[oi-dev] Broadcom BCM4312 wireless on OpenIndiana 64-bit

Fri Mar 8 21:34:06 UTC 2013

On 2013-03-07 16:22, Jean-Pierre wrote:
> Hi all
>
> I have finally been able to activate a 64-bit WiFi driver for a Broadcom
> BCM4312 wireless hardware on OpenIndiana (oi151a5).

Hello, many of the details you've posted in the report (thanks BTW!)
seem like vague pointers for those who know what to google for -
perhaps by the words you've mentioned :)

If there are particular blogs/articles to read and learn in order to
repeat your recipe for the reader's (i.e. my) hardware, could you
please post the URLs?

For example, how did you locate and patch the multimedia registers,
or change linking of the driver, etc. to solve the issues you list?

Or are all these fixes part of your NDIS changes and will be revised,
refined and posted as the updated source code "ready to use" with an
original Windows driver to wrap?

Now, I hope this would also work with my BCM4313... at least, after
reading your detected issues I get a faint idea why my wrap panicked
the kernel when attaching ;)

Thanks again,
//Jim

>
> pci bus 0x0004 cardnum 0x00 function 0x00: vendor 0x14e4 device 0x4315
>   Broadcom Corporation BCM4312 802.11b/g LP-PHY
>   CardVendor 0x103c card 0x137d (Hewlett-Packard Company, BCM4312
> 802.11b/g Wire$
>
> I used an old Windows driver (bcmwl564.sys) which I knew compatible
> with my hardware, because I used it a few years ago on Linux with
> ndiswrapper (this driver is still available on the Internet as part of a
> file named R174291.exe), and I used the NDIS emulator ndis-1.2.2 by Bill
> Paul. (the most recent one I found on the Internet).
>
> I did not have to make fundamental changes, but below are the main
> issues I had to face :
>
> - issue 1 : the Windows driver is linked at run-time
>     The ndis emulator is linked at build-time with a binary image of the
>     Windows driver. In order to create the links, the driver image was
>     copied at run time to a zalloc'ed memory. As a consequence the copy
>     needed to be executable, which AFAIK is sometimes not possible (at
>     least for functions triggered by timers.)
>
>     Work around : dissociate the driver code from its variables and only
>     dynamically allocate the variables and the indirect pointers to
>     external code (ie the ndis emulator). By dissociating code and
>     variables, linking at build time became possible (along with type
>     type checking at compile time.)
>
> - issue 2 : the Windows driver makes use of multimedia registers (xmm)
>     The multimedia instructions "movaps" and "movdqu" were used to
>     optimize memcpy().
>
>     Work around : patch the 71 multimedia instructions at build time.
>
> - issue 3 : a few emulation functions were called directly with a
>     Windows ABI. This mostly happened for indirect calls, when a
>     pointer to a function was set in a structure within the Windows
>     driver.
>
>     Work around : make a strict typing of emulation functions, detect
>     the wrong ABIs at compile time, and fix them.
>
> - issue 4 : a few generic Windows wrapper functions did not reserve 32
> bytes on the stack.
>     When these 32 bytes are not needed to save arguments passed in
> registers,
>     they can be used in the Windows driver for local variables. When the
>     space is not reserved, the return pointer is overwritten with obvious
>     consequences.
>
> - issue 5 : the layout of a few structures did not match the Windows
> driver ones.
>     This was somewhat expected because the alignment rules are compiler
>     specific. More surprisingly, in at least one case, the Windows driver
>     wanted a structure different from what is described on msdn. Maybe the
>     structure has been shortened since it was used to compile the driver.
>
>
> March  7, 2013 03:18:43 PM CET
> #ifconfig bcmndis0 plumb
> March  7, 2013 03:19:03 PM CET
> #wificonfig -i bcmndis0 setparam essid=(not shown)
> March  7, 2013 03:19:23 PM CET
> #wificonfig -i bcmndis0 setparam encryption=wep
> March  7, 2013 03:19:43 PM CET
> #wificonfig -i bcmndis0 setparam wepkey1=(not shown)
> March  7, 2013 03:20:03 PM CET
> #wificonfig -i bcmndis0 setparam authmode=shared_key
> March  7, 2013 03:20:23 PM CET
> #wificonfig -i bcmndis0 connect (not shown)
> wificonfig: connecting to essid '(not shown)'
> March  7, 2013 03:20:46 PM CET
> #wificonfig -i bcmndis0 showstatus
>      linkstatus: connected
>      active profile: none
>      essid: (not shown)
>      bssid: 00:90:4b:18:93:ea
>      encryption: wep
>      signal strength: medium(11)
> March  7, 2013 03:21:07 PM CET
> #wificonfig -i bcmndis0 getparam
>      bssid: 00:90:4b:18:93:ea
>      essid: DW-B-200-2827b
>      bsstype: bss(ap, infrastructure)
>      createibss: no
>      channel: 3
>      powermode: fast
>      authmode: shared_key
>      encryption: wep
>      wepkeyindex: 1
>      signal: 11
>      radio: on
> March  7, 2013 03:21:27 PM CET
> # ifconfig bcmndis0 dhcp
>
> # netstat -rn
>
> Routing Table: IPv4
>    Destination           Gateway           Flags  Ref     Use     Interface
> -------------------- -------------------- ----- ----- ---------- ---------
> default              10.0.0.1             UG        1          0 bcmndis0
> 10.0.0.0             10.130.123.16        U         2          0 bcmndis0
> 127.0.0.1            127.0.0.1            UH        4        160 lo0
>
> Routing Table: IPv6
>    Destination/Mask            Gateway                   Flags Ref   Use
>     If
> --------------------------- --------------------------- ----- ---
> ------- -----
> ::1                         ::1                         UH      2 16 lo0
> fe80::/10                   fe80::21e:ecff:fee5:b68a    U       2 0 rge0
> # ping 10.0.0.1
> 10.0.0.1 is alive
>
> I can access the configuration page of the router, but for some reason
> I cannot get the name service.
>
> I will make the modified NDIS emulator source code available
> when I have done the necessary cleaning.
>
> Regards
>
> Jean-Pierre