On Wednesday 05 August 2009 00:23:23 gregor kowski wrote:

On 8/5/09, Michael Buesch [off-list ref] wrote:

quoted

On Wednesday 05 August 2009 00:03:11 gregor kowski wrote:

quoted

On 8/4/09, Michael Buesch [off-list ref] wrote:

quoted

On Tuesday 04 August 2009 23:54:42 gregor kowski wrote:

You always talk about "bugs". What are these "bugs"? Is it just the
wrong
max_nr_keys assignment? That's trivial to fix.

So [1] is ok ?

Could you answer my question?

That's [1]. But may be my description wasn't good.
I will try a new one.
we can have up to 50 pairwise keys (due to RCMTA and tkip stuff).

In the case of new API :
- max_nr_keys is set to 58
- in b43_key_clear we call do_key_write for index in [0 ... 58]
- in do_key_write we call keymac_write for index in [4 ... 58]
- in keymac_write write to RCMTA index [0 ... 54]
We write too much pairwise keys.

- max_nr_keys is set to 58
- b43_key_write generate pairwise keys in [sta_keys_start ...
max_nr_keys] = [0 ... 58] and call do_key_write
- in do_key_write we call keymac_write for index in [4 ... 58]
- in keymac_write write to RCMTA index [0 ... 54]
We write too much pairwise keys.

Yeah, I do understand this bug. My question was if that is the only bug.

So max_nr_keys seems wrong in case of new API.

It's not that simple, actually.
The max_nr_keys thing was never meant to specify which API we're on.
It was invented to do the RCMTA vs *oldcrappymechanismiforgotthenameof*.
max_nr_keys essentially became obsolete and dead code when b43 did not
support <rev5 anymore. I will submit a patch which removes it and cleans up
the code alltogether.

-- 
Greetings, Michael.