From: Mikael Pettersson <mikpe@xxxxxxxx> Date: Thu, 14 Jun 2007 10:13:46 +0200 > Same failure here with xorg-6.8.1 and 2.6.22-rc4 + this patch, unfortunately. > > I've made an Xorg strace log for this kernel, and regular Xorg log files > for 2.6.21 and this kernel, available in: > > http://user.it.uu.se/~mikpe/linux/to-davem/xorg-log-2.6.22-rc4-davem1 > http://user.it.uu.se/~mikpe/linux/to-davem/Xorg.0.log-2.6.21 > http://user.it.uu.se/~mikpe/linux/to-davem/Xorg.0.log-2.6.22-rc4-davem1 I can see what's going wrong but not why :-) X is trying to read the PCI_CLASS_REVISION, PCI_REVISION_ID, etc. for the PCI controller and that's returning '0' in the byte which should be 0x06 to indicate that it's a PCI host controller. In fact, in the trace, each final byte is messed up somehow for 4-byte reads. I wrote a test program and I cannot reproduce this weird behavior on my workstation (SB1500) and on every single system these PCI config space accesses go through: arch/sparc64/kernel/pci.c:pci_host_bridge_read_pci_cfg() which returns constant data back, independant of the actual PCI controller type underneath. Puzzling eh? :-) I'm going to fire back up my ultra5 and see if I can reproduce this there in order to further debug. What we see in the X server strace log is: open("/proc/bus/pci/0000:00/00.0", O_RDWR) = 5 lseek(5, 0, SEEK_SET) = 0 read(5, "\216\20\0 ", 4) = 4 lseek(5, 8, SEEK_SET) = 8 read(5, "\0\0\0\0", 4) = 4 That second read is the problem, from a good trace it is: open("/proc/bus/pci/0000:00/00.0", O_RDWR) = 5 lseek(5, 0, SEEK_SET) = 0 read(5, "\216\20\0\240", 4) = 4 lseek(5, 8, SEEK_SET) = 8 read(5, "\0\0\0\6", 4) = 4 Note that "\0\0\0\6" at offset 0x8 as I mentioned above. I wrote a test program which tries to do exactly that sequence: #include <sys/types.h> #include <sys/stat.h> #include <fcntl.h> #include <unistd.h> #include <stdio.h> int main(void) { unsigned int x0, x8; int fd; fd = open("/proc/bus/pci/0000:00/00.0", O_RDONLY); lseek(fd, 0, SEEK_SET); read(fd, &x0, 4); lseek(fd, 8, SEEK_SET); read(fd, &x8, 4); close(fd); printf("x0:0x%08x\n", x0); printf("x8:0x%08x\n", x8); return 0; } But that works properly on my SB1500, so I'll try this test program next on my ultra5. It should print: x0:0x8e100080 x8:0x00000006 and an strace of that test program shows good reads: open("/proc/bus/pci/0000:00/00.0", O_RDONLY) = 3 lseek(3, 0, SEEK_SET) = 0 read(3, "\216\20\0\200", 4) = 4 lseek(3, 8, SEEK_SET) = 8 read(3, "\0\0\0\6", 4) = 4 close(3) = 0 - To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
