When powersaving is enabled, assocaition times are very high
(for WPA2 networks, the time can easily be around the 3 seconds).
This is caused, because the flushing of the queues takes
too much time. Without the flushing callback mac80211 assumes
a timeout of 100ms while scanning. Limit all flush waiting
loops to the same maximum.
We can apply this maximum by passing the drop status to the
driver, which makes sure the driver performs extra actions
during the waiting for the queue to become empty.
After these changes, association times fall within the
healthy range of ~0.6 seconds with powersaving enabled.
The difference between association time between powersaving
enabled and disabled is now only ~0.1 second (which can also
be due to the measuring method).
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The number of flags defined for the rt2x00dev->flags field,
has been growing over the years. Currently we are approaching
the maximum number of bits which are available in the field.
A secondary problem, is that one part of the field are initialized only
during boot, because the driver requirements are initialized or device
requirements are loaded from the EEPROM. In both cases, the flags are
fixed and will not change during device operation. The other flags are
the device state, and will change frequently. So far this resulted in the fact
that for some flags, the atomic bit accessors are used, while for the others
the non-atomic variants are used.
By splitting the flags up into a "flags" and "cap_flags" we can put all flags
which are fixed inside "cap_flags". This field can then be read non-atomically.
In the "flags" field we keep the device state, which is going to be read atomically.
This adds more room for more flags in the future, and sanitizes the field access methods.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Instead of receiving an unlimited number of frames, stop after 16
entries and reschedule the rxdone tasklet. This allows other tasklets
to be run inbetween.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In the spirit of optimizing the code to get the queue structure of TX queues,
also optimize the code to get beacon queues. We can simply use the bcn queue
field of the rt2x00_dev structure instead of using the rt2x00queue_get_queue
function.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The ATIM queue is considered to be a TX queue by the drivers that support
the queue. Therefore include support for the ATIM queue to the
rt2x00queue_get_tx_queue function so that the drivers that support the ATIM
queue can also use that function.
Add the support in such a way that drivers that do not support the ATIM
queue are not penalized in their efficiency.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Current code for the atim queue is strange, as it is considered in the
rt2x00_dev structure as a second beacon queue.
Normalize this by letting the atim queue have its own struct data_queue
pointer in the rt2x00_dev structure.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The PCI device irqmask is locked by a spin_lock. Currently
spin_lock_irqsave is used everywhere. To reduce the locking overhead
replace spin_lock_irqsave in hard irq context with spin_lock and in
soft irq context with spin_lock_irq.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
"ifs" is only used by no-HT devices. Move it into the plcp substruct and
fill in the value only for no-HT devices.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
HT and no-HT rt2x00 devices use a partly different TX descriptor.
Optimize the tx desciptor memory layout by putting the PLCP and HT
substructs into a union and introduce a new driver flag to decide which
TX desciptor format is used by the device.
This saves us the expensive PLCP calculation fOr HT devices and the HT
descriptor setup on no-HT devices.
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Newer devices like rt2800* own a hardware sequence counter and thus
don't need to use a software sequence counter at all. Add a new driver
flag to shortcut the software sequence number generation on devices that
don't need it.
rt61pci, rt73usb and rt2800* seem to make use of a hw sequence counter
while rt2400pci and rt2500* need to do it in software.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Remove STATE_RADIO_IRQ_OFF_ISR and STATE_RADIO_IRQ_ON_ISR as they are
not used anymore.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Fix interrupt processing on slow machines by using individual tasklets
for each different device interrupt. This ensures that while a RX or TX
status tasklet is scheduled only the according device interrupt is
masked and other interrupts such as TBTT can still be processed.
Also, this allows us to use tasklet_hi_schedule for TBTT processing
which is required to not send out beacons with a wrong DTIM count (due
to delayed periodic beacon updates). Furthermore, this improves the
latency between the TBTT and sending out buffered multi- and broadcast
traffic.
As a nice bonus, the interrupt handling overhead should be much lower.
Compile-tested only.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch allows to dynamically remove beaconing interfaces without
shutting beaconing down on all interfaces.
The only place to start and stop beaconing are now the start- and
stop_queue callbacks. Hence, we can remove some register writes during
interface bring up (config_intf) and only write the correct sync mode
to the register there.
When multiple beaconing interfaces are present we should enable
beaconing as soon as mac80211 enables beaconing on at least one of
them. The beacon queue gets stopped when the last beaconing
interface was stopped by mac80211. Therefore, introduce another
interface counter to keep track ot the number of enabled beaconing
interfaces and start or stop the beacon queue accordingly.
To allow single interfaces to stop beaconing, add a new driver
callback clear_beacon to clear a single interface's beacon without
affecting the other interfaces. Don't overload the clear_entry callback
for clearing beacons as that would introduce additional overhead
(check for each TX queue) into the clear_entry callback which is used
on the drivers TX/RX hotpaths.
Furthermore, the write beacon callback doesn't need to enable beaconing
anymore but since beaconing should be disabled while a new beacon is
written or cleared we still disable beacon generation and enable it
afterwards again in the driver specific callbacks. However, beacon
related interrupts should not be disabled/enabled here, that's solely
done from the start- and stop queue callbacks. It would be nice to stop
the beacon queue just before the beacon update and enable it afterwards
in rt2x00queue itself instead of the current implementation that relies
on the driver doing the right thing. However, since start- and
stop_queue are mutex protected we cannot use them for atomic beacon
updates.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The Queue names were incorrectly copied from the legacy drivers,
as a result the queue names were inversed to what was expected.
This renames the queues using this mapping:
QID_AC_BK -> QID_AC_VO (priority 0)
QID_AC_BE -> QID_AC_VI (priority 1)
QID_AC_VI -> QID_AC_BE (priority 2)
QID_AC_VO -> QID_AC_BK (priority 3)
Note that this was a naming problem only, which didn't affect
the assignment of frames to their respective queues.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
As part of the queue refactoring, change the queue callback
function names to have 3 different actions: start, kick & stop.
We can now also remove the STATE_RADIO_RX_ON/STATE_RADIO_RX_OFF
device_state flags, and replace the usage with using the
start_queue/stop_queue callback functions.
This streamlines the RX queue handling to the
similar approach as all other queues.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
As part of the queue refactoring, we now introduce
3 queue commands: start, kick, stop.
- Start: will enable a queue, for TX this will
not mean anything, while for beacons and RX
this will update the registers to enable the queue.
- Kick: This will kick all pending frames to
the hardware. This is needed for the TX queue
to push all frames to the HW after the queue
has been started
- Stop: This will stop the queue in the hardware,
and cancel any pending work (So this doesn't
mean the queue is empty after a stop!).
Move all code from the drivers into the appropriate
functions, and link those calls to the old rt2x00lib
callback functions (we will fix this later when we
refactor the queue control inside rt2x00lib).
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Commit 0204464329 "Check for specific changed
flags when updating the erp config" changed the way in which a new beacon
interval gets handled. However, due to a bug in rt2800usb and rt2800pci the
beacon interval was reset during each scan, thus causing problems in AdHoc
mode.
Fix this by not cleaning up the beacon interval when killing the beacon queue
but just prevent the device from sending out beacons.
Reported-by: Wolfgang Kufner <wolfgang.kufner@gmail.com>
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
As part of the queue refactoring, the rt2x00lib_toggle_rx
can be removed and replaced with the call directly to
the set_device_state callback function.
We can remove the STATE_RADIO_RX_ON_LINK and
STATE_RADIO_RX_OFF_LINK, as it was only used for
special behavior inside rt2x00lib rather then the
drivers.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Implement a basic flush callback function, which simply loops
over all TX queues and waits until all frames have been transmitted
and the status reports have been gathered.
At this moment we don't support dropping any frames during the
flush, but mac80211 will only send 'false' for this argument anyway,
so this is not important at this time.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The tx descriptor values qid, cw_min, cw_max and aifs are directly
accessible through the tx entry struct. So there's no need to copy
them into the tx descriptor and passing them to the indiviual drivers.
Instead we can just get the correct value from the tx entry.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
All rt2x00 devices used the same Tx and Rx ring size (24 entries) till
now. Newer devices (like rt2800) can however make use of a larger TX and
RX ring due to 11n capabilities (AMPDUs of size 64 for example).
Hence, bring rt2x00 in sync with the legacy drivers and use the same TX
and RX ring sizes. Also remove the global defines RX_ENTRIES, TX_ENTRIES,
BEACON_ENTRIES and ATIM_ENTRIES and use per driver values.
That is 24 entries for rt2400pci, 32 entries for rt2500pci, rt2500usb,
rt61pci and rt73usb and 128 (RX) and 64 (TX) for rt2800pci and rt2800usb.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
A lot of functions accept a struct rt2x00_dev combined with
either a struct queue_entry or struct data_queue argument.
This can be simplified by only passing on the queue/entry
argument.
In cases where rt2x00_dev and a sk_buff are send together,
we can send the queue_entry instead.
rt2x00usb_alloc_urb and rt2x00usb_free_urb have a bit
of vague naming. Instead they allocate all the data which
belongs to a rt2x00 data queue entry.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Previously rt2x00 was always updating all erp related config variables
even though mac80211 might only have changed one. Hence, pass the
changed flags to the config_erp driver callback so that the driver
can limit the changes to the correct values.
This fixes an issue in AP mode where the beacon interval is not
initialized (and thus zero) but still sent to the hardware causing an
interrupt storm on rt2800pci hanging the system.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
During initialization each driver reads the default TX power
for each individual channel. However mac80211 only accepts the
maximum value (which is also handled as default value).
As a result, the TX power of the device was being limited to
the default value, which is often quite low compared to the
real maximum acceptable value.
This patch allows each driver to set the maximum value on a
per-channel basis which is forwarded to mac80211. The default
value will be preserved for now, in case we want to update
mac80211 to differentiate between the maximum and default txpower.
This fixes bug complaining about limited TX power values like:
https://bugzilla.kernel.org/show_bug.cgi?id=16358
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
write_tx_desc shouldn't pass a rt2x00dev and skb pointer,
instead it should use the same format as other TX frame
callback functions, which is passing the data_entry pointer
which contains all the information which is needed to work
on a TX frame.
Most callers of the kick_tx_queue and kill_tx_queue already
have the data_queue pointer, so rather then sending the QID
with the given function, when the driver requests a new
pointer to the data_queue, it is more efficient to just
send the data_queue pointer directly.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
These changes may be slightly safer in some instances.
There are other kzalloc calls with a multiply, but those
calls are typically "small fixed #" * sizeof(some pointer)"
and those are not converted.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Luciano Coelho <luciano.coelho@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Using the set_tim callback without managing the DTIM count and the
broad- and multicast buffering in hw, fw or the driver results in wrong
DTIM count values being sent out in beacons. Since all PCI drivers
fetch new beacons periodically and hence get an updated TIM we can just
remove the set_tim callback from these.
The rt2x00 USB drivers don't update the beacon periodically and thus
rely on the set_tim callback to get a correct TIM for beacon
transmission. USB devices still suffer from the DTIM count being wrong
under some circumstances but removing the set_tim callback from these
would cause more harm then good.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Use threaded interrupts for all rt2x00 PCI devices.
This has several generic advantages:
- Reduce the time we spend in hard irq context
- Use non-atmic mac80211 functions for rx/tx
Furthermore implementing broad- and multicast buffering will be
much easier in process context while maintaining low latency and
updating the beacon just before transmission (pre tbtt interrupt)
can also be done in process context.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Currently vgc_level_reg and vgc_level are equal to eachother,
while the purpose of vgc_level_reg is the value last written
to the register and is remembered through link tuning resets.
The vgc_level is the currently active level, which is
reset during link tuning resets.
The usage of these variables depends on the drivers, some drivers
need both, while others need only one of the two.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
While scanning the link tuner must be disabled. Otherwise
it will interfere with receiving all beacons for each channel
due to changing sensitivity levels.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Rename CONFIG_DISABLE_LINK_TUNING to DRIVER_SUPPORT_LINK_TUNING
Link tuning support is not only based on EEPROM decisions, but
also if the device actually supports it.
Currently only rt2500usb doesn't support link tuning because
of hardware problems. But rt2800usb is also suspected of having
problems with link tuning.
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Now that the write_tx_data functions are merged, also merge the relevant
parts of the txdone handling into common code, rather than {usb,pci}
specific code.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Now that rt2x00pci_write_tx_data and rt2x00usb_write_tx_data are similar
we can merge them in a single function in rt2x00queue.c.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Instead of fiddling with the skb->data pointer and thereby risking
out of bounds accesses, properly reserve the space needed in an
skb for descriptors.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Not all the devices require a TX descriptor to be written (i.e. rt2800
device don't require them). Push down the creation of the TX descriptor
to the device drivers so that they can decide for themselves whether
a TX descriptor is to be created.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
(Based on a patch created by Ondrej Zary)
In some circumstances the Ralink devices do not properly go to sleep
or wake up, with timeouts occurring.
Fix this by retrying telling the device that it has to wake up or
sleep.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The handling of tx descriptors for beacons can be simplified by updating
write_tx_desc implementations of each driver to write directly to the
queue entry descriptor instead of to a provided memory area.
This is also a preparation for further clean ups where descriptors are
properly reserved in the skb instead of fiddling with the skb data
pointer.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Where possible, write the tx descriptor words from start to end, to
follow a logical ordering of words.
Where this is not possible (in rt2400pci, rt2500pci and rt61pci) add
a comment as to why word 0 needs to be written last.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Preparation to fix rt2800 beaconing.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
All of the driver's kick_tx_queue callback functions treat the TX queue
for beacons in a special manner.
Clean this up by integrating the kicking of the beacon queue into the
write_beacon callback function, and let the generic code no longer call
the kick_tx_queue callback function when updating the beacon.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
And use it consistently in the chipset drivers.
Preparation for further clean ups.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In all drivers ensure that auto wakeup is disabled before waking up the device.
This is needed to prevent connection stability issues and problems in waking up
the device.
Based upon a patch from Ondrej Zary <linux@rainbow-software.org>
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Cc: Ondrej Zary <linux@rainbow-software.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Ivo van Doorn