keep identity maps consistent with mmaps even when pat_disabled

Make reserve_memtype internally take care of pat disabled case and fallback
to default return values.

Remove the specific pat_disabled checks in track_* routines.

Change kernel_map_sync_memtype to sync identity map even when
pat_disabled.

This change ensures that, even for pat_disabled case, we take care of
keeping identity map in sync. Before this patch, in pat disabled case,
ioremap() keeps the identity maps in sync and other APIs like pci and
/dev/mem mmap don't, which is not a very consistent behavior.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, mtrr: make mtrr_aps_delayed_init static bool
H. Peter Anvin [Sat, 22 Aug 2009 00:00:02 +0000 (17:00 -0700)]
x86, mtrr: make mtrr_aps_delayed_init static bool

mtr_aps_delayed_init was declared u32 and made global, but it only
ever takes boolean values and is only ever used in
arch/x86/kernel/cpu/mtrr/main.c.  Declare it "static bool" and remove
external references.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
15 years agox86, pat/mtrr: Rendezvous all the cpus for MTRR/PAT init
Suresh Siddha [Thu, 20 Aug 2009 01:05:36 +0000 (18:05 -0700)]
x86, pat/mtrr: Rendezvous all the cpus for MTRR/PAT init

SDM Vol 3a section titled "MTRR considerations in MP systems" specifies
the need for synchronizing the logical cpu's while initializing/updating
MTRR.

Currently Linux kernel does the synchronization of all cpu's only when
a single MTRR register is programmed/updated. During an AP online
(during boot/cpu-online/resume)  where we initialize all the MTRR/PAT registers,
we don't follow this synchronization algorithm.

This can lead to scenarios where during a dynamic cpu online, that logical cpu
is initializing MTRR/PAT with cache disabled (cr0.cd=1) etc while other logical
HT sibling continue to run (also with cache disabled because of cr0.cd=1
on its sibling).

Starting from Westmere, VMX transitions with cr0.cd=1 don't work properly
(because of some VMX performance optimizations) and the above scenario
(with one logical cpu doing VMX activity and another logical cpu coming online)
can result in system crash.

Fix the MTRR initialization by doing rendezvous of all the cpus. During
boot and resume, we delay the MTRR/PAT init for APs till all the
logical cpu's come online and the rendezvous process at the end of AP's bringup,
will initialize the MTRR/PAT for all AP's.

For dynamic single cpu online, we synchronize all the logical cpus and
do the MTRR/PAT init on the AP that is coming online.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agogeneric-ipi: Allow cpus not yet online to call smp_call_function with irqs disabled
Suresh Siddha [Thu, 20 Aug 2009 01:05:35 +0000 (18:05 -0700)]
generic-ipi: Allow cpus not yet online to call smp_call_function with irqs disabled

Because of deadlock possiblities smp_call_function() is not allowed to
be called with interrupts disabled. Add an exception for the cpu not
yet online, as no one else can send smp call function interrupt to this
cpu that is not yet online and as such deadlock condition is not possible.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86: Fix an incorrect argument of reserve_bootmem()
Amerigo Wang [Fri, 21 Aug 2009 08:34:45 +0000 (04:34 -0400)]
x86: Fix an incorrect argument of reserve_bootmem()

This line looks suspicious, because if this is true, then the
'flags' parameter of function reserve_bootmem_generic() will be
unused when !CONFIG_NUMA. I don't think this is what we want.

Signed-off-by: WANG Cong <amwang@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: akpm@linux-foundation.org
LKML-Reference: <20090821083709.5098.52505.sendpatchset@localhost.localdomain>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: Fix system crash when loading with "reservetop" parameter
Xiao Guangrong [Thu, 20 Aug 2009 12:23:11 +0000 (20:23 +0800)]
x86: Fix system crash when loading with "reservetop" parameter

The system will die if the kernel is booted with "reservetop"
parameter, in present code, parse "reservetop" parameter after
early_ioremap_init(), and some function still use
early_ioremap() after it.

The problem is, "reservetop" parameter can modify
'FIXADDR_TOP', then the virtual address got by early_ioremap()
is base on old 'FIXADDR_TOP', but the page mapping is base on
new 'FIXADDR_TOP', it will occur page fault, and the IDT is not
prepare yet, so, the system is dead.

So, put parse_early_param() in the front of
early_ioremap_init() in this patch.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: yinghai@kernel.org
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <4A8D402F.4080805@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: add vmlinux.lds to targets in arch/x86/boot/compressed/Makefile
Jan Beulich [Thu, 20 Aug 2009 15:14:15 +0000 (16:14 +0100)]
x86: add vmlinux.lds to targets in arch/x86/boot/compressed/Makefile

The absence of vmlinux.lds here keeps .vmlinux.lds.cmd from being
included, which in turn leads to it and all its dependents always
getting rebuilt independent of whether they are already up-to-date.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
LKML-Reference: <4A8D84670200007800010D31@vpn.id2.novell.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agoMerge branch 'bugfix' of git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen...
Ingo Molnar [Thu, 20 Aug 2009 10:05:24 +0000 (12:05 +0200)]
Merge branch 'bugfix' of git://git./linux/kernel/git/jeremy/xen into x86/urgent

15 years agoxen: rearrange things to fix stackprotector
Jeremy Fitzhardinge [Mon, 17 Aug 2009 19:26:53 +0000 (12:26 -0700)]
xen: rearrange things to fix stackprotector

Make sure the stack-protector segment registers are properly set up
before calling any functions which may have stack-protection compiled
into them.

[ Impact: prevent Xen early-boot crash when stack-protector is enabled ]
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
15 years agox86: make sure load_percpu_segment has no stackprotector
Jeremy Fitzhardinge [Mon, 17 Aug 2009 19:25:41 +0000 (12:25 -0700)]
x86: make sure load_percpu_segment has no stackprotector

load_percpu_segment() is used to set up the per-cpu segment registers,
which are also used for -fstack-protector.  Make sure that the
load_percpu_segment() function doesn't have stackprotector enabled.

[ Impact: allow percpu setup before calling stack-protected functions ]

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
15 years agoi386: Fix section mismatches for init code with !HOTPLUG_CPU
Jan Beulich [Tue, 18 Aug 2009 15:41:33 +0000 (16:41 +0100)]
i386: Fix section mismatches for init code with !HOTPLUG_CPU

Commit 0e83815be719d3391bf5ea24b7fe696c07dbd417 changed the
section the initial_code variable gets allocated in, in an
attempt to address a section conflict warning. This, however
created a new section conflict when building without
HOTPLUG_CPU. The apparently only (reasonable) way to address
this is to always use __REFDATA.

Once at it, also fix a second section mismatch when not using
HOTPLUG_CPU.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: Robert Richter <robert.richter@amd.com>
LKML-Reference: <4A8AE7CD020000780001054B@vpn.id2.novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, pat: Allow ISA memory range uncacheable mapping requests
Suresh Siddha [Mon, 17 Aug 2009 20:23:50 +0000 (13:23 -0700)]
x86, pat: Allow ISA memory range uncacheable mapping requests

Max Vozeler reported:
>  Bug 13877 -  bogl-term broken with CONFIG_X86_PAT=y, works with =n
>
>  strace of bogl-term:
>  814   mmap2(NULL, 65536, PROT_READ|PROT_WRITE, MAP_SHARED, 4, 0)
>  = -1 EAGAIN (Resource temporarily unavailable)
>  814   write(2, "bogl: mmaping /dev/fb0: Resource temporarily unavailable\n",
>        57) = 57

PAT code maps the ISA memory range as WB in the PAT attribute, so that
fixed range MTRR registers define the actual memory type (UC/WC/WT etc).

But the upper level is_new_memtype_allowed() API checks are failing,
as the request here is for UC and the return tracked type is WB (Tracked type is
WB as MTRR type for this legacy range potentially will be different for each
4k page).

Fix is_new_memtype_allowed() by always succeeding the ISA address range
checks, as the null PAT (WB) and def MTRR fixed range register settings
satisfy the memory type needs of the applications that map the ISA address
range.

Reported-and-Tested-by: Max Vozeler <xam@debian.org>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
15 years agox86, mce: Don't initialize MCEs on unknown CPUs
Ingo Molnar [Mon, 17 Aug 2009 08:19:00 +0000 (10:19 +0200)]
x86, mce: Don't initialize MCEs on unknown CPUs

An older test-box started hanging at the following point during
bootup:

 [    0.022996] Mount-cache hash table entries: 512
 [    0.024996] Initializing cgroup subsys debug
 [    0.025996] Initializing cgroup subsys cpuacct
 [    0.026995] Initializing cgroup subsys devices
 [    0.027995] Initializing cgroup subsys freezer
 [    0.028995] mce: CPU supports 5 MCE banks

I've bisected it down to commit 4efc0670 ("x86, mce: use 64bit
machine check code on 32bit"), which utilizes the MCE code on
32-bit systems too.

The problem is caused by this detail in my config:

  # CONFIG_CPU_SUP_INTEL is not set

This disables the quirks in mce_cpu_quirks() but still enables
MCE support - which then hangs due to the missing quirk
workaround needed on this CPU:

if (c->x86 == 6 && c->x86_model < 0x1A && banks > 0)
mce_banks[0].init = 0;

The safe solution is to not initialize MCEs if we dont know on
what CPU we are running (or if that CPU's support code got
disabled in the config).

Also be a bit more defensive on 32-bit systems: dont do a
boot-time dump of pending MCEs not just on the specific system
that we found a problem with (Pentium-M), but earlier ones as
well.

Now this problem is probably not common and disabling CPU
support is rare - but still being more defensive in something
we turned on for a wide range of CPUs is prudent.

Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
LKML-Reference: Message-ID: <4A88E3E4.40506@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, mce: don't log boot MCEs on Pentium M (model == 13) CPUs
Bartlomiej Zolnierkiewicz [Tue, 28 Jul 2009 21:52:54 +0000 (23:52 +0200)]
x86, mce: don't log boot MCEs on Pentium M (model == 13) CPUs

On my legacy Pentium M laptop (Acer Extensa 2900) I get bogus MCE on a cold
boot with CONFIG_X86_NEW_MCE enabled, i.e. (after decoding it with mcelog):

MCE 0
HARDWARE ERROR. This is *NOT* a software problem!
Please contact your hardware vendor
CPU 0 BANK 1 MCG status:
MCi status:
Error overflow
Uncorrected error
Error enabled
Processor context corrupt
MCA: Data CACHE Level-1 UNKNOWN Error
STATUS f200000000000195 MCGSTATUS 0

[ The other STATUS values observed: f2000000000001b5 (... UNKNOWN error)
  and f200000000000115 (... READ Error).

  To verify that this is not a CONFIG_X86_NEW_MCE bug I also modified
  the CONFIG_X86_OLD_MCE code (which doesn't log any MCEs) to dump
  content of STATUS MSR before it is cleared during initialization. ]

Since the bogus MCE results in a kernel taint (which in turn disables
lockdep support) don't log boot MCEs on Pentium M (model == 13) CPUs
by default ("mce=bootlog" boot parameter can be be used to get the old
behavior).

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Reviewed-by: Andi Kleen <andi@firstfloor.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: Annotate section mismatch warnings in kernel/apic/x2apic_uv_x.c
Leonardo Potenza [Sun, 16 Aug 2009 16:55:48 +0000 (18:55 +0200)]
x86: Annotate section mismatch warnings in kernel/apic/x2apic_uv_x.c

The function uv_acpi_madt_oem_check() has been marked __init,
the struct apic_x2apic_uv_x has been marked __refdata.

The aim is to address the following section mismatch messages:

WARNING: arch/x86/kernel/apic/built-in.o(.data+0x1368): Section mismatch in reference from the variable apic_x2apic_uv_x to the function .cpuinit.text:uv_wakeup_secondary()
The variable apic_x2apic_uv_x references
the function __cpuinit uv_wakeup_secondary()
If the reference is valid then annotate the
variable with __init* or __refdata (see linux/init.h) or name the variable:
*driver, *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console,

WARNING: arch/x86/kernel/built-in.o(.data+0x68e8): Section mismatch in reference from the variable apic_x2apic_uv_x to the function .cpuinit.text:uv_wakeup_secondary()
The variable apic_x2apic_uv_x references
the function __cpuinit uv_wakeup_secondary()
If the reference is valid then annotate the
variable with __init* or __refdata (see linux/init.h) or name the variable:
*driver, *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console,

WARNING: arch/x86/built-in.o(.text+0x7b36f): Section mismatch in reference from the function uv_acpi_madt_oem_check() to the function .init.text:early_ioremap()
The function uv_acpi_madt_oem_check() references
the function __init early_ioremap().
This is often because uv_acpi_madt_oem_check lacks a __init
annotation or the annotation of early_ioremap is wrong.

WARNING: arch/x86/built-in.o(.text+0x7b38d): Section mismatch in reference from the function uv_acpi_madt_oem_check() to the function .init.text:early_iounmap()
The function uv_acpi_madt_oem_check() references
the function __init early_iounmap().
This is often because uv_acpi_madt_oem_check lacks a __init
annotation or the annotation of early_iounmap is wrong.

WARNING: arch/x86/built-in.o(.data+0x8668): Section mismatch in reference from the variable apic_x2apic_uv_x to the function .cpuinit.text:uv_wakeup_secondary()
The variable apic_x2apic_uv_x references
the function __cpuinit uv_wakeup_secondary()
If the reference is valid then annotate the
variable with __init* or __refdata (see linux/init.h) or name the variable:
*driver, *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console,

Signed-off-by: Leonardo Potenza <lpotenza@inwind.it>
LKML-Reference: <200908161855.48302.lpotenza@inwind.it>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, mce: therm_throt: Don't log redundant normality
Hugh Dickins [Sun, 16 Aug 2009 14:54:37 +0000 (15:54 +0100)]
x86, mce: therm_throt: Don't log redundant normality

0d01f31439c1e4d602bf9fdc924ab66f407f5e38 "x86, mce: therm_throt
- change when we print messages" removed redundant
announcements of "Temperature/speed normal".

They're not worth logging and remove their accompanying
"Machine check events logged" messages as well from the
console.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Dmitry Torokhov <dtor@mail.ru>
LKML-Reference: <Pine.LNX.4.64.0908161544100.7929@sister.anvils>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: Fix UV BAU destination subnode id
Cliff Wickman [Fri, 14 Aug 2009 18:56:37 +0000 (13:56 -0500)]
x86: Fix UV BAU destination subnode id

The SGI UV Broadcast Assist Unit is used to send TLB shootdown
messages to remote nodes of the system.  The header of the
message must contain the subnode id of the block in the
receiving hub that handles such messages.  It should always be
0x10, the id of the "LB" block.

It had previously been documented as a "must be zero" field.

Signed-off-by: Cliff Wickman <cpw@sgi.com>
Acked-by: Jack Steiner <steiner@sgi.com>
LKML-Reference: <E1Mc1x7-0005Ce-6t@eag09.americas.sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoLinux 2.6.31-rc6
Linus Torvalds [Thu, 13 Aug 2009 22:43:34 +0000 (15:43 -0700)]
Linux 2.6.31-rc6

15 years agogenirq: prevent wakeup of freed irq thread
Linus Torvalds [Thu, 13 Aug 2009 20:05:10 +0000 (13:05 -0700)]
genirq: prevent wakeup of freed irq thread

free_irq() can remove an irqaction while the corresponding interrupt
is in progress, but free_irq() sets action->thread to NULL
unconditionally, which might lead to a NULL pointer dereference in
handle_IRQ_event() when the hard interrupt context tries to wake up
the handler thread.

Prevent this by moving the thread stop after synchronize_irq(). No
need to set action->thread to NULL either as action is going to be
freed anyway.

This fixes a boot crash reported against preempt-rt which uses the
mainline irq threads code to implement full irq threading.

[ tglx: removed local irqthread variable ]

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
15 years agoMerge branch 'perfcounters-fixes-for-linus' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Thu, 13 Aug 2009 19:24:33 +0000 (12:24 -0700)]
Merge branch 'perfcounters-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'perfcounters-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  perf_counter: Report the cloning task as parent on perf_counter_fork()
  perf_counter: Fix an ipi-deadlock
  perf: Rework/fix the whole read vs group stuff
  perf_counter: Fix swcounter context invariance
  perf report: Don't show unresolved DSOs and symbols when -S/-d is used
  perf tools: Add a general option to enable raw sample records
  perf tools: Add a per tracepoint counter attribute to get raw sample
  perf_counter: Provide hw_perf_counter_setup_online() APIs
  perf list: Fix large list output by using the pager
  perf_counter, x86: Fix/improve apic fallback
  perf record: Add missing -C option support for specifying profile cpu
  perf tools: Fix dso__new handle() to handle deleted DSOs
  perf tools: Fix fallback to cplus_demangle() when bfd_demangle() is not available
  perf report: Show the tid too in -D
  perf record: Fix .tid and .pid fill-in when synthesizing events
  perf_counter, x86: Fix generic cache events on P6-mobile CPUs
  perf_counter, x86: Fix lapic printk message

15 years agoMerge branch 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 13 Aug 2009 19:09:16 +0000 (12:09 -0700)]
Merge branch 'core-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  futex: Fix handling of bad requeue syscall pairing
  futex: Fix compat_futex to be same as futex for REQUEUE_PI
  locking, sched: Give waitqueue spinlocks their own lockdep classes
  futex: Update futex_q lock_ptr on requeue proxy lock

15 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 13 Aug 2009 19:08:44 +0000 (12:08 -0700)]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86: Fix oops in identify_cpu() on CPUs without CPUID
  x86: Clear incorrectly forced X86_FEATURE_LAHF_LM flag
  x86, mce: therm_throt - change when we print messages
  x86: Add reboot quirk for every 5 series MacBook/Pro

15 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec...
Linus Torvalds [Thu, 13 Aug 2009 18:17:40 +0000 (11:17 -0700)]
Merge branch 'upstream-linus' of git://git./linux/kernel/git/jlbec/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec/ocfs2: (22 commits)
  ocfs2: Fix possible deadlock when extending quota file
  ocfs2: keep index within status_map[]
  ocfs2: Initialize the cluster we're writing to in a non-sparse extend
  ocfs2: Remove redundant BUG_ON in __dlm_queue_ast()
  ocfs2/quota: Release lock for error in ocfs2_quota_write.
  ocfs2: Define credit counts for quota operations
  ocfs2: Remove syncjiff field from quota info
  ocfs2: Fix initialization of blockcheck stats
  ocfs2: Zero out padding of on disk dquot structure
  ocfs2: Initialize blocks allocated to local quota file
  ocfs2: Mark buffer uptodate before calling ocfs2_journal_access_dq()
  ocfs2: Make global quota files blocksize aligned
  ocfs2: Use ocfs2_rec_clusters in ocfs2_adjust_adjacent_records.
  ocfs2: Fix deadlock on umount
  ocfs2: Add extra credits and access the modified bh in update_edge_lengths.
  ocfs2: Fail ocfs2_get_block() immediately when a block needs allocation
  ocfs2: Fix error return in ocfs2_write_cluster()
  ocfs2: Fix compilation warning for fs/ocfs2/xattr.c
  ocfs2: Initialize count in aio_write before generic_write_checks
  ocfs2: log the actual return value of ocfs2_file_aio_write()
  ...

15 years agoMerge branch 'for-linus' of git://neil.brown.name/md
Linus Torvalds [Thu, 13 Aug 2009 17:59:29 +0000 (10:59 -0700)]
Merge branch 'for-linus' of git://neil.brown.name/md

* 'for-linus' of git://neil.brown.name/md:
  md: allow upper limit for resync/reshape to be set when array is read-only
  md/raid5: Properly remove excess drives after shrinking a raid5/6
  md/raid5: make sure a reshape restarts at the correct address.
  md/raid5: allow new reshape modes to be restarted in the middle.
  md: never advance 'events' counter by more than 1.
  Remove deadlock potential in md_open

15 years agoMerge branch 'sh/for-2.6.31' of git://git.kernel.org/pub/scm/linux/kernel/git/lethal...
Linus Torvalds [Thu, 13 Aug 2009 17:57:53 +0000 (10:57 -0700)]
Merge branch 'sh/for-2.6.31' of git://git./linux/kernel/git/lethal/sh-2.6

* 'sh/for-2.6.31' of git://git.kernel.org/pub/scm/linux/kernel/git/lethal/sh-2.6:
  sh: fix i2c init order on ap325rxa V2
  sh: fix i2c init order on Migo-R V2
  sh: convert processor device setup functions to arch_initcall()

15 years agoMake sock_sendpage() use kernel_sendpage()
Linus Torvalds [Thu, 13 Aug 2009 15:28:36 +0000 (08:28 -0700)]
Make sock_sendpage() use kernel_sendpage()

kernel_sendpage() does the proper default case handling for when the
socket doesn't have a native sendpage implementation.

Now, arguably this might be something that we could instead solve by
just specifying that all protocols should do it themselves at the
protocol level, but we really only care about the common protocols.
Does anybody really care about sendpage on something like Appletalk? Not
likely.

Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Julien TINNES <julien@cr0.org>
Acked-by: Tavis Ormandy <taviso@sdf.lonestar.org>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoperf_counter: Report the cloning task as parent on perf_counter_fork()
Peter Zijlstra [Thu, 13 Aug 2009 14:14:42 +0000 (16:14 +0200)]
perf_counter: Report the cloning task as parent on perf_counter_fork()

A bug in (9f498cc: perf_counter: Full task tracing) makes
profiling multi-threaded apps it go belly up.

[ output as: (PID:TID):(PPID:PTID) ]

 # ./perf report -D | grep FORK
0x4b0 [0x18]: PERF_EVENT_FORK: (3237:3237):(3236:3236)
0xa10 [0x18]: PERF_EVENT_FORK: (3237:3238):(3236:3236)
0xa70 [0x18]: PERF_EVENT_FORK: (3237:3239):(3236:3236)
0xad0 [0x18]: PERF_EVENT_FORK: (3237:3240):(3236:3236)
0xb18 [0x18]: PERF_EVENT_FORK: (3237:3241):(3236:3236)

Shows us that the test (27d028d perf report: Update for the new
FORK/EXIT events) in builtin-report.c:

        /*
         * A thread clone will have the same PID for both
         * parent and child.
         */
        if (thread == parent)
                return 0;

Will clearly fail.

The problem is that perf_counter_fork() reports the actual
parent, instead of the cloning thread.

Fixing that (with the below patch), yields:

 # ./perf report -D | grep FORK
0x4c8 [0x18]: PERF_EVENT_FORK: (1590:1590):(1589:1589)
0xbd8 [0x18]: PERF_EVENT_FORK: (1590:1591):(1590:1590)
0xc80 [0x18]: PERF_EVENT_FORK: (1590:1592):(1590:1590)
0x3338 [0x18]: PERF_EVENT_FORK: (1590:1593):(1590:1590)
0x66b0 [0x18]: PERF_EVENT_FORK: (1590:1594):(1590:1590)

Which both makes more sense and doesn't confuse perf report
anymore.

Reported-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: paulus@samba.org
Cc: Anton Blanchard <anton@samba.org>
Cc: Arjan van de Ven <arjan@infradead.org>
LKML-Reference: <1250172882.5241.62.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf_counter: Fix an ipi-deadlock
Peter Zijlstra [Thu, 13 Aug 2009 09:47:54 +0000 (11:47 +0200)]
perf_counter: Fix an ipi-deadlock

perf_pending_counter() is called from IRQ context and will call
perf_counter_disable(), however perf_counter_disable() uses
smp_call_function_single() which doesn't fancy being used with
IRQs disabled due to IPI deadlocks.

Fix this by making it use the local __perf_counter_disable()
call and teaching the counter_sched_out() code about pending
disables as well.

This should cover the case where a counter migrates before the
pending queue gets processed.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey J Ashford <cjashfor@us.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
LKML-Reference: <20090813103655.244097721@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf: Rework/fix the whole read vs group stuff
Peter Zijlstra [Thu, 13 Aug 2009 09:47:53 +0000 (11:47 +0200)]
perf: Rework/fix the whole read vs group stuff

Replace PERF_SAMPLE_GROUP with PERF_SAMPLE_READ and introduce
PERF_FORMAT_GROUP to deal with group reads in a more generic
way.

This allows you to get group reads out of read() as well.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey J Ashford <cjashfor@us.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
LKML-Reference: <20090813103655.117411814@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf_counter: Fix swcounter context invariance
Peter Zijlstra [Thu, 13 Aug 2009 07:51:55 +0000 (09:51 +0200)]
perf_counter: Fix swcounter context invariance

perf_swcounter_is_counting() uses a lock, which means we cannot
use swcounters from NMI or when holding that particular lock,
this is unintended.

The below removes the lock, this opens up race window, but not
worse than the swcounters already experience due to RCU
traversal of the context in perf_swcounter_ctx_event().

This also fixes the hard lockups while opening a lockdep
tracepoint counter.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
Cc: Corey J Ashford <cjashfor@us.ibm.com>
LKML-Reference: <1250149915.10001.66.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf report: Don't show unresolved DSOs and symbols when -S/-d is used
Arnaldo Carvalho de Melo [Wed, 12 Aug 2009 21:19:57 +0000 (18:19 -0300)]
perf report: Don't show unresolved DSOs and symbols when -S/-d is used

We're interested in just those symbols/DSOs, so filter out the
unresolved ones.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <20090812211957.GE3495@ghostprotocols.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf tools: Add a general option to enable raw sample records
Frederic Weisbecker [Thu, 13 Aug 2009 08:27:19 +0000 (10:27 +0200)]
perf tools: Add a general option to enable raw sample records

While we can enable the perf sample records per tracepoint
counter, we may also want to enable this option for every
tracepoint counters to open, so that we don't need to add a
:record flag for all of them.

Add the -R, --raw-samples options for this purpose.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1250152039-7284-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf tools: Add a per tracepoint counter attribute to get raw sample
Frederic Weisbecker [Thu, 13 Aug 2009 08:27:18 +0000 (10:27 +0200)]
perf tools: Add a per tracepoint counter attribute to get raw sample

Add a new flag field while opening a tracepoint perf counter:

-e tracepoint_subsystem:tracepoint_name:flags

This is intended to be generic although for now it only supports the
r[e[c[o[r[d]]]]] flag:

./perf record -e workqueue:workqueue_insertion:record
./perf record -e workqueue:workqueue_insertion:r

will have the same effect: enabling the raw samples record for
the given tracepoint counter.

In the future, we may want to support further flags, separated
by commas.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1250152039-7284-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf_counter: Provide hw_perf_counter_setup_online() APIs
Ingo Molnar [Thu, 13 Aug 2009 08:13:22 +0000 (10:13 +0200)]
perf_counter: Provide hw_perf_counter_setup_online() APIs

Provide weak aliases for hw_perf_counter_setup_online(). This is
used by the BTS patches (for v2.6.32), but it interacts with
fixes so propagate this upstream. (it has no effect as of yet)

Also export perf_counter_output() to architecture code.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf list: Fix large list output by using the pager
Arnaldo Carvalho de Melo [Wed, 12 Aug 2009 17:44:59 +0000 (14:44 -0300)]
perf list: Fix large list output by using the pager

When /sys/kernel/debug is mounted the list can be imense, so
use the pager like the other tools.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <20090812174459.GB3495@ghostprotocols.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agosh: fix i2c init order on ap325rxa V2
Magnus Damm [Fri, 7 Aug 2009 03:52:18 +0000 (03:52 +0000)]
sh: fix i2c init order on ap325rxa V2

Convert the AP325RXA board code to register devices at
arch_initcall() time instead of device_initcall(). This
fix unbreaks pcf8563 RTC driver support.

Signed-off-by: Magnus Damm <damm@igel.co.jp>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
15 years agosh: fix i2c init order on Migo-R V2
Magnus Damm [Thu, 13 Aug 2009 02:39:02 +0000 (11:39 +0900)]
sh: fix i2c init order on Migo-R V2

Convert the Migo-R board code to register devices at
arch_initcall() time instead of __initcall(). This fix
unbreaks migor_ts touch screen driver support.

Signed-off-by: Magnus Damm <damm@igel.co.jp>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
15 years agosh: convert processor device setup functions to arch_initcall()
Magnus Damm [Wed, 22 Jul 2009 15:14:29 +0000 (15:14 +0000)]
sh: convert processor device setup functions to arch_initcall()

Convert the processor platform device setup
functions from __initcall() and sometimes
device_initcall() to arch_initcall().

This makes sure that the platform devices are
registered a bit earlier so the devices are
available when drivers register using initcall
levels earlier than device_initcall().

A good example is platform devices needed by
i2c-sh_mobile.c which registers a bit earlier
using subsys_initcall().

Signed-off-by: Magnus Damm <damm@igel.co.jp>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
15 years agomd: allow upper limit for resync/reshape to be set when array is read-only
NeilBrown [Thu, 13 Aug 2009 00:41:50 +0000 (10:41 +1000)]
md: allow upper limit for resync/reshape to be set when array is read-only

Normally we only allow the upper limit for a reshape to be decreased
when the array not performing a sync/recovery/reshape, otherwise there
could be races.  But if an array is part-way through a reshape when it
is assembled the reshape is started immediately leaving no window
to set an upper bound.

If the array is started read-only, the reshape will be suspended until
the array becomes writable, so that provides a window during which it
is perfectly safe to reduce the upper limit of a reshape.

So: allow the upper limit (sync_max) to be reduced even if the reshape
thread is running, as long as the array is still read-only.

Signed-off-by: NeilBrown <neilb@suse.de>
15 years agomd/raid5: Properly remove excess drives after shrinking a raid5/6
NeilBrown [Thu, 13 Aug 2009 00:41:49 +0000 (10:41 +1000)]
md/raid5: Properly remove excess drives after shrinking a raid5/6

We were removing the drives, from the array, but not
removing symlinks from /sys/.... and not marking the device
as having been removed.

Signed-off-by: NeilBrown <neilb@suse.de>
15 years agomd/raid5: make sure a reshape restarts at the correct address.
NeilBrown [Thu, 13 Aug 2009 00:13:00 +0000 (10:13 +1000)]
md/raid5: make sure a reshape restarts at the correct address.

This "if" don't allow for the possibility that the number of devices
doesn't change, and so sector_nr isn't set correctly in that case.
So change '>' to '>='.

Signed-off-by: NeilBrown <neilb@suse.de>
15 years agomd/raid5: allow new reshape modes to be restarted in the middle.
NeilBrown [Thu, 13 Aug 2009 00:06:24 +0000 (10:06 +1000)]
md/raid5: allow new reshape modes to be restarted in the middle.

md/raid5 doesn't allow a reshape to restart if it involves writing
over the same part of disk that it would be reading from.
This happens at the beginning of a reshape that increases the number
of devices, at the end of a reshape that decreases the number of
devices, and continuously for a reshape that does not change the
number of devices.

The current code is correct for the "increase number of devices"
case as the critical section at the start is handled by userspace
performing a backup.

It does not work for reducing the number of devices, or the
no-change case.
For 'reducing', we need to invert the test.  For no-change we cannot
really be sure things will be safe, so simply require the array
to be read-only, which is how the user-space code which carefully
starts such arrays works.

Signed-off-by: NeilBrown <neilb@suse.de>
15 years agomd: never advance 'events' counter by more than 1.
NeilBrown [Wed, 12 Aug 2009 23:54:02 +0000 (09:54 +1000)]
md: never advance 'events' counter by more than 1.

When assembling arrays, md allows two devices to have different event
counts as long as the difference is only '1'.  This is to cope with
a system failure between updating the metadata on two difference
devices.

However there are currently times when we update the event count by
2.  This was done to keep the event count even when the array is clean
and odd when it is dirty, which allows us to avoid writing common
update to spare devices and so allow those spares to go to sleep.

This is bad for the above reason.  So change it to never increase by
two.  This means that the alignment between 'odd/even' and
'clean/dirty' might take a little longer to attain, but that is only a
small cost.  The spares will get a few more updates but that will
still be spared (;-) most updates and can still go to sleep.

Prior to this patch there was a small chance that after a crash an
array would fail to assemble due to the overly large event count
mismatch.

Signed-off-by: NeilBrown <neilb@suse.de>
15 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
Linus Torvalds [Wed, 12 Aug 2009 16:55:46 +0000 (09:55 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  Remove double removal of blktrace directory

15 years agoRemove double removal of blktrace directory
Alan D. Brunelle [Fri, 7 Aug 2009 16:01:08 +0000 (12:01 -0400)]
Remove double removal of blktrace directory

commit fd51d251e4cdb21f68e9dbc4336514d64a105a79
Author: Stefan Raspl <raspl@linux.vnet.ibm.com>
Date:   Tue May 19 09:59:08 2009 +0200

    blktrace: remove debugfs entries on bad path

added in an explicit invocation of debugfs_remove for bt->dir, in
blk_remove_buf_file_callback we are also getting the directory removed. On
occasion I am seeing memory corruption that I have bisected down to
this commit. [The testing involves a (long) series of I/O benchmarks
with blktrace invoked around the actual runs.] I believe that this
committed patch is correct, but the problem actually lies in the code
in blk_remove_buf_file_callback.

With this patch I am able to consistently get complete runs whereas
previously I could not get a single run to complete.

The first part of the patch simply moves the debugfs_remove below the
relay_close: the relay_close call will remove files under bt->dir, and
so we should not remove the directory until all the files we created
have been removed. (Note: This is not sufficient to fix the problem -
the file system code has ref counts on the directoy, so our invocation
does not cause the directory to actually be removed. Nonetheless, we
should not rely upon that feature.)

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoMerge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
Linus Torvalds [Wed, 12 Aug 2009 15:49:35 +0000 (08:49 -0700)]
Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs

* 'for-linus' of git://oss.sgi.com/xfs/xfs:
  xfs: fix spin_is_locked assert on uni-processor builds
  xfs: check for dinode realtime flag corruption
  use XFS_CORRUPTION_ERROR in xfs_btree_check_sblock
  xfs: switch to NOFS allocation under i_lock in xfs_attr_rmtval_get
  xfs: switch to NOFS allocation under i_lock in xfs_readlink_bmap
  xfs: switch to NOFS allocation under i_lock in xfs_attr_rmtval_set
  xfs: switch to NOFS allocation under i_lock in xfs_buf_associate_memory
  xfs: switch to NOFS allocation under i_lock in xfs_dir_cilookup_result
  xfs: switch to NOFS allocation under i_lock in xfs_da_buf_make
  xfs: switch to NOFS allocation under i_lock in xfs_da_state_alloc
  xfs: switch to NOFS allocation under i_lock in xfs_getbmap
  xfs: avoid memory allocation under m_peraglock in growfs code

15 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Wed, 12 Aug 2009 15:32:47 +0000 (08:32 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  ALSA: hda - Don't override ADC definitions for ALC codecs
  ALSA: hda - Add missing vmaster initialization for ALC269
  ASoC: Add missing DRV_NAME definitions for fsl/* drivers

15 years agoMerge branch 'zerolen' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/misc-2.6
Linus Torvalds [Wed, 12 Aug 2009 15:29:32 +0000 (08:29 -0700)]
Merge branch 'zerolen' of git://git./linux/kernel/git/jgarzik/misc-2.6

* 'zerolen' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/misc-2.6:
  Remove zero-length file drivers/mtd/maps/sbc8240.c

15 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzi...
Linus Torvalds [Wed, 12 Aug 2009 15:24:17 +0000 (08:24 -0700)]
Merge branch 'upstream-linus' of git://git./linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  ahci: add workaround for on-board 5723s on some gigabyte boards
  ahci: Soften up the dmesg on SB600 PMP softreset failure recovery
  Documentation/kernel-parameters.txt: document libata's ignore_hpa option
  sata_nv: MSI support, disabled by default
  libata: OCZ Vertex can't do HPA
  pata_atiixp: fix second channel support
  pata_at91: fix resource release

15 years agoNFS: Fix an O_DIRECT Oops...
Trond Myklebust [Wed, 12 Aug 2009 13:12:30 +0000 (09:12 -0400)]
NFS: Fix an O_DIRECT Oops...

We can't call nfs_readdata_release()/nfs_writedata_release() without
first initialising and referencing args.context. Doing so inside
nfs_direct_read_schedule_segment()/nfs_direct_write_schedule_segment()
causes an Oops.

We should rather be calling nfs_readdata_free()/nfs_writedata_free() in
those cases.

Looking at the O_DIRECT code, the "struct nfs_direct_req" is already
referencing the nfs_open_context for us. Since the readdata and writedata
structures carry a reference to that, we can simplify things by getting rid
of the extra nfs_open_context references, so that we can replace all
instances of nfs_readdata_release()/nfs_writedata_release().

Reported-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Tested-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoperf_counter, x86: Fix/improve apic fallback
Ingo Molnar [Tue, 11 Aug 2009 08:40:08 +0000 (10:40 +0200)]
perf_counter, x86: Fix/improve apic fallback

Johannes Stezenbach reported that his Pentium-M based
laptop does not have the local APIC enabled by default,
and hence perfcounters do not get initialized.

Add a fallback for this case: allow non-sampled counters
and return with an error on sampled counters. This allows
'perf stat' to work out of box - and allows 'perf top'
and 'perf record' to fall back on a hrtimer based sampling
method.

( Passing 'lapic' on the boot line will allow hardware
  sampling to occur - but if the APIC is disabled
  permanently by the hardware then this fallback still
  allows more systems to use perfcounters. )

Also decouple perfcounter support from X86_LOCAL_APIC.

-v2: fix typo breaking counters on all other systems ...

Reported-by: Johannes Stezenbach <js@sig21.net>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf record: Add missing -C option support for specifying profile cpu
Jens Axboe [Wed, 12 Aug 2009 09:18:01 +0000 (11:18 +0200)]
perf record: Add missing -C option support for specifying profile cpu

perf top supports a -C for setting the profile CPU, but perf
record does not. This adds the same option for perf record,
allowing the user to specify a specific target profile CPU.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20090812091801.GC12579@kernel.dk>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf tools: Fix dso__new handle() to handle deleted DSOs
Arnaldo Carvalho de Melo [Tue, 11 Aug 2009 20:04:36 +0000 (17:04 -0300)]
perf tools: Fix dso__new handle() to handle deleted DSOs

It is better than showing the map addr, this way at least we
know that we can't get the symtabs because the DSO was deleted
(system update) while an app still used such DSO.

Yeah, don't do that, but if you do, you'll figure it out
quicker this way.

[acme@doppio linux-2.6-tip]$ perf report | head -15
 # Samples: 3796
 #
 # Overhead  Command                                                        Shared Object  Symbol
 # ........  .......  ...................................................................  ......
 #
    23.55%   pidgin  /lib64/libglib-2.0.so.0.2000.4.#prelink#.Pd98lu (deleted)            [.] 0x00000000038844
    21.55%   pidgin  /lib64/libpthread-2.10.1.so.#prelink#.AFwK8Q (deleted)               [.] 0x0000000000a42d
    10.85%   pidgin  [kernel]                                                             [.] vread_hpet
     7.85%   pidgin  /lib64/libgobject-2.0.so.0.2000.4.#prelink#.o1vpU7 (deleted)         [.] 0x00000000014de8
     3.35%   pidgin  /lib64/libc-2.10.1.so (deleted)                                      [.] 0x0000000007a875
     3.19%   pidgin  /lib64/libdbus-1.so.3.4.0.#prelink#.6mwgZP (deleted)                 [.] 0x0000000001d254
     3.06%   pidgin  /usr/lib64/libgtk-x11-2.0.so.0.1600.5.#prelink#.511hAl (deleted)     [.] 0x000000002334e7
     2.90%   pidgin  /usr/lib64/libgdk-x11-2.0.so.0.1600.5.#prelink#.5qlMo1 (deleted)     [.] 0x00000000037b2d
     1.84%   pidgin  [kernel]                                                             [k] do_sys_poll
     1.45%   pidgin  /usr/lib64/libX11.so.6.2.0.#prelink#.iR59Rx (deleted)                [.] 0x0000000004c751
[acme@doppio linux-2.6-tip]$

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Luis Claudio R. Gonçalves <lclaudio@redhat.com>
Cc: Clark Williams <williams@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20090811200436.GA3478@ghostprotocols.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf tools: Fix fallback to cplus_demangle() when bfd_demangle() is not available
Arnaldo Carvalho de Melo [Tue, 11 Aug 2009 19:22:11 +0000 (16:22 -0300)]
perf tools: Fix fallback to cplus_demangle() when bfd_demangle() is not available

In old binutils we can't access bfd_demangle(), use
cplus_demangle() just like oprofile.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Luis Claudio R. Gonçalves <lclaudio@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20090811192211.GG18061@ghostprotocols.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf report: Show the tid too in -D
Arnaldo Carvalho de Melo [Tue, 11 Aug 2009 19:21:38 +0000 (16:21 -0300)]
perf report: Show the tid too in -D

This made it easier to find the firefox threading related
bug.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20090811192138.GE18061@ghostprotocols.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoperf record: Fix .tid and .pid fill-in when synthesizing events
Arnaldo Carvalho de Melo [Tue, 11 Aug 2009 19:22:00 +0000 (16:22 -0300)]
perf record: Fix .tid and .pid fill-in when synthesizing events

Noticed when trying to record events for a firefox thread. We
were synthesizing both .tid and .pid with the pid passed via
--pid.

Fix it by reading /proc/PID/status and getting the tgid
to use in .pid, .tid gets the specified "pid".

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20090811192200.GF18061@ghostprotocols.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoRemove zero-length file drivers/mtd/maps/sbc8240.c
Jeff Garzik [Wed, 12 Aug 2009 10:29:57 +0000 (06:29 -0400)]
Remove zero-length file drivers/mtd/maps/sbc8240.c

It was "deleted" in commit 2bf961b7ccd69e108ac435c67e2b0522b403c578

Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agoahci: add workaround for on-board 5723s on some gigabyte boards
Tejun Heo [Tue, 4 Aug 2009 05:30:08 +0000 (14:30 +0900)]
ahci: add workaround for on-board 5723s on some gigabyte boards

Some gigabytes have on-board SIMG5723s connected to JMB ahcis.  These
are used to implement hardware raid.  Unfortunately some firmware
revisions on these 5723s don't bring the link down when all the
downstream ports are unoccupied while not responding to reset protocol
which makes libata think that there's device attached to the port but
is not responding and retry.  This results in painfully wrong boot
detection time for these ports when they're empty.

This patch quirks those boards such that ahci gives up after the
initial timeout.  Combined with parallel probing, this gives quick
enough probing and also is safe because SIMG5723 will respond to the
first try if any of the downstream ports is occupied.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Marc Bowes <marcbowes@gmail.com>
Reported-by: Nicolas Mailhot <Nicolas.Mailhot@LaPoste.net>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agoahci: Soften up the dmesg on SB600 PMP softreset failure recovery
Shane Huang [Wed, 5 Aug 2009 02:10:41 +0000 (10:10 +0800)]
ahci: Soften up the dmesg on SB600 PMP softreset failure recovery

Too strong words led to spurious bug reports: Novell bugzilla #527748,
RedHat bugzilla #468800. This patch is used to soften up the dmesg on
SB600 PMP softreset failure recovery, so as to remove the scariness and
concern from community.

Reported-by: pgnet Dev <pgnet.dev@gmail.com>
Signed-off-by: Shane Huang <shane.huang@amd.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agoDocumentation/kernel-parameters.txt: document libata's ignore_hpa option
Michael Prokop [Wed, 5 Aug 2009 22:14:10 +0000 (00:14 +0200)]
Documentation/kernel-parameters.txt: document libata's ignore_hpa option

By default the kernel honors the HPA (host protected area) of hard
drives.  Using libata's ignore_hpa module option it's possible to
change this behaviour.

Document usage and options of libata.ignore_hpa in
Documentation/kernel-parameters.txt.

Signed-off-by: Michael Prokop <mika@grml.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agosata_nv: MSI support, disabled by default
Tony Vroon [Wed, 5 Aug 2009 23:50:09 +0000 (00:50 +0100)]
sata_nv: MSI support, disabled by default

At least the nVidia MCP55 controller quite happily supports MSI.
This adds an option to use it. It is disabled by default.
As per feedback by Robert Hancock, it will honour the user
request as the kernel will not enable MSI where the controller
or the specific system configuration do not support it.

Signed-off-by: Tony Vroon <tony@linx.net>
Cc: Robert Hancock <hancockrwd@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agolibata: OCZ Vertex can't do HPA
Tejun Heo [Thu, 6 Aug 2009 16:59:15 +0000 (01:59 +0900)]
libata: OCZ Vertex can't do HPA

OCZ Vertex SSD can't do HPA and not in a usual way.  It reports HPA,
allows unlocking but then fails all IOs which fall in the unlocked
area.  Quirk it so that HPA unlocking is not used for the device.

Reported by Daniel Perup in bnc#522414.

 https://bugzilla.novell.com/show_bug.cgi?id=522414

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Daniel Perup <probe@spray.se>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agopata_atiixp: fix second channel support
Bartlomiej Zolnierkiewicz [Thu, 6 Aug 2009 15:47:05 +0000 (17:47 +0200)]
pata_atiixp: fix second channel support

PIO and MWDMA timings are never programmed for the second channel
because timing registers are treated as 16-bit long ones.

The bug is an attixp -> pata_atiixp regression and goes back to:

commit 669a5db411d85a14f86cd92bc16bf7ab5b8aa235
Author: Jeff Garzik <jeff@garzik.org>
Date:   Tue Aug 29 18:12:40 2006 -0400

    [libata] Add a bunch of PATA drivers.

Cc: Krystian Juskowiak <jusko@tlen.pl>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bbpetkov@yahoo.de>
Cc: Robert Hancock <hancockrwd@gmail.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agopata_at91: fix resource release
Tejun Heo [Fri, 7 Aug 2009 02:15:20 +0000 (11:15 +0900)]
pata_at91: fix resource release

Julias Lawall discovered that pata_at91 wasn't freeing a memory region
allocated with kzalloc() on init failure paths.  Upon review,
pata_at91 also seems to be doing unnecessary explicit resource
releases for managed resources too.  Convert memory allocation to
managed one and drop unnecessary explicit resource releases.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Julia Lawall <julia@diku.dk>
Cc: Sergey Matyukevich <geomatsi@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agox86: Fix oops in identify_cpu() on CPUs without CPUID
Ondrej Zary [Tue, 11 Aug 2009 18:00:11 +0000 (20:00 +0200)]
x86: Fix oops in identify_cpu() on CPUs without CPUID

Kernel is broken for x86 CPUs without CPUID since 2.6.28. It
crashes with NULL pointer dereference in identify_cpu():

766        generic_identify(c);
767
768-->     if (this_cpu->c_identify)
769               this_cpu->c_identify(c);

this_cpu is NULL. This is because it's only initialized in
get_cpu_vendor() function, which is not called if the CPU has
no CPUID instruction.

Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
LKML-Reference: <200908112000.15993.linux@rainbow-software.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoxfs: fix spin_is_locked assert on uni-processor builds
Christoph Hellwig [Mon, 10 Aug 2009 14:32:44 +0000 (11:32 -0300)]
xfs: fix spin_is_locked assert on uni-processor builds

Without SMP or preemption spin_is_locked always returns false,
so we can't do an assert with it.  Instead use assert_spin_locked,
which does the right thing on all builds.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@sandeen.net>
Reported-by: Johannes Engel <jcnengel@googlemail.com>
Tested-by: Johannes Engel <jcnengel@googlemail.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: check for dinode realtime flag corruption
Christoph Hellwig [Mon, 10 Aug 2009 14:32:18 +0000 (11:32 -0300)]
xfs: check for dinode realtime flag corruption

Ramon tested XFS with a modified version of fsfuzzer and hit a NULL
pointer dereference in __xfs_get_blocks due to the RT device target
pointer being NULL.

To fix this reject inode with the realtime bit set on a a filesystem
without an RT subvolume during inode read.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@sandeen.net>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Reported-by: Ramon de Carvalho Valle <ramon@risesecurity.org>
Tested-by: Ramon de Carvalho Valle <ramon@risesecurity.org>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agouse XFS_CORRUPTION_ERROR in xfs_btree_check_sblock
Eric Sandeen [Mon, 20 Jul 2009 15:52:15 +0000 (10:52 -0500)]
use XFS_CORRUPTION_ERROR in xfs_btree_check_sblock

In Red Hat Bug 512552
 - Can't write to XFS mount during raid5 resync

a user ran into corruption while resyncing a raid, and we failed
a consistency test, but didn't get much more info; it'd be nice
to call XFS_CORRUPTION_ERROR here so we can see the buffer
contents.

Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_attr_rmtval_get
Christoph Hellwig [Sat, 18 Jul 2009 22:15:01 +0000 (18:15 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_attr_rmtval_get

xfs_attr_rmtval_get is always called with i_lock held, but i_lock is taken
in reclaim context so all allocations under it must avoid recursions into
the filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_readlink_bmap
Christoph Hellwig [Sat, 18 Jul 2009 22:15:00 +0000 (18:15 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_readlink_bmap

xfs_readlink_bmap is called with i_lock held, but i_lock is taken in
reclaim context so all allocations under it must avoid recursions into
the filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_attr_rmtval_set
Christoph Hellwig [Sat, 18 Jul 2009 22:14:59 +0000 (18:14 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_attr_rmtval_set

xfs_attr_rmtval_set is always called with i_lock held, and i_lock is taken
in reclaim context so all allocations under it must avoid recursions into
the filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_buf_associate_memory
Christoph Hellwig [Sat, 18 Jul 2009 22:14:58 +0000 (18:14 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_buf_associate_memory

xfs_buf_associate_memory is used for setting up the spare buffer for the
log wrap case in xlog_sync which can happen under i_lock when called from
xfs_fsync. The i_lock mutex is taken in reclaim context so all allocations
under it must avoid recursions into the filesystem.  There are a couple
more uses of xfs_buf_associate_memory in the log recovery code that are
also affected by this, but I'd rather keep the code simple than passing on
a gfp_mask argument.  Longer term we should just stop requiring the memoery
allocation in xlog_sync by some smaller rework of the buffer layer.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_dir_cilookup_result
Christoph Hellwig [Sat, 18 Jul 2009 22:14:57 +0000 (18:14 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_dir_cilookup_result

xfs_dir_cilookup_result is always called with i_lock held, but i_lock is taken
in reclaim context so all allocations under it must avoid recursions into the
filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_da_buf_make
Christoph Hellwig [Sat, 18 Jul 2009 22:14:56 +0000 (18:14 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_da_buf_make

i_lock is taken in the reclaim context so all allocations under it
must avoid recursions into the filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_da_state_alloc
Christoph Hellwig [Sat, 18 Jul 2009 22:14:55 +0000 (18:14 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_da_state_alloc

xfs_da_state_alloc is always called with i_lock held, but i_lock is taken in
reclaim context so all allocations under it must avoid recursions into the
filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: switch to NOFS allocation under i_lock in xfs_getbmap
Christoph Hellwig [Sat, 18 Jul 2009 22:14:54 +0000 (18:14 -0400)]
xfs: switch to NOFS allocation under i_lock in xfs_getbmap

xfs_getbmap allocates memory with i_lock held, but i_lock is taken in
reclaim context so all allocations under it must avoid recursions into
the filesystem.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoxfs: avoid memory allocation under m_peraglock in growfs code
Christoph Hellwig [Sat, 18 Jul 2009 22:14:53 +0000 (18:14 -0400)]
xfs: avoid memory allocation under m_peraglock in growfs code

Allocate the memory for the larger m_perag array before taking the
per-AG lock as the per-AG lock can be taken under the i_lock which
can be taken from reclaim context.

Reported by the new reclaim context tracing in lockdep.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
15 years agoMerge branch 'fix/hda' into for-linus
Takashi Iwai [Wed, 12 Aug 2009 06:05:20 +0000 (08:05 +0200)]
Merge branch 'fix/hda' into for-linus

* fix/hda:
  ALSA: hda - Don't override ADC definitions for ALC codecs
  ALSA: hda - Add missing vmaster initialization for ALC269

15 years agoMerge branch 'fix/asoc' into for-linus
Takashi Iwai [Wed, 12 Aug 2009 06:05:19 +0000 (08:05 +0200)]
Merge branch 'fix/asoc' into for-linus

* fix/asoc:
  ASoC: Add missing DRV_NAME definitions for fsl/* drivers

15 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/fyu/linux-2.6
Linus Torvalds [Wed, 12 Aug 2009 00:06:16 +0000 (17:06 -0700)]
Merge branch 'release' of git://git./linux/kernel/git/fyu/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/fyu/linux-2.6:
  arch/ia64/kernel/iosapic: missing test after ioremap()
  ia64/topology.c: exit cache_add_dev when kobject_init_and_add fails
  arch/ia64/Makefile: Remove -mtune=merced in IA64 kernel build
  IA64: includecheck fix: ia64, pgtable.h
  IA64: includecheck fix: ia64, ia64_ksyms.c
  ia64: boolean __test_and_clear_bit
  Bug Fix arch/ia64/kernel/pci-dma.c: fix recursive dma_supported() call in iommu_dma_supported()

15 years agoarch/ia64/kernel/iosapic: missing test after ioremap()
Roel Kluin [Tue, 11 Aug 2009 21:52:11 +0000 (14:52 -0700)]
arch/ia64/kernel/iosapic: missing test after ioremap()

Missing test after ioremap()

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
15 years agoia64/topology.c: exit cache_add_dev when kobject_init_and_add fails
Fenghua Yu [Tue, 11 Aug 2009 21:52:11 +0000 (14:52 -0700)]
ia64/topology.c: exit cache_add_dev when kobject_init_and_add fails

Make cache_add_dev exit sysfs when kobject_init_and_add returns an error.

Signed-off-by: Xiaotian Feng <dfeng@redhat.com>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
15 years agoarch/ia64/Makefile: Remove -mtune=merced in IA64 kernel build
Fenghua Yu [Tue, 11 Aug 2009 21:52:11 +0000 (14:52 -0700)]
arch/ia64/Makefile: Remove -mtune=merced in IA64 kernel build

Between GCC version 3.4.0 and 4.3.3 (including 3.4.0 and 4.3.3), -mtune=merced
is implemented in GCC. Starting from 4.4.0, -mtune=merced is deprecated.

Even implemented in versions between 3.4.0 and 4.3.3, the -mtune=merced
feature has been broken in some of the versions. For example, GCC 4.1.2 reports
interanl tuning function errors during kernel building with -mtune=merced. Or
GCC Bugzilla 16130 reports another -mtune=merced issue on GCC 3.4.1.

So I would remove the -mtune=merced from IA64 kernel build. Without this option,
kernel on Merced will remain the same except losing an unstable and out-of-date
performance tunning feature.

Since GCC version 3.4.0, -mtune=mckinley has been implemented. The
-mtune=mckinley option functions the same as mtune=itanium2. And mtune=itanium2
is the default option. So we don't need to add mtune=mckinley either since its
been the default option in any GCC version which implements this option.

Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
15 years agoIA64: includecheck fix: ia64, pgtable.h
Jaswinder Singh Rajput [Tue, 11 Aug 2009 21:52:11 +0000 (14:52 -0700)]
IA64: includecheck fix: ia64, pgtable.h

fix the following 'make includecheck' warning:

  arch/ia64/include/asm/pgtable.h: asm/processor.h is included more than once.

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
15 years agoIA64: includecheck fix: ia64, ia64_ksyms.c
Jaswinder Singh Rajput [Tue, 11 Aug 2009 21:52:10 +0000 (14:52 -0700)]
IA64: includecheck fix: ia64, ia64_ksyms.c

fix the following 'make includecheck' warning:

  arch/ia64/kernel/ia64_ksyms.c: asm/page.h is included more than once.

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
15 years agoia64: boolean __test_and_clear_bit
Johannes Weiner [Tue, 11 Aug 2009 21:52:10 +0000 (14:52 -0700)]
ia64: boolean __test_and_clear_bit

__test_and_clear_bit() returns a bitfield with the tested-for bit set.
Make it consistent with the other bitops - of ia64 but also every
other architecture - and return a boolean value.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Fenghua Yu <fenghua.yu@intel.com>