Skip to contentSkip to conversation

DISCUSSION Christine Ames 1 participant 1 message Feb 20, 2002

lore
Share & tools
lore
#0 lore raw
Quoted 3 lines · Jeff Garzik
Quoted 2 lines · David Stroupe

I am creating a PCI driver for a custom card and

want to write 0xffff to

Quoted 11 lines
Quoted 11 lines

a location offset from the base by 0x48 and
have the following code:

<snip>
unsigned int io_addr;
unsigned int io_size;
void* base;
pci_enable_device (pdev)
io_addr = pci_resource_start(pdev, 0);
io_size = pci_resource_len(pdev, 0);
if ((pci_resource_flags(pdev, 0) &

IORESOURCE_MEM)){

Quoted 2 lines
Quoted 2 lines

if(check_mem_region(io_addr, io_size))
DBG("Already In Use");//this is never

reached

Quoted 1 line
Quoted 1 line

request_mem_region(io_addr, io_size ,

"Card Driver");

Quoted 3 lines

In 2.4 and later, check request_mem_region return
value, and never call
check_mem_region.

Gasp! Are you sure we should never call check_mem_region, Jeff?

This check has worked for me under kernels 2.4.7 & 2.4.7-10. If check_mem_region returns < 0 my driver bails with a message to the effect that my chip's memory region is already in use...

Could this lead to screw-ups?
Should I lose the check?
(my pci initialization code follows)

Quoted 4 lines
Quoted 4 lines
        base=ioremap(io_addr, io_size);
        if(base==0)
           DBG("memory mapped wrong\n");//this is

never reached

Quoted 4 lines
Quoted 4 lines
        writew(0xffff, base + 0x48);

<snip>

The card is found, io_addr = 0xe9011000 and

io_size = 0x200.

Quoted 2 lines
Quoted 2 lines

The write is unsuccessful or at least the data

never reaches the card.

Quoted 6 lines
Quoted 1 line

What am I doing incorrectly?

Looks correct to me... maybe you need to do

	readw(base + 0x48)

to flush the transaction?

David, I realize this is a "no-brainer" but, if Jeff's

comments don't help double-check that your I/O space begins at PCI BAR 0 (as opposed to PCI BARs 1-5).

<snip>
// enable the device
if( pci_enable_device( pdev )) {
printk( KERN_ERR "init_pci_device:

pci_enable_device failed\n" );
    return 0;

}

// set dma mask for device
// since we handle 32-bit dma, I don't think this is necessary...
if( pci_set_dma_mask( pdev, DMA_MASK ) != 0 ) {
printk( KERN_ERR "init_pci_device:

pci_set_dma_mask failed\n");
    return 0;

}

// enable bus master bit in PCI_COMMAND register
pci_set_master( pdev );

  #ifdef DEBUG
    debug_display_pci_dev( pdev );
  #endif // DEBUG

// access I/O space
// registers begin at BAR 4
phys_addr = pdev->resource[ 4 ].start;

// make sure nobody else is using the same device
if( check_mem_region( phys_addr, REG_MAP_SIZE ) < 0
) {
printk( KERN_ERR "init_pci_device: memory region

already in use!\n" );
    return 0;

}

if( request_mem_region( phys_addr, REG_MAP_SIZE,
pdc_ptr->name ) == NULL ) {
printk( KERN_ERR "init_pci_device: memory region

not available!\n" );
    return 0;

}

// remap physical pci space into virtual memory space
pdc_ptr->regs = ioremap_nocache( phys_addr,
REG_MAP_SIZE );

// initialize pdev private driver data
pdev->driver_data = pdc_ptr;

return 1;
<snip>

--Christine

__________________________________________________
Do You Yahoo!?
Yahoo! Sports - Coverage of the 2002 Olympic Games
http://sports.yahoo.com