Merge x86/dumpstack into tracing/ftrace because upcoming ftrace changes
depend on cleanups already in x86/dumpstack.
Also merge to latest upstream -rc.
The previous patch from Alan Cox ("nfsd: fix vm overcommit crash",
commit 731572d39f) fixed the problem where
knfsd crashes on exported shmemfs objects and strict overcommit is set.
But the patch forgot supporting the case when CONFIG_SECURITY is
disabled.
This patch copies a part of his fix which is mainly for detecting a bug
earlier.
Acked-by: James Morris <jmorris@namei.org>
Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Junjiro R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It seems that on some nVidia controllers using AltStatus register
can be unreliable so default to Status register if the PCI device
is in Compatibility Mode. In order to achieve this:
* Add ide_pci_is_in_compatibility_mode() inline helper to <linux/ide.h>.
* Add IDE_HFLAG_BROKEN_ALTSTATUS host flag and set it in amd74xx host
driver for nVidia controllers in Compatibility Mode.
* Teach actual_try_to_identify() and drive_is_ready() about the new flag.
This fixes the regression caused by removal of CONFIG_IDEPCI_SHARE_IRQ
config option in 2.6.25 and using AltStatus register unconditionally when
available (kernel.org bugs #11659 and #10216).
[ Moreover for CONFIG_IDEPCI_SHARE_IRQ=y (which is what most people
and distributions use) it never worked correctly. ]
Thanks to Remy LABENE and Lars Winterfeld for help with debugging the problem.
More info at:
http://bugzilla.kernel.org/show_bug.cgi?id=11659http://bugzilla.kernel.org/show_bug.cgi?id=10216
Reported-by: Remy LABENE <remy.labene@free.fr>
Tested-by: Remy LABENE <remy.labene@free.fr>
Tested-by: Lars Winterfeld <lars.winterfeld@tu-ilmenau.de>
Acked-by: Borislav Petkov <petkovbb@gmail.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
2nd part of the fixes needed for
http://bugzilla.kernel.org/show_bug.cgi?id=11796.
When the idr tree is either grown or shrunk, then the update to the number
of layers and the top pointer were not atomic. This race caused crashes.
The attached patch fixes that by replicating the layers counter in each
layer, thus idr_find doesn't need idp->layers anymore.
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Clement Calmels <cboulte@gmail.com>
Cc: Nadia Derbey <Nadia.Derbey@bull.net>
Cc: Pierre Peiffer <peifferp@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It has been thought that the per-user file descriptors limit would also
limit the resources that a normal user can request via the epoll
interface. Vegard Nossum reported a very simple program (a modified
version attached) that can make a normal user to request a pretty large
amount of kernel memory, well within the its maximum number of fds. To
solve such problem, default limits are now imposed, and /proc based
configuration has been introduced. A new directory has been created,
named /proc/sys/fs/epoll/ and inside there, there are two configuration
points:
max_user_instances = Maximum number of devices - per user
max_user_watches = Maximum number of "watched" fds - per user
The current default for "max_user_watches" limits the memory used by epoll
to store "watches", to 1/32 of the amount of the low RAM. As example, a
256MB 32bit machine, will have "max_user_watches" set to roughly 90000.
That should be enough to not break existing heavy epoll users. The
default value for "max_user_instances" is set to 128, that should be
enough too.
This also changes the userspace, because a new error code can now come out
from EPOLL_CTL_ADD (-ENOSPC). The EMFILE from epoll_create() was already
listed, so that should be ok.
[akpm@linux-foundation.org: use get_current_user()]
Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: <stable@kernel.org>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Reported-by: Vegard Nossum <vegardno@ifi.uio.no>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Impact: extend information in /proc/sched_debug
This patch adds uid information in sched_debug for CONFIG_USER_SCHED
Signed-off-by: Arun R Bharadwaj <arun@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
libata: blacklist Seagate drives which time out FLUSH_CACHE when used with NCQ
[libata] pata_rb532_cf: fix signature of the xfer function
[libata] pata_rb532_cf: fix and rename register definitions
ata_piix: add borked Tecra M4 to broken suspend list
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
IB/mlx4: Fix MTT leakage in resize CQ
IB/ehca: Fix problem with generated flush work completions
IB/ehca: Change misleading error message on memory hotplug
mlx4_core: Save/restore default port IB capability mask
Some recent Seagate harddrives have firmware bug which causes FLUSH
CACHE to timeout under certain circumstances if NCQ is being used.
This can be worked around by disabling NCQ and fixed by updating the
firmware. Implement ATA_HORKAGE_FIRMWARE_UPDATE and blacklist these
devices.
The wiki page has been updated to contain information on this issue.
http://ata.wiki.kernel.org/index.php/Known_issues
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
* master.kernel.org:/home/rmk/linux-2.6-arm:
Allow architectures to override copy_user_highpage()
[ARM] pxa/palmtx: misc fixes to use generic GPIO API
ARM: OMAP: Fixes for suspend / resume GPIO wake-up handling
[ARM] pxa/corgi: update default config to exclude tosa from being built
[ARM] pxa/pcm990: use negative number for an invalid GPIO in camera data
ARM: OMAP: Typo fix for clock_allow_idle
ARM: OMAP: Remove broken LCD driver for SX1
[ARM] 5335/1: pxa25x_udc: Fix is_vbus_present to return 1 or 0
[ARM] pxa/MioA701: bluetooth resume fix
[ARM] pxa/MioA701: fix memory corruption.
* 'irq-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
irq.h: fix missing/extra kernel-doc
genirq: __irq_set_trigger: change pr_warning to pr_debug
irq: fix typo
x86: apic honour irq affinity which was set in early boot
genirq: fix the affinity setting in setup_irq
genirq: keep affinities set from userspace across free/request_irq()
All architectures now use the generic compat_sys_ptrace, as should every
new architecture that needs 32bit compat (if we'll ever get another).
Remove the now superflous __ARCH_WANT_COMPAT_SYS_PTRACE define, and also
kill a comment about __ARCH_SYS_PTRACE that was added after
__ARCH_SYS_PTRACE was already gone.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Same as for hotplug_cpu - we want static notifier_block in there in meminitdata,
to avoid false positives whenever it's used.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 7ff93f8b ("mlx4_core: Multiple port type support") introduced
support for different port types. As part of that support, SET_PORT
is invoked to set the port type during driver startup. However, as a
side-effect, for IB ports the invocation of this command also sets the
port's capability mask to zero (losing the default value set by FW).
To fix this, get the default ib port capabilities (via a MAD_IFC Port
Info query) during driver startup, and save them for use in the
mlx4_SET_PORT command when setting the port-type to Infiniband.
This patch fixes problems with subnet manager (SM) failover such as
<https://bugs.openfabrics.org/show_bug.cgi?id=1183>, which occurred
because the IsTrapSupported bit in the capability mask was zeroed.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
This patch adds the power management support into the physical
abstraction layer.
Suspend and resume functions respectively turns on/off the bit 11
into the PHY Basic mode control register.
Generic PHY device starts supporting PM.
In order to support the wake-on LAN and avoid to put in power down
the PHY device, the MDIO is aware of what the Ethernet device wants to do.
Voluntary, no CONFIG_PM defines were added into the sources.
Also generic suspend/resume functions are exported to allow
other drivers use them (such as genphy_config_aneg etc.).
Within the phy_driver_register function, we need to remove the
memset. It overrides the device driver owner and it is not good.
Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add document and comments on marker_synchronize_unregister(): it
should be called before freeing resources that the probes depend on.
Based on comments from Lai Jiangshan and Mathieu Desnoyers.
Signed-off-by: Wu Fengguang <wfg@linux.intel.com>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We merge this branch because x86/debug touches code that we started
cleaning up in x86/irq. The two branches started out independent,
but as unexpected amount of activity went into x86/irq, they became
dependent. Resolve that by this cross-merge.
Impact: remove unused code
__local_bh_enable has been replaced with _local_bh_enable.
As comments says "it always nests inside local_bh_enable() sections"
has not been valid now. Also there is no reason to use __local_bh_enable
anywhere, so we can remove this useless function.
Signed-off-by: Liming Wang <liming.wang@windriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
With aliasing VIPT cache support, the ARM implementation of
clear_user_page() and copy_user_page() sets up a temporary kernel space
mapping such that we have the same cache colour as the userspace page.
This avoids having to consider any userspace aliases from this operation.
However, when highmem is enabled, kmap_atomic() have to setup mappings.
The copy_user_highpage() and clear_user_highpage() call these functions
before delegating the copies to copy_user_page() and clear_user_page().
The effect of this is that each of the *_user_highpage() functions setup
their own kmap mapping, followed by the *_user_page() functions setting
up another mapping. This is rather wasteful.
Thankfully, copy_user_highpage() can be overriden by architectures by
defining __HAVE_ARCH_COPY_USER_HIGHPAGE. However, replacement of
clear_user_highpage() is more difficult because its inline definition
is not conditional. It seems that you're expected to define
__HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE and provide a replacement
__alloc_zeroed_user_highpage() implementation instead.
The allocation itself is fine, so we don't want to override that. What
we really want to do is to override clear_user_highpage() with our own
version which doesn't kmap_atomic() unnecessarily.
Other VIPT architectures (PARISC and SH) would also like to override
this function as well.
Acked-by: Hugh Dickins <hugh@veritas.com>
Acked-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
entry_32.S is now the only user of KPROBE_ENTRY / KPROBE_END,
treewide. This patch reorders entry_64.S and explicitly generates
a separate section for functions that need the protection. The
generated code before and after the patch is equal.
The KPROBE_ENTRY and KPROBE_END macro's are removed too.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to be consistent with NL80211_ATTR_POWER_RULE_MAX_EIRP,
change NL80211_FREQUENCY_ATTR_MAX_TX_POWER to use mBm and U32 instead
of dBm and U8. This is a userspace interface change, but the previous
version had not yet been pushed upstream and there are no userspace
programs using this yet, so there is justification to get this change in
as long as it goes in before the previous version gets out.
Signed-off-by: Jouni Malinen <j@w1.fi>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The rfkill class API requires that the driver connected to a class
call rfkill_force_state() on resume to update the real state of the
rfkill controller, OR that it provides a get_state() hook.
This means there is potentially a hidden call in the resume code flow
that changes rfkill->state (i.e. rfkill_force_state()), so the
previous state of the transmitter was being lost.
The simplest and most future-proof way to fix this is to explicitly
store the pre-sleep state on the rfkill structure, and restore from
that on resume.
Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This is useful information to provide for userspace (e.g., hostapd needs
this to generate Country IE).
Signed-off-by: Jouni Malinen <jouni.malinen@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch replaces __builtin_return_address(0) with _RET_IP_, since a
previous patch moved _RET_IP_ and _THIS_IP_ to include/linux/kernel.h and
they're widely available now. This makes for shorter and easier to read
code.
[penberg@cs.helsinki.fi: remove _RET_IP_ casts to void pointer]
Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Add asl, pzl and di debugfs files to uwb/uwbN/wusbhc for WHCI host
controller. These dump the current ASL, PZL and DI buffer.
Signed-off-by: David Vrabel <david.vrabel@csr.com>
This was a forward port of work done by Mathieu Desnoyers, I changed it to
encode the 'what' parameter on the tracepoint name, so that one can register
interest in specific events and not on classes of events to then check the
'what' parameter.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: new "power-tracer" ftrace plugin
This patch adds a C/P-state ftrace plugin that will generate
detailed statistics about the C/P-states that are being used,
so that we can look at detailed decisions that the C/P-state
code is making, rather than the too high level "average"
that we have today.
An example way of using this is:
mount -t debugfs none /sys/kernel/debug
echo cstate > /sys/kernel/debug/tracing/current_tracer
echo 1 > /sys/kernel/debug/tracing/tracing_enabled
sleep 1
echo 0 > /sys/kernel/debug/tracing/tracing_enabled
cat /sys/kernel/debug/tracing/trace | perl scripts/trace/cstate.pl > out.svg
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: more efficient code for ftrace graph tracer
This patch uses the dynamic patching, when available, to patch
the function graph code into the kernel.
This patch will ease the way for letting both function tracing
and function graph tracing run together.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: feature
This patch sets a C-like output for the function graph tracing.
For this aim, we now call two handler for each function: one on the entry
and one other on return. This way we can draw a well-ordered call stack.
The pid of the previous trace is loosely stored to be compared against
the one of the current trace to see if there were a context switch.
Without this little feature, the call tree would seem broken at
some locations.
We could use the sched_tracer to capture these sched_events but this
way of processing is much more simpler.
2 spaces have been chosen for indentation to fit the screen while deep
calls. The time of execution in nanosecs is printed just after closed
braces, it seems more easy this way to find the corresponding function.
If the time was printed as a first column, it would be not so easy to
find the corresponding function if it is called on a deep depth.
I plan to output the return value but on 32 bits CPU, the return value
can be 32 or 64, and its difficult to guess on which case we are.
I don't know what would be the better solution on X86-32: only print
eax (low-part) or even edx (high-part).
Actually it's thee same problem when a function return a 8 bits value, the
high part of eax could contain junk values...
Here is an example of trace:
sys_read() {
fget_light() {
} 526
vfs_read() {
rw_verify_area() {
security_file_permission() {
cap_file_permission() {
} 519
} 1564
} 2640
do_sync_read() {
pipe_read() {
__might_sleep() {
} 511
pipe_wait() {
prepare_to_wait() {
} 760
deactivate_task() {
dequeue_task() {
dequeue_task_fair() {
dequeue_entity() {
update_curr() {
update_min_vruntime() {
} 504
} 1587
clear_buddies() {
} 512
add_cfs_task_weight() {
} 519
update_min_vruntime() {
} 511
} 5602
dequeue_entity() {
update_curr() {
update_min_vruntime() {
} 496
} 1631
clear_buddies() {
} 496
update_min_vruntime() {
} 527
} 4580
hrtick_update() {
hrtick_start_fair() {
} 488
} 1489
} 13700
} 14949
} 16016
msecs_to_jiffies() {
} 496
put_prev_task_fair() {
} 504
pick_next_task_fair() {
} 489
pick_next_task_rt() {
} 496
pick_next_task_fair() {
} 489
pick_next_task_idle() {
} 489
------------8<---------- thread 4 ------------8<----------
finish_task_switch() {
} 1203
do_softirq() {
__do_softirq() {
__local_bh_disable() {
} 669
rcu_process_callbacks() {
__rcu_process_callbacks() {
cpu_quiet() {
rcu_start_batch() {
} 503
} 1647
} 3128
__rcu_process_callbacks() {
} 542
} 5362
_local_bh_enable() {
} 587
} 8880
} 9986
kthread_should_stop() {
} 669
deactivate_task() {
dequeue_task() {
dequeue_task_fair() {
dequeue_entity() {
update_curr() {
calc_delta_mine() {
} 511
update_min_vruntime() {
} 511
} 2813
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
This patch changes the name of the "return function tracer" into
function-graph-tracer which is a more suitable name for a tracing
which makes one able to retrieve the ordered call stack during
the code flow.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This adds country IE parsing to mac80211 and enables its usage
within the new regulatory infrastructure in cfg80211. We parse
the country IEs only on management beacons for the BSSID you are
associated to and disregard the IEs when the country and environment
(indoor, outdoor, any) matches the already processed country IE.
To avoid following misinformed or outdated APs we build and use
a regulatory domain out of the intersection between what the AP
provides us on the country IE and what CRDA is aware is allowed
on the same country.
A secondary device is allowed to follow only the same country IE
as it make no sense for two devices on a system to be in two
different countries.
In the case the AP is using country IEs for an incorrect country
the user may help compliance further by setting the regulatory
domain before or after the IE is parsed and in that case another
intersection will be performed.
CONFIG_WIRELESS_OLD_REGULATORY is supported but requires CRDA
present.
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Impact: restructure DS memory allocation to be done by the usage site of DS
Require pre-allocated buffers in ds.h.
Move the BTS buffer allocation for ptrace into ptrace.c.
The pointer to the allocated buffer is stored in the traced task's
task_struct together with the handle returned by ds_request_bts().
Removes memory accounting code.
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: generalize the DS code to shared buffers
Change the in-kernel ds.h interface to identify the tracer via a
handle returned on ds_request_~().
Tracers used to be identified via their task_struct.
The changes are required to allow DS to be shared between different
tasks, which is needed for perfmon2 and for ftrace.
For ptrace, the handle is stored in the traced task's task_struct.
This should probably go into a (arch-specific) ptrace context some
time.
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup, move all hrtimer processing into hardirq context
This is an attempt at removing some of the hrtimer complexity by
reducing the number of callback modes to 1.
This means that all hrtimer callback functions will be ran from HARD-irq
context.
I went through all the 30 odd hrtimer callback functions in the kernel
and saw only one that I'm not quite sure of, which is the one in
net/can/bcm.c - hence I'm CC-ing the folks responsible for that code.
Furthermore, the hrtimer core now calls callbacks directly with IRQs
disabled in case you try to enqueue an expired timer. If this timer is a
periodic timer (which should use hrtimer_forward() to advance its time)
then it might be possible to end up in an inf. recursive loop due to the
fact that hrtimer_forward() doesn't round up to the next timer
granularity, and therefore keeps on calling the callback - obviously
this needs a fix.
Aside from that, this seems to compile and actually boot on my dual core
test box - although I'm sure there are some bugs in, me not hitting any
makes me certain :-)
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since the netlink option for DCB is necessary to actually be useful,
simplified the Kconfig option. In addition, added useful help text for the
Kconfig option.
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As a concession to vendors who have to deal with one source for different
kernel versions, add a HAVE_NET_DEVICE_OPS so they don't end up hard
coding ifdef against kernel version.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Stephen Rothwell reported tht this warning started triggering in
linux-next:
In file included from init/main.c:27:
include/linux/tty.h: In function ‘tty_kref_get’:
include/linux/tty.h:330: warning: ‘______f’ is static but declared in inline function ‘tty_kref_get’ which is not static
Which gcc emits for 'extern inline' functions that nevertheless define
static variables. Change it to 'static inline', which is the norm
in the kernel anyway.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
