[PATCH] iommu/amd: Fix event counter availability check

Alexander Monakov amonakov at ispras.ru
Fri May 29 20:07:38 UTC 2020

The driver performs an extra check if the IOMMU's capabilities advertise
presence of performance counters: it verifies that counters are writable
by writing a hard-coded value to a counter and testing that reading that
counter gives back the same value.

Unfortunately it does so quite early, even before pci_enable_device is
called for the IOMMU, i.e. when accessing its MMIO space is not
guaranteed to work. On Ryzen 4500U CPU, this actually breaks the test:
the driver assumes the counters are not writable, and disables the

Moving init_iommu_perf_ctr just after iommu_flush_all_caches resolves
the issue. This is the earliest point in amd_iommu_init_pci where the
call succeeds on my laptop.

Signed-off-by: Alexander Monakov <amonakov at ispras.ru>
Cc: Joerg Roedel <joro at 8bytes.org>
Cc: Suravee Suthikulpanit <suravee.suthikulpanit at amd.com>
Cc: iommu at lists.linux-foundation.org

PS. I'm seeing another hiccup with IOMMU probing on my system:
pci 0000:00:00.2: can't derive routing for PCI INT A
pci 0000:00:00.2: PCI INT A: not connected

Hopefully I can figure it out, but I'd appreciate hints.

 drivers/iommu/amd_iommu_init.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
index 5b81fd16f5fa..1b7ec6b6a282 100644
--- a/drivers/iommu/amd_iommu_init.c
+++ b/drivers/iommu/amd_iommu_init.c
@@ -1788,8 +1788,6 @@ static int __init iommu_init_pci(struct amd_iommu *iommu)
 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
 		amd_iommu_np_cache = true;
-	init_iommu_perf_ctr(iommu);
 	if (is_rd890_iommu(iommu->dev)) {
 		int i, j;
@@ -1891,8 +1889,10 @@ static int __init amd_iommu_init_pci(void)
-	for_each_iommu(iommu)
+	for_each_iommu(iommu) {
+		init_iommu_perf_ctr(iommu);
+	}
 	if (!ret)

base-commit: 75caf310d16cc5e2f851c048cd597f5437013368

More information about the iommu mailing list