Does there exists any 32bit-only PARISC SMP machines?

Commit Message

John David Anglin Sept. 4, 2013, 3:41 p.m. UTC

On 9/4/2013 10:28 AM, Domenico Andreoli wrote:
> BTW, this is the most detailed investigation I'm aware of:
>
> https://lkml.org/lkml/2008/4/20/251
I have the sense that MEM_RENDEZ_HI may not be set correctly for your 
machine:

         /* Set the smp rendezvous address into page zero.
         ** It would be safer to do this in init_smp_config() but
         ** it's just way easier to deal with here because
         ** of 64-bit function ptrs and the address is local to this file.
         */
         load32          PA(smp_slave_stext),%r10
         stw             %r10,0x10(%r0)  /* MEM_RENDEZ */
         stw             %r0,0x28(%r0)   /* MEM_RENDEZ_HI - assume addr 
< 4GB */

This is possibly due to a change in memory range detection:

Initialized PDC Console for debugging.
Determining PDC firmware type: System Map.
model 00005d10 00000491 00000000 00000002 77b406fc 100000f0 00000008 
000000b2 000000b2
@@ -8,41 +9,28 @@
CPUID vers 17 rev 10 (0x0000022a)
capabilities 0x3
model 9000/785/J5600
-Memory Ranges:
- 0) Start 0x0000000000000000 End 0x00000000efffffff Size 3840 MB
- 1) Start 0x00000010f0000000 End 0x00000010ffffffff Size 256 MB
-Total Memory: 4096 MB
-initrd: 7fa75000-7ffee77e
-initrd: reserving 3fa75000-3ffee77e (mem_max 100000000)
-On node 0 totalpages: 983040
- Normal zone: 13440 pages used for memmap
- Normal zone: 0 pages reserved
- Normal zone: 969600 pages, LIFO batch:31
-On node 1 totalpages: 65536
- Normal zone: 896 pages used for memmap
- Normal zone: 0 pages reserved
- Normal zone: 64640 pages, LIFO batch:15
+Total Memory: 3840 MB

Dave

Comments

John David Anglin Sept. 5, 2013, 4:24 p.m. UTC | #1

On 9/4/2013 11:41 AM, John David Anglin wrote:
> On 9/4/2013 10:28 AM, Domenico Andreoli wrote:
>> BTW, this is the most detailed investigation I'm aware of:
>>
>> https://lkml.org/lkml/2008/4/20/251
> I have the sense that MEM_RENDEZ_HI may not be set correctly for your 
> machine:
>
>         /* Set the smp rendezvous address into page zero.
>         ** It would be safer to do this in init_smp_config() but
>         ** it's just way easier to deal with here because
>         ** of 64-bit function ptrs and the address is local to this file.
>         */
>         load32          PA(smp_slave_stext),%r10
>         stw             %r10,0x10(%r0)  /* MEM_RENDEZ */
>         stw             %r0,0x28(%r0)   /* MEM_RENDEZ_HI - assume addr 
> < 4GB */
>
> This is possibly due to a change in memory range detection:
>
> --- dmesg.txt 2008-04-21 11:19:16.000000000 +0200
> +++ dmesg2.txt 2008-04-21 15:27:46.000000000 +0200
> @@ -1,6 +1,7 @@
> -Linux version 2.6.22-3-parisc64-smp (Debian 2.6.22-6) 
> ([e-mail]) (gcc version 4.1.3 20071019 (prerelease) (Debian 
> 4.1.2-17)) #2 SMP Mon Nov 12 21:04:20 CET 2007
> +Linux version 2.6.25 (cavok@ska) (gcc version 4.1.3 20080308 
> (prerelease) (Debian 4.1.2-21)) #8 SMP Mon Apr 21 12:25:50 CES8
Kyle more or less confirmed that smp_slave_stext is not being call.  The 
firmware documentation
indicates that this might occur if memory isn't "properly" initialized.  
It would be helpful if you did a regression search
for the change that broke the cpu release.

Dave

Patch

--- dmesg.txt 2008-04-21 11:19:16.000000000 +0200
+++ dmesg2.txt 2008-04-21 15:27:46.000000000 +0200
@@ -1,6 +1,7 @@ 
-Linux version 2.6.22-3-parisc64-smp (Debian 2.6.22-6) ([e-mail]) 
(gcc version 4.1.3 20071019 (prerelease) (Debian 4.1.2-17)) #2 SMP Mon 
Nov 12 21:04:20 CET 2007
+Linux version 2.6.25 (cavok@ska) (gcc version 4.1.3 20080308 
(prerelease) (Debian 4.1.2-21)) #8 SMP Mon Apr 21 12:25:50 CES8
FP[0] enabled: Rev 1 Model 16
The 64-bit Kernel has started...
+console [ttyB0] enabled