In 2.4.17-pre5, agp_generic_agp_enable() and serverworks_agp_enable()
were changed to look for AGP devices as coprocessors on non-VGA
devices as well as VGA devices.
But those functions contain TWO loops that scan for AGP devices (one
to collect the settings they support, and a second to update the
command registers), and 2.4.17-pre5 only changed the first loop in
each case. Shouldn't both loops be changed, as in the "agp_dev" patch
attached?
While looking at this code, I noticed a possible further improvement.
Currently it looks like this:
pci_for_each_dev(device)
{
if((((device->class >> 16) & 0xFF) != PCI_BASE_CLASS_DISPLAY) &&
(device->class != (PCI_CLASS_PROCESSOR_CO << 8)))
continue;
pci_read_config_dword(device, 0x04, &scratch);
if (!(scratch & 0x00100000))
continue;
pci_read_config_byte(device, 0x34, &cap_ptr);
if (cap_ptr != 0x00) {
do {
pci_read_config_dword(device,
cap_ptr, &cap_id);
if ((cap_id & 0xff) != 0x02)
cap_ptr = (cap_id >> 8) & 0xff;
}
while (((cap_id & 0xff) != 0x02) && (cap_ptr != 0x00));
}
if (cap_ptr != 0x00) {
...
The bulk of the above is just a reimplementation of pci_find_capability(),
so what about changing the whole shooting match to the following:
pci_for_each_dev(device) {
cap_ptr = pci_find_capability(device, PCI_CAP_ID_AGP);
if (cap_ptr != 0x00) {
...
Is there anything gained by checking the device class before looking
for the AGP capability? The "agp_find_cap" patch attached implements
this idea. I've compiled and booted kernels with both patches, but
they're otherwise untested.
-- Bjorn Helgaas - bjorn_helgaas@hp.com Linux Systems Operation R&D Hewlett-Packard
This archive was generated by hypermail 2b29 : Wed Jan 23 2002 - 21:01:07 EST