Skip to contentSkip to conversation

iommu/riscv: Support sharing irq lines between iommu queues

PATCH In review Xu Lu 2 participants 6 messages Jul 1, 2024

Subsystems (3)
lore
Share & tools
lore

Xu Lu

author
#0 lore raw

cover iommu/riscv: Support sharing irq lines between iommu queues

This is an additional patch for Tomasz Jeznach's riscv iommu driver patch series[1].

It seems that when the number of wired irq lines is less than the number of iommu queues, iommu platform driver does not assign virtual irqs for iommu queues whose id exceeds the total number of irq lines.

If it is a problem worth fixing, please feel free to merge this patch into existing commits. And please ignore this patch if we don't think it is a problem.

[1] link: https://lore.kernel.org/all/[email protected]/

Xu Lu (1):
iommu/riscv: Support sharing irq lines between iommu queues

 drivers/iommu/riscv/iommu-platform.c | 4 ++++
 1 file changed, 4 insertions(+)
Signature · 1 line
2.20.1
author
#1 lore raw

PATCH 1/1 iommu/riscv: Support sharing irq lines between iommu queues

When the number of wired interrupt lines is less than the number of iommu queues, we should assign one irq line for several queues and configure csr icvec accordingly.

---
 drivers/iommu/riscv/iommu-platform.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/iommu/riscv/iommu-platform.c +4 -0
index da336863f152..1d0af1260d5b 100644
@@ -60,6 +60,10 @@ static int riscv_iommu_platform_probe(struct platform_device *pdev)
 	for (vec = 0; vec < iommu->irqs_count; vec++)
 		iommu->irqs[vec] = platform_get_irq(pdev, vec);
 
+	for (vec = iommu->irqs_count; vec < RISCV_IOMMU_INTR_COUNT; vec++)
+		iommu->irqs[vec] = platform_get_irq(pdev,
+						    (vec % iommu->irqs_count));
+
 	/* Enable wire-signaled interrupts, fctl.WSI */
 	if (!(iommu->fctl & RISCV_IOMMU_FCTL_WSI)) {
 		iommu->fctl |= RISCV_IOMMU_FCTL_WSI;
Signature · 1 line
2.20.1
#2 lore raw
Quoted 5 lines · Xu Lu' via linux

When the number of wired interrupt lines is less than the number of iommu queues, we should assign one irq line for several queues and configure csr icvec accordingly.

Is this case not already covered by the code in function riscv_iommu_init_check(), assigning cause to vector register (ICVEC) based on available interrupt vectors? Interrupt vector selection is based on accepted ICVEC configuration (see riscv_iommu_queue_vec()).

Relevant code (patch 'Command and fault queue support'):
iommu->icvec = FIELD_PREP(RISCV_IOMMU_ICVEC_FIV, 1 % iommu->irqs_count) |

       FIELD_PREP(RISCV_IOMMU_ICVEC_PIV, 2 % iommu->irqs_count) |
       FIELD_PREP(RISCV_IOMMU_ICVEC_PMIV, 3 % iommu->irqs_count);

Values in iommu->irqs[] should represent actual interrupts allocated for the device.

Best,
- Tomasz

Quoted 28 lines
---
 drivers/iommu/riscv/iommu-platform.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/iommu/riscv/iommu-platform.c +4 -1
index da336863f152..1d0af1260d5b 100644
@@ -60,6 +60,10 @@ static int riscv_iommu_platform_probe(struct platform_device *pdev)
        for (vec = 0; vec < iommu->irqs_count; vec++)
                iommu->irqs[vec] = platform_get_irq(pdev, vec);
 
+       for (vec = iommu->irqs_count; vec < RISCV_IOMMU_INTR_COUNT; vec++)
+               iommu->irqs[vec] = platform_get_irq(pdev,
+                                                   (vec % iommu->irqs_count));
+
        /* Enable wire-signaled interrupts, fctl.WSI */
        if (!(iommu->fctl & RISCV_IOMMU_FCTL_WSI)) {
                iommu->fctl |= RISCV_IOMMU_FCTL_WSI;
--

2.20.1

--
You received this message because you are subscribed to the Google Groups "linux" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
To view this discussion on the web visit https://groups.google.com/a/rivosinc.com/d/msgid/linux/20240701072908.25503-2-luxu.kernel%40bytedance.com.
For more options, visit https://groups.google.com/a/rivosinc.com/d/optout.

author
#3 lore raw

Sorry for not describing it clearly.

Indeed, the hardware interrupt lines have be configured in riscv_iommu_init_check().

What I mean is that virtual interrupts of irqs[iommu->irqs_count] to irqs[RISCV_IOMMU_INTR_COUNT - 1] are not set, which will cause riscv_iommu_queue_enable() fail:
const unsigned int irq =

iommu->irqs[riscv_iommu_queue_vec(iommu, queue->qid)];
       rc = request_threaded_irq(irq, xxx);
       if (rc) {
                 queue->iommu = NULL;
                 return rc;
       }

The virtual irq of queue id 'q' should be the same as irqs[q % iommu->irqs_count] as ICVEC is configured in the same way.

Regards,

Xu Lu

Quoted 52 lines · Tomasz Jeznach
Quoted 5 lines · Xu Lu' via linux

When the number of wired interrupt lines is less than the number of iommu queues, we should assign one irq line for several queues and configure csr icvec accordingly.

Is this case not already covered by the code in function riscv_iommu_init_check(), assigning cause to vector register (ICVEC) based on available interrupt vectors? Interrupt vector selection is based on accepted ICVEC configuration (see riscv_iommu_queue_vec()).

Relevant code (patch 'Command and fault queue support'):
iommu->icvec = FIELD_PREP(RISCV_IOMMU_ICVEC_FIV, 1 % iommu->irqs_count) |

       FIELD_PREP(RISCV_IOMMU_ICVEC_PIV, 2 % iommu->irqs_count) |
       FIELD_PREP(RISCV_IOMMU_ICVEC_PMIV, 3 % iommu->irqs_count);

Values in iommu->irqs[] should represent actual interrupts allocated for the device.

Best,
- Tomasz

Quoted 28 lines
---
 drivers/iommu/riscv/iommu-platform.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/iommu/riscv/iommu-platform.c +4 -1
index da336863f152..1d0af1260d5b 100644
@@ -60,6 +60,10 @@ static int riscv_iommu_platform_probe(struct platform_device *pdev)
        for (vec = 0; vec < iommu->irqs_count; vec++)
                iommu->irqs[vec] = platform_get_irq(pdev, vec);
 
+       for (vec = iommu->irqs_count; vec < RISCV_IOMMU_INTR_COUNT; vec++)
+               iommu->irqs[vec] = platform_get_irq(pdev,
+                                                   (vec % iommu->irqs_count));
+
        /* Enable wire-signaled interrupts, fctl.WSI */
        if (!(iommu->fctl & RISCV_IOMMU_FCTL_WSI)) {
                iommu->fctl |= RISCV_IOMMU_FCTL_WSI;
--

2.20.1

--
You received this message because you are subscribed to the Google Groups "linux" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
To view this discussion on the web visit https://groups.google.com/a/rivosinc.com/d/msgid/linux/20240701072908.25503-2-luxu.kernel%40bytedance.com.
For more options, visit https://groups.google.com/a/rivosinc.com/d/optout.

author
#4 lore raw

Oh, I think I understand what you mean now after checking the implementation of riscv_iommu_queue_vec(). I apologize for not understanding the meaning of the code before, please ignore my disturbance.

Best Regards,

Xu Lu

#5 lore raw
Quoted 9 lines · Xu Lu

Oh, I think I understand what you mean now after checking the implementation of riscv_iommu_queue_vec(). I apologize for not understanding the meaning of the code before, please ignore my disturbance.

Best Regards,

Xu Lu

No worries, thanks for checking that path. LMK if there are any other problems with the driver.

Best,
- Tomasz