Two issues preventing ruffians from booting 2.4 (with bridges patch)
were found and fixed.
- rather trivial one: someone decided that interrupt 4
(irq 20 from builtin scsi) is 'bogus' ;-)
- another issue is way not trivial and cost Michal and me a
lot of sweat. Type 1 PCI configuration space accesses (and to
some degree type 0 accesses to the bridge itself) were subtly
broken. In other words, these accesses were failing randomly
causing all sorts of havoc. At some point I managed to break
them completely just by adding one extra mb()...
Copying the code from 2.2 solved the problem - but it's a kind
of magic.
Interesting, other pyxis machines do not seem to be so sensitive,
so I guess some design problems with ux164 motherboard - all this
looks pretty much like timing issues.
Many thanks to Michal for that painful and time consuming testing
he has done.
Ivan.
--- 2.4.0t11/arch/alpha/kernel/core_cia.c Fri Oct 27 21:55:01 2000
+++ linux/arch/alpha/kernel/core_cia.c Thu Nov 16 17:46:54 2000
@@ -119,6 +119,7 @@ conf_read(unsigned long addr, unsigned c
stat0 = *(vip)CIA_IOC_CIA_ERR;
*(vip)CIA_IOC_CIA_ERR = stat0;
mb();
+ *(vip)CIA_IOC_CIA_ERR; /* re-read to force write */
/* If Type1 access, must set CIA CFG. */
if (type1) {
@@ -128,6 +129,7 @@ conf_read(unsigned long addr, unsigned c
*(vip)CIA_IOC_CFG;
}
+ mb();
draina();
mcheck_expected(0) = 1;
mcheck_taken(0) = 0;
@@ -171,6 +173,7 @@ conf_write(unsigned long addr, unsigned
stat0 = *(vip)CIA_IOC_CIA_ERR;
*(vip)CIA_IOC_CIA_ERR = stat0;
mb();
+ *(vip)CIA_IOC_CIA_ERR; /* re-read to force write */
/* If Type1 access, must set CIA CFG. */
if (type1) {
@@ -180,6 +183,7 @@ conf_write(unsigned long addr, unsigned
*(vip)CIA_IOC_CFG;
}
+ mb();
draina();
mcheck_expected(0) = 1;
mcheck_taken(0) = 0;
@@ -188,7 +192,7 @@ conf_write(unsigned long addr, unsigned
/* Access configuration space. */
*(vip)addr = value;
mb();
- mb(); /* magic */
+ *(vip)addr; /* read back to force the write */
mcheck_expected(0) = 0;
mb();
@@ -606,7 +610,8 @@ do_init_arch(int is_pyxis)
*(vip)CIA_IOC_ERR_MASK = temp;
/* Clear all currently pending errors. */
- *(vip)CIA_IOC_CIA_ERR = 0;
+ temp = *(vip)CIA_IOC_CIA_ERR;
+ *(vip)CIA_IOC_CIA_ERR = temp;
/* Turn on mchecks. */
temp = *(vip)CIA_IOC_CIA_CTRL;
--- 2.4.0t11/arch/alpha/kernel/sys_ruffian.c Sun Nov 5 16:16:43 2000
+++ linux/arch/alpha/kernel/sys_ruffian.c Tue Nov 21 00:25:42 2000
@@ -56,9 +56,9 @@ ruffian_init_irq(void)
init_i8259a_irqs();
- /* Not interested in the bogus interrupts (0,3,4,6),
+ /* Not interested in the bogus interrupts (0,3,6),
NMI (1), HALT (2), flash (5), or 21142 (8). */
- init_pyxis_irqs(0x17f0000);
+ init_pyxis_irqs(0x16f0000);
common_init_isa_dma();
}
On Tue, Nov 21, 2000 at 06:46:09PM +0300, Ivan Kokshaysky wrote:
> Interesting, other pyxis machines do not seem to be so sensitive,
> so I guess some design problems with ux164 motherboard - all this
> looks pretty much like timing issues.
Wow. Thanks for following through on this.
r~
On Tue, Nov 21, 2000 at 10:47:20AM -0800, Richard Henderson wrote:
> On Tue, Nov 21, 2000 at 06:46:09PM +0300, Ivan Kokshaysky wrote:
> > Interesting, other pyxis machines do not seem to be so sensitive,
> > so I guess some design problems with ux164 motherboard - all this
> > looks pretty much like timing issues.
>
> Wow. Thanks for following through on this.
I can now confirm that I can boot using SCSI disks (the fact that
this was possible for a while into IDE was a life-saver here :-)
a Ruffian (pyxis) Alpha using 2.4.0-test11 kernel and two patches
posted by Ivan (bridges-2.4.0t11.gz and extra ruffian fixes).
Here are fragments from 'dmesg':
....
Booting on Ruffian using machine vector Ruffian from MILO
Command line: bootdevice=sda2 bootfile=/vml240o11.gz root=/dev/sda2
....
SCSI subsystem driver Revision: 1.00
sym53c8xx: at PCI bus 1, device 13, function 0
sym53c8xx: 53c875 detected
sym53c875-0: rev 0x3 on pci bus 1 device 13 function 0 irq 20
sym53c875-0: ID 7, Fast-20, Parity Checking
sym53c875-0: on-chip RAM at 0xa101000
sym53c875-0: restart (scsi reset).
sym53c875-0: Downloading SCSI SCRIPTS.
scsi0 : sym53c8xx - version 1.6b
Vendor: IBM Model: DDRS-39130D Rev: DC1B
Type: Direct-Access ANSI SCSI revision: 02
Vendor: TOSHIBA Model: CD-ROM XM-6201TA Rev: 1037
Type: CD-ROM ANSI SCSI revision: 02
Vendor: IBM Model: DDRS-34560D Rev: DC1B
Type: Direct-Access ANSI SCSI revision: 02
sym53c875-0-<2,0>: tagged command queue depth set to 8
sym53c875-0-<10,0>: tagged command queue depth set to 8
Detected scsi disk sda at scsi0, channel 0, id 2, lun 0
Detected scsi disk sdb at scsi0, channel 0, id 10, lun 0
....
VFS: Mounted root (ext2 filesystem) readonly.
Those who posted "me too" could you please test that this is not
only a fluke on my particular machine?
Thanks a bunch, Ivan. Also thanks are extended to Gerard Roudier who
provided a crucial hint in the moment when we appeared to be completly
stuck. :-)
Michal