On Mon, 22 Jan 2007 21:00:04 +0100
Luca Tettamanti [off-list ref] wrote:

Il Sun, Jan 21, 2007 at 09:33:39PM -0600, Jay Cliburn ha scritto: 

quoted

Randy Dunlap wrote:

quoted

On Sun, 21 Jan 2007 15:07:37 -0600 Jay Cliburn wrote:

[snip]

quoted

quoted

+	value = ioread16(hw->hw_addr + REG_PCIE_CAP_LIST);
+	return ((value & 0xFF00) == 0x6C00) ? 0 : 1;

Are there defines or enums for these?
Fewer magic numbers would be nice/helpful/readable.

[snip]

quoted

quoted

+	s32 ret;
+	ret = atl1_write_phy_reg(hw, 29, 0x0029);

Fewer magic numbers?

Unfortunately, we don't have a spec.  This is how the vendor coded it.

[snip]

quoted

quoted

+
+int enable_msi;
+module_param(enable_msi, int, 0444);
+MODULE_PARM_DESC(enable_msi, "Enable PCI MSI");

Hm, I thought that we didn't want individual drivers having MSI config
options...

Luca?  This one was yours IIRC.  Care to chime in?

I remember that discussion, but since there's no sistem-wide MSI
blacklist (or whitelist) I don't think it's safe to enable it
unconditionally. Judging from bug reports on lkml it's not safe to
assume that MSI support is sane on non-Intel chipsets.

Luca

There is MSI blacklisting see drivers/pci/quirks.c code.
But the blacklist isn't complete enough yet.

IMHO the MSI disabling should be removed from drivers and be done
in the PCI core. But it doesn't seem to have gotten widespread
support.
	

-- 
Stephen Hemminger [off-list ref]