diff mbox series

[v4,2/2] net/mlx5: support socket direct mode bonding

Message ID 20211014025801.3055244-3-rongweil@nvidia.com (mailing list archive)
State New
Delegated to: Raslan Darawsheh
Headers show
Series support socket direct mode bonding | expand

Checks

Context Check Description
ci/intel-Testing success Testing PASS
ci/Intel-compilation success Compilation OK
ci/iol-aarch64-compile-testing success Testing PASS
ci/iol-aarch64-unit-testing success Testing PASS
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-x86_64-compile-testing success Testing PASS
ci/github-robot: build success github build: passed
ci/iol-x86_64-unit-testing success Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-broadcom-Performance success Performance Testing PASS
ci/iol-broadcom-Functional success Functional Testing PASS
ci/iol-spell-check-testing warning Testing issues
ci/checkpatch success coding style OK

Commit Message

Rongwei Liu Oct. 14, 2021, 2:58 a.m. UTC
In socket direct mode, it's possible to bind any two (maybe four
in future) PCIe devices with IDs like xxxx:xx:xx.x and
yyyy:yy:yy.y. Bonding member interfaces are unnecessary to have
the same PCIe domain/bus/device ID anymore,

Kernel driver uses "system_image_guid" to identify if devices can
be bound together or not. Sysfs "phys_switch_id" is used to get
"system_image_guid" of each network interface.

OFED 5.4+ is required to support "phys_switch_id".

Signed-off-by: Rongwei Liu <rongweil@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
 doc/guides/nics/mlx5.rst               |  4 +++
 doc/guides/rel_notes/release_21_11.rst |  4 +++
 drivers/net/mlx5/linux/mlx5_os.c       | 43 ++++++++++++++++++++------
 3 files changed, 42 insertions(+), 9 deletions(-)
diff mbox series

Patch

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index bae73f42d8..b58236e00a 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -464,6 +464,10 @@  Limitations
   - In order to achieve best insertion rate, application should manage the flows per lcore.
   - Better to disable memory reclaim by setting ``reclaim_mem_mode`` to 0 to accelerate the flow object allocation and release with cache.
 
+- Bonding under socket direct mode
+
+  - Needs OFED 5.4+.
+
 Statistics
 ----------
 
diff --git a/doc/guides/rel_notes/release_21_11.rst b/doc/guides/rel_notes/release_21_11.rst
index dfc2cbdeed..2a6cc765c2 100644
--- a/doc/guides/rel_notes/release_21_11.rst
+++ b/doc/guides/rel_notes/release_21_11.rst
@@ -106,6 +106,10 @@  New Features
   * Added DES-CBC, AES-XCBC-MAC, AES-CMAC and non-HMAC algo support.
   * Added PDCP short MAC-I support.
 
+* **Updated Mellanox mlx5 driver.**
+
+  * Added socket direct mode bonding support.
+
 * **Updated NXP dpaa2_sec crypto PMD.**
 
   * Added PDCP short MAC-I support.
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 3746057673..1d57b934fc 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -2008,6 +2008,8 @@  mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
 	FILE *bond_file = NULL, *file;
 	int pf = -1;
 	int ret;
+	uint8_t cur_guid[32] = {0};
+	uint8_t guid[32] = {0};
 
 	/*
 	 * Try to get master device name. If something goes
@@ -2022,6 +2024,8 @@  mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
 	np = mlx5_nl_portnum(nl_rdma, ibv_dev->name);
 	if (!np)
 		return -1;
+	if (mlx5_get_device_guid(pci_dev, cur_guid, sizeof(cur_guid)) < 0)
+		return -1;
 	/*
 	 * The Master device might not be on the predefined
 	 * port (not on port index 1, it is not garanted),
@@ -2050,6 +2054,7 @@  mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
 		char tmp_str[IF_NAMESIZE + 32];
 		struct rte_pci_addr pci_addr;
 		struct mlx5_switch_info	info;
+		int ret;
 
 		/* Process slave interface names in the loop. */
 		snprintf(tmp_str, sizeof(tmp_str),
@@ -2080,15 +2085,6 @@  mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
 				tmp_str);
 			break;
 		}
-		/* Match PCI address, allows BDF0+pfx or BDFx+pfx. */
-		if (pci_dev->domain == pci_addr.domain &&
-		    pci_dev->bus == pci_addr.bus &&
-		    pci_dev->devid == pci_addr.devid &&
-		    ((pci_dev->function == 0 &&
-		      pci_dev->function + owner == pci_addr.function) ||
-		     (pci_dev->function == owner &&
-		      pci_addr.function == owner)))
-			pf = info.port_name;
 		/* Get ifindex. */
 		snprintf(tmp_str, sizeof(tmp_str),
 			 "/sys/class/net/%s/ifindex", ifname);
@@ -2105,6 +2101,30 @@  mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
 		bond_info->ports[info.port_name].pci_addr = pci_addr;
 		bond_info->ports[info.port_name].ifindex = ifindex;
 		bond_info->n_port++;
+		/*
+		 * Under socket direct mode, bonding will use
+		 * system_image_guid as identification.
+		 * After OFED 5.4, guid is readable (ret >= 0) under sysfs.
+		 * All bonding members should have the same guid even if driver
+		 * is using PCIe BDF.
+		 */
+		ret = mlx5_get_device_guid(&pci_addr, guid, sizeof(guid));
+		if (ret < 0)
+			break;
+		else if (ret > 0) {
+			if (!memcmp(guid, cur_guid, sizeof(guid)) &&
+			    owner == info.port_name &&
+			    (owner != 0 || (owner == 0 &&
+			    !rte_pci_addr_cmp(pci_dev, &pci_addr))))
+				pf = info.port_name;
+		} else if (pci_dev->domain == pci_addr.domain &&
+		    pci_dev->bus == pci_addr.bus &&
+		    pci_dev->devid == pci_addr.devid &&
+		    ((pci_dev->function == 0 &&
+		      pci_dev->function + owner == pci_addr.function) ||
+		     (pci_dev->function == owner &&
+		      pci_addr.function == owner)))
+			pf = info.port_name;
 	}
 	if (pf >= 0) {
 		/* Get bond interface info */
@@ -2117,6 +2137,11 @@  mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
 			DRV_LOG(INFO, "PF device %u, bond device %u(%s)",
 				ifindex, bond_info->ifindex, bond_info->ifname);
 	}
+	if (owner == 0 && pf != 0) {
+		DRV_LOG(INFO, "PCIe instance %04x:%02x:%02x.%x isn't bonding owner",
+				pci_dev->domain, pci_dev->bus, pci_dev->devid,
+				pci_dev->function);
+	}
 	return pf;
 }