This patch fixes a crash caused by an interrupt coming in when an IRQ stack
is being torn down. When this happens, handle_signal will loop, setting up
the IRQ stack again because the tearing down had finished, and handling
whatever signals had come in.
However, to_irq_stack returns a mask of pending signals to be handled, plus
bit zero is set if the IRQ stack was already active, and thus shouldn't be
torn down. This causes a problem because when handle_signal goes around
the loop, sig will be zero, and to_irq_stack will duly set bit zero in the
returned mask, faking handle_signal into believing that it shouldn't tear
down the IRQ stack and return thread_info pointers back to their original
values.
This will eventually cause a crash, as the IRQ stack thread_info will
continue pointing to the original task_struct and an interrupt will look
into it after it has been freed.
The fix is to stop passing a signal number into to_irq_stack. Rather, the
pending signals mask is initialized beforehand with the bit for sig already
set. References to sig in to_irq_stack can be replaced with references to
the mask.
[akpm@linux-foundation.org: use UL]
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch proposes fixes to the reference counting of memory policy in the
page allocation paths and in show_numa_map(). Extracted from my "Memory
Policy Cleanups and Enhancements" series as stand-alone.
Shared policy lookup [shmem] has always added a reference to the policy,
but this was never unrefed after page allocation or after formatting the
numa map data.
Default system policy should not require additional ref counting, nor
should the current task's task policy. However, show_numa_map() calls
get_vma_policy() to examine what may be [likely is] another task's policy.
The latter case needs protection against freeing of the policy.
This patch adds a reference count to a mempolicy returned by
get_vma_policy() when the policy is a vma policy or another task's
mempolicy. Again, shared policy is already reference counted on lookup. A
matching "unref" [__mpol_free()] is performed in alloc_page_vma() for
shared and vma policies, and in show_numa_map() for shared and another
task's mempolicy. We can call __mpol_free() directly, saving an admittedly
inexpensive inline NULL test, because we know we have a non-NULL policy.
Handling policy ref counts for hugepages is a bit trickier.
huge_zonelist() returns a zone list that might come from a shared or vma
'BIND policy. In this case, we should hold the reference until after the
huge page allocation in dequeue_hugepage(). The patch modifies
huge_zonelist() to return a pointer to the mempolicy if it needs to be
unref'd after allocation.
Kernel Build [16cpu, 32GB, ia64] - average of 10 runs:
w/o patch w/ refcount patch
Avg Std Devn Avg Std Devn
Real: 100.59 0.38 100.63 0.43
User: 1209.60 0.37 1209.91 0.31
System: 81.52 0.42 81.64 0.34
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Andi Kleen <ak@suse.de>
Cc: Christoph Lameter <clameter@sgi.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It turned out, that the user namespace is released during the do_exit() in
exit_task_namespaces(), but the struct user_struct is released only during the
put_task_struct(), i.e. MUCH later.
On debug kernels with poisoned slabs this will cause the oops in
uid_hash_remove() because the head of the chain, which resides inside the
struct user_namespace, will be already freed and poisoned.
Since the uid hash itself is required only when someone can search it, i.e.
when the namespace is alive, we can safely unhash all the user_struct-s from
it during the namespace exiting. The subsequent free_uid() will complete the
user_struct destruction.
For example simple program
#include <sched.h>
char stack[2 * 1024 * 1024];
int f(void *foo)
{
return 0;
}
int main(void)
{
clone(f, stack + 1 * 1024 * 1024, 0x10000000, 0);
return 0;
}
run on kernel with CONFIG_USER_NS turned on will oops the
kernel immediately.
This was spotted during OpenVZ kernel testing.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Acked-by: "Serge E. Hallyn" <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Surprisingly, but (spotted by Alexey Dobriyan) the uid hash still uses
list_heads, thus occupying twice as much place as it could. Convert it to
hlist_heads.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The do_split() function for htree dir blocks is intended to split a leaf
block to make room for a new entry. It sorts the entries in the original
block by hash value, then moves the last half of the entries to the new
block - without accounting for how much space this actually moves. (IOW,
it moves half of the entry *count* not half of the entry *space*). If by
chance we have both large & small entries, and we move only the smallest
entries, and we have a large new entry to insert, we may not have created
enough space for it.
The patch below stores each record size when calculating the dx_map, and
then walks the hash-sorted dx_map, calculating how many entries must be
moved to more evenly split the existing entries between the old block and
the new block, guaranteeing enough space for the new entry.
The dx_map "offs" member is reduced to u16 so that the overall map size
does not change - it is temporarily stored at the end of the new block, and
if it grows too large it may be overwritten. By making offs and size both
u16, we won't grow the map size.
Also add a few comments to the functions involved.
This fixes the testcase reported by hooanon05@yahoo.co.jp on the
linux-ext4 list, "ext3 dir_index causes an error"
Thanks to Andreas Dilger for discussing the problem & solution with me.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Tested-by: Junjiro Okajima <hooanon05@yahoo.co.jp>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: <linux-ext4@vger.kernel.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There is still some confusion and disagreement over what this interface should
actually do. So it is best that we disable it in 2.6.23 until we get that
fully sorted out.
(sys_timerfd() was present in 2.6.22 but it was apparently broken, so here we
assume that nobody is using it yet).
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Davide Libenzi <davidel@xmailserver.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Convert asserts (BUGs) in dx_probe from bad on-disk data to recoverable
errors with helpful warnings. With help catching other asserts from Duane
Griffin <duaneg@dghda.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Acked-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Theodore Ts'o <tytso@mit.edu>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The mask on i830 should be 0x70 always, later chips 0xF0 should be okay.
Signed-off-by: Dave Airlie <airlied@linux.ie>
Acked-by: Zhenyu Wang <zhenyu.z.wang@intel.com>
Cc: Michael Haas <laga@laga.ath.cx>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Xen ignores all updates to cr4, and some versions will kill the domain if
you try to change its value. Just ignore all changes.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix a couple drivers that do not correctly terminate their pci_device_id
lists. This results in garbage being spewed into modules.pcimap when the
module happens to not have 28 NULL bytes following the table, and/or the
last PCI ID is actually truncated from the table when calculating the
modules.alias PCI aliases, cause those unfortunate device IDs to not
auto-load.
Signed-off-by: Kees Cook <kees@ubuntu.com>
Acked-by: Corey Minyard <minyard@acm.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Acked-by: Jeff Garzik <jeff@garzik.org>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The shrinking of a virtual memory area that is mmap(2)'d to a memory
special file (device drivers/char/mspec.c) can cause a panic.
If the mapped size of the vma (vm_area_struct) is very large, mspec allocates
a large vma_data structure with vmalloc(). But such a vma can be shrunk by
an munmap(2). The current driver uses the current size of each vma to
deduce whether its vma_data structure was allocated by kmalloc() or vmalloc().
So if the vma was shrunk it appears to have been allocated by kmalloc(),
and mspec attempts to free it with kfree(). This results in a panic.
This patch avoids the panic (by preserving the type of the allocation) and
also makes mspec work correctly as the vma is split into pieces by the
munmap(2)'s.
All vma's derived from such a split vma share the same vma_data structure that
represents all the pages mapped into this set of vma's. The mpec driver
must be made capable of using the right portion of the structure for each
member vma. In other words, it must index into the array of page addresses
using the portion of the array that represents the current vma. This is
enabled by storing the vma group's vm_start in the vma_data structure.
The shared vma_data's are not protected by mm->mmap_sem in the fork() case
so the reference count is left as atomic_t.
Signed-off-by: Cliff Wickman <cpw@sgi.com>
Acked-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently the rtc driver, rtc-ds1552.c uses an unsigned long to store the
base mmio address of the NVRAM/RTC. This breaks on 32-bit systems with
larger physical addresses.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: David Brownell <david-b@pacbell.net>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently the rtc driver, rtc-ds1742.c uses an unsigned long to store the
base mmio address of the NVRAM/RTC. This breaks on systems like PowerPC
440, which is a 32-bit core with 36-bit physical addresses: IO on the
system, including the RTC, is typically above the 4GB point, and cannot fit
into an unsigned long.
This patch fixes the problem by replacing the unsigned long with a
resource_size_t. Tested on Ebony (PPC440) (with additional patches to
instantiate the ds1742 platform device appropriately).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
struct utsname is copied from master one without any exclusion.
Here is sample output from one proggie doing
sethostname("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa");
sethostname("bbbbbbbbbbbbbbbbbbbbbbbbbbbbbb");
and another
clone(,, CLONE_NEWUTS, ...)
uname()
hostname = 'aaaaaaaaaaaaaaaaaaaaaaaaabbbbb'
hostname = 'bbbaaaaaaaaaaaaaaaaaaaaaaaaaaa'
hostname = 'aaaaaaaabbbbbbbbbbbbbbbbbbbbbb'
hostname = 'aaaaaaaaaaaaaaaaaaaaaaaaaabbbb'
hostname = 'aaaaaaaaaaaaaaaaaaaaaaaaaaaabb'
hostname = 'aaabbbbbbbbbbbbbbbbbbbbbbbbbbb'
hostname = 'bbbbbbbbbbbbbbbbaaaaaaaaaaaaaa'
Hostname is sometimes corrupted.
Yes, even _the_ simplest namespace activity had bug in it. :-(
Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 8314418629 (Freezer: make kernel
threads nonfreezable by default) breaks freezing when attempting to resume
from an initrd, because the init (which is freezeable) spins while waiting
for another thread to run /linuxrc, but doesn't check whether it has been
told to enter the refrigerator. The original patch replaced a call to
try_to_freeze() with a call to yield(). I believe a simple reversion is
wrong because if !CONFIG_PM_SLEEP, try_to_freeze() is a noop. It should
still yield.
Signed-off-by: Nigel Cunningham <nigel@nigel.suspend2.net>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I found a type mismatch in UML that makes host block devices unusable as ubd
devices on x86_64 and other 64 bits systems (segfault of the mm subsystem):
In block/ioctl.c, the following lines show that the BLKGETSIZE ioctl expects
a pointer to a long:
case BLKGETSIZE:
if ((bdev->bd_inode->i_size >> 9) > ~0UL)
return -EFBIG;
return put_ulong(arg, bdev->bd_inode->i_size >> 9);
In arch/um/os-Linux/file.c, os_file_size calls it with an int.
The ioctl_list man page should be fixed as well.
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
sparc32:
drivers/block/DAC960.c: In function 'DAC960_V1_EnableMemoryMailboxInterface':
drivers/block/DAC960.c:1168: error: 'DMA_32BIT_MASK' undeclared (first use in this function)
drivers/block/DAC960.c:1168: error: (Each undeclared identifier is reported only
Cc: <dac@conglom-o.org>
Cc: <stable@kernel.org>
Cc: Alessandro Polverini <alex@nibbles.it>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Name it thinkpad-acpi version 0.16 to avoid any confusion with some 0.15
thinkpad-acpi development snapshots and backports that had input layer
support, but no hotkey_report_mode support.
Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Signed-off-by: Len Brown <len.brown@intel.com>
Revert new 2.6.23 CONFIG_THINKPAD_ACPI_INPUT_ENABLED Kconfig option because
it would create a legacy we don't want to support.
CONFIG_THINKPAD_ACPI_INPUT_ENABLED was added to try to fix an issue that is
now moot with the addition of the netlink ACPI event report interface to
the ACPI core.
Now that ACPI core can send events over netlink, we can use a different
strategy to keep backwards compatibility with older userspace, without the
need for the CONFIG_THINKPAD_ACPI_INPUT_ENABLED games. And it arrived
before CONFIG_THINKPAD_ACPI_INPUT_ENABLED made it to a stable mainline
kernel, even, which is Good.
This patch is in sync with some changes to thinkpad-acpi backports, that
will keep things sane for userspace across different combinations of kernel
versions, thinkpad-acpi backports (or the lack thereof), and userspace
capabilities:
Unless a module parameter is used, thinkpad-acpi will now behave in such a
way that it will work well (by default) with userspace that still uses only
the old ACPI procfs event interface and doesn't care for thinkpad-acpi
input devices.
It will also always work well with userspace that has been updated to use
both the thinkpad-acpi input devices, and ACPI core netlink event
interface, regardless of any module parameter.
The module parameter was added to allow thinkpad-acpi to work with
userspace that has been partially updated to use thinkpad-acpi input
devices, but not the new ACPI core netlink event interface. To use this
mode of hot key reporting, one has to specify the hotkey_report_mode=2
module parameter.
The thinkpad-acpi driver exports the value of hotkey_report_mode through
sysfs, as well. thinkpad-acpi backports to older kernels, that do not
support the new ACPI core netlink interface, have code to allow userspace
to switch hotkey_report_mode at runtime through sysfs. This capability
will not be provided in mainline thinkpad-acpi as it is not needed there.
Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Cc: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Richard Hughes <hughsient@gmail.com>
Signed-off-by: Len Brown <len.brown@intel.com>
* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
[SPARC64]: Warn user if cpu is ignored.
[SPARC64]: Fix lockdep, particularly on SMP.
[SPARC64]: Update defconfig.
* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
[VLAN]: Fix net_device leak.
[PPP] generic: Fix receive path data clobbering & non-linear handling
[PPP] generic: Call skb_cow_head before scribbling over skb
[NET] skbuff: Add skb_cow_head
[BRIDGE]: Kill clone argument to br_flood_*
[PPP] pppoe: Fill in header directly in __pppoe_xmit
[PPP] pppoe: Fix data clobbering in __pppoe_xmit and return value
[PPP] pppoe: Fix skb_unshare_check call position
[SCTP]: Convert bind_addr_list locking to RCU
[SCTP]: Add RCU synchronization around sctp_localaddr_list
[PKT_SCHED]: sch_cbq.c: Shut up uninitialized variable warning
[PKTGEN]: srcmac fix
[IPV6]: Fix source address selection.
[IPV4]: Just increment OutDatagrams once per a datagram.
[IPV6]: Just increment OutDatagrams once per a datagram.
[IPV6]: Fix unbalanced socket reference with MSG_CONFIRM.
[NET_SCHED] protect action config/dump from irqs
[NET]: Fix two issues wrt. SO_BINDTODEVICE.
When CONFIG_ISA is disabled, the isa_driver support will not be compiled
in. Define stubs so that we don't get link-time errors.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In "[VLAN]: Move device registation to seperate function" (commit
e89fe42cd0), a pile of code got moved
to register_vlan_dev(), including grabbing a reference to underlying
device. However, original dev_hold() had been left behind, so we
leak a reference to net_device now...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds missing pskb_may_pull calls to deal with non-linear
packets that may arrive from pppoe or pppol2tp.
It also copies cloned packets before writing over them.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's rude to write over data that other people are still using. So call
skb_cow_head before PPP proceeds to modify the skb data.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds an optimised version of skb_cow that avoids the copy if
the header can be modified even if the rest of the payload is cloned.
This can be used in encapsulating paths where we only need to modify the
header. As it is, this can be used in PPPOE and bridging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The clone argument is only used by one caller and that caller can clone
the packet itself. This patch moves the clone call into the caller and
kills the clone argument.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the hdr variable (which is copied into the skb)
and instead sets the header directly in the skb.
It also uses __skb_push instead of skb_push since we've just checked
using skb_cow for enough head room.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function __pppoe_xmit modifies the skb data and therefore it needs
to copy and skb data if it's cloned.
In fact, it currently allocates a new skb so that it can return 0 in
case of error without freeing the original skb. This is totally wrong
because returning zero is meant to indicate congestion whereupon pppoe
is supposed to wake up the upper layer once the congestion subsides.
This makes sense for ppp_async and ppp_sync but is out-of-place for
pppoe. This patch makes it always return 1 and free the skb.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The skb_unshare_check call needs to be made before pskb_may_pull,
not after.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the sctp_sockaddr_entry is now RCU enabled as part of
the patch to synchronize sctp_localaddr_list, it makes sense to
change all handling of these entries to RCU. This includes the
sctp_bind_addrs structure and it's list of bound addresses.
This list is currently protected by an external rw_lock and that
looks like an overkill. There are only 2 writers to the list:
bind()/bindx() calls, and BH processing of ASCONF-ACK chunks.
These are already seriealized via the socket lock, so they will
not step on each other. These are also relatively rare, so we
should be good with RCU.
The readers are varied and they are easily converted to RCU.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Sridhar Samdurala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sctp_localaddr_list is modified dynamically via NETDEV_UP
and NETDEV_DOWN events, but there is not synchronization
between writer (even handler) and readers. As a result,
the readers can access an entry that has been freed and
crash the sytem.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Sridhar Samdurala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/sch_cbq.c: In function 'cbq_enqueue':
net/sched/sch_cbq.c:383: warning: 'ret' may be used uninitialized in this function
has been verified to be a bogus case. So let's shut it up.
Signed-off-by: Satyam Sharma <satyam@infradead.org>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The commit 95c385 broke proper source address selection for cases in which
there is a address which is makred 'deprecated'. The commit mistakenly
changed ifa->flags to ifa_result->flags (probably copy/paste error from a
few lines above) in the 'Rule 3' address selection code.
The patch restores the previous RFC-compliant behavior.
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
When NR_CPUS is smaller than the cpu probed, let the user
know that the cpu won't be used.
Suggested by Al Viro.
Signed-off-by: David S. Miller <davem@davemloft.net>
As noted by Al Viro, when we try to call prom_set_trap_table()
in the SMP trampoline code we try to take the PROM call spinlock
which doesn't work because the current thread pointer isn't
valid yet and lockdep depends upon that being correct.
Furthermore, we cannot set the current thread pointer register
because it can't be properly dereferenced until we return from
prom_set_trap_table(). Kernel TLB misses only work after that
call.
So do the PROM call to set the trap table directly instead of
going through the OBP library C code, and thus avoid the lock
altogether.
These calls are guarenteed to be serialized fully.
Since there are now no calls to the prom_set_trap_table{_sun4v}()
library functions, they can be deleted.
Signed-off-by: David S. Miller <davem@davemloft.net>
Taking a cpu offline removes the cpu from the online mask before the
CPU_DEAD notification is done. The clock events layer does the cleanup
of the dead CPU from the CPU_DEAD notifier chain. tick_do_timer_cpu is
used to avoid xtime lock contention by assigning the task of jiffies
xtime updates to one CPU. If a CPU is taken offline, then this
assignment becomes stale. This went unnoticed because most of the time
the offline CPU went dead before the online CPU reached __cpu_die(),
where the CPU_DEAD state is checked. In the case that the offline CPU did
not reach the DEAD state before we reach __cpu_die(), the code in there
goes to sleep for 100ms. Due to the stale time update assignment, the
system is stuck forever.
Take the assignment away when a cpu is not longer in the cpu_online_mask.
We do this in the last call to tick_nohz_stop_sched_tick() when the offline
CPU is on the way to the final play_dead() idle entry.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
When a cpu goes offline it is removed from the broadcast masks. If the
mask becomes empty the code shuts down the broadcast device. This is
wrong, because the broadcast device needs to be ready for the online
cpu going idle (into a c-state, which stops the local apic timer).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The jinxed VAIO refuses to resume without hitting keys on the keyboard
when this is not enforced. It is unclear why the cpu ends up in a lower
C State without notifying the clock events layer, but enforcing the
oneshot broadcast here is safe.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reevaluate C/P/T states when a cpu becomes online. This avoids
the caching of the broadcast information in the clockevents layer.
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Len Brown <len.brown@intel.com>
Timekeeping resume adjusts xtime by adding the slept time in seconds and
resets the reference value of the clock source (clock->cycle_last).
clock->cycle last is used to calculate the delta between the last xtime
update and the readout of the clock source in __get_nsec_offset(). xtime
plus the offset is the current time. The resume code ignores the delta
which had already elapsed between the last xtime update and the actual
time of suspend. If the suspend time is short, then we can see time
going backwards on resume.
Suspend:
offs_s = clock->read() - clock->cycle_last;
now = xtime + offs_s;
timekeeping_suspend_time = read_rtc();
Resume:
sleep_time = read_rtc() - timekeeping_suspend_time;
xtime.tv_sec += sleep_time;
clock->cycle_last = clock->read();
offs_r = clock->read() - clock->cycle_last;
now = xtime + offs_r;
if sleep_time_seconds == 0 and offs_r < offs_s, then time goes
backwards.
Fix this by storing the offset from the last xtime update and add it to
xtime during resume, when we reset clock->cycle_last:
sleep_time = read_rtc() - timekeeping_suspend_time;
xtime.tv_sec += sleep_time;
xtime += offs_s; /* Fixup xtime offset at suspend time */
clock->cycle_last = clock->read();
offs_r = clock->read() - clock->cycle_last;
now = xtime + offs_r;
Thanks to Marcelo for tracking this down on the OLPC and providing the
necessary details to analyze the root cause.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: John Stultz <johnstul@us.ibm.com>
Cc: Tosatti <marcelo@kvack.org>
Lockdep complains about the access of rtc in timekeeping_suspend
inside the interrupt disabled region of the write locked xtime lock.
Move the access outside.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: John Stultz <johnstul@us.ibm.com>