Hi peter,
thanks for help.
Your reply is explaining how things should be done.
I am doing the similar steps to access the BARs of a card.
I will try to explain the problem in some more details:-
Start physical addr of BAR is allocated by BIOS.
In linux-2.6.9, whether BAR start phys addr is below OR above 4GB,
i.e. whether 32-bit BAR phys addr(all higher bits 0) OR 64-bit phys
addr(some higher bits non-zero), I am able to access the BAR region
using ioremap() & then readl()/writel().
For linux-2.6.22-8, if BAR start phys addr is below 4GB, things are
working fine here also.
But, for linux-2.6.22-8, when BAR start phys addr is above 4GB, I am
not able to read/write on the BAR region. ioremap() is not failing.
But read data is all 1's instead of actual data.
Any ideas on where to look for?
-Yogeshwar
On 10/3/07, Peter Teoh <htmldeveloper@xxxxxxxxx> wrote:
> On 10/2/07, yogeshwar sonawane <yogyas@xxxxxxxxx> wrote:
> > Hello all,
> >
> > For accessing memory-mapped 64bit-BAR regions of a PCI card, the
> > respective BAR regions has to be made accessible to the kernel using
> > ioremap() function.
> > Then readl()/writel() are working fine on the address returned by ioremap().
> > This is working in linux-2.6.9.
>
> Here is an extraction from 2.6.23-rc5/drivers/pci/msi.c that
> illustrate exactly what you have been saying (my remarks in ***):
>
> static int msix_capability_init(struct pci_dev *dev,
> struct msix_entry *entries, int nvec)
> {
> void __iomem *base;
>
> *** first pci_resource_start() is to get the physical address
> phys_addr = pci_resource_start (dev, bir) + table_offset;
> *** then ioremap() is applied to convert it to virtual address
> base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
>
> *** then assign to the structure
> entry->mask_base = base;
>
> And in the following function readl() and writel() are shown:
>
> static void msix_flush_writes(unsigned int irq)
> {
>
> int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
> PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
> ***Here below is shown how readl() is applied to the virtual address mask_base
> readl(entry->mask_base + offset);
>
> static int msi_free_irqs(struct pci_dev* dev)
> {
>
> list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
> if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
> *** Here writel() is applied to the virtual address mask_base.
> writel(1, entry->mask_base + entry->msi_attrib.entry_nr
> * PCI_MSIX_ENTRY_SIZE
> + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
>
> if (list_is_last(&entry->list, &dev->msi_list))
> iounmap(entry->mask_base);
> *** The mask_base is here unmap, since it has been ioremap() earlier.
> Beyond this point, the virtual address mask_base can therefore not
> be used to write anymore.
> }
> list_del(&entry->list);
> kfree(entry);
> }
>
--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ
