Hans de Goede [Tue, 16 Jun 2015 14:27:49 +0000 (16:27 +0200)]
acpi-video-detect: Unregister acpi_video backlight when dmi quirks are added
Make acpi_video_set_dmi_backlight_type() call
acpi_video_unregister_backlight() when the new dmi quirk results in
the desired backlight interface being of a type other then
acpi_backlight_video.
This avoid the need for the second if in the following construction
which is currently found in many platform/x86 drivers:
if (prefer_vendor_quirk)
acpi_video_dmi_promote_vendor();
if (!acpi_video_backlight_support())
acpi_video_unregister_backlight()
This second if-block will be removed from the platform drivers as part
of their conversion to the new backlight interface selection API.
Currently we have 2 kernel commandline options + dmi-quirks in 3 places all
interacting (in interesting ways) to select which which backlight interface
to use. On the commandline we've acpi_backlight=[video|vendor] and
video.use_native_backlight=[0|1]. DMI quirks we have in
acpi/video-detect.c, acpi/video.c and drivers/platform/x86/*.c .
This commit is the first step to cleaning this up, replacing the 2 cmdline
options with just acpi_backlight=[video|vendor|native|none], and adds a
new API to video_detect.c to reflect this.
Follow up commits will also move other related code, like unregistering the
acpi_video backlight interface if it was registered before other drivers
which take priority over it are loaded, to video_detect.c where this
logic really belongs.
Hans de Goede [Tue, 16 Jun 2015 14:27:47 +0000 (16:27 +0200)]
acpi-video-detect: video: Make video_detect code part of the video module
This is a preparation patch for the backlight interface selection logic
cleanup, there are 2 reasons to not always build the video_detect code
into the kernel:
1) In order for the video_detect.c to also deal with / select native
backlight interfaces on win8 systems, instead of doing this in video.c
where it does not belong, video_detect.c needs to call into the backlight
class code. Which cannot be done if it is builtin and the blacklight class
is not.
2) Currently all the platform/x86 drivers which have quirks to prefer
the vendor driver over acpi-video call acpi_video_unregister_backlight()
to remove the acpi-video backlight interface, this logic really belongs
in video_detect.c, which will cause video_detect.c to depend on symbols of
video.c and video.c already depends on video_detect.c symbols, so they
really need to be a single module.
Note that this commits make 2 changes so as to maintain 100% kernel
commandline compatibility:
1) The __setup call for the acpi_backlight= handling is moved to
acpi/util.c as __setup may only be used by code which is alwasy builtin
2) video.c is renamed to acpi_video.c so that it can be combined with
video_detect.c into video.ko
This commit also makes changes to drivers/platform/x86/Kconfig to ensure
that drivers which use acpi_video_backlight_support() from video_detect.c,
will not be built-in when acpi_video is not built in. This also changes
some "select" uses to "depends on" to avoid dependency loops.
Hans de Goede [Tue, 16 Jun 2015 14:27:46 +0000 (16:27 +0200)]
acpi-video-detect: Move acpi_osi_is_win8 to osl.c
acpi_osi_is_win8 needs access to acpi_gbl_osi_data which is not exported,
so move it to osl.c. Alternatively we could export acpi_gbl_osi_data but
that seems undesirable.
This allows video_detect.c to be build as a module, besides that
acpi_osi_is_win8() is something which does not really belong in
video_detect.c in the first place.
Hans de Goede [Tue, 16 Jun 2015 14:27:45 +0000 (16:27 +0200)]
acpi-video-detect: Move acpi_is_video_device() to acpi/scan.c
This allows video_detect.c to be build as a module, this is a preparation
patch for the backlight interface selection logic cleanup.
Note this commit also causes acpi_is_video_device() to always be build
indepedent of CONFIG_ACPI_VIDEO, as there is no reason to make its
building depend on CONFIG_ACPI_VIDEO.
Hans de Goede [Tue, 16 Jun 2015 14:27:44 +0000 (16:27 +0200)]
acpi-video-detect: Make acpi_video_get_capabilities a private function
acpi_video_get_capabilities() is only used inside video_detect.c so make
it static. While at it also remove the prototype for the non existent
acpi_video_display_switch_support function from acpi.h
Hans de Goede [Tue, 16 Jun 2015 14:27:43 +0000 (16:27 +0200)]
acpi-video-detect: Remove the unused acpi_video_dmi_demote_vendor() function
Remove the now unused acpi_video_dmi_demote_vendor() function, this was
never a proper counter part of acpi_video_dmi_promote_vendor() since
the calls to acpi_video_dmi_promote_vendor() are not counted.
Hans de Goede [Tue, 16 Jun 2015 14:27:42 +0000 (16:27 +0200)]
apple-gmux: Stop using acpi_video_dmi_demote_vendor()
acpi_video_dmi_demote_vendor() is going away as part of the cleanup of
the code for determinging which backlight class driver(s) to register.
The call to acpi_video_dmi_demote_vendor() was meant to undo the call to
acpi_video_dmi_promote_vendor() when the gmux device is removed, this is
questionable though since the promote call sets a flag, not a counter, so
the demote call may undo a promoto done elsewhere. Moreover in practice
this is a nop since the gmux device is never removed, and the flag is only
checked when acpi/video.ko gets loaded, so even if the user manually
removes apple-gmux the demote call is still a nop as video.ko will already
have loaded by this time.
Also note that none of the other users of acpi_video_dmi_promote_vendor()
use acpi_video_dmi_demote_vendor().
If we ever encounter a system with a gmux where the acpi-video interface
should be used, then the proper fix would be to dmi-blacklist the gmux
driver on that system.
Hans de Goede [Tue, 16 Jun 2015 14:27:41 +0000 (16:27 +0200)]
samsung-laptop: Use acpi_video_unregister_backlight instead of acpi_video_unregister
acpi_video_unregister() not only unregisters the acpi-video backlight
interface but also unregisters the acpi video bus event listener, causing
e.g. brightness hotkey presses to no longer generate keypress events.
The unregistering of the acpi video bus event listener usually is
undesirable, which by itself is a good reason to switch to
acpi_video_unregister_backlight().
Another problem with using acpi_video_unregister() rather then using
acpi_video_unregister_backlight() is that on systems with an intel video
opregion (most systems) and a broken_acpi_video quirk, whether or not
the acpi video bus event listener actually gets unregistered depends on
module load ordering:
Scenario a:
1) acpi/video.ko gets loaded (*), does not do acpi_video_register as there
is an intel opregion.
2) intel.ko gets loaded, calls acpi_video_register() which registers both
the listener and the acpi backlight interface
3) samsung-laptop.ko gets loaded, calls acpi_video_unregister() causing
both the listener and the acpi backlight interface to unregister
Scenario b:
1) acpi/video.ko gets loaded (*), does not do acpi_video_register as there
is an intel opregion.
2) samsung-laptop.ko gets loaded, calls acpi_video_dmi_promote_vendor(),
calls acpi_video_unregister(), which is a nop since acpi_video_register
has not yet been called
2) intel.ko gets loaded, calls acpi_video_register() which registers
the listener, but does not register the acpi backlight interface due to
the call to the preciding call to acpi_video_dmi_promote_vendor()
*) acpi/video.ko always loads first as both other modules depend on it.
So we end up with or without an acpi video bus event listener depending
on module load ordering, not good.
Switching to using acpi_video_unregister_backlight() means that independ
of ordering we will always have an acpi video bus event listener fixing
this.
Note that this commit means that systems without an intel video opregion,
and systems which were hitting scenario a wrt module load ordering, are
now getting an acpi video bus event listener while before they were not!
On some systems this may cause the brightness hotkeys to start generating
keypresses while before they were not (good), while on other systems this
may cause the brightness hotkeys to generate multiple keypress events for
a single press (not so good). Since on most systems the acpi video bus is
the canonical source for brightness events I believe that the latter case
will needs to be handled on a case by case basis by filtering out the
duplicate keypresses at the other source for them.
Hans de Goede [Tue, 16 Jun 2015 14:27:40 +0000 (16:27 +0200)]
asus-wmi: Use acpi_video_unregister_backlight instead of acpi_video_unregister
acpi_video_unregister() not only unregisters the acpi-video backlight
interface but also unregisters the acpi video bus event listener, causing
e.g. brightness hotkey presses to no longer generate keypress events.
The unregistering of the acpi video bus event listener usually is
undesirable, which by itself is a good reason to switch to
acpi_video_unregister_backlight().
Another problem with using acpi_video_unregister() rather then using
acpi_video_unregister_backlight() is that on systems with an intel video
opregion (most systems) and a wmi_backlight_power quirk, whether or not
the acpi video bus event listener actually gets unregistered depends on
module load ordering:
Scenario a:
1) acpi/video.ko gets loaded (*), does not do acpi_video_register as there
is an intel opregion.
2) intel.ko gets loaded, calls acpi_video_register() which registers both
the listener and the acpi backlight interface
3) asus-wmi.ko gets loaded, calls acpi_video_unregister() causing both
the listener and the acpi backlight interface to unregister
Scenario b:
1) acpi/video.ko gets loaded (*), does not do acpi_video_register as there
is an intel opregion.
2) asus-wmi.ko gets loaded, calls acpi_video_dmi_promote_vendor(),
calls acpi_video_unregister(), which is a nop since acpi_video_register
has not yet been called
2) intel.ko gets loaded, calls acpi_video_register() which registers
the listener, but does not register the acpi backlight interface due to
the call to the preciding call to acpi_video_dmi_promote_vendor()
*) acpi/video.ko always loads first as both other modules depend on it.
So we end up with or without an acpi video bus event listener depending
on module load ordering, not good.
Switching to using acpi_video_unregister_backlight() means that independ
of ordering we will always have an acpi video bus event listener fixing
this.
Note that this commit means that systems without an intel video opregion,
and systems which were hitting scenario a wrt module load ordering, are
now getting an acpi video bus event listener while before they were not!
On some systems this may cause the brightness hotkeys to start generating
keypresses while before they were not (good), while on other systems this
may cause the brightness hotkeys to generate multiple keypress events for
a single press (not so good). Since on most systems the acpi video bus is
the canonical source for brightness events I believe that the latter case
will needs to be handled on a case by case basis by filtering out the
duplicate keypresses at the other source for them.
Hans de Goede [Tue, 16 Jun 2015 14:27:39 +0000 (16:27 +0200)]
apple_gmux: Use acpi_video_unregister_backlight instead of acpi_video_unregister
acpi_video_unregister() not only unregisters the acpi-video backlight
interface but also unregisters the acpi video bus event listener, causing
e.g. brightness hotkey presses to no longer generate keypress events.
The unregistering of the acpi video bus event listener usually is
undesirable, which by itself is a good reason to switch to
acpi_video_unregister_backlight().
Another problem with using acpi_video_unregister() rather then using
acpi_video_unregister_backlight() is that on systems with an intel video
opregion (most systems) whether or not the acpi video bus event listener
actually gets unregistered depends on module load ordering:
Scenario a:
1) acpi/video.ko gets loaded (*), does not do acpi_video_register as there
is an intel opregion.
2) intel.ko gets loaded, calls acpi_video_register() which registers both
the listener and the acpi backlight interface
3) apple-gmux.ko gets loaded, calls acpi_video_unregister() causing both
the listener and the acpi backlight interface to unregister
Scenario b:
1) acpi/video.ko gets loaded (*), does not do acpi_video_register as there
is an intel opregion.
2) apple-gmux.ko gets loaded, calls acpi_video_dmi_promote_vendor(),
calls acpi_video_unregister(), which is a nop since acpi_video_register
has not yet been called
2) intel.ko gets loaded, calls acpi_video_register() which registers
the listener, but does not register the acpi backlight interface due to
the call to the preciding call to acpi_video_dmi_promote_vendor()
*) acpi/video.ko always loads first as both other modules depend on it.
So we end up with or without an acpi video bus event listener depending
on module load ordering, not good.
Switching to using acpi_video_unregister_backlight() means that independ
of ordering we will always have an acpi video bus event listener fixing
this.
Note that this commit means that systems without an intel video opregion,
and systems which were hitting scenario a wrt module load ordering, are
now getting an acpi video bus event listener while before they were not!
On some systems this may cause the brightness hotkeys to start generating
keypresses while before they were not (good), while on other systems this
may cause the brightness hotkeys to generate multiple keypress events for
a single press (not so good). Since on most systems the acpi video bus is
the canonical source for brightness events I believe that the latter case
will needs to be handled on a case by case basis by filtering out the
duplicate keypresses at the other source for them.
Bjorn Helgaas [Thu, 18 Jun 2015 22:56:45 +0000 (17:56 -0500)]
Merge branches 'pci/host-xgene' and 'pci/hotplug' into next
* pci/host-xgene:
PCI: xgene: Allow config access to Root Port even when link is down
PCI: xgene: Disable Configuration Request Retry Status for v1 silicon
* pci/hotplug:
PCI: pciehp: Inline the "handle event" functions into the ISR
PCI: pciehp: Rename queue_interrupt_event() to pciehp_queue_interrupt_event()
PCI: pciehp: Make queue_interrupt_event() void
PCI: pciehp: Clean up debug logging
Peter Zijlstra [Thu, 11 Jun 2015 12:46:48 +0000 (14:46 +0200)]
hrtimer: Allow hrtimer::function() to free the timer
Currently an hrtimer callback function cannot free its own timer
because __run_hrtimer() still needs to clear HRTIMER_STATE_CALLBACK
after it. Freeing the timer would result in a clear use-after-free.
Solve this by using a scheme similar to regular timers; track the
current running timer in hrtimer_clock_base::running.
Peter Zijlstra [Wed, 17 Jun 2015 12:29:24 +0000 (14:29 +0200)]
seqcount: Introduce raw_write_seqcount_barrier()
Introduce raw_write_seqcount_barrier(), a new construct that can be
used to provide write barrier semantics in seqcount read loops instead
of the usual consistency guarantee.
But avoids issueing two back-to-back smp_wmb() instructions.
This construct works because the read side will 'stall' when observing
odd values. This means that -- referring to the example in the comment
below -- even though there is no (matching) read barrier between the
loads of X and Y, we cannot observe !x && !y, because:
- if we observe Y == false we must observe the first sequence
increment, which makes us loop, until
- we observe !(seq & 1) -- the second sequence increment -- at which
time we must also observe T == true.
Peter Zijlstra [Thu, 11 Jun 2015 12:46:46 +0000 (14:46 +0200)]
seqcount: Rename write_seqcount_barrier()
I'll shortly be introducing another seqcount primitive that's useful
to provide ordering semantics and would like to use the
write_seqcount_barrier() name for that.
Seeing how there's only one user of the current primitive, lets rename
it to invalidate, as that appears what its doing.
While there, employ lockdep_assert_held() instead of
assert_spin_locked() to not generate debug code for regular kernels.
Peter Zijlstra [Thu, 11 Jun 2015 12:46:45 +0000 (14:46 +0200)]
hrtimer: Fix hrtimer_is_queued() hole
A queued hrtimer that gets restarted (hrtimer_start*() while
hrtimer_is_queued()) will briefly appear as unqueued/inactive, even
though the timer has always been active, we just moved it.
Close this hole by preserving timer->state in
hrtimer_start_range_ns()'s remove_hrtimer() call.
Bjorn Helgaas [Mon, 15 Jun 2015 02:35:13 +0000 (21:35 -0500)]
PCI: pciehp: Inline the "handle event" functions into the ISR
The pciehp_handle_*() functions (pciehp_handle_attention_button(), etc.)
only contain a line or two of useful code, so it's clumsy to put
them in separate functions. All they so is add an event to a work queue,
and it's clearer to see that directly in the ISR.
Inline them directly into pcie_isr(). No functional change.
Bjorn Helgaas [Mon, 15 Jun 2015 02:27:52 +0000 (21:27 -0500)]
PCI: pciehp: Make queue_interrupt_event() void
Nobody looks at the return value from queue_interrupt_event(), so errors
were silently ignored. Convert it to a "void" function and note the error
in the dmesg log.
Duc Dang [Thu, 18 Jun 2015 18:45:39 +0000 (11:45 -0700)]
PCI: xgene: Allow config access to Root Port even when link is down
Previously, when a Root Port's link was down, we didn't allow config access
to the Root Port, which meant that if the Root Port led to an empty slot,
"lspci" didn't even show the Root Port.
Allow config access to Root Port even when link is down.
Duc Dang [Sat, 13 Jun 2015 00:35:57 +0000 (17:35 -0700)]
PCI: xgene: Disable Configuration Request Retry Status for v1 silicon
When a CPU reads the Vendor and Device ID of a non-existent device, the
controller should fabricate return data of 0xFFFFFFFF. Configuration
Request Retry Status (CRS) is not applicable in this case because the
device doesn't exist at all.
The X-Gene v1 PCIe controller has a bug in the CRS logic such that when CRS
is enabled, it fabricates return data of 0xFFFF0001 for this case, which
means "the device exists but is not ready." That causes the PCI core to
retry the read until it times out after 60 seconds.
Disable CRS capability advertisement by clearing the CRS Software
Visibility bit in the Root Capabilities Register.
Commit b9a5e5e18fbf "ACPI / init: Fix the ordering of
acpi_reserve_resources()" overlooked the fact that the memory
and/or I/O regions reserved by acpi_reserve_resources() may
conflict with those reserved by the PNP "system" driver.
If that conflict actually takes place, it causes the reservations
made by the "system" driver to fail while before commit b9a5e5e18fbf
all reservations made by it and by acpi_reserve_resources() would be
successful. In turn, that allows the resources that haven't been
reserved by the "system" driver to be used by others (e.g. PCI) which
sometimes leads to functional problems (up to and including boot
failures).
To fix that issue, introduce a common resource reservation routine,
acpi_reserve_region(), to be used by both acpi_reserve_resources()
and the "system" driver, that will track all resources reserved by
it and avoid making conflicting requests.
Back in the days, vgic.c used to have an intimate knowledge of
the actual GICv2. These days, this has been abstracted away into
hardware-specific backends.
Remove the now useless arm-gic.h #include directive, making it
clear that GICv2 specific code doesn't belong here.
Sebastian Ott [Tue, 16 Jun 2015 12:03:37 +0000 (14:03 +0200)]
s390/setup: fix DMA_API_DEBUG warnings
Addresses from the usable space in [_ehead, _stext] lead to false
positives in DMA_API_DEBUG code (which will complain when an address
is in [_text, _etext]).
Avoid these warnings by not using that memory in case of
CONFIG_DMA_API_DEBUG=y.
John Stultz [Wed, 17 Jun 2015 18:16:43 +0000 (11:16 -0700)]
selftest: Timers: Avoid signal deadlock in leap-a-day
In 0c4a5fc95b1df (Add leap-second timer edge testing to
leap-a-day.c), we added a timer to the test which checks to make
sure timers near the leapsecond edge behave correctly.
However, the output generated from the timer uses ctime_r, which
isn't async-signal safe, and should that signal land while the
main test is using ctime_r to print its output, its possible for
the test to deadlock on glibc internal locks.
Thus this patch reworks the output to avoid using ctime_r in
the signal handler.
Russell King [Tue, 16 Jun 2015 22:07:01 +0000 (23:07 +0100)]
irq: spear-shirq: Fix race in installing chained IRQ handler
Fix a race where a pending interrupt could be received and the handler
called before the handler's data has been setup, by converting to
irq_set_chained_handler_and_data().
Russell King [Tue, 16 Jun 2015 22:06:56 +0000 (23:06 +0100)]
irq: irq-keystone: Fix race in installing chained IRQ handler
Fix a race where a pending interrupt could be received and the handler
called before the handler's data has been setup, by converting to
irq_set_chained_handler_and_data().
Russell King [Tue, 16 Jun 2015 22:06:50 +0000 (23:06 +0100)]
gpio: gpio-tegra: Fix race in installing chained IRQ handler
Fix a race where a pending interrupt could be received and the handler
called before the handler's data has been setup, by converting to
irq_set_chained_handler_and_data().
Russell King [Tue, 16 Jun 2015 22:06:45 +0000 (23:06 +0100)]
gpio: gpio-mxs: Fix race in installing chained IRQ handler
Fix a race where a pending interrupt could be received and the handler
called before the handler's data has been setup, by converting to
irq_set_chained_handler_and_data().
Russell King [Tue, 16 Jun 2015 22:06:40 +0000 (23:06 +0100)]
gpio: gpio-mxc: Fix race in installing chained IRQ handler
Fix a race where a pending interrupt could be received and the handler
called before the handler's data has been setup, by converting to
irq_set_chained_handler_and_data().
Russell King [Tue, 16 Jun 2015 22:06:35 +0000 (23:06 +0100)]
ARM: gemini: Fix race in installing GPIO chained IRQ handler
The gemini code was installing its chained interrupt handler (which
enables the interrupt) before it was setting its data, which is bad if
the IRQ was previously pending. Avoid this problem by converting it to
irq_set_chained_handler_and_data().
Russell King [Tue, 16 Jun 2015 22:06:30 +0000 (23:06 +0100)]
GPU: ipu: Fix race in installing IPU chained IRQ handler
The IPU code was installing its chained interrupt handler (which enables
the interrupt) before it was setting its data, which provokes an oops on
kexec. Fix this by converting to irq_set_chained_handler_and_data().
[drm] Initialized drm 1.1.0 20060810
imx-drm display-subsystem: parent device of /soc/aips-bus@02000000/ldb@020e0008/lvds-channel@1 is not available
imx-drm display-subsystem: parent device of /soc/aips-bus@02000000/ldb@020e0008/lvds-channel@1 is not available
Unable to handle kernel NULL pointer dereference at virtual address 00000070
pgd = c0004000
[00000070] *pgd=00000000
Internal error: Oops: 5 [#1] SMP ARM
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.1.0-rc6+ #1693
Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
task: d74c0000 ti: d74aa000 task.ti: d74aa000
PC is at ipu_irq_handle+0x28/0xd8
LR is at ipu_irq_handler+0x6c/0xc0
pc : [<c03c56d8>] lr : [<c03c58a4>] psr: 200001d3
sp : d74abbd0 ip : d74abc00 fp : d74abbfc
r10: 000001e0 r9 : c0085154 r8 : 00000009
r7 : 00000000 r6 : 00000000 r5 : d74abc04 r4 : c0a6b6a8
r3 : 00000000 r2 : 00000009 r1 : d74abc04 r0 : 00000000
Flags: nzCv IRQs off FIQs off Mode SVC_32 ISA ARM Segment kernel
Control: 10c5387d Table: 10004059 DAC: 00000015
Process swapper/0 (pid: 1, stack limit = 0xd74aa210)
Stack: (0xd74abbd0 to 0xd74ac000)
Backtrace:
[<c03c56b0>] (ipu_irq_handle) from [<c03c58a4>] (ipu_irq_handler+0x6c/0xc0)
[<c03c5838>] (ipu_irq_handler) from [<c0080154>] (generic_handle_irq+0x28/0x38)
[<c008012c>] (generic_handle_irq) from [<c0080288>] (__handle_domain_irq+0x5c/0xb8)
[<c008022c>] (__handle_domain_irq) from [<c0009428>] (gic_handle_irq+0x28/0x68)
[<c0009400>] (gic_handle_irq) from [<c0013dc4>] (__irq_svc+0x44/0x5c)
[<c07638fc>] (_raw_spin_unlock_irqrestore) from [<c00803bc>] (__irq_put_desc_unlock+0x1c/0x40)
[<c00803a0>] (__irq_put_desc_unlock) from [<c00841f4>] (__irq_set_handler+0x54/0x5c)
[<c00841a0>] (__irq_set_handler) from [<c03c5f48>] (ipu_probe+0x29c/0x708)
[<c03c5cac>] (ipu_probe) from [<c03d3848>] (platform_drv_probe+0x50/0xac)
[<c03d37f8>] (platform_drv_probe) from [<c03d1f3c>] (driver_probe_device+0x1d4/0x278)
Russell King [Tue, 16 Jun 2015 22:06:20 +0000 (23:06 +0100)]
irq: Add irq_set_chained_handler_and_data()
Driver authors seem to get the ordering of irq_set_chained_handler()
and irq_set_handler_data() wrong - ordering the former before the
latter. This opens a race window where, if there is an interrupt
pending, the handler will be called between these two calls,
potentially resulting in an oops.
Provide a single interface to set both of these together, especially
as that's commonly what is required.
x86/mm/pat, drivers/infiniband/ipath: Use arch_phys_wc_add() and require PAT disabled
We are burrying direct access to MTRR code support on
x86 in order to take advantage of PAT. In the future, we
also want to make the default behaviour of ioremap_nocache()
to use strong UC, use of mtrr_add() on those systems
would make write-combining void.
In order to help both enable us to later make strong
UC default and in order to phase out direct MTRR access
code port the driver over to arch_phys_wc_add() and
annotate that the device driver requires systems to
boot with PAT disabled, with the 'nopat' kernel parameter.
This is a workable compromise given that the ipath device
driver powers the old HTX bus cards that only work in
AMD systems, while the newer IB/qib device driver
powers all PCI-e cards. The ipath device driver is
obsolete, hardware is hard to find and because of this
its a reasonable compromise to require users of ipath
to boot with 'nopat'.
x86/mm/pat, drivers/media/ivtv: Use arch_phys_wc_add() and require PAT disabled
We are burrying direct access to MTRR code support on
x86 in order to take advantage of PAT. In the future, we
also want to make the default behavior of ioremap_nocache()
to use strong UC, at which point the use of mtrr_add() on
those systems would make write-combining void.
In order to help both enable us to later make strong
UC default and in order to phase out direct MTRR access
code, port the driver over to the arch_phys_wc_add() API
and annotate that the device driver requires systems to
boot with PAT disabled, with the 'nopat' kernel parameter.
This is a workable compromise given that the hardware is
really rare these days, and perhaps only some lost souls
stuck with obsolete hardware are expected to be using this
feature of the device driver.
x86/cpu/amd: Give access to the number of nodes in a physical package
Stash the number of nodes in a physical processor package
locally and add an accessor to be called by interested parties.
The first user is the MCE injection module which uses it to find
the node base core in a package for injecting a certain type of
errors.
Feng Tang [Mon, 15 Jun 2015 09:40:01 +0000 (17:40 +0800)]
x86/platform/intel/baytrail: Add comments about why we disabled HPET on Baytrail
This question has been asked many times, and finally I found the
official document which explains the problem of HPET on Baytrail,
that it will halt in deep idle states.
Jason Gerecke [Tue, 16 Jun 2015 01:01:45 +0000 (18:01 -0700)]
HID: wacom: Introduce new 'touch_input' device
Instead of having a single 'input_dev' device that will take either pen
or touch data depending on the type of the device, create seperate devices
devices for each. By splitting things like this, we can support devices
(e.g. the I2C "AES" sensors in some newer tablet PCs) that send both pen
and touch reports from a single endpoint.
This splits the 'wacom_setup_pentouch_input_capabilites' function into
pieces dedicated to doing setup for just the pen interface and just
the touch interface. This makes it easier to focus on the relevant
piece when making changes.
Jason Gerecke [Tue, 16 Jun 2015 01:01:43 +0000 (18:01 -0700)]
HID: wacom: Introduce a new WACOM_DEVICETYPE_PAD device_type
Historically, both the touch and pad tools would have shared the
'BTN_TOOL_FINGER' type. Any time you needed to distinguish the two, you
had to use some other bit of knowledge (e.g. that the pad was on the same
interface as the pen, and thus 'touch_max' would be zero).
To make these checks more readable, we introduce WACOM_DEVICETYPE_PAD.
Although we still have to rely on other bits of knowledge to set this
bit on the right interface (since it cannot be detected from the HID
descriptor), it can be done just once inside 'wacom_setup_device_quirks'.
Jason Gerecke [Tue, 16 Jun 2015 01:01:42 +0000 (18:01 -0700)]
HID: wacom: Treat features->device_type values as flags
The USB devices that this driver has historically supported segregate the
pen and touch portions of the tablet. Oftentimes the segregation would be
done at the interface level, though on occasion (e.g. Cintiq 24HDT) the
tablet would combine two totally independent USB devices behind an internal
USB hub. Because pen and touch never shared the same interface, it made
sense for the 'device_type' to store a single value: "pen" or "touch".
Recently, however, some I2C devices have been created which combine the
two. A first step to accomodating this is to expand 'device_type' so that
it can represent two (or potentially more) types simultaneously. To do
this, we treat it as a bitfield and set/check individual bits rather
than using the '=' and '==' operators.
This should not result in any functional change since no supported devices
(that I'm aware of, at least) have HID descriptors that indicate both
pen and touch reports on a single interface.
Jason Gerecke [Tue, 16 Jun 2015 01:01:41 +0000 (18:01 -0700)]
HID: wacom: Simplify 'wacom_update_name'
A little bit of cleanup work for 'wacom_update_name' to make it easier on
the eyes. Creates a temporary 'name' variable on which we'll perform our
edits. Once the name is in its final form, it will be copied (with
appropriate suffix) to 'wacom_wac->name' and 'wacom_wac->pad_name'.
Ingo Molnar [Thu, 18 Jun 2015 07:36:33 +0000 (09:36 +0200)]
Merge tag 'perf-core-for-mingo-2' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
User visible changes:
- List perf probes to stdout. (Masami Hiramatsu)
- Return error when none of the requested probes were
installed. (Masami Hiramatsu)
- Cut off the gcc optimization postfixes from
function name in 'perf probe'. (Masami Hiramatsu)
- Allow disabling/enabling events dynamicly in 'perf top':
a 'perf top' session can instantly become a 'perf report'
one, i.e. going from dynamic analysis to a static one,
returning to a dynamic one is possible, to toogle the
modes, just press CTRL+z. (Arnaldo Carvalho de Melo)
- Greatly speed up 'perf probe --list' by caching debuginfo.
(Masami Hiramatsu)
- Fix 'perf trace' race condition at the end of started
workloads. (Sukadev Bhattiprolu)
- Fix a problem when opening old perf.data with different
byte order. (Wang Nan)
Infrastructure changes:
- Replace map->referenced & maps->removed_maps with
map->refcnt. (Arnaldo Carvalho de Melo)
- Introduce the xyarray__reset() function. (Jiri Olsa)
John Stultz [Wed, 17 Jun 2015 17:05:53 +0000 (10:05 -0700)]
timekeeping: Copy the shadow-timekeeper over the real timekeeper last
The fix in d151832650ed9 (time: Move clock_was_set_seq update
before updating shadow-timekeeper) was unfortunately incomplete.
The main gist of that change was to do the shadow-copy update
last, so that any state changes were properly duplicated, and
we wouldn't accidentally have stale data in the shadow.
Unfortunately in the main update_wall_time() logic, we update
use the shadow-timekeeper to calculate the next update values,
then while holding the lock, copy the shadow-timekeeper over,
then call timekeeping_update() to do some additional
bookkeeping, (skipping the shadow mirror). The bug with this is
the additional bookkeeping isn't all read-only, and some
changes timkeeper state. Thus we might then overwrite this state
change on the next update.
To avoid this problem, do the timekeeping_update() on the
shadow-timekeeper prior to copying the full state over to
the real-timekeeper.
This avoids problems with both the clock_was_set_seq and
next_leap_ktime being overwritten and possibly the
fast-timekeepers as well.
Many thanks to Prarit for his rigorous testing, which discovered
this problem, along with Prarit and Daniel's work validating this
fix.
Viresh Kumar [Wed, 17 Jun 2015 10:34:46 +0000 (16:04 +0530)]
clockevents: Check state instead of mode in suspend/resume path
CLOCK_EVT_MODE_* macros are present for backward compatibility (as most
of the drivers are still using old ->set_mode() interface).
These macro's shouldn't be used anymore in code, that is common to both
driver interfaces, i.e. ->set_mode() and ->set_state_*().
Drivers implementing ->set_state_*() interface, which have their
clkevt->mode set to 0 (clkevt device structures are normally globally
defined), will not participate in suspend/resume as they will always be
marked as UNUSED.
Fix this by checking state of the clockevent device instead of mode,
which is updated for both the interfaces.
Rabin Vincent [Sun, 14 Jun 2015 17:26:11 +0000 (19:26 +0200)]
mmc: queue: prevent soft lockups on PREEMPT=n
On systems with CONFIG_PREEMPT=n, under certain circumstances, mmcqd
can continuously process requests for several seconds without blocking,
triggering the soft lockup watchdog. For example, this can happen if
mmcqd runs on the CPU which services the controller's interrupt, and
a process on a different CPU continuously writes to the MMC block
device.
NMI watchdog: BUG: soft lockup - CPU#0 stuck for 22s! [mmcqd/0:664]
CPU: 0 PID: 664 Comm: mmcqd/0 Not tainted 4.1.0-rc7+ #4
PC is at _raw_spin_unlock_irqrestore+0x24/0x28
LR is at mmc_start_request+0x104/0x134
...
[<805112a8>] (_raw_spin_unlock_irqrestore) from [<803db664>] (mmc_start_request+0x104/0x134)
[<803db664>] (mmc_start_request) from [<803dc008>] (mmc_start_req+0x274/0x394)
[<803dc008>] (mmc_start_req) from [<803eb2c4>] (mmc_blk_issue_rw_rq+0xd0/0xb98)
[<803eb2c4>] (mmc_blk_issue_rw_rq) from [<803ebe8c>] (mmc_blk_issue_rq+0x100/0x470)
[<803ebe8c>] (mmc_blk_issue_rq) from [<803ecab8>] (mmc_queue_thread+0xd0/0x170)
[<803ecab8>] (mmc_queue_thread) from [<8003fd14>] (kthread+0xe0/0xfc)
[<8003fd14>] (kthread) from [<8000f768>] (ret_from_fork+0x14/0x2c)
Fix it by adding a cond_resched() in the request handling loop so that
other processes get a chance to run.
Linus Torvalds [Thu, 18 Jun 2015 06:56:57 +0000 (20:56 -1000)]
Merge tag 'trace-fix-filter-4.1-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
Pull tracing filter fix from Steven Rostedt:
"Vince Weaver reported a warning when he added perf event filters into
his fuzzer tests. There's a missing check of balanced operations when
parenthesis are used, and this triggers a WARN_ON() and when reading
the failure, the filter reports no failure occurred.
The operands were not being checked if they match, this adds that"
* tag 'trace-fix-filter-4.1-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
tracing: Have filter check for balanced ops
Since when we start discussions about the usage Media Controller for
complex hardware, one thing become clear: the way it is, MC fails to
map anything different than capture/output/m2m video-only streaming.
The point is that MC has entities named as devnodes, but the only
devnode used (before the DVB patches) is MEDIA_ENT_T_DEVNODE_V4L.
Due to the way MC got implemented, however, this entity actually
doesn't represent the devnode, but the hardware I/O engine that
receives data via DMA.
By coincidence, such DMA is associated with the V4L device node
on webcam hardware, but this is not true even for other V4L2
devices. For example, on USB hardware, the DMA is done via the
USB controller. The data passes though a in-kernel filter that
strips off the URB headers. Other V4L2 devices like radio may not
even have DMA. When it have, the DMA is done via ALSA, and not
via the V4L devnode.
In other words, MC is broken as a whole, but tagging it as BROKEN
right now would do more harm than good.
So, instead, let's mark, for now, the DVB part as broken and
block all new changes to MC while we fix this mess, whith
we hopefully will do for the next Kernel version.
Herbert Xu [Wed, 17 Jun 2015 06:58:24 +0000 (14:58 +0800)]
crypto: drivers - Fix Kconfig selects
This patch fixes a number of problems in crypto driver Kconfig
entries:
1. Select BLKCIPHER instead of BLKCIPHER2. The latter is internal
and should not be used outside of the crypto API itself.
2. Do not select ALGAPI unless you use a legacy type like
CRYPTO_ALG_TYPE_CIPHER.
3. Select the algorithm type that you are implementing, e.g., AEAD.
4. Do not select generic C code such as CBC/ECB unless you use them
as a fallback.
5. Remove default n since that is the default default.
Steffen Trumtrar [Tue, 16 Jun 2015 10:59:07 +0000 (12:59 +0200)]
crypto: caam - fix non-64-bit write/read access
The patch
crypto: caam - Add definition of rd/wr_reg64 for little endian platform
added support for little endian platforms to the CAAM driver. Namely a
write and read function for 64 bit registers.
The only user of this functions is the Job Ring driver (drivers/crypto/caam/jr.c).
It uses the functions to set the DMA addresses for the input/output rings.
However, at least in the default configuration, the least significant 32 bits are
always in the base+0x0004 address; independent of the endianness of the bytes itself.
That means the addresses do not change with the system endianness.
DMA addresses are only 32 bits wide on non-64-bit systems, writing the upper 32 bits
of this value to the register for the least significant bits results in the DMA address
being set to 0.
Fix this by always writing the registers in the same way.
Hugh Dickins [Sun, 14 Jun 2015 16:48:09 +0000 (09:48 -0700)]
mm: shmem_zero_setup skip security check and lockdep conflict with XFS
It appears that, at some point last year, XFS made directory handling
changes which bring it into lockdep conflict with shmem_zero_setup():
it is surprising that mmap() can clone an inode while holding mmap_sem,
but that has been so for many years.
Since those few lockdep traces that I've seen all implicated selinux,
I'm hoping that we can use the __shmem_file_setup(,,,S_PRIVATE) which
v3.13's commit c7277090927a ("security: shmem: implement kernel private
shmem inodes") introduced to avoid LSM checks on kernel-internal inodes:
the mmap("/dev/zero") cloned inode is indeed a kernel-internal detail.
This also covers the !CONFIG_SHMEM use of ramfs to support /dev/zero
(and MAP_SHARED|MAP_ANONYMOUS). I thought there were also drivers
which cloned inode in mmap(), but if so, I cannot locate them now.
Bjorn Helgaas [Mon, 15 Jun 2015 21:28:29 +0000 (16:28 -0500)]
PCI: pciehp: Clean up debug logging
The pciehp debug logging is overly verbose and often redundant. Almost all
of the information printed by dbg_ctrl() is also printed by the normal PCI
core enumeration code and by pcie_init().
Remove the redundant debug info.
When claiming a pciehp bridge, we print the slot characteristics, e.g.,
Bjorn Helgaas [Tue, 9 Jun 2015 22:31:38 +0000 (17:31 -0500)]
x86/PCI: Use host bridge _CRS info on systems with >32 bit addressing
We enable _CRS on all systems from 2008 and later. On older systems, we
ignore _CRS and assume the whole physical address space (excluding RAM and
other devices) is available for PCI devices, but on systems that support
physical address spaces larger than 4GB, it's doubtful that the area above
4GB is really available for PCI.
After d56dbf5bab8c ("PCI: Allocate 64-bit BARs above 4G when possible"), we
try to use that space above 4GB *first*, so we're more likely to put a
device there.
On Juan's Toshiba Satellite Pro U200, BIOS left the graphics, sound, 1394,
and card reader devices unassigned (but only after Windows had been
booted). Only the sound device had a 64-bit BAR, so it was the only device
placed above 4GB, and hence the only device that didn't work.
Keep _CRS enabled even on pre-2008 systems if they support physical address
space larger than 4GB.
powerpc/iommu/ioda2: Enable compile with IOV=on and IOMMU_API=off
The pnv_pci_ioda2_unset_window() function is used to do the final
cleanup of a DMA window being released:
- via VFIO ioctl by the guest request;
- via unplugging a virtual PCI function.
However the function was under #ifdef CONFIG_IOMMU_API and was missing.
This moves the helper outside of IOMMU_API block and enables it
for either or both IOMMU_API and PCI_IOV.
Jeremy Kerr [Mon, 15 Jun 2015 09:25:34 +0000 (17:25 +0800)]
powerpc/powernv: fix construction of opal PRD messages
We currently have a bug in the PRD code, where the contents of an
incoming message (beyond the header) will be overwritten by the list
item manipulations when adding to to the prd_msg_queue.
This change reorders struct opal_prd_msg_queue_item, so that the
message body doesn't overlap the list_head.
We also clarify the memcpy of the message, as we're copying unnecessary
bytes at the end of the message data.
The eeh subsystem for powernv requires the opal event irqchip to be
initialised prior to initialisation or the following errors are
produced (and eeh doesn't work as expected):
irq: XICS didn't like hwirq-0x9 to VIRQ17 mapping (rc=-22)
pnv_eeh_post_init: Can't request OPAL event interrupt (0)
On powernv eeh is initialised from a subsys_initcall due to a check
for machine_is(powernv) in eeh_init(). This patch increases the
initcall priority of opal_event_init() to an arch_initcall to ensure
the opal event interface is initialised prior to any users of it.
Now it is possible to press CTRL+z at anytime and that will disable the
events being monitored, essentially turning 'top' into 'report', with
pressing CTRL+z again making it enable the events again, returning to
the 'top' behaviour, i.e. dynamic + decaying of older samples.
One may want, for instance, play with:
-d, --delay <n> number of seconds to delay between refreshes
and:
-z, --zero zero history across updates
Plus CTRL+z to see only the events since last zeroing, etc.
For an upcoming feature in 'perf top' we will have a hotkey to
enable/disable events, so remember if the events in the list are
enabled or disabled and allows toggling this state using a new
method.
perf trace: Fix race condition at the end of started workloads
I get following crash on multiple systems and across several releases
(at least since v3.18).
Core was generated by `/tmp/perf trace sleep 0.2 '.
Program terminated with signal SIGSEGV, Segmentation fault.
#0 perf_mmap__read_head (mm=0x3fff9bf30070) at util/evlist.h:195
195 u64 head = ACCESS_ONCE(pc->data_head);
(gdb) bt
#0 perf_mmap__read_head (mm=0x3fff9bf30070) at util/evlist.h:195
#1 perf_evlist__mmap_read (evlist=0x10027f11910, idx=<optimized out>)
at util/evlist.c:637
#2 0x000000001003ce4c in trace__run (argv=<optimized out>,
argc=<optimized out>, trace=0x3fffd7b28288) at builtin-trace.c:2259
#3 cmd_trace (argc=<optimized out>, argv=<optimized out>,
prefix=<optimized out>) at builtin-trace.c:2799
#4 0x00000000100657b8 in run_builtin (p=0x10176798 <commands+480>, argc=3,
argv=0x3fffd7b2b550) at perf.c:370
#5 0x00000000100063e8 in handle_internal_command (argv=0x3fffd7b2b550, argc=3)
at perf.c:429
#6 run_argv (argv=0x3fffd7b2af70, argcp=0x3fffd7b2af7c) at perf.c:473
#7 main (argc=3, argv=0x3fffd7b2b550) at perf.c:588
The problem seems to be a race condition, when the application has just
exited. Some/all fds associated with the perf-events (tracepoints) go
into a POLLHUP/ POLLERR state and the mmap region associated with those
events are unmapped (in perf_evlist__filter_pollfd()).
But we go back and do a perf_evlist__mmap_read() which assumes that the
mmaps are still valid and we hit the crash.
If the mapping for an event is released, its refcnt is 0 (and ->base
is NULL), so ensure we have non-zero refcount before accessing the map.
Note that perf-record has a similar logic but unlike perf-trace, the
record__mmap_read_all() checks the evlist->mmap[i].base before accessing
the map.
Masami Hiramatsu [Wed, 17 Jun 2015 14:58:54 +0000 (23:58 +0900)]
perf probe: Speed up perf probe --list by caching debuginfo
Speed up the "perf probe --list" by caching the last used debuginfo.
perf probe --list always open and load debuginfo for each entry of probe
list. This takes very a long time.
E.g. with vfs_* events (total 96 probes)
[root@localhost perf]# time ./perf probe -l &> /dev/null
real 0m25.376s
user 0m24.381s
sys 0m1.012s
To solve this issue, this adds debuginfo_cache to cache the
last used debuginfo on memory.
With this fix, the perf-probe --list significantly improves
its speed.
[root@localhost perf]# time ./perf probe -l &> /dev/null
Masami Hiramatsu [Tue, 16 Jun 2015 11:50:57 +0000 (20:50 +0900)]
perf probe: Show usage even if the last event is skipped
When the last part of converted events are blacklisted or out-of-text,
those are skipped and perf probe doesn't show usage examples. This
fixes it to show the example even if the last part of event list is
skipped.
E.g. without this patch, events are added, but suddenly end:
# perf probe vfs_*
vfs_caches_init_early is out of .text, skip it.
vfs_caches_init is out of .text, skip it.
Added new events:
probe:vfs_fallocate (on vfs_*)
probe:vfs_open (on vfs_*)
...
probe:vfs_dentry_acceptable (on vfs_*)
probe:vfs_load_quota_inode (on vfs_*)
#
With this fix:
# perf probe vfs_*
vfs_caches_init_early is out of .text, skip it.
vfs_caches_init is out of .text, skip it.
Added new events:
probe:vfs_fallocate (on vfs_*)
...
probe:vfs_load_quota_inode (on vfs_*)
You can now use it in all perf tools, such as:
perf record -e probe:vfs_load_quota_inode -aR sleep 1
Note that this can be reproduced ONLY IF the vfs_caches_init* is the
last part of matched symbol list. I've checked this happens on
"3.19.0-generic #18-Ubuntu" kernel binary.
Wang Nan [Wed, 17 Jun 2015 14:46:29 +0000 (14:46 +0000)]
perf tools: Move libtraceevent dynamic list to separated LDFLAGS variable
Commit e3d09ec8126fe2c9a3ade661e2126e215ca27a80 ("tools lib traceevent:
Export dynamic symbols used by traceevent plugins") adds libtraceevent
dynamic list directly into LDFLAGS, which makes all targets depend on
that list through LDFLAGS.
This is not good since some of targets like libgtk.so doesn't use plugin
at all, but require the existance of that list because of linker
options.
This patch isolates the -Xlink option into LIBTRACEEVENT_DYNAMIC_LIST_LDFLAGS,
makes only perf and perf.so use it.
The bug is in perf_event__attr_swap(). It swaps all fields in 'struct
perf_event_attr' without checking whether the swapped field exist or
not. In addition, in read_event_desc() allocs memory for attr according
to size read from perf.data.
Therefore, if the perf.data is collected by an old perf (without
aux_watermark, for example), when perf_event__attr_swap() swaping
attr->aux_watermark it destroy malloc's metadata.
This patch introduces boundary checking in perf_event__attr_swap(). It
adds macros bswap_field_64 and bswap_field_32 into
perf_event__attr_swap() to make it only swap exist fields.
Commit fd1d0ddf2ae9 (KVM: arm/arm64: check IRQ number on userland
injection) rightly limited the range of interrupts userspace can
inject in a guest, but failed to consider the (unlikely) case where
a guest is configured with 1024 interrupts.
In this case, interrupts ranging from 1020 to 1023 are unuseable,
as they have a special meaning for the GIC CPU interface.
Make sure that these number cannot be used as an IRQ. Also delete
a redundant (and similarily buggy) check in kvm_set_irq.
projects / linux.git / log