x86/amd_node: Fix PCI device reference counting in amd_smn_init()grafted

The local "root" pointer is a temporary variable used during the device
search. Therefore, refcount related to the search iterators should be cleaned
up after the search is complete.

Use the __free() cleanup macro to ensure the refcount is decremented when the
temporary pointer goes out of scope.

Additionally, increment the refcount when caching a root pointer. This ensures
the in-use refcount is separate from the temporary search refcounting.

Finally, drop the redundant "root = NULL" before the second search loop. The
pci_get_class() iterator always decrements the refcount of its "from"
argument, so the first loop can only fall through with "root" already NULL.

Fixes: 0a4b61d9c2e4 ("x86/amd_node: Fix AMD root device caching")
Closes: https://sashiko.dev/#/patchset/20260806160159.230453-1-jason.andryuk%40amd.com
Reported-by: Sashiko <sashiko-bot@kernel.org>
Assisted-by: LLM
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Cc: <stable@kernel.org>
Link: https://patch.msgid.link/20260903154325.74343-1-yazen.ghannam@amd.com

1 files changed, 164 insertions, 0 deletions

diff --git a/drivers/iommu/intel/perf.c b/drivers/iommu/intel/perf.c
new file mode 100644
index 000000000..bec98dcbb
--- /dev/null
+++ b/drivers/iommu/intel/perf.c
@@ -0,0 +1,164 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * perf.c - performance monitor
+ *
+ * Copyright (C) 2021 Intel Corporation
+ *
+ * Author: Lu Baolu <baolu.lu@linux.intel.com>
+ *         Fenghua Yu <fenghua.yu@intel.com>
+ */
+
+#include <linux/spinlock.h>
+
+#include "iommu.h"
+#include "perf.h"
+
+static DEFINE_SPINLOCK(latency_lock);
+
+bool dmar_latency_enabled(struct intel_iommu *iommu, enum latency_type type)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+
+	return lstat && lstat[type].enabled;
+}
+
+int dmar_latency_enable(struct intel_iommu *iommu, enum latency_type type)
+{
+	struct latency_statistic *lstat;
+	unsigned long flags;
+	int ret = -EBUSY;
+
+	if (dmar_latency_enabled(iommu, type))
+		return 0;
+
+	spin_lock_irqsave(&latency_lock, flags);
+	if (!iommu->perf_statistic) {
+		iommu->perf_statistic = kzalloc_objs(*lstat, DMAR_LATENCY_NUM,
+						     GFP_ATOMIC);
+		if (!iommu->perf_statistic) {
+			ret = -ENOMEM;
+			goto unlock_out;
+		}
+	}
+
+	lstat = iommu->perf_statistic;
+
+	if (!lstat[type].enabled) {
+		lstat[type].enabled = true;
+		lstat[type].counter[COUNTS_MIN] = UINT_MAX;
+		ret = 0;
+	}
+unlock_out:
+	spin_unlock_irqrestore(&latency_lock, flags);
+
+	return ret;
+}
+
+void dmar_latency_disable(struct intel_iommu *iommu, enum latency_type type)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+	unsigned long flags;
+
+	if (!dmar_latency_enabled(iommu, type))
+		return;
+
+	spin_lock_irqsave(&latency_lock, flags);
+	memset(&lstat[type], 0, sizeof(*lstat));
+	spin_unlock_irqrestore(&latency_lock, flags);
+}
+
+void dmar_latency_update(struct intel_iommu *iommu, enum latency_type type, u64 latency)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+	unsigned long flags;
+	u64 min, max;
+
+	if (!dmar_latency_enabled(iommu, type))
+		return;
+
+	spin_lock_irqsave(&latency_lock, flags);
+	if (latency < 100)
+		lstat[type].counter[COUNTS_10e2]++;
+	else if (latency < 1000)
+		lstat[type].counter[COUNTS_10e3]++;
+	else if (latency < 10000)
+		lstat[type].counter[COUNTS_10e4]++;
+	else if (latency < 100000)
+		lstat[type].counter[COUNTS_10e5]++;
+	else if (latency < 1000000)
+		lstat[type].counter[COUNTS_10e6]++;
+	else if (latency < 10000000)
+		lstat[type].counter[COUNTS_10e7]++;
+	else
+		lstat[type].counter[COUNTS_10e8_plus]++;
+
+	min = lstat[type].counter[COUNTS_MIN];
+	max = lstat[type].counter[COUNTS_MAX];
+	lstat[type].counter[COUNTS_MIN] = min_t(u64, min, latency);
+	lstat[type].counter[COUNTS_MAX] = max_t(u64, max, latency);
+	lstat[type].counter[COUNTS_SUM] += latency;
+	lstat[type].samples++;
+	spin_unlock_irqrestore(&latency_lock, flags);
+}
+
+static char *latency_counter_names[] = {
+	"                  <0.1us",
+	"   0.1us-1us", "    1us-10us", "  10us-100us",
+	"   100us-1ms", "    1ms-10ms", "      >=10ms",
+	"     min(us)", "     max(us)", " average(us)"
+};
+
+static char *latency_type_names[] = {
+	"   inv_iotlb", "  inv_devtlb", "     inv_iec",
+	"     svm_prq"
+};
+
+void dmar_latency_snapshot(struct intel_iommu *iommu, char *str, size_t size)
+{
+	struct latency_statistic *lstat = iommu->perf_statistic;
+	unsigned long flags;
+	int bytes = 0, i, j;
+
+	memset(str, 0, size);
+
+	for (i = 0; i < COUNTS_NUM; i++)
+		bytes += scnprintf(str + bytes, size - bytes,
+				  "%s", latency_counter_names[i]);
+
+	spin_lock_irqsave(&latency_lock, flags);
+	for (i = 0; i < DMAR_LATENCY_NUM; i++) {
+		if (!dmar_latency_enabled(iommu, i))
+			continue;
+
+		bytes += scnprintf(str + bytes, size - bytes,
+				  "\n%s", latency_type_names[i]);
+
+		for (j = 0; j < COUNTS_NUM; j++) {
+			u64 val = lstat[i].counter[j];
+
+			switch (j) {
+			case COUNTS_MIN:
+				if (val == UINT_MAX)
+					val = 0;
+				else
+					val = div_u64(val, 1000);
+				break;
+			case COUNTS_MAX:
+				val = div_u64(val, 1000);
+				break;
+			case COUNTS_SUM:
+				if (lstat[i].samples)
+					val = div_u64(val, (lstat[i].samples * 1000));
+				else
+					val = 0;
+				break;
+			default:
+				break;
+			}
+
+			bytes += scnprintf(str + bytes, size - bytes,
+					  "%12lld", val);
+		}
+	}
+	spin_unlock_irqrestore(&latency_lock, flags);
+}