i2c_smbus_write_block_data vs scheduled routine calls (add_timer) - kernel crash

i2c_smbus_write_block_data vs scheduled routine calls (add_timer) - kernel crash

[NiTr0]

I tried to improve voltage/FSB manipulation module for EeePC 
(http://code.google.com/p/eeepc-linux) - and I found that 
i2c_smbus_write_block_data will cause kernel panic when is called from 
function called by timer. I can't see kernel panic message because 
system print very long call trace. Also on EeePC there is no RS232 port 
- but I have USB->RS232 dongle, can it be used as terminal for debug 
messages?

All variables are declared as static, all data is prepared correctly - 
so IMHO trouble is outside of my driver.
There is a part of code:

static struct i2c_client eee_pll_smbus_client = {
        .adapter = NULL,
        .addr = 0x69,
        .flags = 0,
};
static char eee_pll_data[I2C_SMBUS_BLOCK_MAX];
static int eee_pll_datalen = 0;
.......................
        i2c_smbus_write_block_data(&eee_pll_smbus_client,0, 
eee_pll_datalen, eee_pll_data);

Re: i2c_smbus_write_block_data vs scheduled routine calls (add_timer) - kernel crash

[Matthias Kaehlcke]

El Tue, Aug 04, 2009 at 09:42:41PM +0300 NiTr0 ha dit:

> I tried to improve voltage/FSB manipulation module for EeePC  
> (http://code.google.com/p/eeepc-linux) - and I found that  
> i2c_smbus_write_block_data will cause kernel panic when is called from  
> function called by timer. I can't see kernel panic message because  
> system print very long call trace. Also on EeePC there is no RS232 port  
> - but I have USB->RS232 dongle, can it be used as terminal for debug  
> messages?
>
> All variables are declared as static, all data is prepared correctly -  
> so IMHO trouble is outside of my driver.
> There is a part of code:
>
> static struct i2c_client eee_pll_smbus_client = {
>        .adapter = NULL,
>        .addr = 0x69,
>        .flags = 0,
> };
> static char eee_pll_data[I2C_SMBUS_BLOCK_MAX];
> static int eee_pll_datalen = 0;
> .......................
>        i2c_smbus_write_block_data(&eee_pll_smbus_client,0,  
> eee_pll_datalen, eee_pll_data);

a possible source of problems is that i2c_smbus_xfer() (invoked by
i2c_smbus_write_block_data()) tries to aquire a mutex if
adapter->algo->smbus_xfer is set.

http://lxr.linux.no/linux+v2.6.30/drivers/i2c/i2c-core.c#L1994

timers are executed in a softirq, ie in interrupt context, but mutexes
must not be used within interrupts.

i'm neither an i2c-core nor an kernel expert, but this point looks
conflictive to me and might be the reason for the kernel panic you
get.

-- 
Matthias Kaehlcke
Embedded Linux Engineer
Barcelona

        They that can give up essential liberty to obtain a little
           temporary safety deserve neither liberty nor safety
                         (Benjamin Franklin)
                                                                 .''`.
    using free software / Debian GNU/Linux | http://debian.org  : :'  :
                                                                `. `'`
gpg --keyserver pgp.mit.edu --recv-keys 47D8E5D4                  `-

Re: i2c_smbus_write_block_data vs scheduled routine calls (add_timer) - kernel crash

[Mark Brown]

On Tue, Aug 04, 2009 at 09:42:41PM +0300, NiTr0 wrote:

> All variables are declared as static, all data is prepared correctly
> - so IMHO trouble is outside of my driver.
> There is a part of code:

The problem is that your driver is trying to do I2C/SMBus I/O from a
timer, which is run from an interrupt context.  Since I2C buses are
generally very slow the controllers are interrupt driven and the read
and write operations on them will sleep waiting for operations to
complete.  If the code waited for the I/O to complete in an interrupt
context it would cause serious performance problems in the rest of the
system.

You should do the actual I/O from another context, such as a workqueue.
schedule_delayed_work() may be helpful here.