Date: Thu, 07 Nov 1996 11:35:09 -0700 From: "Mike Durian" <durian@plutotech.com> To: freebsd-hackers@freebsd.org Subject: small bugs in pci code Message-ID: <199611071835.LAA25115@pluto.plutotech.com>
next in thread | raw e-mail | index | archive | help
I've discovered three bugs in the pci code (at least I believe they
are bugs). These are in the 2.1.5 release. Since we've made some
local changes to these files, I don't have proper diffs, but the
changes are very small and isolated.
1) In pcireg.h, the definitions of PCI_PPB_IOBASE_EXTRACT and
PCI_PPB_IOLIMIT_EXTRACT should be:
#define PCI_PPB_IOBASE_EXTRACT(x) (((x) << 8) & 0xF000)
#define PCI_PPB_IOLIMIT_EXTRACT(x) (((x) << 0) & 0xF000)
Notice the mask has changed from 0xff00 to 0xf000. This is
because the lower nibble is not part of the address, but instead
determines if the addressing is 16 bit or 32 bit.
2) Related to #1, the mask that is OR'd with the PCI_PPB_IOLIMIT_EXTRACT
value should be 0xfff. It should not be calculated from the
amount of memory requested. I don't have a line number for this,
but it is in pci.c near the PCI_PPB_IOLIMIT_EXTRACT string, which
only occurs once. The PCI standard says, "the upper 4 bits of I/O
Limit register are writeable and correspond to address bits AD[15::12].
For purpose of address decoding the bridge assumes that the lower 12
address bits, AD[11::0], of the I/O limit address (not implemented
in the I/O Limit register) are FFFh."
3) In the same pci_bus_config() function, there is a section of for
loop code that begins with the comment, "Read the current mapping,
and update the pcicb fields." In this for loop, there is a switch
statement that handles the two types of mapping: I/O and memory.
The line:
switch (data & 7) {
should read:
switch (map & 7) {
The lower three bits that determine the mapping type are found in
the address value, not the size value.
mike
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199611071835.LAA25115>