If the sentinel triggers, we do not want the boot loader authors to
just poke it and make the error go away, we want them to actually fix
the problem.
This should help avoid making the incorrect change in non-compliant
bootloaders.
[ hpa: dropped the Documentation/x86/boot.txt hunk pending
clarifications ]
Signed-off-by: Peter Jones <pjones@redhat.com>
Link: http://lkml.kernel.org/r/1362592823-28967-1-git-send-email-pjones@redhat.com
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
When boot_params->sentinel is set, all we really know is that some
undefined set of fields in struct boot_params contain garbage. In the
particular case of efi_info, however, there is a private magic for
that substructure, so it is generally safe to leave it even if the
bootloader is broken.
kexec (for which we did the initial analysis) did not initialize this
field, but of course all the EFI bootloaders do, and most EFI
bootloaders are broken in this respect (and should be fixed.)
Reported-by: Robin Holt <holt@sgi.com>
Link: http://lkml.kernel.org/r/CA%2B5PVA51-FT14p4CRYKbicykugVb=PiaEycdQ57CK2km_OQuRQ@mail.gmail.com
Tested-by: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Henrik reported that his MacAir 3.1 would not boot with
| commit 8d57470d8f
| Date: Fri Nov 16 19:38:58 2012 -0800
|
| x86, mm: setup page table in top-down
It turns out that we do not calculate the real_end properly:
We try to get 2M size with 4K alignment, and later will round down
to 2M, so we will get less then 2M for first mapping, in extreme
case could be only 4K only. In Henrik's system it has (1M-32K) as
last usable rage is [mem 0x7f9db000-0x7fef8fff].
The problem is exposed when EFI booting have several holes and it
will force mapping to use PTE instead as we only map usable areas.
To fix it, just make it be 2M aligned, so we can be guaranteed to be
able to use large pages to map it.
Reported-by: Henrik Rydberg <rydberg@euromail.se>
Bisected-by: Henrik Rydberg <rydberg@euromail.se>
Tested-by: Henrik Rydberg <rydberg@euromail.se>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/CAE9FiQX4nQ7_1kg5RL_vh56rmcSHXUi1ExrZX7CwED4NGMnHfg@mail.gmail.com
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The commit 27be457000
('x86 idle: remove 32-bit-only "no-hlt" parameter, hlt_works_ok
flag') removed the hlt_works_ok flag from struct cpuinfo_x86, but
boot_cpu_data and new_cpu_data initializers were not changed
causing setting f00f_bug flag, instead of fdiv_bug.
If CONFIG_X86_F00F_BUG is not set the f00f_bug flag is never
cleared.
To avoid such problems in future C99-style initialization is now
used.
Signed-off-by: Krzysztof Mazur <krzysiek@podlesie.net>
Acked-by: Borislav Petkov <bp@suse.de>
Cc: len.brown@intel.com
Link: http://lkml.kernel.org/r/1362266082-2227-1-git-send-email-krzysiek@podlesie.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Raid56 merge (merge commit e942f88) had mistakenly removed a call to
__cancel_balance(), which resulted in balance not cleaning up after itself
after a successful finish. (Cleanup includes switching the state, removing
the balance item and releasing mut_ex_op testnset lock.) Bring it back.
Reported-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
Commit 1fe42740 ("ARM: dts: mxs: Add the LCD to the 10049 board") seem
to have been applied with some fuzzyness, and the framebuffer
initialisation code for the CFA-10049 ended up in the CFA-10037
initialisation function.
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
A bunch of fixes, nothing truely horrible:
- Fix PCH irq handling race which resulted in missed gmbus/dp aux irqs
and subsequent fallout (Paulo)
- Fixup off-by-one in our hsw id table (Kenneth)
- Fixup ilk rc6 support (disabled by default), regression introduced in
3.8
- g4x plane w/a from Egbert Eich
- gen2/3/4 dpms suspend/standy fixes for VGA outputs from Patrik Jakobsson
- Workaround dying ivb machines with less aggressive rc6 values (Stéphane
Marchesin)
* 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel:
drm/i915: Turn off hsync and vsync on ADPA when disabling crt
drm/i915: Fix incorrect definition of ADPA HSYNC and VSYNC bits
drm/i915: also disable south interrupts when handling them
drm/i915: enable irqs earlier when resuming
drm/i915: Increase the RC6p threshold.
DRM/i915: On G45 enable cursor plane briefly after enabling the display plane.
drm/i915: Fix Haswell/CRW PCI IDs.
drm/i915: Don't clobber crtc->fb when queue_flip fails
drm/i915: wait_event_timeout's timeout is in jiffies
drm/i915: Fix missing variable initilization
We had a recent fix to fix the release of pagecache pages when
cifs_writev_requeue writes fail. Unfortunately, it releases the page
before trying to unlock it. At that point, the page might be gone by the
time the unlock comes in.
Unlock the page first before checking the value of "rc", and only then
end writeback and release the pages. The page lock isn't required for
any of those operations so this should be safe.
Reported-by: Anton Altaparmakov <aia21@cam.ac.uk>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
cifs_rename_pending_delete() attempts to silly rename file using
CIFSSMBRenameOpenFile(). This uses the SET_FILE_INFORMATION TRANS2
command with information level set to the passthru info-level
SMB_SET_FILE_RENAME_INFORMATION.
We need to check to make sure that the server support passthru
info-levels before attempting the silly rename or else we will fail to
rename the file.
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
Fix check for error condition after setting attributes with
CIFSSMBSetFileInfo().
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Pavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: Steve French <sfrench@us.ibm.com>
WARNING: vmlinux.o(.text+0xfb80): Section mismatch in reference
from the function armpmu_register() to the function
.init.text:armpmu_init()
The function armpmu_register() references
the function __init armpmu_init().
This is often because armpmu_register lacks a __init
annotation or the annotation of armpmu_init is wrong.
Just drop the __init marking on armpmu_init() because
armpmu_register() no longer has an __init marking.
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Before jumping to (position independent) C-code from the decompressor's
assembler world we set-up the C environment. This setup currently does not
set r9, which for arm-none-uclinux-uclibceabi toolchains is by default
expected to be the PIC offset base register (IE should point to the
beginning of the GOT).
Currently, therefore, in order to build working kernels that use the
decompressor it is necessary to use an arm-linux-gnueabi toolchain, or
similar. uClinux toolchains cause a prefetch abort to occur at the beginning
of the decompress_kernel function.
This patch allows uClinux toolchains to build bootable zImages by forcing
the -mno-single-pic-base option, which ensures that the location of the GOT
is re-derived each time it is required, and r9 becomes free for use as a
general purpose register.
This has a small (4% in instruction terms) advantage over the alternative of
setting r9 to point to the GOT before calling into the C-world.
Signed-off-by: Jonathan Austin <jonathan.austin@arm.com>
Acked-by: Nicolas Pitre <nico@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
* pm-fixes:
cpufreq / intel_pstate: Do not load on VM that does not report max P state.
cpufreq / intel_pstate: Fix intel_pstate_init() error path
PM / QoS: Remove device PM QoS sysfs attributes at the right place
PM / QoS: Fix concurrency issues and memory leaks in device PM QoS
cpufreq: highbank: do not initialize array with a loop
PM / OPP: improve introductory documentation
cpufreq: Fix a typo in comment
mailbox, pl320-ipc: remove __init from probe function
It seems some VMs support the P state MSRs but return zeros. Fail
gracefully if we are running in this environment.
References: https://bugzilla.redhat.com/show_bug.cgi?id=916833
Reported-by: Josh Boyer <jwboyer@redhat.com>
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If cpufreq_register_driver() fails just free memory that has been
allocated and return. intel_pstate_exit() function is removed since we
are built-in only now there is no reason for a module exit procedure.
Reported-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
John W. Linville says:
====================
This time just passing along a big batch of fixes from Johannes...
For the mac80211 bits:
"Here I have fixes from Ben Greear for stray work items when deleting
interfaces, another idle handling fix from Felix, a fix from Marco ro a
mesh PS buffering crash and I have a fix for the VHT MCS calculation in
association request frames and more nl80211 feature advertising removal
as well as a workaround to increase the dump size if the SKB overhead is
too large. For 3.10 I already have a complete fix queued, but that also
requires (simple) userspace changes."
And for the iwlwifi bits:
"The patches from Dor fix a bunch of calibration issues in the new MVM
driver, and Emmanuel has a number of fixes there as well. Also, we
decided to disable 8k A-MSDU by default, so that's in there. My own
patches are addressing an issue we found with the new devices but that
seems to also exist on older ones, the DMA writeback the devices do can
be delayed and cause issues. The fix is unfortunately relatively large
and depends on two other changes (to not be hugely conflicting), but I
think it's still worth it at this point."
As Johannes says, it is a bit large. But I hope it is still early
enough in the cycle to make that worthwhile.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The SLIPORT_SEMAPHORE register shadowed in the
config-space may not reflect the correct POST stage after
an EEH reset in BE2/3; it may return FW_READY state even though
FW is not ready. This causes the driver to prematurely
poll the FW mailbox and fail.
For BE2/3 use the CSR-BAR/0xac instead.
Reported-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Hutchings says:
====================
Fix regressions introduced by the last set of fixes (sorry):
1. Potential deadlock when disabling TX queues.
2. RX was broken on architectures other than x86 and powerpc.
I still expect to send one more bug fix for 3.9, but as it sometimes
takes days to reproduce the bug it's going to take a couple of weeks of
testing to be confident that it's really fixed.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
RX DMA buffers start at an offset of EFX_PAGE_IP_ALIGN bytes from the
start of a cache line. This offset obviously needs to be included in
the virtual address, but this was missed in commit b590ace09d
('sfc: Fix efx_rx_buf_offset() in the presence of swiotlb') since
EFX_PAGE_IP_ALIGN is equal to 0 on both x86 and powerpc.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
efx_device_detach_sync() locks all TX queues before marking the device
detached and thus disabling further TX scheduling. But it can still
be interrupted by TX completions which then result in TX scheduling in
soft interrupt context. This will deadlock when it tries to acquire
a TX queue lock that efx_device_detach_sync() already acquired.
To avoid deadlock, we must use netif_tx_{,un}lock_bh().
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
According to PRM we need to disable hsync and vsync even though ADPA is
disabled. The previous code did infact do the opposite so we fix it.
Signed-off-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=56359
Tested-by: max <manikulin@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
krb5 mounts started failing as of
683428fae8 "sunrpc: Update svcgss xdr
handle to rpsec_contect cache".
The problem is that mounts are usually done with some host principal
which isn't normally mapped to any user, in which case svcgssd passes
down uid -1, which the kernel is then expected to map to the
export-specific anonymous uid or gid.
The new uid_valid/gid_valid checks were therefore causing that downcall
to fail.
(Note the regression may not have been seen with older userspace that
tended to map unknown principals to an anonymous id on their own rather
than leaving it to the kernel.)
Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Joseph was hitting a failure case when mounting efivarfs which
resulted in an incorrect error message,
$ sudo mount -v /sys/firmware/efi/efivars mount: Cannot allocate memory
triggered when efivarfs_valid_name() returned -EINVAL.
Make sure we pass accurate return values up the stack if
efivarfs_fill_super() fails to build inodes for EFI variables.
Reported-by: Joseph Yasi <joe.yasi@gmail.com>
Reported-by: Lingzhu Xiang <lxiang@redhat.com>
Cc: Josh Boyer <jwboyer@redhat.com>
Cc: Jeremy Kerr <jk@ozlabs.org>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: <stable@vger.kernel.org> # v3.8
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Stricter validation was introduced with commit da27a24383
("efivarfs: guid part of filenames are case-insensitive") and commit
47f531e8ba ("efivarfs: Validate filenames much more aggressively"),
which is necessary for the guid portion of efivarfs filenames, but we
don't need to be so strict with the first part, the variable name. The
UEFI specification doesn't impose any constraints on variable names
other than they be a NULL-terminated string.
The above commits caused a regression that resulted in users seeing
the following message,
$ sudo mount -v /sys/firmware/efi/efivars mount: Cannot allocate memory
whenever pstore EFI variables were present in the variable store,
since their variable names failed to pass the following check,
/* GUID should be right after the first '-' */
if (s - 1 != strchr(str, '-'))
as a typical pstore filename is of the form, dump-type0-10-1-<guid>.
The fix is trivial since the guid portion of the filename is GUID_LEN
bytes, we can use (len - GUID_LEN) to ensure the '-' character is
where we expect it to be.
(The bogus ENOMEM error value will be fixed in a separate patch.)
Reported-by: Joseph Yasi <joe.yasi@gmail.com>
Tested-by: Joseph Yasi <joe.yasi@gmail.com>
Reported-by: Lingzhu Xiang <lxiang@redhat.com>
Cc: Josh Boyer <jwboyer@redhat.com>
Cc: Jeremy Kerr <jk@ozlabs.org>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: <stable@vger.kernel.org> # v3.8
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
UEFI variables are typically stored in flash. For various reasons, avaiable
space is typically not reclaimed immediately upon the deletion of a
variable - instead, the system will garbage collect during initialisation
after a reboot.
Some systems appear to handle this garbage collection extremely poorly,
failing if more than 50% of the system flash is in use. This can result in
the machine refusing to boot. The safest thing to do for the moment is to
forbid writes if they'd end up using more than half of the storage space.
We can make this more finegrained later if we come up with a method for
identifying the broken machines.
Signed-off-by: Matthew Garrett <matthew.garrett@nebula.com>
Cc: Josh Boyer <jwboyer@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Disable bits for ADPA HSYNC and VSYNC where mixed up resulting in suspend
becoming standby and vice versa. Fixed by swapping their bit position.
Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The number of DMA fault reasons in intel's document are from 1
to 0xD, but in dmar.c fault reason 0xD is not printed out.
In this document:
"Intel Virtualization Technology for Directed I/O Architecture Specification"
http://download.intel.com/technology/computing/vptech/Intel(r)_VT_for_Direct_IO.pdf
Chapter 4. Support For Device-IOTLBs
Table 6. Unsuccessful Translated Requests
There is fault reason for 0xD not listed in kernel:
Present context-entry used to process translation request
specifies blocking of Translation Requests (Translation Type (T)
field value not equal to 01b).
This patch adds reason 0xD as well.
Signed-off-by: Li, Zhen-Hua <zhen-hual@hp.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Donald Dutile <ddutile@redhat.com>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Hannes Reinecke <hare@suse.de>
Link: http://lkml.kernel.org/r/1362537797-6034-1-git-send-email-zhen-hual@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
While PCI card faces EEH errors, reset (usually hot reset) is
expected to recover from the EEH errors. After EEH core finishes
the reset, the driver callback (be_eeh_reset) is called and wait
the firmware to complete POST successfully. The original code would
return with error once detecting failure during POST stage. That
seems not enough.
The patch forces the driver (be_eeh_reset) to wait the firmware
completes POST until timeout, instead of returning error upon
detection POST failure immediately. Also, it would improve the
reliability of the EEH funtionality of the driver.
Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Acked-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a router forwards a packet that contains the IPv4 timestamp option,
if there is no space left in the option for the router to add its own
timestamp, then the router increments the Overflow value in the option.
However, if the addresses of the routers are prespecified in the option,
then the overflow condition cannot happen: the option is structured so
that each prespecified router has a place to write its timestamp. Other
routers do not add a timestamp, so there will never be a lack of space.
This fix ensures that the Overflow value in the IPv4 timestamp option is
not incremented when the addresses of the routers are prespecified, even
if the Pointer value is greater than the Length value.
Signed-off-by: David Ward <david.ward@ll.mit.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
We should use time_after_eq() to get maximum latency of two ticks,
instead of three.
Bug added in commit 24f8b2385 (net: increase receive packet quantum)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix new kernel-doc warnings in net/core/dev.c:
Warning(net/core/dev.c:4788): No description found for parameter 'new_carrier'
Warning(net/core/dev.c:4788): Excess function parameter 'new_carries' description in 'dev_change_carrier'
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
We should reset nf settings bond to the skb as ipip/ipgre do.
If not, the conntrack/nat info bond to the origin packet may continually
redirect the packet to vxlan interface causing a routing loop.
this is the scenario:
VETP VXLAN Gateway
/----\ /---------------\
| | | |
| vx+--+vx --NAT-> eth0+--> Internet
| | | |
\----/ \---------------/
when there are any packet coming from internet to the vetp, there will be lots
of garbage packets coming out the gateway's vxlan interface, but none actually
sent to the physical interface, because they are redirected back to the vxlan
interface in the postrouting chain of NAT rule, and dmesg complains:
Mar 1 21:52:53 debian kernel: [ 8802.997699] Dead loop on virtual device vxlan0, fix it urgently!
Mar 1 21:52:54 debian kernel: [ 8804.004907] Dead loop on virtual device vxlan0, fix it urgently!
Mar 1 21:52:55 debian kernel: [ 8805.012189] Dead loop on virtual device vxlan0, fix it urgently!
Mar 1 21:52:56 debian kernel: [ 8806.020593] Dead loop on virtual device vxlan0, fix it urgently!
the patch should fix the problem
Signed-off-by: Zang MingJie <zealot0630@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
QFQ+ can select for service only 'eligible' aggregates, i.e.,
aggregates that would have started to be served also in the emulated
ideal system. As a consequence, for QFQ+ to be work conserving, at
least one of the active aggregates must be eligible when it is time to
choose the next aggregate to serve.
The set of eligible aggregates is updated through the function
qfq_update_eligible(), which does guarantee that, after its
invocation, at least one of the active aggregates is eligible.
Because of this property, this function is invoked in
qfq_deactivate_agg() to guarantee that at least one of the active
aggregates is still eligible after an aggregate has been deactivated.
In particular, the critical case is when there are other active
aggregates, but the aggregate being deactivated happens to be the only
one eligible.
However, this precaution is not needed for QFQ+ to be work conserving,
because update_eligible() is always invoked also at the beginning of
qfq_choose_next_agg(). This patch removes the additional invocation of
update_eligible() in qfq_deactivate_agg().
Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Reviewed-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
By definition of (the algorithm of) QFQ+, the system virtual time must
be pushed up only if there is no 'eligible' aggregate, i.e. no
aggregate that would have started to be served also in the ideal
system emulated by QFQ+. QFQ+ serves only eligible aggregates, hence
the aggregate currently in service is eligible. As a consequence, to
decide whether there is no eligible aggregate, QFQ+ must also check
whether there is no aggregate in service.
Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Reviewed-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Aggregate budgets are computed so as to guarantee that, after an
aggregate has been selected for service, that aggregate has enough
budget to serve at least one maximum-size packet for the classes it
contains. For this reason, after a new aggregate has been selected
for service, its next packet is immediately dequeued, without any
further control.
The maximum packet size for a class, lmax, can be changed through
qfq_change_class(). In case the user sets lmax to a lower value than
the the size of some of the still-to-arrive packets, QFQ+ will
automatically push up lmax as it enqueues these packets. This
automatic push up is likely to happen with TSO/GSO.
In any case, if lmax is assigned a lower value than the size of some
of the packets already enqueued for the class, then the following
problem may occur: the size of the next packet to dequeue for the
class may happen to be larger than lmax, after the aggregate to which
the class belongs has been just selected for service. In this case,
even the budget of the aggregate, which is an unsigned value, may be
lower than the size of the next packet to dequeue. After dequeueing
this packet and subtracting its size from the budget, the latter would
wrap around.
This fix prevents the budget from wrapping around after any packet
dequeue.
Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Reviewed-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If no aggregate is in service, then the function qfq_dequeue() does
not dequeue any packet. For this reason, to guarantee QFQ+ to be work
conserving, a just-activated aggregate must be set as in service
immediately if it happens to be the only active aggregate.
This is done by the function qfq_enqueue().
Unfortunately, the function qfq_add_to_agg(), used to add a class to
an aggregate, does not perform this important additional operation.
In particular, if: 1) qfq_add_to_agg() is invoked to complete the move
of a class from a source aggregate, becoming, for this move, inactive,
to a destination aggregate, becoming instead active, and 2) the
destination aggregate becomes the only active aggregate, then this
aggregate is not however set as in service. QFQ+ remains then in a
non-work-conserving state until a new invocation of qfq_enqueue()
recovers the situation.
This fix solves the problem by moving the logic for setting an
aggregate as in service directly into the function qfq_activate_agg().
Hence, from whatever point qfq_activate_aggregate() is invoked, QFQ+
remains work conserving. Since the more-complex logic of this new
version of activate_aggregate() is not necessary, in qfq_dequeue(), to
reschedule an aggregate that finishes its budget, then the aggregate
is now rescheduled by invoking directly the functions needed.
Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Reviewed-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Between two invocations of make_eligible, the system virtual time may
happen to grow enough that, in its binary representation, a bit with
higher order than 31 flips. This happens especially with
TSO/GSO. Before this fix, the mask used in make_eligible was computed
as (1UL<<index_of_last_flipped_bit)-1, whose value is well defined on
a 64-bit architecture, because index_of_flipped_bit <= 63, but is in
general undefined on a 32-bit architecture if index_of_flipped_bit > 31.
The fix just replaces 1UL with 1ULL.
Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Reviewed-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
QFQ+ schedules the active aggregates in a group using a bucket list
(one list per group). The bucket in which each aggregate is inserted
depends on the aggregate's timestamps, and the number
of buckets in a group is enough to accomodate the possible (range of)
values of the timestamps of all the aggregates in the group. For this
property to hold, timestamps must however be computed correctly. One
necessary condition for computing timestamps correctly is that the
number of bits dequeued for each aggregate, while the aggregate is in
service, does not exceed the maximum budget budgetmax assigned to the
aggregate.
For each aggregate, budgetmax is proportional to the number of classes
in the aggregate. If the number of classes of the aggregate is
decreased through qfq_change_class(), then budgetmax is decreased
automatically as well. Problems may occur if the aggregate is in
service when budgetmax is decreased, because the current remaining
budget of the aggregate and/or the service already received by the
aggregate may happen to be larger than the new value of budgetmax. In
this case, when the aggregate is eventually deselected and its
timestamps are updated, the aggregate may happen to have received an
amount of service larger than budgetmax. This may cause the aggregate
to be assigned a higher virtual finish time than the maximum
acceptable value for the last bucket in the bucket list of the group.
This fix introduces a cap that addresses this issue.
Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Reviewed-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
DTR/RTS need to be raised, regardless of the open() mode, but not
if the port has already shutdown.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Without a memory and compiler barrier, the task state change
can migrate relative to the condition testing in a blocking loop.
However, the task state change must be visible across all cpus
prior to testing those conditions. Failing to do this can result
in the familiar 'lost wakeup' and this task will hang until killed.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Although tty_lock() already protects concurrent update to
blocked_open, that fails to meet the separation-of-concerns between
tty_port and tty.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Saving the port count bump is unsafe. If the tty is hung up while
this open was blocking, the port count is zeroed.
Explicitly check if the tty was hung up while blocking, and correct
the port count if not.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
