2000-10-12 09:00:20

by Francois romieu

[permalink] [raw]
Subject: Re: ioremap of pci base addresses

The Thu, Oct 12, 2000 at 11:02:50AM +0530, [email protected] wrote :
> once i have got the pci_dev structure( by calling pci_find*), do i
> explicitly need to call ioremap for remapping mmio.
> i think pci_enable_device does this. correct me if i am wrong..

ioremap does not only remap mmio but returns a token that should be
used if you want to further access the mmapped area (readl(token) and
writel(some_data, token) for example).

Part of 'foo' code could look like this :

static struct pci_driver foo_driver = {
name: "foo",
id_table: foo_pci_tbl,
probe: foo_init_one,
remove: foo_remove_one,

static int __init foo_init_one (struct pci_dev *pdev,
struct pci_device_id *ent)
u32 token;

if (pci_enable_device(pdev))
goto err_out;

* Some area (non pci-configuration registers or other) one wants to
* mmap. Ask 'foo' manual for the description of the Base Address
* Register in foo's pci configuration space.
if (!request_mem_region(pci_resource_start(pdev, 0),
pci_resource_len(pdev, 0), "mmaped registers")) {
printk(KERN_ERR "foo: can't reserve MMIO region\n");
goto err_out;
token = = (unsigned long)ioremap(pci_resource_start(pdev, 0),
pci_resource_len(pdev, 0));

* The following will happen if one forgets to balance ioremap
* and iounmap at insmod/rmmod time for example.
if (!token) {
printk(KERN_ERR "foo: cannot remap MMIO region %lx @ %lx\n",
pci_resource_len(pdev, 0), pci_resource_start(pdev, 0));
goto err_out_free_mmio_region;

[more stuff that may fail and should go at least to label err_out_iounmap]

return 0;

iounmap ((void *)token);
release_mem_region(pci_resource_start(pdev, 0),
pci_resource_len(pdev, 0));
return -ENODEV;

static void foo_remove_one (struct pci_dev *pdev)
[some stuff]

iounmap((void *)some_safe_structure->token);

[kfree may appear...]

release_mem_region(pci_resource_start(pdev, 0),
pci_resource_len(pdev, 0));