Kernel Panic during memcpy_toio to PCI card

Matt,

Thanks for the response, I am using a 2.4.18 kernel with a busybox. This is an embedded system with the file system laid out by an 'initrd.gz'. I am new to Linux. Can LKCD still be used in our case ?

Suneeta

-----Original Message-----
From: Matt D. Robinson [mailto:[email protected]]
Sent: Friday, November 01, 2002 3:26 PM
To: Donepudi, Suneeta; Matt D. Robinson
Subject: Re: Kernel Panic during memcpy_toio to PCI card

Hey, Suneeta. Can you try LKCD and see if you can get a crash dump with it? Also, is this 2.4 or 2.5?

--Matt

On Fri, 1 Nov 2002, Donepudi, Suneeta wrote:

|>Hi,
|>
|>I would like help in diagnosing a kernel panic while accessing a PCI

device.
|>
|>Everything runs fine for sometime and in about 1/2 hour I get a Kernel Panic

|>message saying :
|>
|>"Unable to handle kernel paging request at virtual address 0xc2821000"
|>
|>Analysis with Ksymoops shows that it is happening during a memcpy_toio()
|>with the PCI card. The PCI card uses three Base Address Registers with
|>virtual addresses mapped as follows (after ioremap has been issued):
|>
|>BAR0 = 0xc280f000
|>BAR1 = 0xc2811000
|>BAR2 = 0xc2822000
|>
|>It seems like the kernel panic is complaining about an address which is a
|>combination of BAR1 (lower bytes) and BAR2 (upper bytes). It should really
|>be accessing the BAR1 address at the point the crash occurred.
|>
|>I put the following if-statement just before the memcpy_toio():
|>-----------------------------------------------------------
|>if (((long int)pci_bar1) == 0xc2821000)
|>{
|>	printk (KERN_ERR "Illegal address for BAR1\n");
|>	return -1;
|>}
|>memcpy_toio (pci_bar1, in_ptr, len);
|>------------------------------------------------------------
|>
|>It still caused the crash in the same manner and at the same location.
|>Could someone help me with pointers to where I should start looking ?
|>Disabling interrupts around the memcpy_toio() did not make any
|>difference. Is this a hardware problem with the PCI card ? We are using
|>a Xilinx core with out FPGA build into it.
|>Is there a book I could read to learn more about debugging this in the 
|>Kernel ?
|>
|>Thanks a bunch,
|>Suneeta

Yes, grab the 4.1.1 stuff from lkcd.sourceforge.net. Let the lkcd-general list know if you're having problems, one of us should be able to help.

--Matt

On Fri, 1 Nov 2002, Donepudi, Suneeta wrote:

|>Matt,
|>
|>Thanks for the response, I am using a 2.4.18 kernel with
|>a busybox. This is an embedded system with the file system
|>laid out by an 'initrd.gz'. I am new to Linux. Can LKCD
|>still be used in our case ?
|>
|>
|>Suneeta
|>
|>-----Original Message-----
|>From: Matt D. Robinson [mailto:[email protected]]
|>Sent: Friday, November 01, 2002 3:26 PM
|>To: Donepudi, Suneeta; Matt D. Robinson
|>Subject: Re: Kernel Panic during memcpy_toio to PCI card
|>
|>
|>Hey, Suneeta.  Can you try LKCD and see if you can get
|>a crash dump with it?  Also, is this 2.4 or 2.5?
|>
|>--Matt
|>
|>On Fri, 1 Nov 2002, Donepudi, Suneeta wrote:
|>|>Hi,
|>|>
|>|>I would like help in diagnosing a kernel panic while accessing a PCI
|>device.
|>|>
|>|>Everything runs fine for sometime and in about 1/2 hour I get a Kernel
|>Panic
|>|>message saying :
|>|>
|>|>"Unable to handle kernel paging request at virtual address 0xc2821000"
|>|>
|>|>Analysis with Ksymoops shows that it is happening during a memcpy_toio()
|>|>with the PCI card. The PCI card uses three Base Address Registers with
|>|>virtual addresses mapped as follows (after ioremap has been issued):
|>|>
|>|>BAR0 = 0xc280f000
|>|>BAR1 = 0xc2811000
|>|>BAR2 = 0xc2822000
|>|>
|>|>It seems like the kernel panic is complaining about an address which is a
|>|>combination of BAR1 (lower bytes) and BAR2 (upper bytes). It should really
|>|>be accessing the BAR1 address at the point the crash occurred.
|>|>
|>|>I put the following if-statement just before the memcpy_toio():
|>|>-----------------------------------------------------------
|>|>if (((long int)pci_bar1) == 0xc2821000)
|>|>{
|>|>	printk (KERN_ERR "Illegal address for BAR1\n");
|>|>	return -1;
|>|>}
|>|>memcpy_toio (pci_bar1, in_ptr, len);
|>|>------------------------------------------------------------
|>|>
|>|>It still caused the crash in the same manner and at the same location.
|>|>Could someone help me with pointers to where I should start looking ?
|>|>Disabling interrupts around the memcpy_toio() did not make any
|>|>difference. Is this a hardware problem with the PCI card ? We are using
|>|>a Xilinx core with out FPGA build into it.
|>|>Is there a book I could read to learn more about debugging this in the 
|>|>Kernel ?
|>|>
|>|>Thanks a bunch,
|>|>Suneeta
|>-
|>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
|>the body of a message to [email protected]
|>More majordomo info at  http://vger.kernel.org/majordomo-info.html
|>Please read the FAQ at  http://www.tux.org/lkml/
|>

memcpy_toio() in 2.4 is still using memcpy() which could use prefetch() if you compile for certain processors. Prefetching io space could do all sorts of nasties.

2.5 changed this define (in include/asm-i386/io.h) to use __memcpy instead, which doesn't use prefetching.

Dave

Dave

| Dave Jones.        http://www.codemonkey.org.uk

If so that really wants fixing ASAP ie for 2.4.20