| Message ID | 20240930195617.37369-2-jdamato@fastly.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | ena: Link IRQs, queues, and NAPI instances | expand |
> Link IRQs to NAPI instances with netif_napi_set_irq. This information can be > queried with the netdev-genl API. Note that the ENA device appears to > allocate an IRQ for management purposes which does not have a NAPI > associated with it; this commit takes this into consideration to accurately > construct a map between IRQs and NAPI instances. > > Compare the output of /proc/interrupts for my ena device with the output > of netdev-genl after applying this patch: > > $ cat /proc/interrupts | grep enp55s0 | cut -f1 --delimiter=':' > 94 > 95 > 96 > 97 > 98 > 99 > 100 > 101 > > $ ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/netdev.yaml \ > --dump napi-get --json='{"ifindex": 2}' > > [{'id': 8208, 'ifindex': 2, 'irq': 101}, > {'id': 8207, 'ifindex': 2, 'irq': 100}, > {'id': 8206, 'ifindex': 2, 'irq': 99}, > {'id': 8205, 'ifindex': 2, 'irq': 98}, > {'id': 8204, 'ifindex': 2, 'irq': 97}, > {'id': 8203, 'ifindex': 2, 'irq': 96}, > {'id': 8202, 'ifindex': 2, 'irq': 95}, > {'id': 8201, 'ifindex': 2, 'irq': 94}] > > Signed-off-by: Joe Damato <jdamato@fastly.com> > --- > drivers/net/ethernet/amazon/ena/ena_netdev.c | 12 +++++++++++- > 1 file changed, 11 insertions(+), 1 deletion(-) > > diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c > b/drivers/net/ethernet/amazon/ena/ena_netdev.c > index c5b50cfa935a..e88de5e426ef 100644 > --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c > +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c > @@ -1679,7 +1679,7 @@ static int ena_request_io_irq(struct ena_adapter > *adapter) > u32 io_queue_count = adapter->num_io_queues + adapter- > >xdp_num_queues; > unsigned long flags = 0; > struct ena_irq *irq; > - int rc = 0, i, k; > + int rc = 0, i, k, irq_idx; nit: This breaks RCT guidelines, can you please move it to be below io_queue_count? > > if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) { > netif_err(adapter, ifup, adapter->netdev, @@ -1705,6 +1705,16 @@ > static int ena_request_io_irq(struct ena_adapter *adapter) > irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); > } > > + /* Now that IO IRQs have been successfully allocated map them to the > + * corresponding IO NAPI instance. Note that the mgmnt IRQ does not > + * have a NAPI, so care must be taken to correctly map IRQs to NAPIs. > + */ > + for (i = 0; i < io_queue_count; i++) { > + irq_idx = ENA_IO_IRQ_IDX(i); > + irq = &adapter->irq_tbl[irq_idx]; > + netif_napi_set_irq(&adapter->ena_napi[i].napi, irq->vector); > + } > + > return rc; > > err: > -- > 2.43.0 Thank you for uploading this patch.
On Tue, Oct 01, 2024 at 08:57:47AM +0000, Arinzon, David wrote: > > Link IRQs to NAPI instances with netif_napi_set_irq. This information can be > > queried with the netdev-genl API. Note that the ENA device appears to > > allocate an IRQ for management purposes which does not have a NAPI > > associated with it; this commit takes this into consideration to accurately > > construct a map between IRQs and NAPI instances. > > > > Compare the output of /proc/interrupts for my ena device with the output > > of netdev-genl after applying this patch: > > > > $ cat /proc/interrupts | grep enp55s0 | cut -f1 --delimiter=':' > > 94 > > 95 > > 96 > > 97 > > 98 > > 99 > > 100 > > 101 > > > > $ ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/netdev.yaml \ > > --dump napi-get --json='{"ifindex": 2}' > > > > [{'id': 8208, 'ifindex': 2, 'irq': 101}, > > {'id': 8207, 'ifindex': 2, 'irq': 100}, > > {'id': 8206, 'ifindex': 2, 'irq': 99}, > > {'id': 8205, 'ifindex': 2, 'irq': 98}, > > {'id': 8204, 'ifindex': 2, 'irq': 97}, > > {'id': 8203, 'ifindex': 2, 'irq': 96}, > > {'id': 8202, 'ifindex': 2, 'irq': 95}, > > {'id': 8201, 'ifindex': 2, 'irq': 94}] > > > > Signed-off-by: Joe Damato <jdamato@fastly.com> > > --- > > drivers/net/ethernet/amazon/ena/ena_netdev.c | 12 +++++++++++- > > 1 file changed, 11 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c > > b/drivers/net/ethernet/amazon/ena/ena_netdev.c > > index c5b50cfa935a..e88de5e426ef 100644 > > --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c > > +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c > > @@ -1679,7 +1679,7 @@ static int ena_request_io_irq(struct ena_adapter > > *adapter) > > u32 io_queue_count = adapter->num_io_queues + adapter- > > >xdp_num_queues; > > unsigned long flags = 0; > > struct ena_irq *irq; > > - int rc = 0, i, k; > > + int rc = 0, i, k, irq_idx; > > nit: This breaks RCT guidelines, can you please move it to be below io_queue_count? You are right; sorry I missed that. Will fix in the next revision.
diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c b/drivers/net/ethernet/amazon/ena/ena_netdev.c index c5b50cfa935a..e88de5e426ef 100644 --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c @@ -1679,7 +1679,7 @@ static int ena_request_io_irq(struct ena_adapter *adapter) u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues; unsigned long flags = 0; struct ena_irq *irq; - int rc = 0, i, k; + int rc = 0, i, k, irq_idx; if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) { netif_err(adapter, ifup, adapter->netdev, @@ -1705,6 +1705,16 @@ static int ena_request_io_irq(struct ena_adapter *adapter) irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); } + /* Now that IO IRQs have been successfully allocated map them to the + * corresponding IO NAPI instance. Note that the mgmnt IRQ does not + * have a NAPI, so care must be taken to correctly map IRQs to NAPIs. + */ + for (i = 0; i < io_queue_count; i++) { + irq_idx = ENA_IO_IRQ_IDX(i); + irq = &adapter->irq_tbl[irq_idx]; + netif_napi_set_irq(&adapter->ena_napi[i].napi, irq->vector); + } + return rc; err:
Link IRQs to NAPI instances with netif_napi_set_irq. This information can be queried with the netdev-genl API. Note that the ENA device appears to allocate an IRQ for management purposes which does not have a NAPI associated with it; this commit takes this into consideration to accurately construct a map between IRQs and NAPI instances. Compare the output of /proc/interrupts for my ena device with the output of netdev-genl after applying this patch: $ cat /proc/interrupts | grep enp55s0 | cut -f1 --delimiter=':' 94 95 96 97 98 99 100 101 $ ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/netdev.yaml \ --dump napi-get --json='{"ifindex": 2}' [{'id': 8208, 'ifindex': 2, 'irq': 101}, {'id': 8207, 'ifindex': 2, 'irq': 100}, {'id': 8206, 'ifindex': 2, 'irq': 99}, {'id': 8205, 'ifindex': 2, 'irq': 98}, {'id': 8204, 'ifindex': 2, 'irq': 97}, {'id': 8203, 'ifindex': 2, 'irq': 96}, {'id': 8202, 'ifindex': 2, 'irq': 95}, {'id': 8201, 'ifindex': 2, 'irq': 94}] Signed-off-by: Joe Damato <jdamato@fastly.com> --- drivers/net/ethernet/amazon/ena/ena_netdev.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)