Moving target

Moving target

[Guillaume Dargaud]

Hello all,
I had a homebrew driver that compiled and ran well on 2.6.38 and I 
recently upgraded to the latest xilinx kernel and had some trouble 
recompiling without adjustments (of_driver/driver, 
of_platform_driver/platform_driver and some such).

But now, after those adjustments, it crashes on run.

My first generic question is: is there a clean canonical example of how 
to write each part (module init/exit, /dev handling, probing, physical 
memory access, interrupt handling, etc) of a basic driver ? Because each 
upgrade breaks my code and as driver writing is only one of my duties, I 
have a hard time keeping up with the moving target of the 
requirements... To say that books are obsolete is an understatement.

And now to the specifics:

[   16.655868] {xad_init 519} Module xad: loading...
[   16.661122] {xad_init 528} Module xad: Major=241, Minor=0, Count=1
[   16.666231] Unable to handle kernel paging request for data at 
address 0x00000030
[   16.673599] Faulting instruction address: 0xc0116fc0
[   16.678495] Oops: Kernel access of bad area, sig: 11 [#1]
[   16.683745] Xilinx Virtex
[   16.686337] Modules linked in: xad(+)
[   16.689978] NIP: c0116fc0 LR: c939a0f8 CTR: c0116fa8
[   16.694912] REGS: c789fd90 TRAP: 0300   Not tainted  (3.0.0-14.1-build3+)
[   16.701629] MSR: 00029030 <EE,ME,CE,IR,DR>  CR: 42000022  XER: 20000000
[   16.708199] DEAR: 00000030, ESR: 00000000
[   16.712185] TASK = c7851420[281] 'insmod' THREAD: c789e000
[   16.717441] GPR00: c939a0f8 c789fe40 c7851420 c9397b70 000046d7 
ffffffff c0110314 00000000
[   16.725736] GPR08: c0337380 00000000 00004000 c0116fa8 22000024 
100b99d4 ffffffff ffffffff
[   16.734030] GPR16: ffffffff ffffffff 100b5aed bfa54388 100844d8 
100844b8 000007a0 00000000
[   16.742324] GPR24: c0047edc 00000124 0000001c c9390000 c93973da 
c9397b08 c93971e8 c9397b70
[   16.750870] NIP [c0116fc0] driver_register+0x18/0x144
[   16.755901] LR [c939a0f8] xad_init+0xf8/0x1e8 [xad]
[   16.760638] Call Trace:
[   16.763119] [c789fe40] [c93973da] xad_device_id+0x1da/0xfffffef8 
[xad] (unreliable)
[   16.770719] [c789fe60] [c939a0f8] xad_init+0xf8/0x1e8 [xad]
[   16.776230] [c789fe80] [c0002508] do_one_initcall+0xe0/0x1c0
[   16.781871] [c789feb0] [c004a08c] sys_init_module+0x15cc/0x17c0
[   16.787720] [c789ff40] [c000c95c] ret_from_syscall+0x0/0x3c
[   16.793212] Instruction dump:
[   16.796155] 80630034 4bfcd669 80010014 38210010 7c0803a6 4e800020 
9421ffe0 7c0802a6
[   16.803842] bf410008 90010024 7c7f1b78 81230004
[   16.808257]  7c000034 5400d97e 0f000000
[   16.813362] ---[ end trace e40dbcf43f9d5d57 ]---

I don't understand the part about address 0x00000030. Does it mean I'm 
trying to wrongly access physical/virtual address 0x30 somewhere in my code?

Here are the very basics from my code (I've removed all printk, error 
checking and what the driver itself does for clarity). I really need a 
hand with the changes of of_something:

static dev_t first;
static unsigned int count = 1;
static int my_major = 241, my_minor = 0;

struct cdev *my_cdev=NULL;

typedef struct XadDevice {
//	struct platform_device *of_dev;
	struct device *dev;
	int irq;
	resource_size_t hw_phy_base, hw_length;	// Address of the IP resource
	void *hw_virt;							// Virtual of above
#ifdef DEBUG
	void *buffer_virt;						// Virtual address of the above
#endif
	struct resource *hw_region;				// Nothing is done with this
	...
} tXadDevice;
tXadDevice Xad;


static irqreturn_t XadIsr(int irq, void *dev_id) { ... }

static int xad_driver_probe(struct device * dev)
	struct resource res;
	struct device_node *dn = dev->of_node;
	int rc=0;

	printk(KERN_INFO SD "Probing %s\n" FL, dn->full_name);

	rc = of_address_to_resource(dn, 0, &res);

	Xad.irq = irq_of_parse_and_map(dn, 0);
	rc=request_irq(Xad.irq, XadIsr, IRQF_TRIGGER_RISING  /* | 
IRQF_DISABLED*/ | IRQF_SHARED | IRQF_SAMPLE_RANDOM, "XadIsr", &Xad);

	init_waitqueue_head(&Xad.wait);

	Xad.hw_region = request_mem_region((unsigned int)Xad.hw_phy_base, 
(unsigned int)(Xad.hw_length),  "xps-acqui-data");

	Xad.hw_virt= ioremap( (unsigned int)Xad.hw_phy_base, (unsigned 
int)(Xad.hw_length) );

	dev_set_drvdata(Xad.dev, &Xad);
	return 0;
}


///////////////////////////////////////////////////////////////////////////////
// There should be something in:
// ll /sys/devices/plb.0/c9800000.xps-acqui-data
static const struct of_device_id xad_device_id[] = {
	{ .compatible     = "xlnx,xps-acqui-data-4.00.a" },	// Must match the DTS
	{}
};

MODULE_DEVICE_TABLE(of, xad_device_id);	// Is this necessary ?

static struct device_driver xad_driver = {
	.probe  = xad_driver_probe,
	.remove = xad_driver_remove,
	.name = "xad-driver",
	.of_match_table = xad_device_id,
};

...

static struct file_operations fops = {
	.owner   = THIS_MODULE,
	.read    = xad_read,
	.open    = xad_open,
	.release = xad_close,
	.unlocked_ioctl   = xad_ioctl,
};

static int __init xad_init(void) {
	first = MKDEV (my_major, my_minor);
	register_chrdev_region(first, count, DEVNAME);
	my_cdev = cdev_alloc ();
	
	cdev_init(my_cdev, &fops);
	rc=cdev_add (my_cdev, first, count);
	
	rc = driver_register(&xad_driver);

	printk(KERN_INFO SD "Driver ready!\n" FL);
	return 0;
}

module_init(xad_init);


Thanks for reading so far !!!
-- 
Guillaume Dargaud
http://www.gdargaud.net/

Re: Moving target

[Andreas Schwab]

Guillaume Dargaud <dargaud@lpsc.in2p3.fr> writes:

> I don't understand the part about address 0x00000030. Does it mean I'm
> trying to wrongly access physical/virtual address 0x30 somewhere in my
> code?

More likely a NULL pointer dereference.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Moving target

[Michael Ellerman]

On Thu, 2012-08-30 at 16:14 +0200, Guillaume Dargaud wrote:
> Hello all,

Hi Guillaume,


> I don't understand the part about address 0x00000030. Does it mean I'm 
> trying to wrongly access physical/virtual address 0x30 somewhere in my code?

> [   16.666231] Unable to handle kernel paging request for data at address 0x00000030

So yes some code is trying to access address 0x30, that is typically a
symptom of dereferencing a NULL pointer.

The 0x30 can be a clue in that it is probably accessing an element of a
structure that is 0x30 past the base of the struct.


> [   16.673599] Faulting instruction address: 0xc0116fc0

This can be useful if you have the vmlinux around you can look at the
exact instruction that caused the fault.


> [   16.678495] Oops: Kernel access of bad area, sig: 11 [#1]
> [   16.683745] Xilinx Virtex
> [   16.686337] Modules linked in: xad(+)
> [   16.689978] NIP: c0116fc0 LR: c939a0f8 CTR: c0116fa8
> [   16.694912] REGS: c789fd90 TRAP: 0300   Not tainted  (3.0.0-14.1-build3+)
> [   16.701629] MSR: 00029030 <EE,ME,CE,IR,DR>  CR: 42000022  XER: 20000000
> [   16.708199] DEAR: 00000030, ESR: 00000000
> [   16.712185] TASK = c7851420[281] 'insmod' THREAD: c789e000
> [   16.717441] GPR00: c939a0f8 c789fe40 c7851420 c9397b70 000046d7 ffffffff c0110314 00000000
> [   16.725736] GPR08: c0337380 00000000 00004000 c0116fa8 22000024 100b99d4 ffffffff ffffffff
> [   16.734030] GPR16: ffffffff ffffffff 100b5aed bfa54388 100844d8 100844b8 000007a0 00000000
> [   16.742324] GPR24: c0047edc 00000124 0000001c c9390000 c93973da c9397b08 c93971e8 c9397b70

> [   16.750870] NIP [c0116fc0] driver_register+0x18/0x144

That is the exact instruction that faulted. You can see it's not
actually in your code, it's in driver_register().


> ///////////////////////////////////////////////////////////////////////////////
> // There should be something in:
> // ll /sys/devices/plb.0/c9800000.xps-acqui-data
> static const struct of_device_id xad_device_id[] = {
> 	{ .compatible     = "xlnx,xps-acqui-data-4.00.a" },	// Must match the DTS
> 	{}
> };
> 
> MODULE_DEVICE_TABLE(of, xad_device_id);	// Is this necessary ?
> 
> static struct device_driver xad_driver = {
> 	.probe  = xad_driver_probe,
> 	.remove = xad_driver_remove,
> 	.name = "xad-driver",
> 	.of_match_table = xad_device_id,
> };

> static int __init xad_init(void) {
> 	first = MKDEV (my_major, my_minor);
> 	register_chrdev_region(first, count, DEVNAME);
> 	my_cdev = cdev_alloc ();
> 	
> 	cdev_init(my_cdev, &fops);
> 	rc=cdev_add (my_cdev, first, count);
> 	
> 	rc = driver_register(&xad_driver);
> 
> 	printk(KERN_INFO SD "Driver ready!\n" FL);
> 	return 0;
> }

I don't think you should be using driver_register().

I think instead you want this to be a platform_driver, and use
platform_driver_register() ?

If you grep for platform_driver_register() in arch/powerpc you will see
several examples.

cheers

Re: Moving target

[Guillaume Dargaud]

On 08/31/2012 01:30 AM, Michael Ellerman wrote:

> I don't think you should be using driver_register().
>
> I think instead you want this to be a platform_driver, and use
> platform_driver_register() ?

Thank you for that. The test engineer was on vacation so it took a while 
before I got feedback on whether it worked or not, and it seems so.
-- 
Guillaume Dargaud
http://www.gdargaud.net/Antarctica/