2.6.26 b43 speed issue (redux) and new firmware drivers

2.6.26 b43 speed issue (redux) and new firmware drivers

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 1258 bytes --]

Hi All,

I'm writing to inquire about a "new" firmware drivers for Broadcom 43xx:
On April 6, 2008 it looks like broadcom-wl-4.150.10.5.2.tar.bz2 appeared
on http://mirror2.openwrt.org/sources/. The install guide for b43
recommends you use b43-fwcutter-011 and broadcom-wl-4.150.10.5.tar.bz2

I tried to use broadcom-wl-4.150.10.5.2.tar.bz2 with b43-fwcutter-011,
but it complained that the md5 sum was not recognized :). Just wondering
if there is any way to use this new firmware?

Thanks in advance,

GI

PS: The main reason I ask is for "performance": A while ago I reported
    that my Broadcom 4312 gave 2/3 the speed under b43 (2.6.25), when
    compared to ndiswrapper. (I tested by running ttcp, and measuring
    the throughput. I got roughly 3MB/s with ndiswrapper, and
    2 MB/s with b43).

    I just upgraded to 2.6.26, and found that my b43 speed *increased*
    to 2.4MB/s :). So thanks to whomever is responsible! I was also
    hoping that if I upgrade even the firmware I would be able to
    "toast" ndiswrapper. (I transfer large files on my home network
    often, which is why I care about the speed so much.)

-- 
'Cigarette' -- A pinch of tobacco, wrapped in paper, fire at one end,
fool at the other.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: 2.6.26 b43 speed issue (redux) and new firmware drivers

[Larry Finger]

Gautam Iyer wrote:
> Hi All,
> 
> I'm writing to inquire about a "new" firmware drivers for Broadcom 43xx:
> On April 6, 2008 it looks like broadcom-wl-4.150.10.5.2.tar.bz2 appeared
> on http://mirror2.openwrt.org/sources/. The install guide for b43
> recommends you use b43-fwcutter-011 and broadcom-wl-4.150.10.5.tar.bz2
> 
> I tried to use broadcom-wl-4.150.10.5.2.tar.bz2 with b43-fwcutter-011,
> but it complained that the md5 sum was not recognized :). Just wondering
> if there is any way to use this new firmware?
> 
> Thanks in advance,
> 
> GI
> 
> PS: The main reason I ask is for "performance": A while ago I reported
>     that my Broadcom 4312 gave 2/3 the speed under b43 (2.6.25), when
>     compared to ndiswrapper. (I tested by running ttcp, and measuring
>     the throughput. I got roughly 3MB/s with ndiswrapper, and
>     2 MB/s with b43).
> 
>     I just upgraded to 2.6.26, and found that my b43 speed *increased*
>     to 2.4MB/s :). So thanks to whomever is responsible! I was also
>     hoping that if I upgrade even the firmware I would be able to
>     "toast" ndiswrapper. (I transfer large files on my home network
>     often, which is why I care about the speed so much.)

I do not know what changed between 4.150.10.5 and 4.150.10.2, but I doubt that 
the firmware in the new driver will make any appreciable difference in the 
performance of the BCM4312. My guess is that they found some errors in the 
handling of the N PHY or the LP PHY, which are the two newest devices, and fixed 
them. Of course, we do need to update fwcutter to test my hypothesis. As a 
reverse engineer, I also need to see what changed. What version is the Windows 
driver that you use?

It is a fact of life that this kind of reverse engineering rarely will perform 
better than the original driver. After all, if we make an exact copy, then we 
repeat their mistakes. Any mistakes of ours have a small chance of improving 
performance. Where we do improve on the original is in the error handling.

I get the same performance with my BCM4312 as you do.

Larry

Re: 2.6.26 b43 speed issue (redux) and new firmware drivers

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 2851 bytes --]

On Wed, Jul 23, 2008 at 10:22:05AM -0500, Larry Finger wrote:

>> I'm writing to inquire about a "new" firmware drivers for Broadcom 43xx:
>> On April 6, 2008 it looks like broadcom-wl-4.150.10.5.2.tar.bz2 appeared
>> on http://mirror2.openwrt.org/sources/. The install guide for b43
>> recommends you use b43-fwcutter-011 and broadcom-wl-4.150.10.5.tar.bz2
>> I tried to use broadcom-wl-4.150.10.5.2.tar.bz2 with b43-fwcutter-011,
>> but it complained that the md5 sum was not recognized :). Just wondering
>> if there is any way to use this new firmware?
>> Thanks in advance,
>> GI
>> PS: The main reason I ask is for "performance": A while ago I reported
>>     that my Broadcom 4312 gave 2/3 the speed under b43 (2.6.25), when
>>     compared to ndiswrapper. (I tested by running ttcp, and measuring
>>     the throughput. I got roughly 3MB/s with ndiswrapper, and
>>     2 MB/s with b43).
>>     I just upgraded to 2.6.26, and found that my b43 speed *increased*
>>     to 2.4MB/s :). So thanks to whomever is responsible! I was also
>>     hoping that if I upgrade even the firmware I would be able to
>>     "toast" ndiswrapper. (I transfer large files on my home network
>>     often, which is why I care about the speed so much.)
>
> I do not know what changed between 4.150.10.5 and 4.150.10.2, but I
> doubt that the firmware in the new driver will make any appreciable
> difference in the performance of the BCM4312. My guess is that they
> found some errors in the handling of the N PHY or the LP PHY, which
> are the two newest devices, and fixed them. Of course, we do need to
> update fwcutter to test my hypothesis. As a reverse engineer, I also
> need to see what changed. What version is the Windows driver that you
> use?

Hi Larry,

Thanks for the response. The windows driver I used can be found at

    http://h20000.www2.hp.com/bizsupport/TechSupport/SoftwareDescription.jsp?lang=en&cc=us&prodTypeId=321957&prodSeriesId=3355644&prodNameId=3355645&swEnvOID=1059&swLang=8&mode=2&taskId=135&swItem=ob-55703-1

It looks like HP released a few newer versions of this driver, which I
haven't tested yet. I use the b43 drivers all the time anyway.

> It is a fact of life that this kind of reverse engineering rarely will
> perform better than the original driver. After all, if we make an
> exact copy, then we repeat their mistakes. Any mistakes of ours have a
> small chance of improving performance. Where we do improve on the
> original is in the error handling.

Time to fire up the petition to get Broadcom to open up their specs for
Linux devs...

Hopefully the new firmware 4.150.10.2 helps with b43 somehow.
Nevertheless, thanks for all you guys have done with the b43 driver!! It
really is wonderful!

GI

-- 
Linux for Network, Mac for Work, Windows for Solitaire

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: 2.6.26 b43 speed issue (redux) and new firmware drivers

[Gautam Iyer]

[-- Attachment #1: Type: text/plain, Size: 2458 bytes --]

On Wed, Jul 23, 2008 at 01:32:49PM -0700, Gautam Iyer wrote:

>>> I'm writing to inquire about a "new" firmware drivers for Broadcom 43xx:
>>> On April 6, 2008 it looks like broadcom-wl-4.150.10.5.2.tar.bz2 appeared
>>> on http://mirror2.openwrt.org/sources/. The install guide for b43
>>> recommends you use b43-fwcutter-011 and broadcom-wl-4.150.10.5.tar.bz2
>>> I tried to use broadcom-wl-4.150.10.5.2.tar.bz2 with b43-fwcutter-011,
>>> but it complained that the md5 sum was not recognized :). Just wondering
>>> if there is any way to use this new firmware?
>>> Thanks in advance,
>>> GI
>>> PS: The main reason I ask is for "performance": A while ago I reported
>>>     that my Broadcom 4312 gave 2/3 the speed under b43 (2.6.25), when
>>>     compared to ndiswrapper. (I tested by running ttcp, and measuring
>>>     the throughput. I got roughly 3MB/s with ndiswrapper, and
>>>     2 MB/s with b43).
>>>     I just upgraded to 2.6.26, and found that my b43 speed *increased*
>>>     to 2.4MB/s :). So thanks to whomever is responsible! I was also
>>>     hoping that if I upgrade even the firmware I would be able to
>>>     "toast" ndiswrapper. (I transfer large files on my home network
>>>     often, which is why I care about the speed so much.)
>>
>> I do not know what changed between 4.150.10.5 and 4.150.10.2, but I
>> doubt that the firmware in the new driver will make any appreciable
>> difference in the performance of the BCM4312. My guess is that they
>> found some errors in the handling of the N PHY or the LP PHY, which
>> are the two newest devices, and fixed them. Of course, we do need to
>> update fwcutter to test my hypothesis. As a reverse engineer, I also
>> need to see what changed. What version is the Windows driver that you
>> use?
> 
> Hi Larry,
> 
> Thanks for the response. The windows driver I used can be found at
> 
>     http://h20000.www2.hp.com/bizsupport/TechSupport/SoftwareDescription.jsp?lang=en&cc=us&prodTypeId=321957&prodSeriesId=3355644&prodNameId=3355645&swEnvOID=1059&swLang=8&mode=2&taskId=135&swItem=ob-55703-1
> 
> It looks like HP released a few newer versions of this driver, which I
> haven't tested yet. I use the b43 drivers all the time anyway.

If it makes any difference -- clicking on the enhancements tab in the
above link says that this driver provides broadcoms 4.170.25.12...

GI

-- 
Bakers trade bread recipes on a knead-to-know basis.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: 2.6.26 b43 speed issue (redux) and new firmware drivers

[Michael Buesch]

On Wednesday 23 July 2008 17:22:05 Larry Finger wrote:
> It is a fact of life that this kind of reverse engineering rarely will perform 
> better than the original driver. After all, if we make an exact copy, then we 
> repeat their mistakes. Any mistakes of ours have a small chance of improving 
> performance. Where we do improve on the original is in the error handling.

Well, there are parts in the b43 code that are subject to optimization.
The power adjustment code, for example. I have a patch in development for this,
but didn't work on it for quite some time.

-- 
Greetings Michael.