Driver for pca9541 bus master selector

Driver for pca9541 bus master selector

[Guenter Roeck]

Hi all,

I am in the process of writing a driver for the pca9541 bus master selector. 
I using the i2c multiplexer framework, which fits the requirements quite nicely.

If anyone happens to know about an existing driver for this chip, please let me know.
I would love to get some input for the arbitration mechanism needed to determine
if and when to switch bus ownership from one master to the other.

Thanks,
Guenter

Re: Driver for pca9541 bus master selector

[Jean Delvare]

Hi Guenter,

On Tue, 28 Sep 2010 07:33:02 -0700, Guenter Roeck wrote:
> I am in the process of writing a driver for the pca9541 bus master selector. 
> I using the i2c multiplexer framework, which fits the requirements quite nicely.
> 
> If anyone happens to know about an existing driver for this chip, please let me know.
> I would love to get some input for the arbitration mechanism needed to determine
> if and when to switch bus ownership from one master to the other.

See drivers/i2c/muxes/pca954x.c in 2.6.36-rc2 and later.

-- 
Jean Delvare

Re: Driver for pca9541 bus master selector

[Guenter Roeck]

On Tue, Sep 28, 2010 at 11:16:07AM -0400, Jean Delvare wrote:
> Hi Guenter,
> 
> On Tue, 28 Sep 2010 07:33:02 -0700, Guenter Roeck wrote:
> > I am in the process of writing a driver for the pca9541 bus master selector. 
> > I using the i2c multiplexer framework, which fits the requirements quite nicely.
> > 
> > If anyone happens to know about an existing driver for this chip, please let me know.
> > I would love to get some input for the arbitration mechanism needed to determine
> > if and when to switch bus ownership from one master to the other.
> 
> See drivers/i2c/muxes/pca954x.c in 2.6.36-rc2 and later.
> 
Yes, I know. I use it as basis. But it doesn't support pca9541, unfortunately.
I use the select/deselect mechanism to trigger arbitration, but I still need
a mechanism to communicate between the masters. After all, the problem to solve
is that I have two masters and one slave, not one master and multiple slaves
as with the other chips.

Note that I do have a solution which I think will work. Just figured it might
save time if someone else already solved the same problem, and especially it 
might help me to determine arbitration time constants. There was mention of
a pca9541 driver several years ago, but I have not been able to find details.

Thanks,
Guenter

Re: Driver for pca9541 bus master selector

[Jean Delvare]

On Tue, 28 Sep 2010 09:07:46 -0700, Guenter Roeck wrote:
> On Tue, Sep 28, 2010 at 11:16:07AM -0400, Jean Delvare wrote:
> > Hi Guenter,
> > 
> > On Tue, 28 Sep 2010 07:33:02 -0700, Guenter Roeck wrote:
> > > I am in the process of writing a driver for the pca9541 bus master selector. 
> > > I using the i2c multiplexer framework, which fits the requirements quite nicely.
> > > 
> > > If anyone happens to know about an existing driver for this chip, please let me know.
> > > I would love to get some input for the arbitration mechanism needed to determine
> > > if and when to switch bus ownership from one master to the other.
> > 
> > See drivers/i2c/muxes/pca954x.c in 2.6.36-rc2 and later.
> > 
> Yes, I know. I use it as basis. But it doesn't support pca9541, unfortunately.

Oh, my bad. I had not noticed the PCA9541 was of a totally different kind.

> I use the select/deselect mechanism to trigger arbitration, but I still need
> a mechanism to communicate between the masters. After all, the problem to solve
> is that I have two masters and one slave, not one master and multiple slaves
> as with the other chips.
> 
> Note that I do have a solution which I think will work. Just figured it might
> save time if someone else already solved the same problem, and especially it 
> might help me to determine arbitration time constants. There was mention of
> a pca9541 driver several years ago, but I have not been able to find details.

-- 
Jean Delvare

Re: Driver for pca9541 bus master selector

[Guenter Roeck]

Hi Jean,

On Tue, Sep 28, 2010 at 12:42:15PM -0400, Jean Delvare wrote:
> On Tue, 28 Sep 2010 09:07:46 -0700, Guenter Roeck wrote:
> > On Tue, Sep 28, 2010 at 11:16:07AM -0400, Jean Delvare wrote:
> > > Hi Guenter,
> > > 
> > > On Tue, 28 Sep 2010 07:33:02 -0700, Guenter Roeck wrote:
> > > > I am in the process of writing a driver for the pca9541 bus master selector. 
> > > > I using the i2c multiplexer framework, which fits the requirements quite nicely.
> > > > 
> > > > If anyone happens to know about an existing driver for this chip, please let me know.
> > > > I would love to get some input for the arbitration mechanism needed to determine
> > > > if and when to switch bus ownership from one master to the other.
> > > 
> > > See drivers/i2c/muxes/pca954x.c in 2.6.36-rc2 and later.
> > > 
> > Yes, I know. I use it as basis. But it doesn't support pca9541, unfortunately.
> 
> Oh, my bad. I had not noticed the PCA9541 was of a totally different kind.
> 
Actually, I fell into the same trap. I was quite happy that someone else wrote 
the driver for me ... until I realized that the chips solve the opposite problem.

Time goes by ... have a look at

http://lists.lm-sensors.org/pipermail/lm-sensors/2004-June/008023.html

I wasn't sure if you (or someone else) ever had a driver for 9541, or if you referred
to 9540 in above post. The links in the post don't work anymore.

Thanks,
Guenter

Re: Driver for pca9541 bus master selector

[Jean Delvare]

On Tue, 28 Sep 2010 10:16:27 -0700, Guenter Roeck wrote:
> On Tue, Sep 28, 2010 at 12:42:15PM -0400, Jean Delvare wrote:
> > Oh, my bad. I had not noticed the PCA9541 was of a totally different kind.
>
> Actually, I fell into the same trap. I was quite happy that someone else wrote 
> the driver for me ... until I realized that the chips solve the opposite problem.
> 
> Time goes by ... have a look at
> 
> http://lists.lm-sensors.org/pipermail/lm-sensors/2004-June/008023.html

I didn't remember that discussion (unsurprisingly.)

> I wasn't sure if you (or someone else) ever had a driver for 9541, or if you referred
> to 9540 in above post. The links in the post don't work anymore.

I was speaking of the PCA9540. I don't think anyone ever wrote a driver
for the PCA9541. And in all honestly I don't think the new i2c-mux
infrastructure is ready for this. It was written with single-master
topologies in mind. You'll have to rethink a lot of things if you want
to add support for "inverted" multiplexers. And I don't have the time
to help, sorry.

I don't quite see the point of such chips, BTW. I2C supports
multi-master already, so two masters talking to the same slave
sequentially is already supported.

-- 
Jean Delvare

Re: Driver for pca9541 bus master selector

[Guenter Roeck]

On Wed, Sep 29, 2010 at 03:19:28AM -0400, Jean Delvare wrote:
[ ... ]
> 
> I was speaking of the PCA9540. I don't think anyone ever wrote a driver
> for the PCA9541. And in all honestly I don't think the new i2c-mux
> infrastructure is ready for this. It was written with single-master
> topologies in mind. You'll have to rethink a lot of things if you want
> to add support for "inverted" multiplexers. And I don't have the time
> to help, sorry.
> 
Actually, the infrastructure works quite nicely. I use select_chan
to acquire the slave, and release_chan to initiate release.
It behaves like a one-channel multiplexer.

> I don't quite see the point of such chips, BTW. I2C supports
> multi-master already, so two masters talking to the same slave
> sequentially is already supported.
> 
Tell that to our HW team, and to them:

	http://www.lineagepower.com/oem/pdf/CPLI2C.pdf

Guenter

Re: Driver for pca9541 bus master selector

[Michael Lawnick]

Guenter Roeck said the following:
> On Wed, Sep 29, 2010 at 03:19:28AM -0400, Jean Delvare wrote:
> [ ... ]
>> 
>> I was speaking of the PCA9540. I don't think anyone ever wrote a driver
>> for the PCA9541. And in all honestly I don't think the new i2c-mux
>> infrastructure is ready for this. It was written with single-master
>> topologies in mind. You'll have to rethink a lot of things if you want
>> to add support for "inverted" multiplexers. And I don't have the time
>> to help, sorry.
>> 
> Actually, the infrastructure works quite nicely. I use select_chan
> to acquire the slave, and release_chan to initiate release.
> It behaves like a one-channel multiplexer.
> 
>> I don't quite see the point of such chips, BTW. I2C supports
>> multi-master already, so two masters talking to the same slave
>> sequentially is already supported.
>> 
> Tell that to our HW team, and to them:
> 
> 	http://www.lineagepower.com/oem/pdf/CPLI2C.pdf
> 
I only took a rough look, but it seemed to me that the pca is part of
the main device.
>From abstraction side of view your problem seems a matter of
synchronization of two independent computer systems. You will need a
shared resource. The access to it could either be part of driver code or
by introducing a 'one-channel multiplexer' as you did. But the sync
needs to be done by your code via a shared data object.

-- 
KR
Michael

Re: Driver for pca9541 bus master selector

[Guenter Roeck]

On Thu, Sep 30, 2010 at 01:59:44AM -0400, Michael Lawnick wrote:
[ ... ]

> I only took a rough look, but it seemed to me that the pca is part of
> the main device.

Yes

> From abstraction side of view your problem seems a matter of
> synchronization of two independent computer systems. You will need a
> shared resource. The access to it could either be part of driver code or
> by introducing a 'one-channel multiplexer' as you did. But the sync
> needs to be done by your code via a shared data object.
> 
Actually, the pca9541 has a bit in its register set which I can use.
See NTESTON in its control register. That is visible as interrupt status
at the other master. That is all I need, and reduces complexity significantly.

Thanks,
Guenter

Re: Driver for pca9541 bus master selector

[Guenter Roeck]

Jean Delvare <khali@...> writes:
[...]
> 
> I don't quite see the point of such chips, BTW. I2C supports
> multi-master already, so two masters talking to the same slave
> sequentially is already supported.
> 
To follow up on this, turns out there are (at least) two good reasons for having
such a chip.

Firts, not all i2c masters support multi-master configuration.

The second reason is much more important. In a redundant system, one would not
want redundant masters to be on the same i2c bus. If one of the two fails and
blocks the bus, access to all slaves would be blocked. Not a good idea. Much
better to have two i2c busses, and select the current master on the slave side.

Turns out both reasons apply in my use case. Testing shows that master selection
works quite well with the driver I just submitted. I can persistently access all
slaves from both masters even under high contention (both masters trying to
access the same slave at the same time). If one of the masters fails while
having access to the connected slaves, the other simply takes over after
arbitration times out.

Guenter

Re: Driver for pca9541 bus master selector

[Guenter Roeck]

On Wed, Sep 29, 2010 at 03:19:28AM -0400, Jean Delvare wrote:
[ ... ]
> I don't quite see the point of such chips, BTW. I2C supports
> multi-master already, so two masters talking to the same slave
> sequentially is already supported.

To follow up on this, turns out there are (at least) two good reasons for having
such a chip.

First, not all i2c masters support multi-master configuration.

The second reason is much more important. In a redundant system, one would not
want redundant masters to be on the same i2c bus. If one of the two fails and
blocks the bus, access to all slaves would be blocked. Not a good idea. Much
better to have two i2c busses, and select the current master on the slave side.

Turns out both reasons apply in my use case. Testing shows that master selection
works quite well with the driver I just submitted. I can persistently access all
slaves from both masters even under high contention (both masters trying to
access the same slave at the same time). If one of the masters fails while
having access to the connected slaves, the other simply takes over after
arbitration times out.

Guenter