The address based work estimate for unmapping (for lockbreak) is and always
was horribly inefficient for sparse mappings. The problem is most simply
explained with an example:
If we find a pgd is clear, we still have to call into unmap_page_range
PGDIR_SIZE / ZAP_BLOCK_SIZE times, each time checking the clear pgd, in
order to progress the working address to the next pgd.
The fundamental way to solve the problem is to keep track of the end
address we've processed and pass it back to the higher layers.
From: Nick Piggin <npiggin@suse.de>
Modification to completely get away from address based work estimate
and instead use an abstract count, with a very small cost for empty
entries as opposed to present pages.
On 2.6.14-git2, ppc64, and CONFIG_PREEMPT=y, mapping and unmapping 1TB
of virtual address space takes 1.69s; with the following patch applied,
this operation can be done 1000 times in less than 0.01s
From: Andrew Morton <akpm@osdl.org>
With CONFIG_HUTETLB_PAGE=n:
mm/memory.c: In function `unmap_vmas':
mm/memory.c:779: warning: division by zero
Due to
zap_work -= (end - start) /
(HPAGE_SIZE / PAGE_SIZE);
So make the dummy HPAGE_SIZE non-zero
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Changed jobs and the Freescale address is no longer valid.
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Since few people need the support anymore, this moves the legacy
pm_xxx functions to CONFIG_PM_LEGACY, and include/linux/pm_legacy.h.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The file_lock spinlock sits close to mostly read fields of 'struct
files_struct'
In SMP (and NUMA) environments, each time a thread wants to open or close
a file, it has to acquire the spinlock, thus invalidating the cache line
containing this spinlock on other CPUS. So other threads doing
read()/write()/... calls that use RCU to access the file table are going
to ask further memory (possibly NUMA) transactions to read again this
memory line.
Move the spinlock to another cache line, so that concurrent threads can
share the cache line containing 'count' and 'fdt' fields.
It's worth up to 9% on a microbenchmark using a 4-thread 2-package x86
machine. See
http://marc.theaimsgroup.com/?l=linux-kernel&m=112680448713342&w=2
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
On Fri, Nov 11, 2005 at 12:58:40PM -0800, David S. Miller wrote:
>
> This change:
>
> diff-tree 8ca2bdc7a9 (from feee207e44d3643d19e648aAuthor: Christoph Hellwig <hch@lst.de>
> Date: Wed Nov 9 12:07:18 2005 -0800
>
> [SPARC] sbus rtc: implement ->compat_ioctl
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Signed-off-by: David S. Miller <davem@davemloft.net>
>
> results in the console now getting spewed on sparc64 systems
> with messages like:
>
> [ 11.968298] ioctl32(hwclock:464): Unknown cmd fd(3) cmd(401c7014){00} arg(efc
> What's happening is hwclock tries first the SBUS rtc device ioctls
> then the normal rtc driver ones.
>
> So things actually worked better when we had the SBUS rtc compat ioctl
> directly handled via the generic compat ioctl code.
>
> There are _so_ many rtc drivers in the kernel implementing the
> generic rtc ioctls that I don't think putting a ->compat_ioctl
> into all of them to fix this problem is feasible. Unless we
> write a single rtc_compat_ioctl(), export it to modules, and hook
> it into all of those somehow.
>
> But even that doesn't appear to have any pretty implementation.
>
> Any better ideas?
We had similar problems with other ioctls where userspace did things
like that. What we did there was to put the compat handler to generic
code. The patch below does that, adding a big comment about what's
going on and removing the COMPAT_IOCTL entires for these on powerpc
that not only weren't ever useful but are duplicated now aswell.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds request_queue->nr_sorted which keeps the number of
requests in the iosched and implement elv_drain_elevator which
performs forced dispatching. elv_drain_elevator checks whether
iosched actually dispatches all requests it has and prints error
message if it doesn't. As buggy forced dispatching can result in
wrong barrier operations, I think this extra check is worthwhile.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jens Axboe <axboe@suse.de>
Also introduces a sysctl option to configure the receive buffer
accounting policy to be either at socket or association level.
Default is all the associations on the same socket share the
receive buffer.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On ia64, it is possible to get NaT Consumption Fault and a kernel panic
when initializing sctp sideeffect commands arguments. The union
sctp_arg_t contains different sized elements and when loading a smaller
sized element (32 or 16 bits), it is possible for a speculative load to
fail and result in a NaT bit set which causes a kernel crash. The easy
way to get around it is to load the largerst member of the union.
Signed-off-by: Vladislav Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The socket level timeout values are maintained in sctp_sock and
association level timeouts are in sctp_association. So there is
no need for ep->timeouts.
Signed-off-by: Vladislav Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch introduces 4-level page tables to ia64. I have run
some benchmarks and found nothing interesting. Performance has
consistently fallen within the noise range.
It also introduces a config option (setting the default to 3
levels). The config option prevents having 4 level page
tables with 64k base page size.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This will let me chop the code size of several drivers right down. In
many cases the actual private data is very useful and constant for a
given host controller so being able to just pass it at probe time would
be very useful indeed (eg with the via driver would could pass the udma
clocking and reduce the code size, or with the AMD one the UDMA
multiplier and the offset)
Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
This patch moves the vdso's to arch/powerpc, adds support for the 32
bits vdso to the 32 bits kernel, rename systemcfg (finally !), and adds
some new (still untested) routines to both vdso's: clock_gettime() with
support for CLOCK_REALTIME and CLOCK_MONOTONIC, clock_getres() (same
clocks) and get_tbfreq() for glibc to retreive the timebase frequency.
Tom,Steve: The implementation of get_tbfreq() I've done for 32 bits
returns a long long (r3, r4) not a long. This is such that if we ever
add support for >4Ghz timebases on ppc32, the userland interface won't
have to change.
I have tested gettimeofday() using some glibc patches in both ppc32 and
ppc64 kernels using 32 bits userland (I haven't had a chance to test a
64 bits userland yet, but the implementation didn't change and was
tested earlier). I haven't tested yet the new functions.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Since the udbg code in ppc64 has no ppc32 equivalent, move it straight
over into arch/powerpc (and include/asm-powerpc for udbg.h). In time,
we probably want to meld the various bits and pieces of 32-bit early
debugging code into udbg, but for now only include it on
CONFIG_PPC64=y builds. The only change during the move is to
standardise the protecting #ifdef/#define in udbg.h, and move its
banner comment above the initial #ifdef (which seems to be normal
practice).
Built and booted on POWER5 LPAR (ARCH=powerpc and ARCH=ppc64). Built
for 32bit multiplatform (ARCH=powerpc).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
The definitions in sparsemem.h arent sufficient. We currently sell
machines with 2TB of RAM, and in order to give us room for a few years
growth lets set it to 16TB.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Convert to sparsemem and remove all the discontigmem code in the
process. This has a few advantages:
- The old numa_memory_lookup_table can go away
- All the arch specific discontigmem magic can go away
We also remove the triple pass of memory properties and instead create a
list of per node extents that we iterate through. A final cleanup would
be to change our lmb code to store extents per node, then we can reuse
that information in the numa code.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Remove ppc64 specific version of nr_cpus_node and use the generic one
provided.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Remove an unused numa define and move a discontigmem specific define
inside the relevant ifdef.
I will submit a separate patch to remove them from other architectures,
but the ppc64 patches to follow depend on this.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
The bit position in the status register corresponding to the
PCI DMA interrupt was incorrect. Additionally, we did not
have a define for the PCI DMA interrupt.
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
On Alpha:
include/linux/libata.h: In function `ata_pad_alloc':
include/linux/libata.h:785: warning: implicit declaration of function `dma_alloc_coherent'
include/linux/libata.h:786: warning: assignment makes pointer from integer without a cast
include/linux/libata.h: In function `ata_pad_free':
include/linux/libata.h:792: warning: implicit declaration of function `dma_free_coherent'
(I have a decouple-some-header-files cleanup in -mm, so it's causing some
fallout of this nature)
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Use "hints" to speed up the SACK processing. Various forms
of this have been used by TCP developers (Web100, STCP, BIC)
to avoid the 2x linear search of outstanding segments.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is an updated version of the RFC3465 ABC patch originally
for Linux 2.6.11-rc4 by Yee-Ting Li. ABC is a way of counting
bytes ack'd rather than packets when updating congestion control.
The orignal ABC described in the RFC applied to a Reno style
algorithm. For advanced congestion control there is little
change after leaving slow start.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move all the code that does linear TCP slowstart to one
inline function to ease later patch to add ABC support.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP peformance with TSO over networks with delay is awful.
On a 100Mbit link with 150ms delay, we get 4Mbits/sec with TSO and
50Mbits/sec without TSO.
The problem is with TSO, we intentionally do not keep the maximum
number of packets in flight to fill the window, we hold out to until
we can send a MSS chunk. But, we also don't update the congestion window
unless we have filled, as per RFC2861.
This patch replaces the check for the congestion window being full
with something smarter that accounts for TSO.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
MSI hardcoded delivery mode to use logical delivery mode. Recently
x86_64 moved to use physical mode addressing to support physflat mode.
With this mode enabled noticed that my eth with MSI werent working.
msi_address_init() was hardcoded to use logical mode for i386 and x86_64.
So when we switch to use physical mode, things stopped working.
Since anyway we dont use lowest priority delivery with MSI, its always
directed to just a single CPU. Its safe and simpler to use
physical mode always, even when we use logical delivery mode for IPI's
or other ioapic RTE's.
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
A nice feature of sysfs is that it can create the symlink from the
driver to the module that is contained in it.
It requires that the device_driver.owner is set, what is not the
case for many PCI drivers.
This patch allows pci_register_driver to set automatically the
device_driver.owner for any PCI driver.
Credits to Al Viro who suggested the method.
Signed-off-by: Laurent Riffard <laurent.riffard@free.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
--
drivers/ide/setup-pci.c | 12 +++++++-----
drivers/pci/pci-driver.c | 9 +++++----
include/linux/ide.h | 3 ++-
include/linux/pci.h | 10 ++++++++--
4 files changed, 22 insertions(+), 12 deletions(-)
This patch tweaks the way pciehp requests control of the hotplug
hardware from BIOS. It now tries to invoke the ACPI _OSC method
for a specific hotplug controller only, rather than walking the
entire acpi namespace invoking all possible _OSC methods under
all host bridges. This allows us to gain control of each hotplug
controller individually, even if BIOS fails to give us control of
some other hotplug controller in the system.
Signed-off-by: Rajesh Shah <rajesh.shah@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Some devices have more than one capability of the same type. For
example, the PCI header for the PathScale InfiniPath looks like:
04:01.0 InfiniBand: Unknown device 1fc1:000d (rev 02)
Subsystem: Unknown device 1fc1:000d
Flags: bus master, fast devsel, latency 0, IRQ 193
Memory at fea00000 (64-bit, non-prefetchable) [size=2M]
Capabilities: [c0] HyperTransport: Slave or Primary Interface
Capabilities: [f8] HyperTransport: Interrupt Discovery and Configuration
There are _two_ HyperTransport capabilities, and the PathScale driver
wants to look at both of them.
The current pci_find_capability() API doesn't work for this, since it
only allows us to get to the first capability of a given type. The
patch below introduces a new pci_find_next_capability(), which can be
used in a loop like
for (pos = pci_find_capability(pdev, <ID>);
pos;
pos = pci_find_next_capability(pdev, pos, <ID>)) {
/* ... */
}
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The protocol field in ethernet headers is big-endian and should be
annotated as such. This patch allows detection of missing ntohs() calls
on the ethernet protocol field when sparse is run with __CHECK_ENDIAN__
defined.
This is a revised version that includes <linux/types.h> so that the
userspace programs are not confused by __be16. Thanks to David S.
Miller.
Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Here is the patch that introduces the generic skb_checksum_complete
which also checks for hardware RX checksum faults. If that happens,
it'll call netdev_rx_csum_fault which currently prints out a stack
trace with the device name. In future it can turn off RX checksum.
I've converted every spot under net/ that does RX checksum checks to
use skb_checksum_complete or __skb_checksum_complete with the
exceptions of:
* Those places where checksums are done bit by bit. These will call
netdev_rx_csum_fault directly.
* The following have not been completely checked/converted:
ipmr
ip_vs
netfilter
dccp
This patch is based on patches and suggestions from Stephen Hemminger
and David S. Miller.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>