Since the uCode hasn't been released (yet?),
warn only if using older than API 4, but load
anything up to API 6.
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
With the embed percpu first chunk allocator, x86 uses either PAGE_SIZE
or PMD_SIZE for atom_size. PMD_SIZE is used when CPU supports PSE so
that percpu areas are aligned to PMD mappings and possibly allow using
PMD mappings in vmalloc areas in the future. Using larger atom_size
doesn't waste actual memory; however, it does require larger vmalloc
space allocation later on for !first chunks.
With reasonably sized vmalloc area, PMD_SIZE shouldn't be a problem
but x86_32 at this point is anything but reasonable in terms of
address space and using larger atom_size reportedly leads to frequent
percpu allocation failures on certain setups.
As there is no reason to not use PMD_SIZE on x86_64 as vmalloc space
is aplenty and most x86_64 configurations support PSE, fix the issue
by always using PMD_SIZE on x86_64 and PAGE_SIZE on x86_32.
v2: drop cpu_has_pse test and make x86_64 always use PMD_SIZE and
x86_32 PAGE_SIZE as suggested by hpa.
Signed-off-by: Tejun Heo <tj@kernel.org> Reported-by: Yanmin Zhang <yanmin.zhang@intel.com> Reported-by: ShuoX Liu <shuox.liu@intel.com> Acked-by: H. Peter Anvin <hpa@zytor.com>
LKML-Reference: <4F97BA98.6010001@intel.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The accessing PCI configuration space with the PCI BIOS32 service does
not work in PV guests.
On systems without MMCONFIG or where the BIOS hasn't marked the
MMCONFIG region as reserved in the e820 map, the BIOS service is
probed (even though direct access is preferred) and this hangs.
Acked-by: Jan Beulich <jbeulich@suse.com> Signed-off-by: David Vrabel <david.vrabel@citrix.com>
[v1: Fixed compile error when CONFIG_PCI is not set] Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
[bwh: Backported to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
which is due to the fact we are trying to access a PFN that is not
accessible to us. The reason (at least in this case) was that
PGD[256] is set to __HYPERVISOR_VIRT_START which was setup (by the
hypervisor) to point to a read-only linear map of the MFN->PFN array.
During our parsing we would get the MFN (a valid one), try to look
it up in the MFN->PFN tree and find it invalid and return ~0 as PFN.
Then pte_mfn_to_pfn would happilly feed that in, attach the flags
and return it back to the caller. 'ptdump_show' bitshifts it and
gets and invalid value that it tries to dereference.
Instead of doing all of that, we detect the ~0 case and just
return !_PAGE_PRESENT.
This bug has been in existence .. at least until 2.6.37 (yikes!)
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Chris Wilson dug out a hw erratum saying that there's noise on the
interrupt line on i945G chips. We also have a bug report from a i945GM
chip with an sdvo hotplug interrupt storm (and no apparent cause).
Play it safe and disable sdvo hotplug on all i945 variants.
Note that this is a regression that has been introduced in 3.1,
when we've enabled sdvo hotplug support with
drm/i915: Enable SDVO hotplug interrupts for HDMI and DVI
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=38442 Reported-and-tested-by: Dominik Köppl <dominik@devwork.org> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Commit 4e8ee7de227e3ab9a72040b448ad728c5428a042 (ARM: SMP: use
idmap_pgd for mapping MMU enable during secondary booting)
switched secondary boot to use idmap_pgd, which is initialized
during early_initcall, instead of a page table initialized during
__cpu_up. This causes idmap_pgd to contain the static mappings
but be missing all dynamic mappings.
If a console is registered that creates a dynamic mapping, the
printk in secondary_start_kernel will trigger a data abort on
the missing mapping before the exception handlers have been
initialized, leading to a hang. Initial boot is not affected
because no consoles have been registered, and resume is usually
not affected because the offending console is suspended.
Onlining a cpu with hotplug triggers the problem.
A workaround is to the printk in secondary_start_kernel until
after the page tables have been switched back to init_mm.
Signed-off-by: Colin Cross <ccross@android.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The inline assembly in kernel_execve() uses r8 and r9. Since this
code sequence does not return, it usually doesn't matter if the
register clobber list is accurate. However, I saw a case where a
particular version of gcc used r8 as an intermediate for the value
eventually passed to r9. Because r8 is used in the inline
assembly, and not mentioned in the clobber list, r9 was set
to an incorrect value.
This resulted in a kernel panic on execution of the first user-space
program in the system. r9 is used in ret_to_user as the thread_info
pointer, and if it's wrong, bad things happen.
Signed-off-by: Tim Bird <tim.bird@am.sony.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
* commit f9dfbf9 "ASoC: tlv320aic23: convert to soc-cache" leads to
a bug preventing resumeof the codec as regmap expects a 9 bits data
register but 0xFFFF is passed in tlv320aic23_set_bias_level and this
values gets cached preventing any write to the TLV320AIC23_PWR
register as the final value produced by regmap is (register << 9) | value
* this patch solves the problem by only working on the 9 bits the
register contains.
Signed-off-by: Eric Bénard <eric@eukrea.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Commit ec81aecb2966 ("hfs: fix a potential buffer overflow") fixed a few
potential buffer overflows in the hfs filesystem. But as Timo Warns
pointed out, these changes also need to be made on the hfsplus
filesystem as well.
Reported-by: Timo Warns <warns@pre-sense.de> Acked-by: WANG Cong <amwang@redhat.com> Cc: Alexey Khoroshilov <khoroshilov@ispras.ru> Cc: Miklos Szeredi <mszeredi@suse.cz> Cc: Sage Weil <sage@newdream.net> Cc: Eugene Teo <eteo@redhat.com> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@lst.de> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Dave Anderson <anderson@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Newer devices have 20 (5000 series) or 30 (6000 series)
hardware queues, rather than the 16 that 4965 had. This
was added to the driver a long time ago, but improperly:
the queue registers for the higher queues aren't just
continuations of the registers for the first 16 queues,
they are in other places. Therefore, the hardware would
lock up when trying to activate queue 16 or above and
the device would have to be restarted.
Thanks goes to Emmanuel who identified this and told me
how the queue programming should be done.
Note that we don't use queues 20 and higher today and
doing so needs more work than this.
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
ctx->vif is dereferenced in different part of iwlwifi code, so do not
nullify it.
This should address at least one of the possible reasons of WARNING at
iwlagn_mac_remove_interface, and perhaps some random crashes when
firmware reset is performed.
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com> Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
[bwh: Backported to 3.2:
- Change filename iwl-mac80211.c to iwl-core.c
- Change context in iwlagn_prepare_restart()] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This driver currently leaves elp_work behind when stopping, which
occasionally results in data corruption because work function ends
up accessing freed memory, typical symptoms of this are various
worker_thread crashes. Fix it by cancelling elp_work.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Currently SDIO glue frees it's own structure before calling
wl1251_free_hw(), which in turn calls ieee80211_unregister_hw().
The later call may result in a need to communicate with the chip
to stop it (as it happens now if the interface is still up before
rmmod), which means calls are made back to the glue, resulting in
freed memory access.
Fix this by freeing glue data last.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This oops is due to interrupts not being disabled in this particular path.
Reported-by: Dave Airlie <airlied@gmail.com> Tested-by: Dave Airlie <airlied@gmail.com> Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
EAP frames for stations in an AP VLAN are sent on the main AP interface
to avoid race conditions wrt. moving stations.
For that to work properly, sta_info_get_bss must be used instead of
sta_info_get when sending EAP packets.
Previously this was only done for cooked monitor injected packets, so
this patch adds a check for tx->skb->protocol to the same place.
Signed-off-by: Felix Fietkau <nbd@openwrt.org> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Driver incorrectly validates command completion: instead of waiting
for a command to be acknowledged it continues execution. Most of the
time driver gets acknowledge of the command completion in a tasklet
before it executes the next one. But sometimes it sends the next
command before it gets acknowledge for the previous one. In such a
case one of the following error messages appear in the log:
Failed to send SYSTEM_CONFIG: Already sending a command.
Failed to send ASSOCIATE: Already sending a command.
Failed to send TX_POWER: Already sending a command.
After that you need to reload the driver to get it working again.
This bug occurs during roaming (reported by Sam Varshavchik)
https://bugzilla.redhat.com/show_bug.cgi?id=738508
and machine booting (reported by Tom Gundersen and Mads Kiilerich)
https://bugs.archlinux.org/task/28097
https://bugzilla.redhat.com/show_bug.cgi?id=802106
This patch doesn't fix the delay issue during firmware load.
But at least device now works as usual after boot.
Signed-off-by: Stanislav Yakovlev <stas.yakovlev@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Commit 2a19032 (b43: reload phy and bss settings after core restarts)
introduced an unconditional call to b43_op_config() at the end of
b43_op_start(). When firmware fails to load this can wedge the system.
There's no need to reload the configuration after a failed
initialization anyway, so only make the call if initialization was
successful.
BugLink: http://bugs.launchpad.net/bugs/950295 Cc: Felix Fietkau <nbd@openwrt.org> Signed-off-by: Seth Forshee <seth.forshee@canonical.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Commit d902747("[libata] Add ATA transport class") introduced
ATA_EFLAG_OLD_ER to mark entries in the error ring as cleared.
But ata_count_probe_trials_cb() didn't check this flag and it still
counts the old error history. So wrong probe trials count is returned
and it causes problem, for example, SATA link speed is slowed down from
3.0Gbps to 1.5Gbps.
Fix it by checking ATA_EFLAG_OLD_ER in ata_count_probe_trials_cb().
Signed-off-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Jeff Garzik <jgarzik@redhat.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
CPU core ID is used to index the core_data[] array. The core ID is, however, not
sequential; 10-core CPUS can have a core ID as high as 25. Increase the limit to
32 to be able to deal with current CPUs.
Signed-off-by: Guenter Roeck <guenter.roeck@ericsson.com> Acked-by: Jean Delvare <khali@linux-fr.org> Acked-by: Durgadoss R <durgadoss.r@intel.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
We really need to use a ACCESS_ONCE() on the sequence value read in
__read_seqcount_begin(), because otherwise the compiler might end up
reloading the value in between the test and the return of it. As a
result, it might end up returning an odd value (which means that a write
is in progress).
If the reader is then fast enough that that odd value is still the
current one when the read_seqcount_retry() is done, we might end up with
a "successful" read sequence, even despite the concurrent write being
active.
In practice this probably never really happens - there just isn't
anything else going on around the read of the sequence count, and the
common case is that we end up having a read barrier immediately
afterwards.
So the code sequence in which gcc might decide to reaload from memory is
small, and there's no reason to believe it would ever actually do the
reload. But if the compiler ever were to decide to do so, it would be
incredibly annoying to debug. Let's just make sure.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The problem was that the first referral was parsed more than once
and so the caller tried the same referrals multiple times.
The problem was introduced partly by commit 066ce6899484d9026acd6ba3a8dbbedb33d7ae1b,
where 'ref += le16_to_cpu(ref->Size);' got lost,
but that was also wrong...
Signed-off-by: Stefan Metzmacher <metze@samba.org> Tested-by: Björn Jacke <bj@sernet.de> Reviewed-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com>
[bwh: Backport to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Ben Hutchings pointed out that the validation in efivars was inadequate -
most obviously, an entry with size 0 would server as a DoS against the
kernel. Improve this based on his suggestions.
Signed-off-by: Matthew Garrett <mjg@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Various people reported nohz load tracking still being wrecked, but Doug
spotted the actual problem. We fold the nohz remainder in too soon,
causing us to loose samples and under-account.
So instead of playing catch-up up-front, always do a single load-fold
with whatever state we encounter and only then fold the nohz remainder
and play catch-up.
Reported-by: Doug Smythies <dsmythies@telus.net> Reported-by: LesÅ=82aw Kope=C4=87 <leslaw.kopec@nasza-klasa.pl> Reported-by: Aman Gupta <aman@tmm1.net> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/n/tip-4v31etnhgg9kwd6ocgx3rxl8@git.kernel.org Signed-off-by: Ingo Molnar <mingo@elte.hu>
[bwh: Backported to 3.2: change filename] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
While testing with the intel_infoframes tool on gen4, I see that when
video DIP is disabled, what we write to the DATA memory is not exactly
what we read back later.
That commit was setting VIDEO_DIP_CTL to 0 when initializing, which
caused the problem.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=43947 Tested-by: Yang Guang <guang.a.yang@intel.com> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
[danvet: Pimped commit message by using the usual commit citation
layout.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Calculate the number of required free pages based on non-high memory
pages only, because that is where the buffers will come from.
Commit 081a9d043c983f161b78fdc4671324d1342b86bc introduced a new buffer
page allocation logic during hibernation, in order to improve the
performance. The amount of pages allocated was calculated based on total
amount of pages available, although only non-high memory pages are
usable for this purpose. This caused hibernation code to attempt to over
allocate pages on platforms that have high memory, which led to hangs.
Signed-off-by: Bojan Smojver <bojan@rexursive.com> Signed-off-by: Rafael J. Wysocki <rjw@suse.de> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
A common flaw in UEFI systems is a refusal to POST triggered by a malformed
boot variable. Once in this state, machines may only be restored by
reflashing their firmware with an external hardware device. While this is
obviously a firmware bug, the serious nature of the outcome suggests that
operating systems should filter their variable writes in order to prevent
a malicious user from rendering the machine unusable.
Signed-off-by: Matthew Garrett <mjg@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Normalize phy->attached_sas_addr to return a zero-address in the case
when device-type == NO_DEVICE or the linkrate is invalid to handle
expanders that put non-zero sas addresses in the discovery response:
Reported-by: Andrzej Jakowski <andrzej.jakowski@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: James Bottomley <JBottomley@Parallels.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
If an expander reports 'PHY VACANT' for a phy index prior to the one
that generated a BCN libsas fails rediscovery. Since a vacant phy is
defined as a valid phy index that will never have an attached device
just continue the search.
Signed-off-by: Thomas Jackson <thomas.p.jackson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: James Bottomley <JBottomley@Parallels.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
When a CPU is hotplugged off, we migrate any IRQs currently affine to it
away and onto another online CPU by calling the irq_set_affinity
function of the relevant interrupt controller chip. This function
returns either IRQ_SET_MASK_OK or IRQ_SET_MASK_OK_NOCOPY, to indicate
whether irq_data.affinity was updated.
If we are forcefully migrating an interrupt (because the affinity mask
no longer identifies any online CPUs) then we should update the IRQ
affinity mask to reflect the new CPU set. Failure to do so can
potentially leave /proc/irq/n/smp_affinity identifying only offline
CPUs, which may confuse userspace IRQ balancing daemons.
This patch updates migrate_one_irq to copy the affinity mask when
the interrupt chip returns IRQ_SET_MASK_OK after forcefully changing the
affinity of an interrupt.
Reported-by: Leif Lindholm <leif.lindholm@arm.com> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
TPIDRURW is a user read/write register forming part of the group of
thread registers in more recent versions of the ARM architecture (~v6+).
Currently, the kernel does not touch this register, which allows tasks
to communicate covertly by reading and writing to the register without
context-switching affecting its contents.
This patch clears TPIDRURW when TPIDRURO is updated via the set_tls
macro, which is called directly from __switch_to. Since the current
behaviour makes the register useless to userspace as far as thread
pointers are concerned, simply clearing the register (rather than saving
and restoring it) will not cause any problems to userspace.
Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
PL310 errata #588369 and #727915 require writes to the debug registers
of the cache controller to work around known problems. Writing these
registers on L220 may cause deadlock, so ensure that we only perform
this operation when we identify a PL310 at probe time.
Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The workaround for PL310 erratum #753970 can lead to deadlock on systems
with an L220 cache controller.
This patch makes the workaround effective only when the cache controller
is identified as a PL310 at probe time.
Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
[bwh: Backported to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Erratum #326103 ("FSR write bit incorrect on a SWP to read-only memory")
only affects the ARM 1136 core prior to r1p0. The workaround
disassembles the faulting instruction to determine whether it was a read
or write access on all v6 cores.
An issue has been reported on the ARM 11MPCore whereby loading the
faulting instruction may happen in parallel with that page being
unmapped, resulting in a deadlock due to the lack of TLB broadcasting
in hardware:
This patch limits the workaround so that it is only used on affected
cores, which are known to be UP only. Other v6 cores can rely on the
FSR to indicate the access type correctly.
Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The autofs packet size has had a very unfortunate size problem on x86:
because the alignment of 'u64' differs in 32-bit and 64-bit modes, and
because the packet data was not 8-byte aligned, the size of the autofsv5
packet structure differed between 32-bit and 64-bit modes despite
looking otherwise identical (300 vs 304 bytes respectively).
We first fixed that up by making the 64-bit compat mode know about this
problem in commit a32744d4abae ("autofs: work around unhappy compat
problem on x86-64"), and that made a 32-bit 'systemd' work happily on a
64-bit kernel because everything then worked the same way as on a 32-bit
kernel.
But it turned out that 'automount' had actually known and worked around
this problem in user space, so fixing the kernel to do the proper 32-bit
compatibility handling actually *broke* 32-bit automount on a 64-bit
kernel, because it knew that the packet sizes were wrong and expected
those incorrect sizes.
As a result, we ended up reverting that compatibility mode fix, and
thus breaking systemd again, in commit fcbf94b9dedd.
With both automount and systemd doing a single read() system call, and
verifying that they get *exactly* the size they expect but using
different sizes, it seemed that fixing one of them inevitably seemed to
break the other. At one point, a patch I seriously considered applying
from Michael Tokarev did a "strcmp()" to see if it was automount that
was doing the operation. Ugly, ugly.
However, a prettier solution exists now thanks to the packetized pipe
mode. By marking the communication pipe as being packetized (by simply
setting the O_DIRECT flag), we can always just write the bigger packet
size, and if user-space does a smaller read, it will just get that
partial end result and the extra alignment padding will simply be thrown
away.
This makes both automount and systemd happy, since they now get the size
they asked for, and the kernel side of autofs simply no longer needs to
care - it could pad out the packet arbitrarily.
Of course, if there is some *other* user of autofs (please, please,
please tell me it ain't so - and we haven't heard of any) that tries to
read the packets with multiple writes, that other user will now be
broken - the whole point of the packetized mode is that one system call
gets exactly one packet, and you cannot read a packet in pieces.
Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The actual internal pipe implementation is already really about
individual packets (called "pipe buffers"), and this simply exposes that
as a special packetized mode.
When we are in the packetized mode (marked by O_DIRECT as suggested by
Alan Cox), a write() on a pipe will not merge the new data with previous
writes, so each write will get a pipe buffer of its own. The pipe
buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn
will tell the reader side to break the read at that boundary (and throw
away any partial packet contents that do not fit in the read buffer).
End result: as long as you do writes less than PIPE_BUF in size (so that
the pipe doesn't have to split them up), you can now treat the pipe as a
packet interface, where each read() system call will read one packet at
a time. You can just use a sufficiently big read buffer (PIPE_BUF is
sufficient, since bigger than that doesn't guarantee atomicity anyway),
and the return value of the read() will naturally give you the size of
the packet.
NOTE! We do not support zero-sized packets, and zero-sized reads and
writes to a pipe continue to be no-ops. Also note that big packets will
currently be split at write time, but that the size at which that
happens is not really specified (except that it's bigger than PIPE_BUF).
Currently that limit is the system page size, but we might want to
explicitly support bigger packets some day.
The main user for this is going to be the autofs packet interface,
allowing us to stop having to care so deeply about exact packet sizes
(which have had bugs with 32/64-bit compatibility modes). But user
space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will
fail with an EINVAL on kernels that do not support this interface.
Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The field is used to pass the UVC request data length, but can also be
used to signal an error when setting it to a negative value. Switch from
unsigned int to __s32.
pullup() is already called properly by udc-core.c and
there's no need to call it from udc_stop(), in fact that
will cause issues.
Reviewed-by: Alexander Shishkin <alexander.shishkin@linux.intel.com> Acked-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Felipe Balbi <balbi@ti.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This patch (as1539) fixes a minor bug in the mass-storage gadget
drivers. When an unknown command is received, the error code sent
back is "Invalid Field in CDB" rather than "Invalid Command". This is
because the bitmask of CDB bytes allowed to be nonzero is incorrect.
When handling an unknown command, we don't care which command bytes
are nonzero. All the bits in the mask should be set, not just eight
of them.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> CC: <Michal Nazarewicz <mina86@mina86.com> Signed-off-by: Felipe Balbi <balbi@ti.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This patch (as1545) fixes a problem affecting several ASUS computers:
The machine crashes or corrupts memory when going into suspend if the
ehci-hcd driver is bound to any controllers. Users have been forced
to unbind or unload ehci-hcd before putting their systems to sleep.
After extensive testing, it was determined that the machines don't
like going into suspend when any EHCI controllers are in the PCI D3
power state. Presumably this is a firmware bug, but there's nothing
we can do about it except to avoid putting the controllers in D3
during system sleep.
The patch adds a new flag to indicate whether the problem is present,
and avoids changing the controller's power state if the flag is set.
Runtime suspend is unaffected; this matters only for system suspend.
However as a side effect, the controller will not respond to remote
wakeup requests while the system is asleep. Hence USB wakeup is not
functional -- but of course, this is already true in the current state
of affairs.
This fixes Bugzilla #42728.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Tested-by: Steven Rostedt <rostedt@goodmis.org> Tested-by: Andrey Rahmatullin <wrar@wrar.name> Tested-by: Oleksij Rempel (fishor) <bug-track@fisher-privat.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[bwh: Backported to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This patch fixes a race whereby a pointer to a buffer
would be overwritten while the buffer was in use leading
to a double free and a memory leak. This causes crashes.
This bug was introduced in 2.6.34
Signed-off-by: Oliver Neukum <oneukum@suse.de> Tested-by: Bjørn Mork <bjorn@mork.no> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
According to the reporter, external mic starts to work if the
laptop-dmic model is used. According to BIOS pin config, all
pins are consistent with the alc269vb_laptop_dmic fixup, except
for the external mic, which is not present.
BugLink: https://bugs.launchpad.net/bugs/950490 Signed-off-by: David Henningsson <david.henningsson@canonical.com> Signed-off-by: Takashi Iwai <tiwai@suse.de>
[bwh: Backported to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The nl80211 handling code should ensure as much as
it can that the interface is in a valid state, it
can certainly ensure the interface is running.
Not doing so can cause calls through mac80211 into
the driver that result in warnings and unspecified
behaviour in the driver.
Reported-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
On 32-bit systems, a large args->num_cliprects from userspace via ioctl
may overflow the allocation size, leading to out-of-bounds access.
This vulnerability was introduced in commit 432e58ed ("drm/i915: Avoid
allocation for execbuffer object list").
Signed-off-by: Xi Wang <xi.wang@gmail.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
On 32-bit systems, a large args->buffer_count from userspace via ioctl
may overflow the allocation size, leading to out-of-bounds access.
This vulnerability was introduced in commit 8408c282 ("drm/i915:
First try a normal large kmalloc for the temporary exec buffers").
Signed-off-by: Xi Wang <xi.wang@gmail.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[bwh: Backported to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Clearing bit 5 of CACHE_MODE_0 is necessary to prevent GPU hangs in
OpenGL programs such as Google MapsGL, Google Earth, and gzdoom when
using separate stencil buffers. Without it, the GPU tries to use the
LRA eviction policy, which isn't supported. This was supposed to be off
by default, but seems to be on for many machines.
This cannot be done in gen6_init_clock_gating with most of the other
workaround bits; the render ring needs to exist. Otherwise, the
register write gets dropped on the floor (one printk will show it
changed, but a second printk immediately following shows the value
reverts to the old one).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47535 Cc: Rob Castle <futuredub@gmail.com> Cc: Eric Appleman <erappleman@gmail.com> Cc: aaron667@gmx.net Cc: Keith Packard <keithp@keithp.com> Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The docs say this is required for Gen7, and since the bit was added for
Gen6, we are also setting it there pit pf paranoia. Particularly as
Chris points out, if PIPE_CONTROL counts as a 3d state packet.
This was found through doc inspection by Ken and applies to Gen6+;
Reported-by: Kenneth Graunke <kenneth@whitecape.org> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Eric Anholt <eric@anholt.net> Signed-off-by: Keith Packard <keithp@keithp.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
dev_priv keeps track of the current addressing mode that gets set at
execbuffer time. Unfortunately the existing code was doing this before
acquiring struct_mutex which leaves a race with another thread also
doing an execbuffer. If that wasn't bad enough, relocate_slow drops
struct_mutex which opens a much more likely error where another thread
comes in and modifies the state while relocate_slow is being slow.
The solution here is to just defer setting this state until we
absolutely need it, and we know we'll have struct_mutex for the
remainder of our code path.
v2: Keith noticed a bug in the original patch.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by: Keith Packard <keithp@keithp.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
We seem to have a decent confusion between the output timings and the
input timings of the sdvo encoder. If I understand the code correctly,
we use the original mode unchanged for the output timings, safe for
the lvds case. And we should use the adjusted mode for input timings.
Clarify the situation by adding an explicit output_dtd to the sdvo
mode_set function and streamline the code-flow by moving the input and
output mode setting in the sdvo encode together.
Furthermore testing showed that the sdvo input timing needs the
unadjusted dotclock, the sdvo chip will automatically compute the
required pixel multiplier to get a dotclock above 100 MHz.
Fix this up when converting a drm mode to an sdvo dtd.
> diff --git a/drivers/gpu/drm/i915/intel_sdvo.c
> b/drivers/gpu/drm/i915/intel_sdvo.c
> index 093e914..62d22ae 100644
> --- a/drivers/gpu/drm/i915/intel_sdvo.c
> +++ b/drivers/gpu/drm/i915/intel_sdvo.c
> @@ -1122,11 +1123,9 @@ static void intel_sdvo_mode_set(struct drm_encoder *encoder,
>
> /* We have tried to get input timing in mode_fixup, and filled into
> adjusted_mode */
> - if (intel_sdvo->is_tv || intel_sdvo->is_lvds) {
> - intel_sdvo_get_dtd_from_mode(&input_dtd, adjusted_mode);
> + intel_sdvo_get_dtd_from_mode(&input_dtd, adjusted_mode);
> + if (intel_sdvo->is_tv || intel_sdvo->is_lvds)
> input_dtd.part2.sdvo_flags = intel_sdvo->sdvo_flags;
> - } else
> - intel_sdvo_get_dtd_from_mode(&input_dtd, mode);
>
> /* If it's a TV, we already set the output timing in mode_fixup.
> * Otherwise, the output timing is equal to the input timing.
Due to questions raised in review, below a more elaborate analysis of
the bug at hand:
Sdvo seems to have two timings, one is the output timing which will be
sent over whatever is connected on the other side of the sdvo chip (panel,
hdmi screen, tv), the other is the input timing which will be generated by
the gmch pipe. It looks like sdvo is expected to scale between the two.
To make things slightly more complicated, we have a bunch of special
cases:
- For lvds panel we always use a fixed output timing, namely
intel_sdvo->sdvo_lvds_fixed_mode, hence that special case.
- Sdvo has an interface to generate a preferred input timing for a given
output timing. This is the confusing thing that I've tried to clear up
with the follow-on patches.
- A special requirement is that the input pixel clock needs to be between
100MHz and 200MHz (likely to keep it within the electromechanical design
range of PCIe), 270MHz on later gen4+. Lower pixel clocks are
doubled/quadrupled.
The thing this patch tries to fix is that the pipe needs to be
explicitly instructed to double/quadruple the pixels and needs the
correspondingly higher pixel clock, whereas the sdvo adaptor seems to
do that itself and needs the unadjusted pixel clock. For the sdvo
encode side we already set the pixel mutliplier with a different
command (0x21).
This patch tries to fix this mess by:
- Keeping the output mode timing in the unadjusted plain mode, safe
for the lvds case.
- Storing the input timing in the adjusted_mode with the adjusted
pixel clock. This way we don't need to frob around with the core
crtc mode set code.
- Fixing up the pixelclock when constructing the sdvo dtd timing
struct. This is why the first hunk of the patch is an integral part
of the series.
- Dropping the is_tv special case because input_dtd is equivalent to
adjusted_mode after these changes. Follow-up patches clear this up
further (by simply ripping out intel_sdvo->input_dtd because it's
not needed).
v2: Extend commit message with an in-depth bug analysis.
Reported-and-Tested-by: Bernard Blackham <b-linuxgit@largestprime.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=48157 Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[bwh: Indented the hunk quoted above so quilt doesn't try to apply it] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Noticed by Jerome Glisse (after weeks of debugging).
Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Dave Airlie <airlied@redhat.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Newer BKDG[1] versions recommend a different initialization value for
the running average range register in the northbridge. This improves
the power reading by avoiding counter saturations resulting in bogus
values for anything below about 80% of TDP power consumption.
Updated BIOSes will have this new value set up from the beginning,
but meanwhile we correct this value ourselves.
This needs to be done on all northbridges, even on those where the
driver itself does not register at.
This fixes the driver on all current machines to provide proper
values for idle load.
While debugging a latency with someone on IRC (mirage335) on #linux-rt (OFTC),
we discovered that the stacktrace output of the latency tracers
(preemptirqsoff) was empty.
This bug was caused by the creation of the dynamic length stack trace
again (like commit 12b5da3 "tracing: Fix ent_size in trace output" was).
This bug is caused by the latency tracers requiring the next event
to determine the time between the current event and the next. But by
grabbing the next event, the iter->ent_size is set to the next event
instead of the current one. As the stacktrace event is the last event,
this makes the ent_size zero and causes nothing to be printed for
the stack trace. The dynamic stacktrace uses the ent_size to determine
how much of the stack can be printed. The ent_size of zero means
no stack.
The simple fix is to save the iter->ent_size before finding the next event.
Note, mirage335 asked to remain anonymous from LKML and git, so I will
not add the Reported-by and Tested-by tags, even though he did report
the issue and tested the fix.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
With this fix applied build_sched_domains() will return -ENOMEM and
the suspend attempt fails.
Signed-off-by: he, bo <bo.he@intel.com> Reviewed-by: Zhang, Yanmin <yanmin.zhang@intel.com> Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1335355161.5892.17.camel@hebo
[ So, we fail to deallocate a CPU because we cannot allocate RAM :-/
I don't like that kind of sad behavior but nevertheless it should
not crash under high memory load. ] Signed-off-by: Ingo Molnar <mingo@kernel.org>
[bwh: Backported to 3.2: change filename] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This loop on EBCISR register was designed to clear IRQ sources before enabling
a DMA channel. This register is clear-on-read so a race condition can appear if
another channel is already active and has just finished its transfer.
Removing this read on EBCISR is fixing the issue as there is no case where an IRQ
could be pending: we already make sure that this register is drained at probe()
time and during resume.
Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com> Signed-off-by: Vinod Koul <vinod.koul@linux.intel.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Line widgets had not been included in either the power up or power down
sequences so if a widget had an event associated with it that event would
never be run. Fix this minimally by adding them to the sequences, we
should probably be doing away with the specific widget types as they all
have the same priority anyway.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
When we boot on a machine that can hotplug CPUs and we
are using 'dom0_max_vcpus=X' on the Xen hypervisor line
to clip the amount of CPUs available to the initial domain,
we get this:
(XEN) Command line: com1=115200,8n1 dom0_mem=8G noreboot dom0_max_vcpus=8 sync_console mce_verbosity=verbose console=com1,vga loglvl=all guest_loglvl=all
.. snip..
DMI: Intel Corporation S2600CP/S2600CP, BIOS SE5C600.86B.99.99.x032.072520111118 07/25/2011
.. snip.
SMP: Allowing 64 CPUs, 32 hotplug CPUs
installing Xen timer for CPU 7
cpu 7 spinlock event irq 361
NMI watchdog: disabled (cpu7): hardware events not enabled
Brought up 8 CPUs
.. snip..
[acpi processor finds the CPUs are not initialized and starts calling
arch_register_cpu, which creates /sys/devices/system/cpu/cpu8/online]
CPU 8 got hotplugged
CPU 9 got hotplugged
CPU 10 got hotplugged
.. snip..
initcall 1_acpi_battery_init_async+0x0/0x1b returned 0 after 406 usecs
calling erst_init+0x0/0x2bb @ 1
[and the scheduler sticks newly started tasks on the new CPUs, but
said CPUs cannot be initialized b/c the hypervisor has limited the
amount of vCPUS to 8 - as per the dom0_max_vcpus=8 flag.
The spinlock tries to kick the other CPU, but the structure for that
is not initialized and we crash.]
BUG: unable to handle kernel paging request at fffffffffffffed8
IP: [<ffffffff81035289>] xen_spin_lock+0x29/0x60
PGD 180d067 PUD 180e067 PMD 0
Oops: 0002 [#1] SMP
CPU 7
Modules linked in:
In xen_restore_fl_direct(), xen_force_evtchn_callback() was being
called even if no events were pending. This resulted in (depending on
workload) about a 100 times as many xen_version hypercalls as
necessary.
Fix this by correcting the sense of the conditional jump.
This seems to give a significant performance benefit for some
workloads.
There is some subtle tricksy "..since the check here is trying to
check both pending and masked in a single cmpw, but I think this is
correct. It will call check_events now only when the combined
mask+pending word is 0x0001 (aka unmasked, pending)." (Ian)
Acked-by: Ian Campbell <ian.campbell@citrix.com> Signed-off-by: David Vrabel <david.vrabel@citrix.com> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
While that commit was technically the right thing to do, and made the
x86-64 compat mode work identically to native 32-bit mode (and thus
fixing the problem with a 32-bit systemd install on a 64-bit kernel), it
turns out that the automount binaries had workarounds for this compat
problem.
Now, the workarounds are disgusting: doing an "uname()" to find out the
architecture of the kernel, and then comparing it for the 64-bit cases
and fixing up the size of the read() in automount for those. And they
were confused: it's not actually a generic 64-bit issue at all, it's
very much tied to just x86-64, which has different alignment for an
'u64' in 64-bit mode than in 32-bit mode.
But the end result is that fixing the compat layer actually breaks the
case of a 32-bit automount on a x86-64 kernel.
There are various approaches to fix this (including just doing a
"strcmp()" on current->comm and comparing it to "automount"), but I
think that I will do the one that teaches pipes about a special "packet
mode", which will allow user space to not have to care too deeply about
the padding at the end of the autofs packet.
That change will make the compat workaround unnecessary, so let's revert
it first, and get automount working again in compat mode. The
packetized pipes will then fix autofs for systemd.
Reported-and-requested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Ian Kent <raven@themaw.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[bwh: Backported to 3.2: adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Current APIC code assumes MSR_IA32_APICBASE is present for all systems.
Pentium Classic P5 and friends didn't have this MSR. MSR_IA32_APICBASE
was introduced as an architectural MSR by Intel @ P6.
Code paths that can touch this MSR invalidly are when vendor == Intel &&
cpu-family == 5 and APIC bit is set in CPUID - or when you simply pass
lapic on the kernel command line, on a P5.
The below patch stops Linux incorrectly interfering with the
MSR_IA32_APICBASE for P5 class machines. Other code paths exist that
touch the MSR - however those paths are not currently reachable for a
conformant P5.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linux.intel.com> Link: http://lkml.kernel.org/r/4F8EEDD3.1080404@linux.intel.com Signed-off-by: H. Peter Anvin <hpa@linux.intel.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This is due to commit 8a25a2fd126c ("cpu: convert 'cpu' and
'machinecheck' sysdev_class to a regular subsystem") which renders
commit 6c53cbfced04 ("x86, microcode: Correct sysdev_add error path")
useless.
When hostname contains colon (e.g. when it is an IPv6 address) it needs
to be enclosed in brackets to make parsing of NFS device string possible.
Fix nfs_do_root_mount() to enclose hostname properly when needed. NFS code
actually does not need this as it does not parse the string passed by
nfs_do_root_mount() but the device string is exposed to userspace in
/proc/mounts.
CC: Josh Boyer <jwboyer@redhat.com> CC: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Commit f5fff5d forgot to fix TCP_MAXSEG behavior IPv6 sockets, so IPv6
TCP server sockets that used TCP_MAXSEG would find that the advmss of
child sockets would be incorrect. This commit mirrors the advmss logic
from tcp_v4_syn_recv_sock in tcp_v6_syn_recv_sock. Eventually this
logic should probably be shared between IPv4 and IPv6, but this at
least fixes this issue.
Signed-off-by: Neal Cardwell <ncardwell@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
While reviewing the sysctl code in ax25 I spotted races in ax25_exit
where it is possible to receive notifications and packets after already
freeing up some of the data structures needed to process those
notifications and updates.
Call unregister_netdevice_notifier early so that the rest of the cleanup
code does not need to deal with network devices. This takes advantage
of my recent enhancement to unregister_netdevice_notifier to send
unregister notifications of all network devices that are current
registered.
Move the unregistration for packet types, socket types and protocol
types before we cleanup any of the ax25 data structures to remove the
possibilities of other races.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
MAX_ADDR_LEN is 32. ETH_ALEN is 6. mac->sa_data is a 14 byte array, so
the memcpy() is doing a read past the end of the array. I asked about
this on netdev and Ben Hutchings told me it's supposed to be copying
ETH_ALEN bytes (thanks Ben).
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
ops_init should free the net_generic data on
init failure and __register_pernet_operations should not
call ops_free when NET_NS is not enabled.
Signed-off-by: Julian Anastasov <ja@ssi.bg> Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
tcp_grow_window() has to grow rcv_ssthresh up to window_clamp, allowing
sender to increase its window.
tcp_grow_window() still assumes a tcp frame is under MSS, but its no
longer true with LRO/GRO.
This patch fixes one of the performance issue we noticed with GRO on.
Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Neal Cardwell <ncardwell@google.com> Cc: Tom Herbert <therbert@google.com> Acked-by: Neal Cardwell <ncardwell@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
In register_netdevice(), when ndo_init() is successful and later
some error occurred, ndo_uninit() will be called.
So dummy deivce is desirable to implement ndo_uninit() method
to free percpu stats for this case.
And, ndo_uninit() is also called along with dev->destructor() when
device is unregistered, so in order to prevent dev->dstats from
being freed twice, dev->destructor is modified to free_netdev().
Signed-off-by: Hiroaki SHIMODA <shimoda.hiroaki@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Stephane Fillod <fillods@users.sf.net> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
A parameter set exists for WRED mode, called wred_set, to hold the same
values for qavg and qidlestart across all VQs. The WRED mode values had
been previously held in the VQ for the default DP. After these values
were moved to wred_set, the VQ for the default DP was no longer created
automatically (so that it could be omitted on purpose, to have packets
in the default DP enqueued directly to the device without using RED).
However, gred_dump() was overlooked during that change; in WRED mode it
still reads qavg/qidlestart from the VQ for the default DP, which might
not even exist. As a result, this command sequence will cause an oops:
This fixes gred_dump() in WRED mode to use the values held in wred_set.
Signed-off-by: David Ward <david.ward@ll.mit.edu> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
At the beginning of ks_rcv(), a for loop retrieves the
header information relevant to all the frames stored
in the mac's internal buffers. The number of pending
frames is stored as an 8 bits field in KS_RXFCTR.
If interrupts are disabled long enough to allow for more than
32 frames to accumulate in the MAC's internal buffers, a buffer
overflow occurs.
This patch fixes the problem by making the
driver's frame_head_info buffer big enough.
Well actually, since the chip appears to have 12K of
internal rx buffers and the shortest ethernet frame should
be 64 bytes long, maybe the limit could be set to
12*1024/64 = 192 frames, but 255 should be safer.
Signed-off-by: Davide Ciminaghi <ciminaghi@gnudd.com> Signed-off-by: Raffaele Recalcati <raffaele.recalcati@bticino.it> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
The SMSC911x driver resets the ->head, ->data and ->tail pointers in the
skb on the reset path in order to avoid buffer overflow due to packet
padding performed by the hardware.
This patch fixes the receive path so that the skb pointers are fixed up
after the data has been read from the device, The error path is also
fixed to use number of words consistently and prevent erroneous FIFO
fastforwarding when skipping over bad data.
Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
We set intr mask before its handler is registered, this does not work well when
8139cp is sharing irq line with other devices. As the irq could be enabled by
the device before 8139cp's hander is registered which may lead unhandled
irq. Fix this by introducing an helper cp_irq_enable() and call it after
request_irq().
Signed-off-by: Jason Wang <jasowang@redhat.com> Reviewed-by: Flavio Leitner <fbl@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Problem:
There was two separate work_struct structures which share one
handler. Unfortunately getting atl1_adapter structure from
work_struct in case of DMA error was done from incorrect
offset which cause kernel panics.
Solution:
The useless work_struct for DMA error removed and
handler name changed to more generic one.
Signed-off-by: Tony Zelenoff <antonz@parallels.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Marc Merlin reported many order-1 allocations failures in TX path on its
wireless setup, that dont make any sense with MTU=1500 network, and non
SG capable hardware.
After investigation, it turns out TCP uses sk_stream_alloc_skb() and
used as a convention skb_tailroom(skb) to know how many bytes of data
payload could be put in this skb (for non SG capable devices)
Note : these skb used kmalloc-4096 (MTU=1500 + MAX_HEADER +
sizeof(struct skb_shared_info) being above 2048)
Later, mac80211 layer need to add some bytes at the tail of skb
(IEEE80211_ENCRYPT_TAILROOM = 18 bytes) and since no more tailroom is
available has to call pskb_expand_head() and request order-1
allocations.
This patch changes sk_stream_alloc_skb() so that only
sk->sk_prot->max_header bytes of headroom are reserved, and use a new
skb field, avail_size to hold the data payload limit.
This way, order-0 allocations done by TCP stack can leave more than 2 KB
of tailroom and no more allocation is performed in mac80211 layer (or
any layer needing some tailroom)
avail_size is unioned with mark/dropcount, since mark will be set later
in IP stack for output packets. Therefore, skb size is unchanged.
Reported-by: Marc MERLIN <marc@merlins.org> Tested-by: Marc MERLIN <marc@merlins.org> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Correct commit hash for follow-on bug fix] Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit f07d960df3 (tcp: avoid frag allocation for small frames)
breaked assumption in tcp stack that skb is either linear (skb->data_len
== 0), or fully fragged (skb->data_len == skb->len)
tcp_trim_head() made this assumption, we must fix it.
Thanks to Vijay for providing a very detailed explanation.
Reported-by: Vijay Subramanian <subramanian.vijay@gmail.com> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Marc Merlin reported many order-1 allocations failures in TX path on its
wireless setup, that dont make any sense with MTU=1500 network, and non
SG capable hardware.
Turns out part of the problem comes from pskb_expand_head() not using
ksize() to get exact head size given by kmalloc(). Doing the same thing
than __alloc_skb() allows more tailroom in skb and can prevent future
reallocations.
As a bonus, struct skb_shared_info becomes cache line aligned.
Reported-by: Marc MERLIN <marc@merlins.org> Tested-by: Marc MERLIN <marc@merlins.org> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Fix a code path in tcp_rcv_rtt_update() that was comparing scaled and
unscaled RTT samples.
The intent in the code was to only use the 'm' measurement if it was a
new minimum. However, since 'm' had not yet been shifted left 3 bits
but 'new_sample' had, this comparison would nearly always succeed,
leading us to erroneously set our receive-side RTT estimate to the 'm'
sample when that sample could be nearly 8x too high to use.
The overall effect is to often cause the receive-side RTT estimate to
be significantly too large (up to 40% too large for brief periods in
my tests).
Signed-off-by: Neal Cardwell <ncardwell@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
As soon as an skb is queued into socket error queue, another thread
can consume it, so we are not allowed to reference skb anymore, or risk
use after free.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
As soon as an skb is queued into socket receive_queue, another thread
can consume it, so we are not allowed to reference skb anymore, or risk
use after free.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
This happened on a machine with a custom hotplug script calling nameif,
probably due to slow firmware loading. At the time nameif uses ethtool
to gather interface information, i2400m->fw_name is zero and so a null
pointer dereference occurs from within i2400m_get_drvinfo().
Signed-off-by: Phil Sutter <phil.sutter@viprinet.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
When a slave comes up, we're unsetting the current_arp_slave without
removing active flags from it, which can lead to situations where we have
more than one slave with active flags in active-backup mode.
To avoid this situation we must remove the active flags from a slave before
removing it as a current_arp_slave.
Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Marcelo Ricardo Leitner <mleitner@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
A phonet packet is limited to USHRT_MAX bytes, this is never checked during
tx which means that the user can specify any size he wishes, and the kernel
will attempt to allocate that size.
In the good case, it'll lead to the following warning, but it may also cause
the kernel to kick in the OOM and kill a random task on the server.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com> Acked-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Convert array index from the loop bound to the loop index.
And remove the void type conversion to ip6_mc_del1_src() return
code, seem it is unnecessary, since ip6_mc_del1_src() does not
use __must_check similar attribute, no compiler will report the
warning when it is removed.
v2: enrich the commit header
Signed-off-by: RongQing.Li <roy.qing.li@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
As it stands the bridge IGMP snooping system will respond to
group leave messages with queries for remaining membership.
This is both unnecessary and undesirable. First of all any
multicast routers present should be doing this rather than us.
What's more the queries that we send may end up upsetting other
multicast snooping swithces in the system that are buggy.
In fact, we can simply remove the code that send these queries
because the existing membership expiry mechanism doesn't rely
on them anyway.
So this patch simply removes all code associated with group
queries in response to group leave messages.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
getsockopt(..., SCTP_EVENTS, ...) performs a length check and returns
an error if the user provides less bytes than the size of struct
sctp_event_subscribe.
Struct sctp_event_subscribe needs to be extended by an u8 for every
new event or notification type that is added.
This obviously makes getsockopt fail for binaries that are compiled
against an older versions of <net/sctp/user.h> which do not contain
all event types.
This patch changes getsockopt behaviour to no longer return an error
if not enough bytes are being provided by the user. Instead, it
returns as much of sctp_event_subscribe as fits into the provided buffer.
This leads to the new behavior that users see what they have been aware
of at compile time.
The setsockopt(..., SCTP_EVENTS, ...) API is already behaving like this.
Signed-off-by: Thomas Graf <tgraf@suug.ch> Acked-by: Vlad Yasevich <vladislav.yasevich@hp.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
vmsplice()/splice(pipe, socket) call do_tcp_sendpages() one page at a
time, adding at most 4096 bytes to an skb. (assuming PAGE_SIZE=4096)
The call to tcp_push() at the end of do_tcp_sendpages() forces an
immediate xmit when pipe is not already filled, and tso_fragment() try
to split these skb to MSS multiples.
4096 bytes are usually split in a skb with 2 MSS, and a remaining
sub-mss skb (assuming MTU=1500)
This makes slow start suboptimal because many small frames are sent to
qdisc/driver layers instead of big ones (constrained by cwnd and packets
in flight of course)
In fact, applications using sendmsg() (adding an additional memory copy)
instead of vmsplice()/splice()/sendfile() are a bit faster because of
this anomaly, especially if serving small files in environments with
large initial [c]wnd.
Call tcp_push() only if MSG_MORE is not set in the flags parameter.
This bit is automatically provided by splice() internals but for the
last page, or on all pages if user specified SPLICE_F_MORE splice()
flag.
In some workloads, this can reduce number of sent logical packets by an
order of magnitude, making zero-copy TCP actually faster than
one-copy :)
Reported-by: Tom Herbert <therbert@google.com> Cc: Nandita Dukkipati <nanditad@google.com> Cc: Neal Cardwell <ncardwell@google.com> Cc: Tom Herbert <therbert@google.com> Cc: Yuchung Cheng <ycheng@google.com> Cc: H.K. Jerry Chu <hkchu@google.com> Cc: Maciej Żenczykowski <maze@google.com> Cc: Mahesh Bandewar <maheshb@google.com> Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
For every transmitted packet, ppp_start_xmit() will stop the netdev
queue and then, if appropriate, restart it. This causes the TX softirq
to run, entirely gratuitously.
This is "only" a waste of CPU time in the normal case, but it's actively
harmful when the PPP device is a TEQL slave — the wakeup will cause the
offending device to receive the next TX packet from the TEQL queue, when
it *should* have gone to the next slave in the list. We end up seeing
large bursts of packets on just *one* slave device, rather than using
the full available bandwidth over all slaves.
This patch fixes the problem by *not* unconditionally stopping the queue
in ppp_start_xmit(). It adds a return value from ppp_xmit_process()
which indicates whether the queue should be stopped or not.
It *doesn't* remove the call to netif_wake_queue() from
ppp_xmit_process(), because other code paths (especially from
ppp_output_wakeup()) need it there and it's messy to push it out to the
other callers to do it based on the return value. So we leave it in
place — it's a no-op in the case where the queue wasn't stopped, so it's
harmless in the TX path.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
projects / karo-tx-linux.git / log