Re: [PATCH 1/7] genirq/msi: Add a new field in msi_desc to store an IOMMU cookie

From: Robin Murphy
Date: Tue Apr 23 2019 - 09:19:52 EST


On 23/04/2019 12:46, Marc Zyngier wrote:
On 23/04/2019 11:51, Julien Grall wrote:
On 4/23/19 11:23 AM, Marc Zyngier wrote:
Hi Julien,

Hi Marc,

On 18/04/2019 18:26, Julien Grall wrote:
When an MSI doorbell is located downstream of an IOMMU, it is required
to swizzle the physical address with an appropriately-mapped IOVA for any
device attached to one of our DMA ops domain.

At the moment, the allocation of the mapping may be done when composing
the message. However, the composing may be done in non-preemtible
context while the allocation requires to be called from preemptible
context.

A follow-up patch will split the current logic in two functions
requiring to keep an IOMMU cookie per MSI.

This patch introduces a new field in msi_desc to store an IOMMU cookie
when CONFIG_IOMMU_DMA is selected.

Signed-off-by: Julien Grall <julien.grall@xxxxxxx>
---
include/linux/msi.h | 3 +++
1 file changed, 3 insertions(+)

diff --git a/include/linux/msi.h b/include/linux/msi.h
index 7e9b81c3b50d..d7907feef1bb 100644
--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -77,6 +77,9 @@ struct msi_desc {
struct device *dev;
struct msi_msg msg;
struct irq_affinity_desc *affinity;
+#ifdef CONFIG_IOMMU_DMA
+ const void *iommu_cookie;
+#endif
union {
/* PCI MSI/X specific data */


Given that this is the only member in this structure that is dependent
on a config option, you could also add a couple of accessors that would
do nothing when IOMMU_DMA is not selected (and use that in the DMA code).

I haven't seen any use of the helpers so far because the DMA code is
also protected by IOMMU_DMA.

I can add the helpers in the next version if you see any use outside of
the DMA code.

There may not be any user user yet, but I'd surely like to see the
accessors. This isn't very different from the stub functions you add in
patch #2.

If you foresee this being useful in general, do you reckon it would be worth decoupling it under its own irqchip-layer Kconfig which can then be selected by IOMMU_DMA?

Robin.