Re: Support on vendor id and device id

[Arend van Spriel]

+ Greg, +wilc1000 maintainers

On 3/1/2018 11:10 AM, Harsha Rao wrote:

On Wed, Feb 28, 2018 at 3:08 PM, Arend van Spriel
 wrote:

On 2/28/2018 12:14 AM, Harsha Rao wrote:


My suspicion is that your device, is fundamentally a wilc1000 and that

the existing wilc1000 driver will likely largely work for it and all
you really need to do is modify the existing driver to handle the
quirks of your particular implementation of the wilc1000 chip. And,
often WiFi chips will let you change the VID/PID somewhere within
whatever non-volatile storage it has (like where it stores the MAC
address).





So it seems the wilc1000 devices from Microchip/Atmel are also using a
vendor id they did not buy. Could be that the mentioned 3rd party
providing
the SDIO IP actually owns that vendor id, but if you are building your
wifi
chip on that you should better buy you own vendor id from the SD
Association. Now if Harsha is actually working for Microchip (unclear to
me)
there is basically one party that should go shopping.



I would like to clarify that I am not building anything on top of
microchip wifi device.
We have a different HW . Its been just that 3rd party vendor providing
SDIO IP has given
same ID to different customers.



So it is as I said, ie. you are using the 3rd party SDIO IP as is and add
your own wifi IP to it? So what does the term "SDIO IP" mean here. Is it a
piece of hardware that you hook up to your wifi hardware or is it
VHDL/verilog in which the vendor id is defined. If it is VHDL you should
really get your own vendor id from the SD Association and fix it. Otherwise,
the 3rd party hardware should have means to change it. If not, you better
find another party.

Regards,
Arend


Thank you folks for your comments.
The SDIO IP is VHDL IP core integrated on our SoC. And we figured out
a way to update vendor ID at run-time during boot.
We would get our own vendor ID  from SD association and proceed .


Coming back to this thread as it seems that wilc1000 has exactly the 
same issue or blindly reusing the SDIO vendor id. Below excerpt from 
earlier email in this thread:


"""
>> In theory the vendor IDs shouldn't be duplicated on fundamentally
>> different devices, assuming that the manufacturers are doing things
>> "right". The VID is paid for by buying it from the SD Association.

Indeed. And this is fun already. If the sdio.ids file in systemd has any 
value the vendor id 0x296 is assigned to:


0296  GCT Semiconductor, Inc.
5347  GDM72xx WiMAX
"""

So claiming it for wilc1000 seems wrong unless GCT and Microchip are 
actually the same company, but I could not find any evidence for that. 
The bad news is probably that this device is already on the market :-(


Regards,
Arend

Re: Support on vendor id and device id

[Harsha Rao]

On Wed, Feb 28, 2018 at 3:08 PM, Arend van Spriel
 wrote:
> On 2/28/2018 12:14 AM, Harsha Rao wrote:
>
> My suspicion is that your device, is fundamentally a wilc1000 and that
> >>>the existing wilc1000 driver will likely largely work for it and all
> >>>you really need to do is modify the existing driver to handle the
> >>>quirks of your particular implementation of the wilc1000 chip. And,
> >>>often WiFi chips will let you change the VID/PID somewhere within
> >>>whatever non-volatile storage it has (like where it stores the MAC
> >>>address).
>>>
>>> >
>>> >
>>> >So it seems the wilc1000 devices from Microchip/Atmel are also using a
>>> >vendor id they did not buy. Could be that the mentioned 3rd party
>>> > providing
>>> >the SDIO IP actually owns that vendor id, but if you are building your
>>> > wifi
>>> >chip on that you should better buy you own vendor id from the SD
>>> >Association. Now if Harsha is actually working for Microchip (unclear to
>>> > me)
>>> >there is basically one party that should go shopping.
>>> >
>>
>> I would like to clarify that I am not building anything on top of
>> microchip wifi device.
>> We have a different HW . Its been just that 3rd party vendor providing
>> SDIO IP has given
>> same ID to different customers.
>
>
> So it is as I said, ie. you are using the 3rd party SDIO IP as is and add
> your own wifi IP to it? So what does the term "SDIO IP" mean here. Is it a
> piece of hardware that you hook up to your wifi hardware or is it
> VHDL/verilog in which the vendor id is defined. If it is VHDL you should
> really get your own vendor id from the SD Association and fix it. Otherwise,
> the 3rd party hardware should have means to change it. If not, you better
> find another party.
>
> Regards,
> Arend

Thank you folks for your comments.
The SDIO IP is VHDL IP core integrated on our SoC. And we figured out
a way to update vendor ID at run-time during boot.
We would get our own vendor ID  from SD association and proceed .

-
Harsha

Re: Support on vendor id and device id

[Arend van Spriel]

On 2/28/2018 12:14 AM, Harsha Rao wrote:

My suspicion is that your device, is fundamentally a wilc1000 and that
>>>the existing wilc1000 driver will likely largely work for it and all
>>>you really need to do is modify the existing driver to handle the
>>>quirks of your particular implementation of the wilc1000 chip. And,
>>>often WiFi chips will let you change the VID/PID somewhere within
>>>whatever non-volatile storage it has (like where it stores the MAC
>>>address).

>
>
>So it seems the wilc1000 devices from Microchip/Atmel are also using a
>vendor id they did not buy. Could be that the mentioned 3rd party providing
>the SDIO IP actually owns that vendor id, but if you are building your wifi
>chip on that you should better buy you own vendor id from the SD
>Association. Now if Harsha is actually working for Microchip (unclear to me)
>there is basically one party that should go shopping.
>

I would like to clarify that I am not building anything on top of
microchip wifi device.
We have a different HW . Its been just that 3rd party vendor providing
SDIO IP has given
same ID to different customers.


So it is as I said, ie. you are using the 3rd party SDIO IP as is and 
add your own wifi IP to it? So what does the term "SDIO IP" mean here. 
Is it a piece of hardware that you hook up to your wifi hardware or is 
it VHDL/verilog in which the vendor id is defined. If it is VHDL you 
should really get your own vendor id from the SD Association and fix it. 
Otherwise, the 3rd party hardware should have means to change it. If 
not, you better find another party.


Regards,
Arend

Re: Support on vendor id and device id

[Harsha Rao]

On Wed, Feb 28, 2018 at 1:34 AM, Arend van Spriel
 wrote:
> On 2/27/2018 6:34 PM, Steve deRosier wrote:
>>
>> Hi Harsha,
>>
>> Part 2 - the darn cmd-enter shortcut in gmail got me again. Sorry for
>> the premature send.
>>
>>>
>>> As both Larry and Arend mention, while it's possible to have multiple
>>> drivers with the same IDs, for obvious reasons it's not desirable.
>>>
>>> In theory the vendor IDs shouldn't be duplicated on fundamentally
>>> different devices, assuming that the manufacturers are doing things
>>> "right". The VID is paid for by buying it from the SD Association.
>
>
> Indeed. And this is fun already. If the sdio.ids file in systemd has any
> value the vendor id 0x296 is assigned to:
>
> 0296  GCT Semiconductor, Inc.
> 5347  GDM72xx WiMAX
>
>>> My suspicion is that your device, is fundamentally a wilc1000 and that
>>> the existing wilc1000 driver will likely largely work for it and all
>>> you really need to do is modify the existing driver to handle the
>>> quirks of your particular implementation of the wilc1000 chip. And,
>>> often WiFi chips will let you change the VID/PID somewhere within
>>> whatever non-volatile storage it has (like where it stores the MAC
>>> address).
>
>
> So it seems the wilc1000 devices from Microchip/Atmel are also using a
> vendor id they did not buy. Could be that the mentioned 3rd party providing
> the SDIO IP actually owns that vendor id, but if you are building your wifi
> chip on that you should better buy you own vendor id from the SD
> Association. Now if Harsha is actually working for Microchip (unclear to me)
> there is basically one party that should go shopping.
>
I would like to clarify that I am not building anything on top of
microchip wifi device.
We have a different HW . Its been just that 3rd party vendor providing
SDIO IP has given
same ID to different customers.


>
>> What I was going to mention is this is how it's worked for me many
>> times on several chips from QCA, Broadcom and Marvell included. The
>> use-case in this case would be building a custom WiFi module using a
>> vendor's chip. Very common - Silex, Laird and others do this all the
>> time: buy a chip from QCA (or Broadcom or Marvell or...), put it on a
>> board with lna, pa, passives, antennas, etc...  And you package it and
>> sell that often with value-add software.  Sometimes it's fine if it
>> represents like the original chip (for example on the ath6kl chips
>> Laird sells, it just uses the QCA IDs), but sometimes you want it to
>> represent as "my chip" (again, as Laird has to do on the chips they
>> sell to the Windows market to avoid the kind of driver conflicts
>> you're encountering). All of these chips usually have a way to store a
>> MAC address and at least all that I've worked on will also allow you
>> to change the SDIO VID in the same one-time-programable memory.
>> Obviously the method changes depending on the chip.
>>
>> However, in nearly every case, the chip is still fundamentally
>> whatever it was and we still use the upstream driver, though often
>> with tweaks and customizations based on our implementation. In the
>> cases where we've added a custom VID, we've also had to add the new
>> VID to the original driver and then key our quirks off that.
>>
>> I'm not saying this is definitely applicable in your case, but it is
>> common and maybe it will help.
>>

Definitely I will have a look at it .

>> Also thanks for bringing this issue up. Something I haven't thought
>> about in a long time but I'm adding it to my talk at ELC on WiFi
>> module interfacing. http://sched.co/DXn3
>
>
> Hah. Keep plugging that talk and seats may become scarce :-p
>
> Regards,
> Arend

Re: Support on vendor id and device id

[Arend van Spriel]

On 2/27/2018 6:34 PM, Steve deRosier wrote:

Hi Harsha,

Part 2 - the darn cmd-enter shortcut in gmail got me again. Sorry for
the premature send.



As both Larry and Arend mention, while it's possible to have multiple
drivers with the same IDs, for obvious reasons it's not desirable.

In theory the vendor IDs shouldn't be duplicated on fundamentally
different devices, assuming that the manufacturers are doing things
"right". The VID is paid for by buying it from the SD Association.


Indeed. And this is fun already. If the sdio.ids file in systemd has any 
value the vendor id 0x296 is assigned to:


0296  GCT Semiconductor, Inc.
5347  GDM72xx WiMAX


My suspicion is that your device, is fundamentally a wilc1000 and that
the existing wilc1000 driver will likely largely work for it and all
you really need to do is modify the existing driver to handle the
quirks of your particular implementation of the wilc1000 chip. And,
often WiFi chips will let you change the VID/PID somewhere within
whatever non-volatile storage it has (like where it stores the MAC
address).


So it seems the wilc1000 devices from Microchip/Atmel are also using a 
vendor id they did not buy. Could be that the mentioned 3rd party 
providing the SDIO IP actually owns that vendor id, but if you are 
building your wifi chip on that you should better buy you own vendor id 
from the SD Association. Now if Harsha is actually working for Microchip 
(unclear to me) there is basically one party that should go shopping.



What I was going to mention is this is how it's worked for me many
times on several chips from QCA, Broadcom and Marvell included. The
use-case in this case would be building a custom WiFi module using a
vendor's chip. Very common - Silex, Laird and others do this all the
time: buy a chip from QCA (or Broadcom or Marvell or...), put it on a
board with lna, pa, passives, antennas, etc...  And you package it and
sell that often with value-add software.  Sometimes it's fine if it
represents like the original chip (for example on the ath6kl chips
Laird sells, it just uses the QCA IDs), but sometimes you want it to
represent as "my chip" (again, as Laird has to do on the chips they
sell to the Windows market to avoid the kind of driver conflicts
you're encountering). All of these chips usually have a way to store a
MAC address and at least all that I've worked on will also allow you
to change the SDIO VID in the same one-time-programable memory.
Obviously the method changes depending on the chip.

However, in nearly every case, the chip is still fundamentally
whatever it was and we still use the upstream driver, though often
with tweaks and customizations based on our implementation. In the
cases where we've added a custom VID, we've also had to add the new
VID to the original driver and then key our quirks off that.

I'm not saying this is definitely applicable in your case, but it is
common and maybe it will help.

Also thanks for bringing this issue up. Something I haven't thought
about in a long time but I'm adding it to my talk at ELC on WiFi
module interfacing. http://sched.co/DXn3


Hah. Keep plugging that talk and seats may become scarce :-p

Regards,
Arend

Re: Support on vendor id and device id

[Steve deRosier]

Hi Harsha,

Part 2 - the darn cmd-enter shortcut in gmail got me again. Sorry for
the premature send.

>
> As both Larry and Arend mention, while it's possible to have multiple
> drivers with the same IDs, for obvious reasons it's not desirable.
>
> In theory the vendor IDs shouldn't be duplicated on fundamentally
> different devices, assuming that the manufacturers are doing things
> "right". The VID is paid for by buying it from the SD Association.
>
> My suspicion is that your device, is fundamentally a wilc1000 and that
> the existing wilc1000 driver will likely largely work for it and all
> you really need to do is modify the existing driver to handle the
> quirks of your particular implementation of the wilc1000 chip. And,
> often WiFi chips will let you change the VID/PID somewhere within
> whatever non-volatile storage it has (like where it stores the MAC
> address).

What I was going to mention is this is how it's worked for me many
times on several chips from QCA, Broadcom and Marvell included. The
use-case in this case would be building a custom WiFi module using a
vendor's chip. Very common - Silex, Laird and others do this all the
time: buy a chip from QCA (or Broadcom or Marvell or...), put it on a
board with lna, pa, passives, antennas, etc...  And you package it and
sell that often with value-add software.  Sometimes it's fine if it
represents like the original chip (for example on the ath6kl chips
Laird sells, it just uses the QCA IDs), but sometimes you want it to
represent as "my chip" (again, as Laird has to do on the chips they
sell to the Windows market to avoid the kind of driver conflicts
you're encountering). All of these chips usually have a way to store a
MAC address and at least all that I've worked on will also allow you
to change the SDIO VID in the same one-time-programable memory.
Obviously the method changes depending on the chip.

However, in nearly every case, the chip is still fundamentally
whatever it was and we still use the upstream driver, though often
with tweaks and customizations based on our implementation. In the
cases where we've added a custom VID, we've also had to add the new
VID to the original driver and then key our quirks off that.

I'm not saying this is definitely applicable in your case, but it is
common and maybe it will help.

Also thanks for bringing this issue up. Something I haven't thought
about in a long time but I'm adding it to my talk at ELC on WiFi
module interfacing. http://sched.co/DXn3

- Steve

--
Steve deRosier
Cal-Sierra Consulting LLC
https://www.cal-sierra.com/

Re: Support on vendor id and device id

[Steve deRosier]

Hi Harsha,

On Tue, Feb 27, 2018 at 7:29 AM, Harsha Rao  wrote:
> On Tue, Feb 27, 2018 at 8:44 PM, Larry Finger  
> wrote:
>> On 02/27/2018 07:30 AM, Harsha Rao wrote:
>>>
>>> On Tue, Feb 27, 2018 at 4:45 PM, Arend van Spriel
>>>  wrote:

 On 2/27/2018 11:16 AM, Harsha Rao wrote:
>
>
> Hi folks,
> I am developing a new wifi driver for our sdio based wifi device.
>
> I see that SDIO_VENDOR_ID and SDIO_DEVICE_ID for our device (We had
> bought the SDIO IP from 3rd party)  is already  been used by some
> other vendor and its already into staging .
>
> Please suggest me how can I move forward to submit the driver for
> staging.



 Can you be more specific about the conflicting vendor id and device id?
>>>
>>>
>>> Hi,
>>> My driver kicks in  with vendor id 0x296 and device id 0x5347.
>>> But when I grepped the the kernel source I could see an other driver
>>> wilc1000 using the same vendor ID and device ID
>>> ( We have a different hw than them !)
>>>
> Is there a way to reconfigure the values in the SDIO host for
> different value or does Linux allow two drivers with same values to
> survive mutual exclusively



 First come first serve. When a device is detected, the driver core looks
 for
 drivers supporting it based on device table and the first that
 successfully
 returns from the .probe() callback is bound to the device.

 Regards,
 Arend
>>>
>>>
>>> Does Linux allow two drivers with conflicting device and vendor IDs to
>>> stay on in kernel ?
>>
>>
>> Yes. We have a similar situation with the rtl8192e driver in staging and the
>> rtl8192se driver in the wireless tree, which share PCI ID 0bda:8192. Our
>> solution was to insert code in the probe routine that tests some value in
>> addition to the PCI ID. In our case, that was the PCI revision id. Line 2496
>> of file drivers/staging/rtl8192e/rtl8192e/rtl_core.c shows the test in the
>> staging driver.
>>
>> The downsides of this unfortunate duplication are that if the wrong driver
>> loads first, then it remains loaded, you will need to cooperate with the
>> maintainer of the other driver to insert the extra detection code, and you
>> may need to do quite a bit of processing to be able to determine whether you
>> have the correct device.
>>
>> Larry
>>

As both Larry and Arend mention, while it's possible to have multiple
drivers with the same IDs, for obvious reasons it's not desirable.

In theory the vendor IDs shouldn't be duplicated on fundamentally
different devices, assuming that the manufacturers are doing things
"right". The VID is paid for by buying it from the SD Association.

My suspicion is that your device, is fundamentally a wilc1000 and that
the existing wilc1000 driver will likely largely work for it and all
you really need to do is modify the existing driver to handle the
quirks of your particular implementation of the wilc1000 chip. And,
often WiFi chips will let you change the VID/PID somewhere within
whatever non-volatile storage it has (like where it stores the MAC
address).

Re: Support on vendor id and device id

[Harsha Rao]

On Tue, Feb 27, 2018 at 8:44 PM, Larry Finger  wrote:
> On 02/27/2018 07:30 AM, Harsha Rao wrote:
>>
>> On Tue, Feb 27, 2018 at 4:45 PM, Arend van Spriel
>>  wrote:
>>>
>>> On 2/27/2018 11:16 AM, Harsha Rao wrote:


 Hi folks,
 I am developing a new wifi driver for our sdio based wifi device.

 I see that SDIO_VENDOR_ID and SDIO_DEVICE_ID for our device (We had
 bought the SDIO IP from 3rd party)  is already  been used by some
 other vendor and its already into staging .

 Please suggest me how can I move forward to submit the driver for
 staging.
>>>
>>>
>>>
>>> Can you be more specific about the conflicting vendor id and device id?
>>
>>
>> Hi,
>> My driver kicks in  with vendor id 0x296 and device id 0x5347.
>> But when I grepped the the kernel source I could see an other driver
>> wilc1000 using the same vendor ID and device ID
>> ( We have a different hw than them !)
>>
 Is there a way to reconfigure the values in the SDIO host for
 different value or does Linux allow two drivers with same values to
 survive mutual exclusively
>>>
>>>
>>>
>>> First come first serve. When a device is detected, the driver core looks
>>> for
>>> drivers supporting it based on device table and the first that
>>> successfully
>>> returns from the .probe() callback is bound to the device.
>>>
>>> Regards,
>>> Arend
>>
>>
>> Does Linux allow two drivers with conflicting device and vendor IDs to
>> stay on in kernel ?
>
>
> Yes. We have a similar situation with the rtl8192e driver in staging and the
> rtl8192se driver in the wireless tree, which share PCI ID 0bda:8192. Our
> solution was to insert code in the probe routine that tests some value in
> addition to the PCI ID. In our case, that was the PCI revision id. Line 2496
> of file drivers/staging/rtl8192e/rtl8192e/rtl_core.c shows the test in the
> staging driver.
>
> The downsides of this unfortunate duplication are that if the wrong driver
> loads first, then it remains loaded, you will need to cooperate with the
> maintainer of the other driver to insert the extra detection code, and you
> may need to do quite a bit of processing to be able to determine whether you
> have the correct device.
>
> Larry
>
Thanks Larry and Arend for your valuable time in clarifying my doubts..

-
Harsha

Re: Support on vendor id and device id

[Larry Finger]

On 02/27/2018 07:30 AM, Harsha Rao wrote:

On Tue, Feb 27, 2018 at 4:45 PM, Arend van Spriel
 wrote:

On 2/27/2018 11:16 AM, Harsha Rao wrote:


Hi folks,
I am developing a new wifi driver for our sdio based wifi device.

I see that SDIO_VENDOR_ID and SDIO_DEVICE_ID for our device (We had
bought the SDIO IP from 3rd party)  is already  been used by some
other vendor and its already into staging .

Please suggest me how can I move forward to submit the driver for staging.



Can you be more specific about the conflicting vendor id and device id?


Hi,
My driver kicks in  with vendor id 0x296 and device id 0x5347.
But when I grepped the the kernel source I could see an other driver
wilc1000 using the same vendor ID and device ID
( We have a different hw than them !)


Is there a way to reconfigure the values in the SDIO host for
different value or does Linux allow two drivers with same values to
survive mutual exclusively



First come first serve. When a device is detected, the driver core looks for
drivers supporting it based on device table and the first that successfully
returns from the .probe() callback is bound to the device.

Regards,
Arend


Does Linux allow two drivers with conflicting device and vendor IDs to
stay on in kernel ?


Yes. We have a similar situation with the rtl8192e driver in staging and the 
rtl8192se driver in the wireless tree, which share PCI ID 0bda:8192. Our 
solution was to insert code in the probe routine that tests some value in 
addition to the PCI ID. In our case, that was the PCI revision id. Line 2496 of 
file drivers/staging/rtl8192e/rtl8192e/rtl_core.c shows the test in the staging 
driver.


The downsides of this unfortunate duplication are that if the wrong driver loads 
first, then it remains loaded, you will need to cooperate with the maintainer of 
the other driver to insert the extra detection code, and you may need to do 
quite a bit of processing to be able to determine whether you have the correct 
device.


Larry

Re: Support on vendor id and device id

[Harsha Rao]

On Tue, Feb 27, 2018 at 4:45 PM, Arend van Spriel
 wrote:
> On 2/27/2018 11:16 AM, Harsha Rao wrote:
>>
>> Hi folks,
>> I am developing a new wifi driver for our sdio based wifi device.
>>
>> I see that SDIO_VENDOR_ID and SDIO_DEVICE_ID for our device (We had
>> bought the SDIO IP from 3rd party)  is already  been used by some
>> other vendor and its already into staging .
>>
>> Please suggest me how can I move forward to submit the driver for staging.
>
>
> Can you be more specific about the conflicting vendor id and device id?

Hi,
My driver kicks in  with vendor id 0x296 and device id 0x5347.
But when I grepped the the kernel source I could see an other driver
wilc1000 using the same vendor ID and device ID
( We have a different hw than them !)

>> Is there a way to reconfigure the values in the SDIO host for
>> different value or does Linux allow two drivers with same values to
>> survive mutual exclusively
>
>
> First come first serve. When a device is detected, the driver core looks for
> drivers supporting it based on device table and the first that successfully
> returns from the .probe() callback is bound to the device.
>
> Regards,
> Arend

Does Linux allow two drivers with conflicting device and vendor IDs to
stay on in kernel ?

-
Harsha

Re: Support on vendor id and device id

[Arend van Spriel]

On 2/27/2018 11:16 AM, Harsha Rao wrote:

Hi folks,
I am developing a new wifi driver for our sdio based wifi device.

I see that SDIO_VENDOR_ID and SDIO_DEVICE_ID for our device (We had
bought the SDIO IP from 3rd party)  is already  been used by some
other vendor and its already into staging .

Please suggest me how can I move forward to submit the driver for staging.


Can you be more specific about the conflicting vendor id and device id?


Is there a way to reconfigure the values in the SDIO host for
different value or does Linux allow two drivers with same values to
survive mutual exclusively


First come first serve. When a device is detected, the driver core looks 
for drivers supporting it based on device table and the first that 
successfully returns from the .probe() callback is bound to the device.


Regards,
Arend