[PATCH V1 0/2] Enable TPH support in BNXT driver

[PATCH V1 0/2] Enable TPH support in BNXT driver

[Wei Huang]

Hi All,

TPH (TLP Processing Hints) is a PCIe feature that allows endpoint devices
to provide optimization hints for memory-targeted requests. TPH support
is now available in pci.git/next and ready for v6.13.

This patchset enables TPH in the Broadcom bnxt driver. Through the IRQ
affinity notifier, steering tags (ST) are updated when the driver's IRQ
affinity changes. On Broadcom NICs with compatible firmware, this
implementation improves memory bandwidth efficiency and network
performance in real-world benchmarks.

Note: These patches are follow-ups from the earlier TPH enablement
series. Patch #4 (IRQ affinity) has been updated based on feedback from
Jakub Kicinski. Since net-next currently lacks TPH support, we propose
these patches be merged via pci.git/next after review by Jakub and
others.

Thanks,
-Wei

Manoj Panicker (1):
  bnxt_en: Add TPH support in BNXT driver

Michael Chan (1):
  bnxt_en: Pass NQ ID to the FW when allocating RX/RX AGG rings

 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 111 +++++++++++++++++++++-
 drivers/net/ethernet/broadcom/bnxt/bnxt.h |   7 ++
 net/core/netdev_rx_queue.c                |   1 +
 3 files changed, 117 insertions(+), 2 deletions(-)


base-commit: e031efeb3a3a65673086d8f148da53f22c6f8ae5
-- 
2.46.0

[PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Wei Huang]

From: Manoj Panicker <manoj.panicker2@amd.com>

Add TPH support to the Broadcom BNXT device driver. This allows the
driver to utilize TPH functions for retrieving and configuring Steering
Tags when changing interrupt affinity. With compatible NIC firmware,
network traffic will be tagged correctly with Steering Tags, resulting
in significant memory bandwidth savings and other advantages as
demonstrated by real network benchmarks on TPH-capable platforms.

Co-developed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Co-developed-by: Wei Huang <wei.huang2@amd.com>
Signed-off-by: Wei Huang <wei.huang2@amd.com>
Signed-off-by: Manoj Panicker <manoj.panicker2@amd.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 103 ++++++++++++++++++++++
 drivers/net/ethernet/broadcom/bnxt/bnxt.h |   7 ++
 net/core/netdev_rx_queue.c                |   1 +
 3 files changed, 111 insertions(+)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 6e422e24750a..beabc4b4a913 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -55,6 +55,8 @@
 #include <net/page_pool/helpers.h>
 #include <linux/align.h>
 #include <net/netdev_queues.h>
+#include <net/netdev_rx_queue.h>
+#include <linux/pci-tph.h>
 
 #include "bnxt_hsi.h"
 #include "bnxt.h"
@@ -10865,6 +10867,81 @@ int bnxt_reserve_rings(struct bnxt *bp, bool irq_re_init)
 	return 0;
 }
 
+static void bnxt_irq_affinity_notify(struct irq_affinity_notify *notify,
+				     const cpumask_t *mask)
+{
+	struct bnxt_irq *irq;
+	u16 tag;
+	int err;
+
+	irq = container_of(notify, struct bnxt_irq, affinity_notify);
+
+	if (!irq->bp->tph_mode)
+		return;
+
+	cpumask_copy(irq->cpu_mask, mask);
+
+	if (pcie_tph_get_cpu_st(irq->bp->pdev, TPH_MEM_TYPE_VM,
+				cpumask_first(irq->cpu_mask), &tag))
+		return;
+
+	if (pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, tag))
+		return;
+
+	rtnl_lock();
+	if (netif_running(irq->bp->dev)) {
+		err = netdev_rx_queue_restart(irq->bp->dev, irq->ring_nr);
+		if (err)
+			netdev_err(irq->bp->dev,
+				   "RX queue restart failed: err=%d\n", err);
+	}
+	rtnl_unlock();
+}
+
+static void bnxt_irq_affinity_release(struct kref __always_unused *ref)
+{
+	struct irq_affinity_notify *notify =
+		(struct irq_affinity_notify *)
+		container_of(ref, struct irq_affinity_notify, kref);
+	struct bnxt_irq *irq;
+
+	irq = container_of(notify, struct bnxt_irq, affinity_notify);
+
+	if (!irq->bp->tph_mode)
+		return;
+
+	if (pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, 0)) {
+		netdev_err(irq->bp->dev,
+			   "Setting ST=0 for MSIX entry %d failed\n",
+			   irq->msix_nr);
+		return;
+	}
+}
+
+static void bnxt_release_irq_notifier(struct bnxt_irq *irq)
+{
+	irq_set_affinity_notifier(irq->vector, NULL);
+}
+
+static void bnxt_register_irq_notifier(struct bnxt *bp, struct bnxt_irq *irq)
+{
+	struct irq_affinity_notify *notify;
+
+	irq->bp = bp;
+
+	/* Nothing to do if TPH is not enabled */
+	if (!bp->tph_mode)
+		return;
+
+	/* Register IRQ affinity notifier */
+	notify = &irq->affinity_notify;
+	notify->irq = irq->vector;
+	notify->notify = bnxt_irq_affinity_notify;
+	notify->release = bnxt_irq_affinity_release;
+
+	irq_set_affinity_notifier(irq->vector, notify);
+}
+
 static void bnxt_free_irq(struct bnxt *bp)
 {
 	struct bnxt_irq *irq;
@@ -10887,11 +10964,18 @@ static void bnxt_free_irq(struct bnxt *bp)
 				free_cpumask_var(irq->cpu_mask);
 				irq->have_cpumask = 0;
 			}
+
+			bnxt_release_irq_notifier(irq);
+
 			free_irq(irq->vector, bp->bnapi[i]);
 		}
 
 		irq->requested = 0;
 	}
+
+	/* Disable TPH support */
+	pcie_disable_tph(bp->pdev);
+	bp->tph_mode = 0;
 }
 
 static int bnxt_request_irq(struct bnxt *bp)
@@ -10911,6 +10995,12 @@ static int bnxt_request_irq(struct bnxt *bp)
 #ifdef CONFIG_RFS_ACCEL
 	rmap = bp->dev->rx_cpu_rmap;
 #endif
+
+	/* Enable TPH support as part of IRQ request */
+	rc = pcie_enable_tph(bp->pdev, PCI_TPH_ST_IV_MODE);
+	if (!rc)
+		bp->tph_mode = PCI_TPH_ST_IV_MODE;
+
 	for (i = 0, j = 0; i < bp->cp_nr_rings; i++) {
 		int map_idx = bnxt_cp_num_to_irq_num(bp, i);
 		struct bnxt_irq *irq = &bp->irq_tbl[map_idx];
@@ -10934,8 +11024,11 @@ static int bnxt_request_irq(struct bnxt *bp)
 
 		if (zalloc_cpumask_var(&irq->cpu_mask, GFP_KERNEL)) {
 			int numa_node = dev_to_node(&bp->pdev->dev);
+			u16 tag;
 
 			irq->have_cpumask = 1;
+			irq->msix_nr = map_idx;
+			irq->ring_nr = i;
 			cpumask_set_cpu(cpumask_local_spread(i, numa_node),
 					irq->cpu_mask);
 			rc = irq_set_affinity_hint(irq->vector, irq->cpu_mask);
@@ -10945,6 +11038,16 @@ static int bnxt_request_irq(struct bnxt *bp)
 					    irq->vector);
 				break;
 			}
+
+			bnxt_register_irq_notifier(bp, irq);
+
+			/* Init ST table entry */
+			if (pcie_tph_get_cpu_st(irq->bp->pdev, TPH_MEM_TYPE_VM,
+						cpumask_first(irq->cpu_mask),
+						&tag))
+				continue;
+
+			pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, tag);
 		}
 	}
 	return rc;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index 69231e85140b..641d25646367 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -1227,6 +1227,11 @@ struct bnxt_irq {
 	u8		have_cpumask:1;
 	char		name[IFNAMSIZ + BNXT_IRQ_NAME_EXTRA];
 	cpumask_var_t	cpu_mask;
+
+	struct bnxt	*bp;
+	int		msix_nr;
+	int		ring_nr;
+	struct irq_affinity_notify affinity_notify;
 };
 
 #define HWRM_RING_ALLOC_TX	0x1
@@ -2183,6 +2188,8 @@ struct bnxt {
 	struct net_device	*dev;
 	struct pci_dev		*pdev;
 
+	u8			tph_mode;
+
 	atomic_t		intr_sem;
 
 	u32			flags;
diff --git a/net/core/netdev_rx_queue.c b/net/core/netdev_rx_queue.c
index e217a5838c87..10e95d7b6892 100644
--- a/net/core/netdev_rx_queue.c
+++ b/net/core/netdev_rx_queue.c
@@ -79,3 +79,4 @@ int netdev_rx_queue_restart(struct net_device *dev, unsigned int rxq_idx)
 
 	return err;
 }
+EXPORT_SYMBOL_GPL(netdev_rx_queue_restart);
-- 
2.46.0

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Jakub Kicinski]

On Fri, 15 Nov 2024 14:04:11 -0600 Wei Huang wrote:
> +static void bnxt_irq_affinity_release(struct kref __always_unused *ref)

unused? you're using it now

> +{
> +	struct irq_affinity_notify *notify =
> +		(struct irq_affinity_notify *)
> +		container_of(ref, struct irq_affinity_notify, kref);

this is ugly, and cast is unnecessary.

> +	struct bnxt_irq *irq;
> +
> +	irq = container_of(notify, struct bnxt_irq, affinity_notify);

since you init irq out of line you can as well init notify here

> +	if (pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, 0)) {

You checked this function can sleep, right? Because rtnl_lock()
will sleep.


Bjorn, do you have a strong preference to have a user of the TPH code
merged as part of 6.13?  We're very close to the merge window, I'm not
sure build bots etc. will have enough time to hammer this code.
My weak preference would be to punt these driver changes to 6.14
avoid all the conflicts and risks (unless Linus gives us another week.)
-- 
pw-bot: nap

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Michael Chan]

[-- Attachment #1: Type: text/plain, Size: 507 bytes --]

On Fri, Nov 15, 2024 at 2:04 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> Bjorn, do you have a strong preference to have a user of the TPH code
> merged as part of 6.13?  We're very close to the merge window, I'm not
> sure build bots etc. will have enough time to hammer this code.
> My weak preference would be to punt these driver changes to 6.14
> avoid all the conflicts and risks (unless Linus gives us another week.)

Driver changes going in through net-next for 6.14 sounds good to us.

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4209 bytes --]

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Bjorn Helgaas]

On Fri, Nov 15, 2024 at 02:04:34PM -0800, Jakub Kicinski wrote:
> ...
> Bjorn, do you have a strong preference to have a user of the TPH code
> merged as part of 6.13?  We're very close to the merge window, I'm not
> sure build bots etc. will have enough time to hammer this code.
> My weak preference would be to punt these driver changes to 6.14
> avoid all the conflicts and risks (unless Linus gives us another week.)

I do not have a preference.  The PCI core changes are queued for
v6.13, so driver changes will be able to go the normal netdev route
for v6.14.

I agree it seems late to add significant things for v6.13.

Bjorn

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Andy Gospodarek]

On Fri, Nov 15, 2024 at 04:20:38PM -0600, Bjorn Helgaas wrote:
> On Fri, Nov 15, 2024 at 02:04:34PM -0800, Jakub Kicinski wrote:
> > ...
> > Bjorn, do you have a strong preference to have a user of the TPH code
> > merged as part of 6.13?  We're very close to the merge window, I'm not
> > sure build bots etc. will have enough time to hammer this code.
> > My weak preference would be to punt these driver changes to 6.14
> > avoid all the conflicts and risks (unless Linus gives us another week.)
> 
> I do not have a preference.  The PCI core changes are queued for
> v6.13, so driver changes will be able to go the normal netdev route
> for v6.14.
> 
> I agree it seems late to add significant things for v6.13.
> 

Excellent.  Thank you!

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Jiri Slaby]

On 15. 11. 24, 23:04, Jakub Kicinski wrote:
> On Fri, 15 Nov 2024 14:04:11 -0600 Wei Huang wrote:
>> +static void bnxt_irq_affinity_release(struct kref __always_unused *ref)
> 
> unused? you're using it now
> 
>> +{
>> +	struct irq_affinity_notify *notify =
>> +		(struct irq_affinity_notify *)
>> +		container_of(ref, struct irq_affinity_notify, kref);
> 
> this is ugly, and cast is unnecessary.
> 
>> +	struct bnxt_irq *irq;
>> +
>> +	irq = container_of(notify, struct bnxt_irq, affinity_notify);
> 
> since you init irq out of line you can as well init notify here
> 
>> +	if (pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, 0)) {
> 
> You checked this function can sleep, right? Because rtnl_lock()
> will sleep.

Based on the above, I assume a new version was expected, but I cannot 
find any. So, Wei Huang, what's the status of this?

thanks,
-- 
js
suse labs

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Wei Huang]

On 1/8/25 05:33, Jiri Slaby wrote:
> On 15. 11. 24, 23:04, Jakub Kicinski wrote:
>> On Fri, 15 Nov 2024 14:04:11 -0600 Wei Huang wrote:
>>> +static void bnxt_irq_affinity_release(struct kref __always_unused *ref)
>>
>> unused? you're using it now
>>
>>> +{
>>> +    struct irq_affinity_notify *notify =
>>> +        (struct irq_affinity_notify *)
>>> +        container_of(ref, struct irq_affinity_notify, kref);
>>
>> this is ugly, and cast is unnecessary.
>>
>>> +    struct bnxt_irq *irq;
>>> +
>>> +    irq = container_of(notify, struct bnxt_irq, affinity_notify);
>>
>> since you init irq out of line you can as well init notify here
>>
>>> +    if (pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, 0)) {
>>
>> You checked this function can sleep, right? Because rtnl_lock()
>> will sleep.
> 
> Based on the above, I assume a new version was expected, but I cannot
> find any. So, Wei Huang, what's the status of this?

Currently the Broadcom team is driving the changes for upstream bnxt. I
will leave this question to them (Andy, Somnath, Michael).

> 
> thanks,

Re: [PATCH V1 1/2] bnxt_en: Add TPH support in BNXT driver

[Michael Chan]

[-- Attachment #1: Type: text/plain, Size: 620 bytes --]

On Thu, Jan 9, 2025 at 11:09 AM Wei Huang <wei.huang2@amd.com> wrote:
>
> On 1/8/25 05:33, Jiri Slaby wrote:
> >
> > Based on the above, I assume a new version was expected, but I cannot
> > find any. So, Wei Huang, what's the status of this?
>
> Currently the Broadcom team is driving the changes for upstream bnxt. I
> will leave this question to them (Andy, Somnath, Michael).
>

We'll be sending the next version around the end of the week.  It will
include the NAPI disable/enable that we discussed here:

https://lore.kernel.org/netdev/5336d624-8d8b-40a6-b732-b020e4a119a2@davidwei.uk/

Thanks.

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4209 bytes --]

[PATCH V1 2/2] bnxt_en: Pass NQ ID to the FW when allocating RX/RX AGG rings

[Wei Huang]

From: Michael Chan <michael.chan@broadcom.com>

Newer firmware can use the NQ ring ID associated with each RX/RX AGG
ring to enable PCIe Steering Tags.  When allocating RX/RX AGG rings,
pass along NR ring ID for the firmware to use.  This information helps
optimize DMA writes by directing them to the cache closer to the CPU
consuming the data, potentially improving the processing speed.  This
change is backward-compatible with older firmware, which will simply
disregard the information.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index beabc4b4a913..b609bb0b87be 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -6811,10 +6811,12 @@ static int hwrm_ring_alloc_send_msg(struct bnxt *bp,
 
 			/* Association of rx ring with stats context */
 			grp_info = &bp->grp_info[ring->grp_idx];
+			req->nq_ring_id = cpu_to_le16(grp_info->cp_fw_ring_id);
 			req->rx_buf_size = cpu_to_le16(bp->rx_buf_use_size);
 			req->stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
 			req->enables |= cpu_to_le32(
-				RING_ALLOC_REQ_ENABLES_RX_BUF_SIZE_VALID);
+				RING_ALLOC_REQ_ENABLES_RX_BUF_SIZE_VALID |
+				RING_ALLOC_REQ_ENABLES_NQ_RING_ID_VALID);
 			if (NET_IP_ALIGN == 2)
 				flags = RING_ALLOC_REQ_FLAGS_RX_SOP_PAD;
 			req->flags = cpu_to_le16(flags);
@@ -6826,11 +6828,13 @@ static int hwrm_ring_alloc_send_msg(struct bnxt *bp,
 			/* Association of agg ring with rx ring */
 			grp_info = &bp->grp_info[ring->grp_idx];
 			req->rx_ring_id = cpu_to_le16(grp_info->rx_fw_ring_id);
+			req->nq_ring_id = cpu_to_le16(grp_info->cp_fw_ring_id);
 			req->rx_buf_size = cpu_to_le16(BNXT_RX_PAGE_SIZE);
 			req->stat_ctx_id = cpu_to_le32(grp_info->fw_stats_ctx);
 			req->enables |= cpu_to_le32(
 				RING_ALLOC_REQ_ENABLES_RX_RING_ID_VALID |
-				RING_ALLOC_REQ_ENABLES_RX_BUF_SIZE_VALID);
+				RING_ALLOC_REQ_ENABLES_RX_BUF_SIZE_VALID |
+				RING_ALLOC_REQ_ENABLES_NQ_RING_ID_VALID);
 		} else {
 			req->ring_type = RING_ALLOC_REQ_RING_TYPE_RX;
 		}
-- 
2.46.0