Use spin_lock_irqsave_nested() instead of spin_lock_irqsave(). Use a
separate counter for the locking subclass, so that we don't exceed the
maximal value of 8. We have 10 queues, but inactive queues don't get a
locking subclass number.
Provide a replacement for spin_lock_irqsave_nested() for kernels older
than 2.6.19.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3995 0192ed92-7a03-0410-a25b-9323aeb14dbd
Linux 2.6.30 doesn't define IRQ_NONE as a macro. Assume irqreturn_t,
IRQ_NONE and IRQ_HANDLED to be present on Linux 2.6.29 and newer.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3985 0192ed92-7a03-0410-a25b-9323aeb14dbd
Use netdev_priv() for Linux 2.4.27 and newer. Provide compatibility
wrapper for older kernels.
Imply CONFIG_KMOD for Linux 2.6.27 and newer.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3902 0192ed92-7a03-0410-a25b-9323aeb14dbd
due to unlucky timing of beacon timer configuration (which we try to avoid) and
due to unlucky timing of local TSF updates (triggered by the reception of a
beacon with the same BSSID - something we can't avoid) the beacon timers (we
have 4) can be updated seperately, leaving one of them in the past, not beeing
updated until the timers wrap around. due to the fact that the beacon interval
does not fit into the timer period (16 bit) a whole number of times the size of
the ATIM window can get bigger than desired (we don't use ATIM right now so the
window size should always be 1).
this results in a phaenome described as "ramping" (ticket #1154) which is
actually a transmission delay, since the hardware is not allowed to transmit
data in the ATIM period.
the problem is described and discussed in length at
http://thread.gmane.org/gmane.linux.drivers.madwifi.devel/6066
since we don't know of a way to disable ATIM alltogether, the following adds
some workarounds to this problem:
1.) disable interrupts in timing critical function ath_beacon_config
2.) stop beacons before reconfiguring them
3.) check ATIM window after critical code which might have changed the timers
3a.) beacon configuration
3b.) reception of a beacon with the same BSSID. the hardware will have updated
the local TSF and this might have left one of the beacon timers in the past.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3867 0192ed92-7a03-0410-a25b-9323aeb14dbd
av_beacon_alloc should be unsigned long, not unsigned int to match
signatures of bitmask manupulation functions on i386.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3856 0192ed92-7a03-0410-a25b-9323aeb14dbd
AR2424 and AR2425 share the same PCI ID. Probing AR2425 with HAL that
doesn't support it is problematic. It causes an unhandled NMI and puts
the device to a non-functional state. After that, even HAL that
supports AR2425 won't be able to use the device until the system is
rebooted.
Reject AR2425 devices if HAL doesn't positively indicate support for
them.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3825 0192ed92-7a03-0410-a25b-9323aeb14dbd
Instead of simply alternating antenna, send multicast frames out each antenna
based on the proportion of sucessfully transmitted unicast frames. This helps
in situations where one antenna is significantly better than another.
Signed-off-by: Perry Lorier <perry@cs.waikato.ac.nz>
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3776 0192ed92-7a03-0410-a25b-9323aeb14dbd
This fixes a bug when renaming multiple wireless cards on slow devices.
The pci registration causes udev to rename the device but the sysctl is
not registered yet so the device renames but the corresponding sysctl
entries are not renamed.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3746 0192ed92-7a03-0410-a25b-9323aeb14dbd
use of strncat (my bad) and needlessly complicated the debug
output with prefixes.
Thanks, Pavel! :-)
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3738 0192ed92-7a03-0410-a25b-9323aeb14dbd
* Tidy, including refactoring and variable rename for clarity and consistency
* Move some variable initialisation out of sysctl register where it does not belong.
* Only change antenna based on RX traffic if both diversity and an RX antenna is not set
* Set HW/HAL diversity setting on HW/HAL reset
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3717 0192ed92-7a03-0410-a25b-9323aeb14dbd
Remove kbuild by-products in regression/Makefile. Also allow
regression/tkip/test_tkip.c to compile (too few arguments to function
ieee80211_crypto_demic).
Closes: #1973
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3699 0192ed92-7a03-0410-a25b-9323aeb14dbd
* Convert last SKB_CB()->ni to SKB_NI
* Lots of comments
* Refactor various pieces of code for consistency, readability, and removal of goto
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3666 0192ed92-7a03-0410-a25b-9323aeb14dbd
similar code was removed in r3503, but it was wrong before as well, because
the packet needs to be dropped instead of requeued to actually achieve
prioritization between different packet types (WME classes). if we allow
one queue to consume all TX buffers then there is no buffer left to service
packets on higher priority queues.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3648 0192ed92-7a03-0410-a25b-9323aeb14dbd
The register defines were taken from ath5k.
This patch has been only lightly tested, I'm afraid - trying for more coverage.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3641 0192ed92-7a03-0410-a25b-9323aeb14dbd
* More pedanting of comments
* Update sysctl to return early if no change is made and OPNOTSUPP if !sc_hasintmit
* Tidy call to set HAL int. mit. status. Remove chatty printf as it is known that this does not work.
* Preprocessor out HAL call as the int. mit. status is unconditionally set manually.
* Misc. churn.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3639 0192ed92-7a03-0410-a25b-9323aeb14dbd
process discarded frames in ath_recv_mgmt(). this reduces the number of times
we attempt to merge IBSS to invalid beacons.
this patch is based on the work of Derek Smithies
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3634 0192ed92-7a03-0410-a25b-9323aeb14dbd
The last descriptor is now implicit in the structure and meta-data of the buffers; axq_link does not need to be maintained separately.
Add macros to to return the last buffer, descriptor, and link them as necessary.
Make the queue moving macro implicitly link descriptors, and remove the explicit descriptor linking in this case.
ath_node_move_data is still desperately in need of a review; it is used in XR.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3620 0192ed92-7a03-0410-a25b-9323aeb14dbd
Move txqactive() implementation before it's used. Let the compiler
decide whether to inline that function.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3606 0192ed92-7a03-0410-a25b-9323aeb14dbd
During the second stage of the module build, Linux 2.6.26 does not
define $(obj). Try using $(SUBDIRS) as $(obj) before falling back to
the current directory.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3603 0192ed92-7a03-0410-a25b-9323aeb14dbd
* Formatting
* Add a function to remove SKB from ath_buf.
* Call aforementioned function early if we are accepting a frame, so that there's not chance the data can get stomped on by DMA. We should probably do this for ignored frames as well, but I got fed up trying to sort out the stack of goto.
* Remove some KASSERT for skb users, as the SKB are always copied.
git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@3594 0192ed92-7a03-0410-a25b-9323aeb14dbd
