Skip to contentSkip to conversation

Missing code in bootmem.c for discontigmem

DISCUSSION Xavier Bru 2 participants 3 messages Nov 27, 2003 – Dec 1, 2003

lore
Share & tools
lore

Xavier Bru

author
#0 lore raw

Hello all,

Booting 2.6.0-test10 on Numa system, we fall in:

    Kernel panic: map_single: bounce buffer is not DMA'ble
due to alloc_bootmem_low_pages() returning a highmem address at sw IO TLB
initialization:
    Placing software IO TLB between 0xe0000030014e4000 - 0xe0000030016e4000

This was OK in test9 due to init_bootmem_core() sorting the pgdat_list. Restoring the previous code, system boots OK (using the last scsi patches :-)
Herafter the patch:

diff --exclude-from /home2/xb/proc/diff.exclude -Nur 0t10/mm/bootmem.c linux-2.6.0-test10.old/mm/bootmem.c
--- 0t10/mm/bootmem.c 2003-11-24 02:31:28.000000000 +0100
+++ linux-2.6.0-test10.old/mm/bootmem.c 2003-11-27 15:19:53.000000000
+0100
@@ -48,8 +48,24 @@

     bootmem_data_t *bdata = pgdat->bdata;
     unsigned long mapsize = ((end - start)+7)/8;

- pgdat->pgdat_next = pgdat_list;
- pgdat_list = pgdat;
+
+ /*

+     * sort pgdat_list so that the lowest one comes first,
+     * which makes alloc_bootmem_low_pages work as desired.
+     */

+ if (!pgdat_list || pgdat_list->node_start_pfn >
pgdat->node_start_pfn) {
+ pgdat->pgdat_next = pgdat_list;
+ pgdat_list = pgdat;
+ } else {
+ pg_data_t *tmp = pgdat_list;
+ while (tmp->pgdat_next) {
+ if (tmp->pgdat_next->node_start_pfn > pgdat->node_start_pfn)
+ break;
+ tmp = tmp->pgdat_next;

+        }
+        pgdat->pgdat_next = tmp->pgdat_next;
+        tmp->pgdat_next = pgdat;

+ }

     mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
     bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
Signature · 8 lines
 Sincères salutations.
_____________________________________________________________________
 
Xavier BRU                 BULL ISD/R&D/INTEL office:     FREC B1-422
tel : +33 (0)4 76 29 77 45                    http://www-frec.bull.fr
fax : +33 (0)4 76 29 77 70                 mailto:[email protected]
addr: BULL, 1 rue de Provence, BP 208, 38432 Echirolles Cedex, FRANCE
_____________________________________________________________________
#1 lore raw

Yeah, sorry about that, I knew there would be a problem. We need to fix the ia64 discontig code to init the boot memory in the correct order instead since the below change breaks other discontig platforms.

Jesse

Quoted 62 lines · Xavier Bru

Hello all,

Booting 2.6.0-test10 on Numa system, we fall in:

   Kernel panic: map_single: bounce buffer is not DMA'ble

due to alloc_bootmem_low_pages() returning a highmem address at sw IO TLB initialization:
Placing software IO TLB between 0xe0000030014e4000 - 0xe0000030016e4000
This was OK in test9 due to init_bootmem_core() sorting the pgdat_list. Restoring the previous code, system boots OK (using the last scsi patches :-)
Herafter the patch:

diff --exclude-from /home2/xb/proc/diff.exclude -Nur 0t10/mm/bootmem.c linux-2.6.0-test10.old/mm/bootmem.c

mm/bootmem.c 2003-11-27 15:19:53.000000000 renamed from mm/bootmem.c 2003-11-24 02:31:28.000000000 +0100 +19 -2
+0100
@@ -48,8 +48,24 @@
    bootmem_data_t *bdata = pgdat->bdata;
    unsigned long mapsize = ((end - start)+7)/8;
 
-    pgdat->pgdat_next = pgdat_list;
-    pgdat_list = pgdat;
+
+    /*
+     * sort pgdat_list so that the lowest one comes first,
+     * which makes alloc_bootmem_low_pages work as desired.
+     */
+    if (!pgdat_list || pgdat_list->node_start_pfn > 
 pgdat->node_start_pfn) {
+        pgdat->pgdat_next = pgdat_list;
+        pgdat_list = pgdat;
+    } else {
+        pg_data_t *tmp = pgdat_list;
+        while (tmp->pgdat_next) {
+            if (tmp->pgdat_next->node_start_pfn > pgdat->node_start_pfn)
+                break;
+            tmp = tmp->pgdat_next;
+        }
+        pgdat->pgdat_next = tmp->pgdat_next;
+        tmp->pgdat_next = pgdat;
+    }
 
    mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
    bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
Signature · 14 lines
Sinc?res salutations.
_____________________________________________________________________

Xavier BRU                 BULL ISD/R&D/INTEL office:     FREC B1-422
tel : +33 (0)4 76 29 77 45                    http://www-frec.bull.fr
fax : +33 (0)4 76 29 77 70                 mailto:[email protected]
addr: BULL, 1 rue de Provence, BP 208, 38432 Echirolles Cedex, FRANCE
_____________________________________________________________________


-
To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#2 lore raw
In reply to Xavier Bru
Quoted 3 lines · Jesse Barnes

Yeah, sorry about that, I knew there would be a problem. We need to fix the ia64 discontig code to init the boot memory in the correct order instead since the below change breaks other discontig platforms.

It looks like we should break out the second half of discontig.c:find_pernode_space() into a second function that inits the bootmaps in reverse order. As an added bonus, we could check to make sure that all of the pernode spaces actually got allocated and print a warning and/or disable the node if not. That would make debugging weird init time failures a bit easier (I think Alex ran into a problem that would have been quickly solved if we had such checks). Wanna take a stab at it?

Thanks,
Jesse