kern/49181: i386 Kernel compilation error (warning -> error) if no PCIBIOSVERBOSE

[kre%munnari.OZ.AU@localhost]

>Number:         49181
>Category:       kern
>Synopsis:       i386 Kernel compilation error (warning -> error) if no 
>PCIBIOSVERBOSE
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Sep 07 23:35:00 +0000 2014
>Originator:     Robert Elz
>Release:        NetBSD 7_BETA (i386)
>Organization:
        Prince of Songkla University
>Environment:
System: NetBSD munnari.OZ.AU 6.99.30 NetBSD 6.99.30 (MUNNARI-DomU) #0: Mon Feb 
3 19:19:20 ICT 2014 
kre%onyx.coe.psu.ac.th@localhost:/usr/obj/current/kernels/amd64/MUNNARI-DomU 
amd64
Architecture: x86_64
Machine: amd64
>Description:
        If PCIBIOSVERBOSE is not set in the kernel config, then ...

/release/7/src/sys/arch/i386/pci/pci_intr_fixup.c: In function 'pciintr_do_heade
r_fixup':
/release/7/src/sys/arch/i386/pci/pci_intr_fixup.c:669:17: error: variable 'id' s
et but not used [-Werror=unused-but-set-variable]
  pcireg_t intr, id;
                   ^
cc1: all warnings being treated as errors

        That's because "id" is used only inside a section of code
        that is $ifdef PCIBIOSVERBOSE (though it is set unconditionally)>

>How-To-Repeat:
        Attempt to build a NetBSD 7_BETA kernel with a kernel config file
        that disables PCIBIOSVERBOSE   (the same happens on current,
        I might have reported that one before, but if so, it was obviously
        not fixed -- if I didn't report it, apologies).

>Fix:
        For my NetBSD 7 kernel sources, I just added this ludicrous piece of
        code (a stupid solution to a truly stupid problem -- who really cares
        if a variable is set but not used... at least if we know why that's
        being done and it isn't just an accident)

Index: pci_intr_fixup.c
===================================================================
RCS file: /cvsroot/NetBSD/src/sys/arch/i386/pci/pci_intr_fixup.c,v
retrieving revision 1.49
diff -u -r1.49 pci_intr_fixup.c
--- pci_intr_fixup.c    1 Jul 2011 17:37:26 -0000       1.49
+++ pci_intr_fixup.c    7 Sep 2014 23:24:07 -0000
@@ -719,6 +719,9 @@
                        PCIBIOS_PRINTV((" %3d", l->irq));
                PCIBIOS_PRINTV(("  %d   ", l->fixup_stage));
        }
+#else
+       if (id == (pcireg_t)7)
+               printf("Lucky 7!\n");
 #endif
        
        /*

        And no, there's nothing at all special about 7 (except that
        I do not expect id to ever have that value).

        In my current kernel sources, I handle it like this ...

Index: pci_intr_fixup.c
===================================================================
RCS file: /cvsroot/NetBSD/src/sys/arch/i386/pci/pci_intr_fixup.c,v
retrieving revision 1.49
diff -u -r1.49 pci_intr_fixup.c
--- pci_intr_fixup.c    1 Jul 2011 17:37:26 -0000       1.49
+++ pci_intr_fixup.c    19 Apr 2014 16:37:10 -0000
@@ -666,10 +666,15 @@
        struct pciintr_link_map *l;
        int pin, irq, link;
        int bus, device, function;
-       pcireg_t intr, id;
+       pcireg_t intr;
+#ifdef PCIBIOSVERBOSE
+       pcireg_t id;
+#endif
 
        pci_decompose_tag(pc, tag, &bus, &device, &function);
+#ifdef PCIBIOSVERBOSE
        id = pci_conf_read(pc, tag, PCI_ID_REG);
+#endif
 
        intr = pci_conf_read(pc, tag, PCI_INTERRUPT_REG);
        pin = PCI_INTERRUPT_PIN(intr);


        That's riskier, as it is no longer calling pci_conf_read(),
        which could potentially have some side effects (and perhaps
        only on some systems, so that I have not seen any ill effects
        from running kernels compiled that way doesn't mean no-one
        ever will.)