Skip to contentSkip to conversation

The new PCI fixup code ate my IDE controller

DISCUSSION Matthew Wilcox 2 participants 4 messages Sep 22, 2004

lore
Share & tools
lore
#0 lore raw

The new DECLARE_PCI_FIXUP_HEADER() now makes the order of fixups depend on link order. This is a bad thing because it's now one more thing that link order is used to determine. Eventually, we're going to have a knot where you can't move link order to make everything work.

The specific problem here is that quirk_ide_bases() is called before superio_fixup_pci(). This is only a problem on PA-RISC. I can see a few solutions to this. First, we can move the whole suckyio driver to arch/parisc, ensuring it gets linked before drivers/pci. Second, just move the DECLARE_PCI_FIXUP_HEADER to arch/parisc. Third, link drivers/parisc ahead of drivers/pci (what other fun ordering problems might we have?) Fourth, move the IDE quirk from drivers/pci/quirks.c to drivers/ide somewhere (which is linked after drivers/parisc so would happen to fix our problem). Fifth, back out the quirk changes. Sixth, change the pci fixup stuff to sort the quirks by vendor ID (PCI_ANY_ID sorts after any other ID).

Suggestions?

Signature · 6 lines
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain
#1 lore raw
Quoted 17 lines · Matthew Wilcox

The new DECLARE_PCI_FIXUP_HEADER() now makes the order of fixups depend on link order. This is a bad thing because it's now one more thing that link order is used to determine. Eventually, we're going to have a knot where you can't move link order to make everything work.

The specific problem here is that quirk_ide_bases() is called before superio_fixup_pci(). This is only a problem on PA-RISC. I can see a few solutions to this. First, we can move the whole suckyio driver to arch/parisc, ensuring it gets linked before drivers/pci. Second, just move the DECLARE_PCI_FIXUP_HEADER to arch/parisc. Third, link drivers/parisc ahead of drivers/pci (what other fun ordering problems might we have?) Fourth, move the IDE quirk from drivers/pci/quirks.c to drivers/ide somewhere (which is linked after drivers/parisc so would happen to fix our problem). Fifth, back out the quirk changes. Sixth, change the pci fixup stuff to sort the quirks by vendor ID (PCI_ANY_ID sorts after any other ID).

Quoted 1 line

Suggestions?

Hmm. We already have two passes through the fixup stuff. Add a third?

But sorting PCI_ANY_ID to go last seems like a reasonable first stab at an answer, if that's sufficient for your purposes.

Signature · 1 line
dwmw2
#2 lore raw
Quoted 3 lines · David Woodhouse

Hmm. We already have two passes through the fixup stuff. Add a third? But sorting PCI_ANY_ID to go last seems like a reasonable first stab at an answer, if that's sufficient for your purposes.

I don't think we need to go that far. Something like the following should work ... (untested, uncompiled, whitespace-damaged)

Index: drivers/pci/quirks.c
===================================================================
RCS file: /var/cvs/linux-2.6/drivers/pci/quirks.c,v retrieving revision 1.16

pci/quirks.c 22 Sep 2004 18:37:53 -0000 renamed from pci/quirks.c 13 Sep 2004 15:23:21 -0000 1.16 +1 -1
diff -u -p -r1.16 quirks.c
@@ -661,7 +661,7 @@ static void __devinit quirk_ide_bases(st
        printk(KERN_INFO "PCI: Ignoring BAR%d-%d of IDE controller %s\n",
               first_bar, last_bar, pci_name(dev));
 }
-DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID,             PCI_ANY_ID,                   
  quirk_ide_bases );
+DECLARE_PCI_FIXUP_HEADER_ALL(quirk_ide_bases);
 
 /*

* Ensure C0 rev restreaming is off. This is normally done by
Index: include/linux/pci.h
===================================================================
RCS file: /var/cvs/linux-2.6/include/linux/pci.h,v retrieving revision 1.18

linux/pci.h 22 Sep 2004 18:37:54 -0000 renamed from linux/pci.h 13 Sep 2004 15:24:12 -0000 1.18 +2 -0
diff -u -p -r1.18 pci.h
@@ -1016,6 +1016,11 @@ enum pci_fixup_pass {
        __attribute__((__section__(".pci_fixup_header"))) = {                   
        \
                vendor, device, hook };
 
+#define DECLARE_PCI_FIXUP_HEADER_ALL(hook) \
+       static struct pci_fixup __pci_fixup_PCI_ANY_IDPCI_ANY_ID##hook __attribu

te_used__ \
+ __attribute__((__section__(".pci_fixup_header_all"))) = {
+ PCI_ANY_ID, PCI_ANY_ID, hook };
+
#define DECLARE_PCI_FIXUP_FINAL(vendor, device, hook)
\
static struct pci_fixup __pci_fixup_##vendor##device##hook __attribute_u

sed__   \
        __attribute__((__section__(".pci_fixup_final"))) = {                    
        \

Index: include/asm-generic/vmlinux.lds.h
===================================================================
RCS file: /var/cvs/linux-2.6/include/asm-generic/vmlinux.lds.h,v retrieving revision 1.4

asm-generic/vmlinux.lds.h 22 Sep 2004 18:37:54 -0000 renamed from asm-generic/vmlinux.lds.h 13 Sep 2004 15:24:02 -0000 1.4 +1 -0
diff -u -p -r1.4 vmlinux.lds.h
@@ -20,6 +20,7 @@
        .pci_fixup        : AT(ADDR(.pci_fixup) - LOAD_OFFSET) {        \
                VMLINUX_SYMBOL(__start_pci_fixups_header) = .;          \
                *(.pci_fixup_header)                                    \
+               *(.pci_fixup_header_all)                                \
                VMLINUX_SYMBOL(__end_pci_fixups_header) = .;            \
                VMLINUX_SYMBOL(__start_pci_fixups_final) = .;           \
                *(.pci_fixup_final)                                     \
Signature · 6 lines
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain
#3 lore raw
Quoted 7 lines · Matthew Wilcox
Quoted 3 lines · David Woodhouse

Hmm. We already have two passes through the fixup stuff. Add a third? But sorting PCI_ANY_ID to go last seems like a reasonable first stab at an answer, if that's sufficient for your purposes.

I don't think we need to go that far. Something like the following should work ... (untested, uncompiled, whitespace-damaged)

Looks good to me.

Signature · 1 line
dwmw2