Resolves: RHEL-216318, Memory remove operation failed with continuous traces
This commit is contained in:
Than Ngo
parent
20983f36bf
commit
61df47fda3
@ -0,0 +1,68 @@
|
||||
commit 8e0a9410baba62f16e9b9ac1a7096dc6c704ffa1
|
||||
Author: Haren Myneni <haren@linux.ibm.com>
|
||||
Date: Mon Feb 9 22:18:59 2026 -0800
|
||||
|
||||
drmgr: Remove only available LMBs from CPU less NUMA node
|
||||
|
||||
The current logic in remove_cpuless_lmbs() may remove more LMB
|
||||
than requested.
|
||||
|
||||
todo = (count * node->ratio) / 100; --> can be 0
|
||||
todo = min(todo, node->n_lmbs);
|
||||
if (!todo && node->n_lmbs)
|
||||
todo = (count - this_loop); -> todo may be > node->n_lmbs
|
||||
|
||||
Removing more lmbs than requested caused a decrement of count
|
||||
(unsigned int) integer overflow, eventually leading to the drmgr
|
||||
removing all available memory in the system. Existing testing
|
||||
exposed the out of memory issue triggered by this counter
|
||||
overflow.
|
||||
|
||||
This patch fixed this logic by adjusting todo per node dynamically
|
||||
based on the count value and then determine todo value based on
|
||||
number of available LMBs in the node.
|
||||
|
||||
Signed-off-by: Ryan Whittaker <ryancwmd@linux.ibm.com>
|
||||
Reviewed-by: Mingming Cao <mmc@linux.ibm.com>
|
||||
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
|
||||
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
|
||||
|
||||
diff --git a/src/drmgr/drslot_chrp_mem.c b/src/drmgr/drslot_chrp_mem.c
|
||||
index 179fb17..4a36c73 100644
|
||||
--- a/src/drmgr/drslot_chrp_mem.c
|
||||
+++ b/src/drmgr/drslot_chrp_mem.c
|
||||
@@ -1565,11 +1565,16 @@ static int remove_cpuless_lmbs(uint32_t count)
|
||||
continue;
|
||||
|
||||
todo = (count * node->ratio) / 100;
|
||||
- todo = min(todo, node->n_lmbs);
|
||||
- /* Fix rounded value to 0 */
|
||||
- if (!todo && node->n_lmbs)
|
||||
+ /*
|
||||
+ * Fix rounded value to 0 and fix if a 0 ratio has
|
||||
+ * been processed
|
||||
+ */
|
||||
+ if ((!todo && node->n_lmbs) || (count - this_loop < todo))
|
||||
todo = (count - this_loop);
|
||||
|
||||
+ /* Donot request more than available */
|
||||
+ todo = min(todo, node->n_lmbs);
|
||||
+
|
||||
if (todo)
|
||||
todo = remove_lmb_from_node(node, todo);
|
||||
|
||||
@@ -1583,7 +1588,13 @@ static int remove_cpuless_lmbs(uint32_t count)
|
||||
if (!this_loop)
|
||||
break;
|
||||
|
||||
- count -= this_loop;
|
||||
+ /*
|
||||
+ * Should not happen, but in case prevent integer overflow
|
||||
+ */
|
||||
+ if (this_loop < count)
|
||||
+ count -= this_loop;
|
||||
+ else
|
||||
+ count = 0;
|
||||
}
|
||||
|
||||
say(DEBUG, "%d / %d LMBs removed from the CPU less nodes\n",
|
||||
@ -0,0 +1,31 @@
|
||||
commit f74288cf94d73f58a338fc58d8e2907673bb36e5
|
||||
Author: Haren Myneni <haren@linux.ibm.com>
|
||||
Date: Mon Feb 9 22:18:58 2026 -0800
|
||||
|
||||
drmgr: Update cpuless_lmb_count NUMA counter during LMB removal
|
||||
|
||||
After removing LMBs from a NUMA node, LMB counters should be
|
||||
decremented. In the case of CPU less node, cpuless_lmb_count
|
||||
should be decremented. But the current code reduces
|
||||
cpuless_node_count instead of cpuless_lmb_count. This patch
|
||||
reduces cpuless_lmb_count NUMA counter based on number of LMBs
|
||||
selected from that specific CPU less NUMA node.
|
||||
|
||||
Signed-off-by: Ryan Whittaker <ryancwmd@linux.ibm.com>
|
||||
Reviewed-by: Mingming Cao <mmc@linux.ibm.com>
|
||||
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
|
||||
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
|
||||
|
||||
diff --git a/src/drmgr/drslot_chrp_mem.c b/src/drmgr/drslot_chrp_mem.c
|
||||
index d37ee80..179fb17 100644
|
||||
--- a/src/drmgr/drslot_chrp_mem.c
|
||||
+++ b/src/drmgr/drslot_chrp_mem.c
|
||||
@@ -1502,7 +1502,7 @@ static int remove_lmb_from_node(struct ppcnuma_node *node, uint32_t count)
|
||||
if (node->n_cpus)
|
||||
numa.lmb_count -= unlinked;
|
||||
else
|
||||
- numa.cpuless_node_count -= unlinked;
|
||||
+ numa.cpuless_lmb_count -= unlinked;
|
||||
|
||||
if (!node->n_lmbs) {
|
||||
node->ratio = 0; /* for sanity only */
|
||||
@ -1,6 +1,6 @@
|
||||
Name: powerpc-utils
|
||||
Version: 1.3.10
|
||||
Release: 9%{?dist}
|
||||
Release: 9%{?dist}.1
|
||||
Summary: PERL-based scripts for maintaining and servicing PowerPC systems
|
||||
|
||||
Group: System Environment/Base
|
||||
@ -43,6 +43,12 @@ Patch15: powerpc-utils-cpu_info_helpers-add-helper-function-to-retrieve-present-
|
||||
# ppc64_cpu: Fix handling of non-contiguous CPU IDs
|
||||
Patch16: powerpc-utils-ppc64_cpu-fix-handling-of-non-contiguous-CPU-IDs.patch
|
||||
|
||||
# drmgr: Update cpuless_lmb_count NUMA counter during LMB removal
|
||||
Patch17: powerpc-utils-1.3.10-drmgr-Update-cpuless_lmb_count-NUMA-counter-during-LMB-removal.patch
|
||||
|
||||
# drmgr: Remove only available LMBs from CPU less NUMA node
|
||||
Patch18: powerpc-utils-1.3.10-drmgr-Remove-only-available-LMBs-from-CPU-less-NUMA-node.patch
|
||||
|
||||
ExclusiveArch: ppc %{power64}
|
||||
|
||||
BuildRequires: gcc
|
||||
@ -224,6 +230,9 @@ systemctl enable hcn-init.service >/dev/null 2>&1 || :
|
||||
%{_mandir}/man8/lparnumascore.8*
|
||||
|
||||
%changelog
|
||||
* Mon Aug 10 2026 Than Ngo <than@redhat.com> - 1.3.10-9.1
|
||||
- Resolves: RHEL-216318, Memory remove operation failed with continuous traces
|
||||
|
||||
* Wed Apr 23 2025 Than Ngo <than@redhat.com> - 1.3.10-9
|
||||
- Resolves: RHEL-86462, Fix handling of non contiguous CPU IDs
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user