* [PATCH 0/2] iommu/vsi: Fix the runtime PM of the IOMMU
@ 2026-11-13 20:04 Arun Iyengar via B4 Relay
2026-11-13 20:04 ` [PATCH 1/2] iommu/vsi: Don't leak a runtime PM reference in the TLB flush Arun Iyengar via B4 Relay
2026-11-13 20:04 ` [PATCH 2/2] iommu/vsi: Don't resume the IOMMU in its interrupt handler Arun Iyengar via B4 Relay
0 siblings, 2 replies; 5+ messages in thread
From: Arun Iyengar via B4 Relay @ 2026-11-13 20:04 UTC (permalink / raw)
To: Benjamin Gaignard, Joerg Roedel (AMD), Will Deacon, Robin Murphy
Cc: iommu, linux-kernel, Arun Iyengar
Patch 1 lets the IOMMU suspend again: a failed pm_runtime_get() in the
TLB flush leaked a usage count. Patch 2 stops the interrupt handler
from resuming the IOMMU in hard IRQ context.
Tested on RK3588S (Orange Pi 5), v7.3-based kernel that also has
hantro-robust v3 [1] and the vsi-iommu suspend patch I will send next:
without patch 1 the AV1 IOMMU stayed active after decoding. With both
it is runtime suspended with usage count 0 after AV1 fluster (203/242,
unchanged), 20 AV1 decoder unbind/bind cycles and 3 s2idle cycles. No
IOMMU fault occurred, so patch 2's handler path did not run.
No conflict with patch 10 of [1], which touches the probe.
Developed with Claude (Anthropic), which wrote the code and the commit
messages. I reviewed every change.
[1] https://lore.kernel.org/r/20261113-hantro-robust-v3-0-82383c722ad2@gmail.com
Signed-off-by: Arun Iyengar <aiyengar1990@gmail.com>
---
Arun Iyengar (2):
iommu/vsi: Don't leak a runtime PM reference in the TLB flush
iommu/vsi: Don't resume the IOMMU in its interrupt handler
drivers/iommu/vsi-iommu.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
---
base-commit: 2cf4be57b301ed2d7af82e70cff5dfd96894d8de
change-id: 20261113-vsi-iommu-pm-ref-22c83a5473be
Best regards,
--
Arun Iyengar <aiyengar1990@gmail.com>
^ permalink raw reply [flat|nested] 5+ messages in thread* [PATCH 1/2] iommu/vsi: Don't leak a runtime PM reference in the TLB flush 2026-11-13 20:04 [PATCH 0/2] iommu/vsi: Fix the runtime PM of the IOMMU Arun Iyengar via B4 Relay @ 2026-11-13 20:04 ` Arun Iyengar via B4 Relay 2026-11-16 8:11 ` Benjamin Gaignard 2026-11-13 20:04 ` [PATCH 2/2] iommu/vsi: Don't resume the IOMMU in its interrupt handler Arun Iyengar via B4 Relay 1 sibling, 1 reply; 5+ messages in thread From: Arun Iyengar via B4 Relay @ 2026-11-13 20:04 UTC (permalink / raw) To: Benjamin Gaignard, Joerg Roedel (AMD), Will Deacon, Robin Murphy Cc: iommu, linux-kernel, Arun Iyengar From: Arun Iyengar <aiyengar1990@gmail.com> pm_runtime_get() takes a usage reference even when it fails, e.g. with -EINPROGRESS while a resume runs, and vsi_iommu_flush_tlb() then skips the put. The IOMMU then never suspends again. Flush only an active IOMMU and flush the TLB when the IOMMU is enabled, which covers resume. Fixes: 917ace84b770 ("iommu: Add verisilicon IOMMU driver") Signed-off-by: Arun Iyengar <aiyengar1990@gmail.com> --- drivers/iommu/vsi-iommu.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/drivers/iommu/vsi-iommu.c b/drivers/iommu/vsi-iommu.c index 5dac20b143a5..5d581be95333 100644 --- a/drivers/iommu/vsi-iommu.c +++ b/drivers/iommu/vsi-iommu.c @@ -369,7 +369,8 @@ static void vsi_iommu_flush_tlb(struct iommu_domain *domain) struct vsi_iommu *iommu; list_for_each_entry(iommu, &vsi_domain->iommus, node) { - if (pm_runtime_get(iommu->dev) < 0) + /* A suspended IOMMU is flushed when it resumes */ + if (pm_runtime_get_if_active(iommu->dev) <= 0) continue; spin_lock(&iommu->lock); @@ -545,6 +546,8 @@ static void vsi_iommu_enable(struct vsi_iommu *iommu, struct iommu_domain *domai writel(VSI_MMU_OUT_OF_BOUND, iommu->regs + VSI_MMU_CONFIG1_BASE); writel(VSI_MMU_BIT_ENABLE, iommu->regs + VSI_MMU_AHB_EXCEPTION_BASE); writel(VSI_MMU_BIT_ENABLE, iommu->regs + VSI_MMU_AHB_CONTROL_BASE); + writel(VSI_MMU_BIT_FLUSH, iommu->regs + VSI_MMU_FLUSH_BASE); + writel(0, iommu->regs + VSI_MMU_FLUSH_BASE); iommu->enable = true; } @@ -564,8 +567,6 @@ static int vsi_iommu_attach_device(struct iommu_domain *domain, spin_lock(&iommu->lock); vsi_iommu_enable(iommu, domain); - writel(VSI_MMU_BIT_FLUSH, iommu->regs + VSI_MMU_FLUSH_BASE); - writel(0, iommu->regs + VSI_MMU_FLUSH_BASE); list_del_init(&iommu->node); list_add_tail(&iommu->node, &vsi_domain->iommus); -- 2.55.0 ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 1/2] iommu/vsi: Don't leak a runtime PM reference in the TLB flush 2026-11-13 20:04 ` [PATCH 1/2] iommu/vsi: Don't leak a runtime PM reference in the TLB flush Arun Iyengar via B4 Relay @ 2026-11-16 8:11 ` Benjamin Gaignard 0 siblings, 0 replies; 5+ messages in thread From: Benjamin Gaignard @ 2026-11-16 8:11 UTC (permalink / raw) To: aiyengar1990, Joerg Roedel (AMD), Will Deacon, Robin Murphy Cc: iommu, linux-kernel Le 13/11/2026 à 21:04, Arun Iyengar via B4 Relay a écrit : > From: Arun Iyengar <aiyengar1990@gmail.com> > > pm_runtime_get() takes a usage reference even when it fails, e.g. with > -EINPROGRESS while a resume runs, and vsi_iommu_flush_tlb() then skips > the put. The IOMMU then never suspends again. > > Flush only an active IOMMU and flush the TLB when the IOMMU is enabled, > which covers resume. > > Fixes: 917ace84b770 ("iommu: Add verisilicon IOMMU driver") > Signed-off-by: Arun Iyengar <aiyengar1990@gmail.com> Reviewed-by: Benjamin Gaignard <benjamin.gaignard@collabora.com> > --- > drivers/iommu/vsi-iommu.c | 7 ++++--- > 1 file changed, 4 insertions(+), 3 deletions(-) > > diff --git a/drivers/iommu/vsi-iommu.c b/drivers/iommu/vsi-iommu.c > index 5dac20b143a5..5d581be95333 100644 > --- a/drivers/iommu/vsi-iommu.c > +++ b/drivers/iommu/vsi-iommu.c > @@ -369,7 +369,8 @@ static void vsi_iommu_flush_tlb(struct iommu_domain *domain) > struct vsi_iommu *iommu; > > list_for_each_entry(iommu, &vsi_domain->iommus, node) { > - if (pm_runtime_get(iommu->dev) < 0) > + /* A suspended IOMMU is flushed when it resumes */ > + if (pm_runtime_get_if_active(iommu->dev) <= 0) > continue; > > spin_lock(&iommu->lock); > @@ -545,6 +546,8 @@ static void vsi_iommu_enable(struct vsi_iommu *iommu, struct iommu_domain *domai > writel(VSI_MMU_OUT_OF_BOUND, iommu->regs + VSI_MMU_CONFIG1_BASE); > writel(VSI_MMU_BIT_ENABLE, iommu->regs + VSI_MMU_AHB_EXCEPTION_BASE); > writel(VSI_MMU_BIT_ENABLE, iommu->regs + VSI_MMU_AHB_CONTROL_BASE); > + writel(VSI_MMU_BIT_FLUSH, iommu->regs + VSI_MMU_FLUSH_BASE); > + writel(0, iommu->regs + VSI_MMU_FLUSH_BASE); > iommu->enable = true; > } > > @@ -564,8 +567,6 @@ static int vsi_iommu_attach_device(struct iommu_domain *domain, > spin_lock(&iommu->lock); > > vsi_iommu_enable(iommu, domain); > - writel(VSI_MMU_BIT_FLUSH, iommu->regs + VSI_MMU_FLUSH_BASE); > - writel(0, iommu->regs + VSI_MMU_FLUSH_BASE); > > list_del_init(&iommu->node); > list_add_tail(&iommu->node, &vsi_domain->iommus); > ^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH 2/2] iommu/vsi: Don't resume the IOMMU in its interrupt handler 2026-11-13 20:04 [PATCH 0/2] iommu/vsi: Fix the runtime PM of the IOMMU Arun Iyengar via B4 Relay 2026-11-13 20:04 ` [PATCH 1/2] iommu/vsi: Don't leak a runtime PM reference in the TLB flush Arun Iyengar via B4 Relay @ 2026-11-13 20:04 ` Arun Iyengar via B4 Relay 2026-11-16 8:11 ` Benjamin Gaignard 1 sibling, 1 reply; 5+ messages in thread From: Arun Iyengar via B4 Relay @ 2026-11-13 20:04 UTC (permalink / raw) To: Benjamin Gaignard, Joerg Roedel (AMD), Will Deacon, Robin Murphy Cc: iommu, linux-kernel, Arun Iyengar From: Arun Iyengar <aiyengar1990@gmail.com> The interrupt handler calls pm_runtime_resume_and_get(), which can sleep, in hard IRQ context. Handle the interrupt only if the IOMMU is active. Fixes: 917ace84b770 ("iommu: Add verisilicon IOMMU driver") Signed-off-by: Arun Iyengar <aiyengar1990@gmail.com> --- drivers/iommu/vsi-iommu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/iommu/vsi-iommu.c b/drivers/iommu/vsi-iommu.c index 5d581be95333..3dec58224953 100644 --- a/drivers/iommu/vsi-iommu.c +++ b/drivers/iommu/vsi-iommu.c @@ -196,7 +196,7 @@ static irqreturn_t vsi_iommu_irq(int irq, void *dev_id) dma_addr_t iova; u32 status; - if (pm_runtime_resume_and_get(iommu->dev) < 0) + if (pm_runtime_get_if_active(iommu->dev) <= 0) return IRQ_NONE; spin_lock_irqsave(&iommu->lock, flags); -- 2.55.0 ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 2/2] iommu/vsi: Don't resume the IOMMU in its interrupt handler 2026-11-13 20:04 ` [PATCH 2/2] iommu/vsi: Don't resume the IOMMU in its interrupt handler Arun Iyengar via B4 Relay @ 2026-11-16 8:11 ` Benjamin Gaignard 0 siblings, 0 replies; 5+ messages in thread From: Benjamin Gaignard @ 2026-11-16 8:11 UTC (permalink / raw) To: aiyengar1990, Joerg Roedel (AMD), Will Deacon, Robin Murphy Cc: iommu, linux-kernel Le 13/11/2026 à 21:04, Arun Iyengar via B4 Relay a écrit : > From: Arun Iyengar <aiyengar1990@gmail.com> > > The interrupt handler calls pm_runtime_resume_and_get(), which can > sleep, in hard IRQ context. Handle the interrupt only if the IOMMU is > active. > > Fixes: 917ace84b770 ("iommu: Add verisilicon IOMMU driver") > Signed-off-by: Arun Iyengar <aiyengar1990@gmail.com> Reviewed-by: Benjamin Gaignard <benjamin.gaignard@collabora.com> > --- > drivers/iommu/vsi-iommu.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/iommu/vsi-iommu.c b/drivers/iommu/vsi-iommu.c > index 5d581be95333..3dec58224953 100644 > --- a/drivers/iommu/vsi-iommu.c > +++ b/drivers/iommu/vsi-iommu.c > @@ -196,7 +196,7 @@ static irqreturn_t vsi_iommu_irq(int irq, void *dev_id) > dma_addr_t iova; > u32 status; > > - if (pm_runtime_resume_and_get(iommu->dev) < 0) > + if (pm_runtime_get_if_active(iommu->dev) <= 0) > return IRQ_NONE; > > spin_lock_irqsave(&iommu->lock, flags); > ^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-11-16 8:11 UTC | newest] Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2026-11-13 20:04 [PATCH 0/2] iommu/vsi: Fix the runtime PM of the IOMMU Arun Iyengar via B4 Relay 2026-11-13 20:04 ` [PATCH 1/2] iommu/vsi: Don't leak a runtime PM reference in the TLB flush Arun Iyengar via B4 Relay 2026-11-16 8:11 ` Benjamin Gaignard 2026-11-13 20:04 ` [PATCH 2/2] iommu/vsi: Don't resume the IOMMU in its interrupt handler Arun Iyengar via B4 Relay 2026-11-16 8:11 ` Benjamin Gaignard
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®