[Xenomai-help] xeno_can_mem driver

[Xenomai-help] xeno_can_mem driver

[Nils Armbruster]

Hi everybody,

I'm currently trying to get the can-interfaces of an Samsung S3C2440 based board working (NetDCU10).
The used Linux kernel version is 2.6.19, because this was deliverd in the BSP.
I patched this kernel with Xenomai 2.4.0., as this was the newest Version I managed the Kernel to compile with.

The standard driver delivered with the board (lincan 0.3.3) mappes the can devices to 
10000000-10000007 : onboard-can                                                 
10000008-1000000f : 
onboard-can

So I tried to use the driver xeno_can_mem.

I use 
modprobe xen_can_mem mem=0x10000000 irq=17 clock=48000000
to install the module but it complains that there is no such device...

As far as I know, this board uses the an sja1000p chip for can. Do you know if there are any differences to the sja1000 chip?
Are there any changes necessary in the driver sources?

I figured out, that the installation of the driver stops in the function int rtcan_sja1000_register(struct rtcan_device *dev)
And there in that part:

    printk("chip->read_reg(dev, SJA_SR) = %d \n",chip->read_reg(dev, SJA_SR));
    if ((chip->read_reg(dev, SJA_SR) &
	 (SJA_SR_RBS | SJA_SR_DOS | SJA_SR_TBS)) != SJA_SR_TBS) {
	printk("ERROR! No SJA1000 device found!\n");
	return -ENODEV;
    } 

When I comment this out the installation works and in /proc/iomem sja1000-mem is visible.
But I'm not able to send Can-messages with rtcansen.
The program seems to work, but no messages are send as I checked it with a scope.

Does anybody have an hint for me, how to get the can driver working?
Thank you in advice!

best regards,

Nils
                                                 
   
-- 
GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT!
Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01

Re: [Xenomai-help] xeno_can_mem driver

[Sebastian Smolorz]

Nils Armbruster wrote:
> I'm currently trying to get the can-interfaces of an Samsung S3C2440
>  based board working (NetDCU10). The used Linux kernel version is 2.6.19,
>  because this was deliverd in the BSP. I patched this kernel with Xenomai
>  2.4.0., as this was the newest Version I managed the Kernel to compile
>  with.
> 
> The standard driver delivered with the board (lincan 0.3.3) mappes the
>  can devices to 10000000-10000007 : onboard-can
> 10000008-1000000f :
> onboard-can

Hmm, the SJA1000 has 32 registers, not just 8.

> So I tried to use the driver xeno_can_mem.
> 
> I use
> modprobe xen_can_mem mem=0x10000000 irq=17 clock=48000000

Are you sure you give the right clock frequency? Seems to be too high.

> to install the module but it complains that there is no such device...
> 
> As far as I know, this board uses the an sja1000p chip for can. Do you
>  know if there are any differences to the sja1000 chip? Are there any
>  changes necessary in the driver sources?
> 
> I figured out, that the installation of the driver stops in the function
>  int rtcan_sja1000_register(struct rtcan_device *dev) And there in that
>  part:
> 
>     printk("chip->read_reg(dev, SJA_SR) = %d \n",chip->read_reg(dev,
>  SJA_SR)); if ((chip->read_reg(dev, SJA_SR) &
> 	 (SJA_SR_RBS | SJA_SR_DOS | SJA_SR_TBS)) != SJA_SR_TBS) {
> 	printk("ERROR! No SJA1000 device found!\n");
> 	return -ENODEV;
>     }
> 
> When I comment this out the installation works and in /proc/iomem
>  sja1000-mem is visible. But I'm not able to send Can-messages with
>  rtcansen.
> The program seems to work, but no messages are send as I checked it with
>  a scope.

There is no point in commenting out the driver's check for the existence of 
the SJA1000 and complain that it does not work. This check is there for a 
good reason.

> 
> Does anybody have an hint for me, how to get the can driver working?

First find out where the 32 registers of the SJA1000 are mapped and also 
check the clock frequency.

-- 
Sebastian

Re: [Xenomai-help] xeno_can_mem driver

[Nils Armbruster]

Thank you for your reply !

>>  I'm currently trying to get the can-interfaces of an Samsung S3C2440
>>   based board working (NetDCU10). The used Linux kernel version is 
>> 2.6.19,
>>   because this was deliverd in the BSP. I patched this kernel with 
>> Xenomai
>>   2.4.0., as this was the newest Version I managed the Kernel to compile
>>   with.
>>
>>  The standard driver delivered with the board (lincan 0.3.3) mappes the
>>   can devices to 10000000-10000007 : onboard-can
>>  10000008-1000000f :
>>  onboard-can
> 
> Hmm, the SJA1000 has 32 registers, not just 8.
> 

This is strange, cause the original driver just maps these 8 registers.

>>  So I tried to use the driver xeno_can_mem.
>>
>>  I use
>>  modprobe xen_can_mem mem=0x10000000 irq=17 clock=48000000
> 
> Are you sure you give the right clock frequency? Seems to be too high.
> 

I used this clock frequency because I noticed, that the frequency is divided by 2. So I just tried it.
The frequency I found in the original driver is 24MHz.
But even with clock=24000000 it appears "No such device".

>>  to install the module but it complains that there is no such device...
>>
>>  As far as I know, this board uses the an sja1000p chip for can. Do you
>>   know if there are any differences to the sja1000 chip? Are there any
>>   changes necessary in the driver sources?
>>
>>  I figured out, that the installation of the driver stops in the function
>>   int rtcan_sja1000_register(struct rtcan_device *dev) And there in that
>>   part:
>>
>>      printk("chip->read_reg(dev, SJA_SR) = %d \n",chip->read_reg(dev,
>>   SJA_SR)); if ((chip->read_reg(dev, SJA_SR)&
>>      (SJA_SR_RBS | SJA_SR_DOS | SJA_SR_TBS)) != SJA_SR_TBS) {
>>      printk("ERROR! No SJA1000 device found!\n");
>>      return -ENODEV;
>>      }
>>
>>  When I comment this out the installation works and in /proc/iomem
>>   sja1000-mem is visible. But I'm not able to send Can-messages with
>>   rtcansen.
>>  The program seems to work, but no messages are send as I checked it with
>>   a scope.
> 
> There is no point in commenting out the driver's check for the existence of
> the SJA1000 and complain that it does not work. This check is there for a
> good reason.
> 

It was just a try to find out where the problem comes from.

>>
>>  Does anybody have an hint for me, how to get the can driver working?
> 
> First find out where the 32 registers of the SJA1000 are mapped and also
> check the clock frequency.
>

Sorry for my stupid question, but how can I find out where the registers are mapped?
In the original driver the base address is defined as 0x10000000.
Port size ist defined as 4 and when the memory region is requested twice of the port size is requested:
request_mem_region(mem_addr, nx_can->if_portsize * 2, "onboard-can")

Do you have any suggestions how I should proceed?

Thank you!
 
Best regards,

Nils 

-- 
GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT!
Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01

Re: [Xenomai-help] xeno_can_mem driver

[Wolfgang Grandegger]

On 06/09/2010 10:48 AM, Nils Armbruster wrote:
> Thank you for your reply !
> 
>>>  I'm currently trying to get the can-interfaces of an Samsung S3C2440
>>>   based board working (NetDCU10). The used Linux kernel version is 
>>> 2.6.19,
>>>   because this was deliverd in the BSP. I patched this kernel with 
>>> Xenomai
>>>   2.4.0., as this was the newest Version I managed the Kernel to compile
>>>   with.
>>>
>>>  The standard driver delivered with the board (lincan 0.3.3) mappes the
>>>   can devices to 10000000-10000007 : onboard-can
>>>  10000008-1000000f :
>>>  onboard-can
>>
>> Hmm, the SJA1000 has 32 registers, not just 8.
>>
> 
> This is strange, cause the original driver just maps these 8 registers.

How does the read/write register access function look like in your
lincan driver?

> 
>>>  So I tried to use the driver xeno_can_mem.
>>>
>>>  I use
>>>  modprobe xen_can_mem mem=0x10000000 irq=17 clock=48000000

Where did you get irq=17 from? From /proc/interrups?

> Sorry for my stupid question, but how can I find out where the registers are mapped?
> In the original driver the base address is defined as 0x10000000.
> Port size ist defined as 4 and when the memory region is requested twice of the port size is requested:
> request_mem_region(mem_addr, nx_can->if_portsize * 2, "onboard-can")
> 
> Do you have any suggestions how I should proceed?

I assume that the lincan driver source code is under the GPL. Feel free
to post it here so we can have a closer look.

Wolfgang.

Re: [Xenomai-help] xeno_can_mem driver

[Sebastian Smolorz]

Nils Armbruster wrote:
> >>  So I tried to use the driver xeno_can_mem.
> >>
> >>  I use
> >>  modprobe xen_can_mem mem=0x10000000 irq=17 clock=48000000
> >
> > Are you sure you give the right clock frequency? Seems to be too high.
> 
> I used this clock frequency because I noticed, that the frequency is
>  divided by 2. So I just tried it. The frequency I found in the original
>  driver is 24MHz.

You definitely have to define 24MHz here, not 48MHz.

> But even with clock=24000000 it appears "No such device".

Sure, this is another problem: The SJA1000 registers are not accessible (see 
below).

> >>  Does anybody have an hint for me, how to get the can driver working?
> >
> > First find out where the 32 registers of the SJA1000 are mapped and
> > also check the clock frequency.
> 
> Sorry for my stupid question, but how can I find out where the registers
>  are mapped? In the original driver the base address is defined as
>  0x10000000. Port size ist defined as 4 and when the memory region is
>  requested twice of the port size is requested:
>  request_mem_region(mem_addr, nx_can->if_portsize * 2, "onboard-can")
> 
> Do you have any suggestions how I should proceed?

Either you study the appropriate datasheet to learn how the SJA1000 
registers are accessible from your board, or you go through the code of your 
original driver and search for the way the SJA registers are read and write. 
For example, the SJA_CDR register is the last of the 32 registers.

-- 
Sebastian

Re: [Xenomai-help] xeno_can_mem driver

[Nils Armbruster]

-------- Original-Nachricht --------
> Datum: Wed, 09 Jun 2010 11:03:03 +0200
> Von: Wolfgang Grandegger <wg@domain.hid>
> An: Nils Armbruster <nils-a@domain.hid>
> CC: smolorz@domain.hid, xenomai@xenomai.org
> Betreff: Re: [Xenomai-help] xeno_can_mem driver

> On 06/09/2010 10:48 AM, Nils Armbruster wrote:
> > Thank you for your reply !
> > 
> >>>  I'm currently trying to get the can-interfaces of an Samsung S3C2440
> >>>   based board working (NetDCU10). The used Linux kernel version is 
> >>> 2.6.19,
> >>>   because this was deliverd in the BSP. I patched this kernel with 
> >>> Xenomai
> >>>   2.4.0., as this was the newest Version I managed the Kernel to
> compile
> >>>   with.
> >>>
> >>>  The standard driver delivered with the board (lincan 0.3.3) mappes
> the
> >>>   can devices to 10000000-10000007 : onboard-can
> >>>  10000008-1000000f :
> >>>  onboard-can
> >>
> >> Hmm, the SJA1000 has 32 registers, not just 8.
> >>
> > 
> > This is strange, cause the original driver just maps these 8 registers.
> 
> How does the read/write register access function look like in your
> lincan driver?

The registers in lincan driver are written as follows:

extern inline void can_write_reg(const struct canchip_t *chip, unsigned char data, unsigned address)
{
	unsigned long address_to_write;
	address_to_write = chip->chip_base_addr+address;
	chip->write_register(chip, data, address_to_write);
}

where through chip->write_register the following function is called:

void netdcux_can_write_register(const struct canchip_t *chip, unsigned data, unsigned long address)
{
	struct candevice_t *candev = chip->hostdevice;
	struct netdcux_can *nx_can = (struct netdcux_can*)candev->private_data;
	int chipnr;

	chipnr = (address & (0xF << 28)) >> 28;
	nx_can->send_byte_can(chipnr, (0xFF & address), data, (void *)nx_can);
}

adress should be 0x10000000 + Reg so chipnr should be just 0x10000000  after the bit operations.
I think nx_can->send_byte_can calls this function:

void FS_OnBoard_can_data_out(int if_nr, unsigned char addr, unsigned char data, void *dev)
{
	struct netdcux_can *nx_can = (struct netdcux_can *)dev;
	unsigned long can_flags;

#ifdef FS_CAN_DEBUG
	printk("OnBoard_can_out[%d] addr: %x value: %x\n", if_nr, addr, data);
#endif
	spin_lock_irqsave(&fscan_atomic_lock, can_flags);
	iowrite8(addr, nx_can->if_virtaddr_addr[if_nr]);
	iowrite8(data, nx_can->if_virtaddr_data[if_nr]);
	spin_unlock_irqrestore(&fscan_atomic_lock, can_flags);
}

So I think the address of the can controller should still be 0x10000000.
Or am I thinking wrong?

> 
> > 
> >>>  So I tried to use the driver xeno_can_mem.
> >>>
> >>>  I use
> >>>  modprobe xen_can_mem mem=0x10000000 irq=17 clock=48000000
> 
> Where did you get irq=17 from? From /proc/interrups?
> 
> > Sorry for my stupid question, but how can I find out where the registers
> are mapped?
> > In the original driver the base address is defined as 0x10000000.
> > Port size ist defined as 4 and when the memory region is requested twice
> of the port size is requested:
> > request_mem_region(mem_addr, nx_can->if_portsize * 2, "onboard-can")
> > 
> > Do you have any suggestions how I should proceed?
> 
> I assume that the lincan driver source code is under the GPL. Feel free
> to post it here so we can have a closer look.
> 
> Wolfgang.

-- 
GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT!
Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01

Re: [Xenomai-help] xeno_can_mem driver

[Sebastian Smolorz]

Nils Armbruster wrote:
> void FS_OnBoard_can_data_out(int if_nr, unsigned char addr, unsigned char
>  data, void *dev) {
> 	struct netdcux_can *nx_can = (struct netdcux_can *)dev;
> 	unsigned long can_flags;
> 
> #ifdef FS_CAN_DEBUG
> 	printk("OnBoard_can_out[%d] addr: %x value: %x\n", if_nr, addr, data);
> #endif
> 	spin_lock_irqsave(&fscan_atomic_lock, can_flags);
> 	iowrite8(addr, nx_can->if_virtaddr_addr[if_nr]);
> 	iowrite8(data, nx_can->if_virtaddr_data[if_nr]);
> 	spin_unlock_irqrestore(&fscan_atomic_lock, can_flags);
> }
> 

Here we have the reason. the SJA1000 registers aren't linearly mapped into 
the memory space of the processor; it seems that every write to such a 
registers needs two iowrite8s: First the address, then the data. So I would 
say the xeno_can_mem driver isn't useful for you. Your CAN hardware needs a 
special RTCAN driver which does not exist yet.

-- 
Sebastian