On Wednesday 10 September 2008 21:42:57 Francesco Gringoli wrote:
Hello everybody,
we found a strange behavior problem with the b43 driver. We begin
investigating this when doing some access tests with several boards:
we discovered that the Broadcom boards usually acquire the channel
usage with higher priority when the b43 driver is used as respect to
other vendor boards.
After investigating we saw that the intra-queue contention procedure
is carried out by the firmware which stores information, such as AIFS
and current backoffs for each queue, inside the shared memory. The
AIFS for each queue is used to compute the initial backoff so that
internal contention usually gives more priority to VO queue (smallest
AIFS) and less to background (biggest AIFS). AIFS parameters are
initialized by the driver which is told to use a 22 words long
structures. The firmware instead use 16 words structures: this means
that AIFS parameters initialized by the driver do not match the
corresponding values read by the firmware. We saw that the firmware
reads (2,0,0,0) instead of (2,2,3,7) which results in very short
backoff procedures.
Has anyone noticed similar behavior? For the sake of clarity, I'm
referring to latest (30 minutes ago) version of wireless-git and
broadcom-wl-4.150.10.5 driver version. I verified that all ucode5.s,
ucode9.s, ucode11.s, ucode13.s and ucode14.s firmwares use 16 words
structures, and the driver instead still uses 22 words long
structures. Probably the attached patch sorts out the problem.
I also noticed that the specs say that these structures are 16 words:
could not someone read 16 as 0x16 that finally became 0x16 = 22?
The specs are ambiguous here.
The specs table for QoS clearly has 0x16 fields. So I implemented it
this way and read the foobar has 16 words text in hex.
How many people can read hex, if only you and dead people can read hex? ;)
--- b43.h.old 2008-09-10 21:12:59.0 +0200
+++ b43.h 2008-09-10 21:13:09.0 +0200
@@ -569,7 +569,7 @@
#define B43_QOS_VOICE B43_QOS_PARAMS(3)
/* QOS parameter hardware data structure offsets. */
-#define B43_NR_QOSPARAMS 22
+#define B43_NR_QOSPARAMS 16
enum {
B43_QOSPARAM_TXOP = 0,
B43_QOSPARAM_CWMIN,
PS: Can you please avoid starting yet another thread for the topic?
We already have one.
Please, guys, keep _threads_. It's important to _not_ start new threads
all the time, so email archive searching is made possible. This is _also_
the case for private mail from you.
You're not the only one in my inbox, but every time I have to search for
your old mail it's a headache.
So good subjects (this one is _not_ a good one) and not breaking threads
is very important.
--
Greetings Michael.
___
Bcm43xx-dev mailing list
Bcm43xx-dev@lists.berlios.de
https://lists.berlios.de/mailman/listinfo/bcm43xx-dev
