[PATCH v2] iommu/amd: Do not clear unobserved log overflow status bits

From: Guanghui Feng

Date: Thu Nov 19 2026 - 09:14:21 EST


amd_iommu_handle_irq() clears the interrupt status with an unconditional
writel(mask, MMIO_STATUS_OFFSET), where mask is int_mask | overflow_mask.
If hardware sets a log overflow bit after the status register is sampled
at the top of the loop but before that write, the write clears the
freshly-set bit while the stale sampled status does not contain it, so
overflow_handler() is skipped for an overflow that really happened.

This is fatal for the GA log: GALOverflow generates no interrupt of its
own (MMIO Offset 2020h bit 9), amd_iommu_restart_ga_log() is the only
recovery path, and per spec section 2.7.4 logging stays halted until
GALogEn is toggled -- a bare W1C of GALOverflow is not part of that
recovery. The GA log then stays halted, GAInt never re-asserts and the
GA log interrupt line goes permanently silent, losing all AVIC
non-running vCPU wakeups on that IOMMU.

Clear only the bits observed this round by writing status & mask. An
overflow bit set concurrently is then left asserted and is re-observed by
the existing ERBT1312 status read-back at the bottom of the loop, which
re-enters the loop and runs overflow_handler().

The unconditional write has cleared an overflow bit since commit
5ce97f4ec5e0 ("iommu/amd: Recover from event log overflow"); commit
af47b0a24058 ("iommu/amd: Handle GALog overflows") and commit
274c2218b8b2 ("iommu/amd: Handle PPR log overflow") later added the GA
and PPR overflow bits to the same write.

Fixes: 5ce97f4ec5e0 ("iommu/amd: Recover from event log overflow")
Signed-off-by: Guanghui Feng <guanghuifeng@xxxxxxxxxxxxxxxxx>
Reviewed-by: Ankit Soni <Ankit.Soni@xxxxxxx>
---
v2:
- Correct the Fixes tag: 2379f3485239 only moved the existing clear into
amd_iommu_handle_irq(); the unconditional write started clearing an
overflow bit in 5ce97f4ec5e0. Record the GA/PPR follow-up commits.
(Ankit Soni)
- Tighten the commit message and the in-code comment: scope the "no
interrupt of its own" statement to GALOverflow, since EventOverflow
and PprOverflow do have their own interrupt source. The functional
change (writel(status & mask)) is unchanged from v1, so Ankit's
Reviewed-by still applies.
- Collect Ankit's Reviewed-by.

drivers/iommu/amd/iommu.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/amd/iommu.c b/drivers/iommu/amd/iommu.c
index d0d55c93a480..530f97729d68 100644
--- a/drivers/iommu/amd/iommu.c
+++ b/drivers/iommu/amd/iommu.c
@@ -1168,8 +1168,18 @@ static void amd_iommu_handle_irq(void *data, const char *evt_type,
u32 mask = int_mask | overflow_mask;

while (status & mask) {
- /* Enable interrupt sources again */
- writel(mask, iommu->mmio_base + MMIO_STATUS_OFFSET);
+ /*
+ * Clear only the status bits observed in this round. The log
+ * overflow bits are RW1C; clearing one unconditionally here
+ * would drop an overflow that hardware set between the readl()
+ * above and this write, without ever running overflow_handler().
+ * Leaving such a bit asserted lets the existing ERBT1312
+ * read-back below re-observe it and re-enter the loop so
+ * overflow_handler() runs. This matters most for the GA log,
+ * whose GALOverflow bit generates no interrupt of its own, so a
+ * dropped overflow can otherwise halt GA logging permanently.
+ */
+ writel(status & mask, iommu->mmio_base + MMIO_STATUS_OFFSET);

if (int_handler) {
pr_devel("Processing IOMMU (ivhd%d) %s Log\n",
--
2.43.7