Skip to contentSkip to conversation

[4.17] x86/NUMA: correct off-by-1 in node map population

PATCH In review Jan Beulich 3 participants 4 messages Oct 4, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 [4.17] x86/NUMA: correct off-by-1 in node map population

As it turns out populate_memnodemap() so far "relied" on extract_lsb_from_nodes() setting memnodemapsize one too high in edge cases. Correct the issue there as well, by changing "epdx" to be an inclusive PDX and adjusting the respective relational operators.

While there also limit the scope of both related variables.

Fixes b1f4b45d02ca ("x86/NUMA: correct off-by-1 in node map size calculation")

---
For symmetry it might be helpful to change "epdx" to be inclusive in extract_lsb_from_nodes() as well. I actually had it that way first in the earlier change, but then thought the smaller diff would be better.

xen/arch/x86/numa.c +5 -5
@@ -65,15 +65,15 @@ int srat_disabled(void)
 static int __init populate_memnodemap(const struct node *nodes,
                                       int numnodes, int shift, nodeid_t *nodeids)
 {
-    unsigned long spdx, epdx;
     int i, res = -1;
 
     memset(memnodemap, NUMA_NO_NODE, memnodemapsize * sizeof(*memnodemap));
     for ( i = 0; i < numnodes; i++ )
     {
-        spdx = paddr_to_pdx(nodes[i].start);
-        epdx = paddr_to_pdx(nodes[i].end - 1) + 1;
-        if ( spdx >= epdx )
+        unsigned long spdx = paddr_to_pdx(nodes[i].start);
+        unsigned long epdx = paddr_to_pdx(nodes[i].end - 1);
+
+        if ( spdx > epdx )
             continue;
         if ( (epdx >> shift) >= memnodemapsize )
             return 0;
@@ -88,7 +88,7 @@ static int __init populate_memnodemap(co
                 memnodemap[spdx >> shift] = nodeids[i];
 
             spdx += (1UL << shift);
-        } while ( spdx < epdx );
+        } while ( spdx <= epdx );
         res = 1;
     }
 
Acked-by on v1 1/1
#1 lore raw
Quoted 10 lines · Jan Beulich

As it turns out populate_memnodemap() so far "relied" on extract_lsb_from_nodes() setting memnodemapsize one too high in edge cases. Correct the issue there as well, by changing "epdx" to be an inclusive PDX and adjusting the respective relational operators.

While there also limit the scope of both related variables.

Fixes b1f4b45d02ca ("x86/NUMA: correct off-by-1 in node map size calculation")

Thanks, Roger.

#2 lore raw
In reply to Jan Beulich

Hi Jan,

Quoted 9 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Subject: [PATCH][4.17] x86/NUMA: correct off-by-1 in node map population

As it turns out populate_memnodemap() so far "relied" on extract_lsb_from_nodes() setting memnodemapsize one too high in edge cases. Correct the issue there as well, by changing "epdx" to be an inclusive PDX and adjusting the respective relational operators.

Quoted 6 lines

While there also limit the scope of both related variables.

Fixes b1f4b45d02ca ("x86/NUMA: correct off-by-1 in node map size

calculation")

Thanks for the patch.

IIUC this patch is for a regression that we must fix for the 4.17, so:

Kind regards,
Henry

#3 lore raw
In reply to Jan Beulich
Quoted 10 lines · Jan Beulich

As it turns out populate_memnodemap() so far "relied" on extract_lsb_from_nodes() setting memnodemapsize one too high in edge cases. Correct the issue there as well, by changing "epdx" to be an inclusive PDX and adjusting the respective relational operators.

While there also limit the scope of both related variables.

Fixes b1f4b45d02ca ("x86/NUMA: correct off-by-1 in node map size calculation")

With this fix sabro boxes now report:

Oct  4 12:05:25.087462 (XEN) System RAM: 32429MB (33208204kB)
Oct  4 12:05:25.087482 (XEN) SRAT: Node 0 PXM 0 [0000000000000000, 000000007fffffff]
Oct  4 12:05:25.171468 (XEN) SRAT: Node 0 PXM 0 [0000000100000000, 000000047fffffff]
Oct  4 12:05:25.171489 (XEN) SRAT: Node 1 PXM 1 [0000000480000000, 000000087fffffff]
Oct  4 12:05:25.183432 (XEN) NUMA: Using 19 for the hash shift.
Oct  4 12:05:25.183453 (XEN) Domain heap initialised DMA width 32 bits

Thanks, Roger.