[PATCH v2 0/3] cgroup/cpuset: Fix some more bugs in cpuset partition code

[PATCH v2 0/3] cgroup/cpuset: Fix some more bugs in cpuset partition code

[Waiman Long]

 v2:
  - Add two fix patches for problems reported by sashiko and Guopeng.

This is another series of 3 fix patches for bugs in cpuset partition
code.

Waiman Long (3):
  cgroup/cpuset: Fix incorrect CPU accounting with parent's
    effective_xcpus expansion
  cgroup/cpuset: Recompute effective cpumask in update_cpumasks_hier()
    after calling update_parent_effective_cpumask()
  cgroup/cpuset: Do housekeeping check when isolated child with root
    parent may get more xcpus

 kernel/cgroup/cpuset.c                        | 32 ++++++++++++++++---
 .../selftests/cgroup/test_cpuset_prs.sh       | 25 +++++++++++++--
 2 files changed, 51 insertions(+), 6 deletions(-)

-- 
2.56.0

[PATCH v2 1/3] cgroup/cpuset: Fix incorrect CPU accounting with parent's effective_xcpus expansion

[Waiman Long]

Commit ec0254b9eb49 ("cgroup/cpuset: Release invalid isolated CPUs with
a shrunk root partition parent") fixes incorrect isolated_cpus accounting
when contraction of a parent's effective_xcpus can cause a child partition
to lose isolated CPUs. Expansion of parent's effective_cpus can also allow
a child partition to gain more effective_xcpus. However, the current code
doesn't handle this case correctly leading to incorrect effective_cpus
and isolated_cpus accounting. Fix that by looking for this possible
expansion of effective_xcpus and perform the proper accounting action.

New test cases are added to test this effective_xcpus expansion scenarios.

Reported-by: Guopeng Zhang <guopeng.zhang@linux.dev>
Link: https://lore.kernel.org/lkml/20261027123837.100909-1-guopeng.zhang@linux.dev
Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c                            | 12 ++++++++++++
 tools/testing/selftests/cgroup/test_cpuset_prs.sh | 15 +++++++++++++++
 2 files changed, 27 insertions(+)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 3219f927c35e..39c81a0a8ab9 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2097,6 +2097,7 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
 							     parent->effective_xcpus);
 			if (!cpumask_empty(tmp->addmask))
 				isolated_cpus_release(cs, tmp->addmask);
+			/* Expansion of parent's effective_xcpus is handled below */
 		}
 		if (nocpu) {
 			part_error = PERR_NOCPUS;
@@ -2106,11 +2107,22 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
 						     parent->effective_xcpus);
 		} else if ((parent_prs == PRS_ISOLATED) &&
 			   ((old_prs == PRS_ROOT) || (old_prs == -PRS_ROOT))) {
+			/* Parent partition state changed to PRS_ISOLATED */
 			part_error = PERR_ISOPARENT;
 			if (old_prs == PRS_ROOT) {
 				cpumask_copy(tmp->addmask, cs->effective_xcpus);
 				adding = true;
 			}
+		} else if (is_partition_valid(cs) &&
+			   !cpumask_equal(tmp->delmask, cs->effective_xcpus)) {
+			/*
+			 * If there is an expansion of parent's effective_xcpus,
+			 * we may take more xcpus from parent. The parent's
+			 * effective_cpus should still have active CPUs or nocpu
+			 * should have been set.
+			 */
+			deleting = cpumask_andnot(tmp->delmask, tmp->delmask,
+						  cs->effective_xcpus);
 		} else if (is_partition_invalid(cs)) {
 			struct cgroup_subsys_state *css;
 			struct cpuset *child;
diff --git a/tools/testing/selftests/cgroup/test_cpuset_prs.sh b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
index 83ef6c781af3..844efdf029cd 100755
--- a/tools/testing/selftests/cgroup/test_cpuset_prs.sh
+++ b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
@@ -531,6 +531,21 @@ TEST_MATRIX=(
 	" C1-4:P1 C3-4:P2   .      .    T:C4-6   .      .      .     0 A1:5-6|A2:4 A1:P1|A2:P2 4"
 	" C1-4:P1 C1-2:P2   .      .    T:C4-6   .      .      .     0 A1:4-6|A2:4-6 A1:P1|A2:P-2 ."
 
+	# The extra CPUs gained by a child partition when its parent expands the
+	# exclusive CPUs have to be taken away from the parent.
+	" C2-6:X2-5:P1 CX3-6:P1 \
+	                    .      .    X2-6     .      .      .     0 A1:2|A2:3-6|XA2:3-6 A1:P1|A2:P1 ."
+	" C2-6:X2-5:P1 CX3-6:P2 \
+	                    .      .    X2-6     .      .      .     0 A1:2|A2:3-6|XA2:3-6 A1:P1|A2:P2 3-6"
+
+	# Both expansion and contraction of parent's effective_xcpus will be
+	# correctly handled in child partition and the child can be invalidated
+	# if a populated parent could lose all its effective_cpus.
+	" C2-6:X2-5:P1 CX3,5-6:P2 \
+	                    .      .    X4-6     .      .      .     0 A1:4|A2:5-6|XA2:5-6 A1:P1|A2:P2 5-6"
+	" C2-6:X2-5:P1 CX3-6:P2 \
+	                    .      .   T:X4-6    .      .      .     0 A1:4-6|A2:4-6 A1:P1|A2:P-2 ."
+
 	#
 	#  old-A1 old-A2 old-A3 old-B1 new-A1 new-A2 new-A3 new-B1 fail ECPUs Pstate ISOLCPUS
 	#  ------ ------ ------ ------ ------ ------ ------ ------ ---- ----- ------ --------
-- 
2.56.0

[PATCH v2 2/3] cgroup/cpuset: Recompute effective cpumask in update_cpumasks_hier() after calling update_parent_effective_cpumask()

[Waiman Long]

It is reported by sashiko that calling update_parent_effective_cpumask()
may clobber tmp->new_cpus when update_sibling_cpumasks() needs to be
called due to possible change in the parent's effective_cpus value.
Therefore the new effective cpumask has to be recomputed after calling
update_parent_effective_cpumask() to get a correct value back into
tmp->new_cpus.

Fixes: f0af1bfc27b5 ("cgroup/cpuset: Relax constraints to partition & cpus changes")
Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 39c81a0a8ab9..ed014ecb770c 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2359,12 +2359,12 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
 			update_parent_effective_cpumask(cp, partcmd_update, NULL, tmp);
 			/*
 			 * Capture current partition_root_state as it may have
-			 * changed. Recompute effective cpumask if changed.
+			 * changed. Always recompute effective cpumask as tmp->new_cpus
+			 * may be clobbered when update_sibling_cpumasks() is called
+			 * within update_parent_effective_cpumask().
 			 */
 			new_prs = cp->partition_root_state;
-			if (new_prs != old_prs) {
-				compute_effective_cpumask(tmp->new_cpus, cp, parent);
-			}
+			compute_effective_cpumask(tmp->new_cpus, cp, parent);
 		}
 
 		spin_lock_irq(&callback_lock);
-- 
2.56.0

[PATCH v2 3/3] cgroup/cpuset: Do housekeeping check when isolated child with root parent may get more xcpus

[Waiman Long]

It is reported by sashiko and Guopeng that when an isolated child with
root parent gets more isolated CPUs from the expansion of the parent's
effective_xcpus, it is possible that all the active housekeeping CPUs
may be gone. So a prstate_housekeeping_conflict() should be performed
when that happens. Add the check to make sure that we can't exhaust
all the active housekeeping CPUs.

Test cases are added to the NOHZ_FULL_TEST_MATRIX of test_cpuset_prs.sh
to test that.

Reported-by: Guopeng Zhang <guopeng.zhang@linux.dev>
Link: https://lore.kernel.org/lkml/ed6fecb4-38e9-4a73-8c1a-7bdd80175a69@linux.dev
Fixes: e2ffe502ba45 ("cgroup/cpuset: Add cpuset.cpus.exclusive for v2")
Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c                         | 18 +++++++++++++++---
 .../selftests/cgroup/test_cpuset_prs.sh        | 10 ++++++++--
 2 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index ed014ecb770c..6e136888fc8e 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2119,10 +2119,22 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
 			 * If there is an expansion of parent's effective_xcpus,
 			 * we may take more xcpus from parent. The parent's
 			 * effective_cpus should still have active CPUs or nocpu
-			 * should have been set.
+			 * should have been set. We also need to do a
+			 * housekeeping check for an isolated child with root
+			 * parent to make sure that it won't exhaust all the
+			 * active housekeeping CPUs.
 			 */
-			deleting = cpumask_andnot(tmp->delmask, tmp->delmask,
-						  cs->effective_xcpus);
+			cpumask_andnot(tmp->addmask, cs->effective_xcpus, tmp->delmask);
+			if ((parent_prs == PRS_ROOT) && (old_prs == PRS_ISOLATED) &&
+			    prstate_housekeeping_conflict(old_prs, parent_prs,
+							  tmp->delmask, tmp->addmask)) {
+				adding = cpumask_and(tmp->addmask, cs->effective_xcpus,
+						     tmp->delmask);
+				part_error = PERR_HKEEPING;
+			} else {
+				deleting = cpumask_andnot(tmp->delmask, tmp->delmask,
+							  cs->effective_xcpus);
+			}
 		} else if (is_partition_invalid(cs)) {
 			struct cgroup_subsys_state *css;
 			struct cpuset *child;
diff --git a/tools/testing/selftests/cgroup/test_cpuset_prs.sh b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
index 844efdf029cd..bdca6c471f87 100755
--- a/tools/testing/selftests/cgroup/test_cpuset_prs.sh
+++ b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
@@ -137,7 +137,8 @@ LAST_PRESENT_CPU=$(cat /sys/devices/system/cpu/present | sed -e "s/.*[,-]//")
 # If isolated CPUs have been reserved at boot time (as shown in
 # /sys/devices/system/cpu/isolated), these isolated CPUs should be outside of
 # CPUs 0-8 that will be used by this script for testing purpose. If not, some
-# of the tests may fail incorrectly.
+# of the tests may fail incorrectly. The first boot isolated CPU should be
+# 10 or above.
 #
 # The current set of isolated CPUs (as shown in cpuset.cpus.isolated) should
 # be the same as the boot value. If not, wait a bit and retry again just in
@@ -156,7 +157,7 @@ TEST_ISOLCPUS=
 [[ -n "$BOOT_ISOLCPUS" ]] && {
 	FIRST_ISOLCPUS=$(echo $BOOT_ISOLCPUS | sed -e "s/[,-].*//")
 	echo "Boot isolated CPUs: $BOOT_ISOLCPUS"
-	[[ $FIRST_ISOLCPUS -le 8 ]] &&
+	[[ $FIRST_ISOLCPUS -le 9 ]] &&
 		skip_test "Boot isolated CPUs overlap CPUs to be tested"
 }
 
@@ -602,6 +603,7 @@ ISOLCPUS_TEST_MATRIX=(
 )
 
 # Test matrix to be run when boot time nohz_full CPUs cover all CPUs except CPUs 0-7
+# This also means that CPU 8 is available and is a nohz_full CPU.
 NOHZ_FULL_TEST_MATRIX=(
 	#  old-A1 old-A2 old-A3 old-B1 new-A1 new-A2 new-A3 new-B1 fail ECPUs Pstate ISOLCPUS
 	#  ------ ------ ------ ------ ------ ------ ------ ------ ---- ----- ------ --------
@@ -610,6 +612,10 @@ NOHZ_FULL_TEST_MATRIX=(
 	" C2-7:P2    .      .      .     C0-4    .      .      .     0 A1:0-4 A1:P2 0-4"
 	" C1-7:P2 C5-7:P2   .      .       .     .      .    C0:P2   0 A1:1-4|A2:5-7|B1:0
 								       A1:P2|A2:P2|B1:P-2 1-7"
+	# An isolated child partition cannot grow to take over the last
+	# housekeeping CPU when its parent expands the exclusive CPUs.
+	" C0-8:X1-8:P1 CX0-8:P2 .  .     X0-7    .      .      .     0 A1:0-7|A2:0-7 A1:P1|A2:P-2 ."
+	" C0-8:X0-6:P1 CX0-8:P2 .  .     X2-8    .      .      .     0 A1:|A2:2-8 A1:P1|A2:P2 2-8"
 )
 
 #
-- 
2.56.0