[PATCH] interrupts: add uio type to support multiple interrupt vector

[PATCH] interrupts: add uio type to support multiple interrupt vector

[Tianli Lai]

vmxnet3 device use MSI-X interrupt type.
vmxnet3 driver configure the hardware to generate MSI-X interrupts
would fail when call rte_intr_cap_multiple function return zero.

Signed-off-by: Tianli Lai <laitianli@tom.com>
---
 lib/eal/linux/eal_interrupts.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
index d52ec8eb4c..4b4dffa20c 100644
--- a/lib/eal/linux/eal_interrupts.c
+++ b/lib/eal/linux/eal_interrupts.c
@@ -1596,6 +1596,9 @@ rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
 	if (rte_intr_type_get(intr_handle) == RTE_INTR_HANDLE_VDEV)
 		return 1;
 
+	if (rte_intr_type_get(intr_handle) == RTE_INTR_HANDLE_UIO)
+		return 1;
+
 	return 0;
 }
 
-- 
2.27.0

Re: [PATCH] interrupts: add uio type to support multiple interrupt vector

[David Marchand]

On Thu, Apr 7, 2022 at 8:45 AM Tianli Lai <laitianli@tom.com> wrote:
>
> vmxnet3 device use MSI-X interrupt type.
> vmxnet3 driver configure the hardware to generate MSI-X interrupts
> would fail when call rte_intr_cap_multiple function return zero.

Iiuc, the issue is that you are using vmxnet3 devices bound to igb_uio.
Out of curiosity, why not use vfio-pci?


>
> Signed-off-by: Tianli Lai <laitianli@tom.com>
> ---
>  lib/eal/linux/eal_interrupts.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/lib/eal/linux/eal_interrupts.c b/lib/eal/linux/eal_interrupts.c
> index d52ec8eb4c..4b4dffa20c 100644
> --- a/lib/eal/linux/eal_interrupts.c
> +++ b/lib/eal/linux/eal_interrupts.c
> @@ -1596,6 +1596,9 @@ rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
>         if (rte_intr_type_get(intr_handle) == RTE_INTR_HANDLE_VDEV)
>                 return 1;
>
> +       if (rte_intr_type_get(intr_handle) == RTE_INTR_HANDLE_UIO)
> +               return 1;
> +

Doing this will let drivers think there are multiple fds, ok.

Though (I am not entirely sure) with this patch only, are not all
vectors pointing to a single fd.
For example, did you test Rx interrupts? with more than 1 rxq?


-- 
David Marchand