[PATCH v26 09/26] drivers: add process-type guards for secondary process

[Hemant Agrawal]

From: Prashant Gupta <prashant.gupta_3 at nxp.com>

The DPAA QDMA and net/dpaa drivers run primary-only code from paths that
a secondary process also executes.

In dma/dpaa, struct fsl_qdma_engine is allocated from shared memory, but
the register bases stored in it are process-private mmap()s of the QDMA
CCSR region created by the primary. A secondary that probes the device
would advertise a dmadev whose MMIO pointers are not mapped in its own
address space. Reject the probe with -ENOTSUP until a secondary can
establish its own mapping.

In net/dpaa, rte_dpaa_remove() also does primary-only bookkeeping: it
decrements the driver device count and frees the shared Tx
scatter-gather mempool when that count reaches zero. Both are set up
only on the primary probe path, so a secondary must not run them. Let a
secondary release just its local ethdev port and return.

Signed-off-by: Prashant Gupta <prashant.gupta_3 at nxp.com>
---
 drivers/bus/dpaa/dpaa_bus.c    |  3 ++-
 drivers/dma/dpaa/dpaa_qdma.c   | 14 ++++++++++++++
 drivers/net/dpaa/dpaa_ethdev.c |  6 +++++-
 3 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index 5442f8f893..e9bfc7b956 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -804,7 +804,8 @@ dpaa_bus_probe_device(struct rte_driver *drv, struct rte_device *dev)
 	int ret;
 
 	ret = dpaa_drv->probe(dpaa_drv, dpaa_dev);
-	if (ret != 0)
+	/* A positive return means the driver declined the device. */
+	if (ret < 0)
 		DPAA_BUS_ERR("unable to probe: %s", dpaa_dev->name);
 
 	return ret;
diff --git a/drivers/dma/dpaa/dpaa_qdma.c b/drivers/dma/dpaa/dpaa_qdma.c
index e2cb157c43..0234042ded 100644
--- a/drivers/dma/dpaa/dpaa_qdma.c
+++ b/drivers/dma/dpaa/dpaa_qdma.c
@@ -1433,6 +1433,20 @@ dpaa_qdma_probe(__rte_unused struct rte_dpaa_driver *dpaa_drv,
 	struct rte_dma_dev *dmadev;
 	int ret;
 
+	/*
+	 * struct fsl_qdma_engine lives in memory shared with the primary, and
+	 * reg_base in it is a process-private mmap of the QDMA CCSR region that
+	 * is only valid in the primary. Until a secondary can establish its own
+	 * mapping, do not advertise the device here: registering it would hand
+	 * out a usable dmadev whose MMIO pointers are not mapped in this
+	 * process.
+	 */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
+		DPAA_QDMA_WARN("%s: not supported in secondary process",
+			dpaa_dev->device.name);
+		return -ENOTSUP;
+	}
+
 	dmadev = rte_dma_pmd_allocate(dpaa_dev->device.name,
 				      rte_socket_id(),
 				      sizeof(struct fsl_qdma_engine));
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 1688acfb30..049723cb19 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -1,7 +1,7 @@
 /* SPDX-License-Identifier: BSD-3-Clause
  *
  *   Copyright 2016 Freescale Semiconductor, Inc. All rights reserved.
- *   Copyright 2017-2020,2022-2025 NXP
+ *   Copyright 2017-2020,2022-2026 NXP
  *
  */
 /* System headers */
@@ -2691,6 +2691,10 @@ rte_dpaa_remove(struct rte_dpaa_device *dpaa_dev)
 	PMD_INIT_FUNC_TRACE();
 
 	eth_dev = rte_eth_dev_allocated(dpaa_dev->device.name);
+
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return eth_dev != NULL ? rte_eth_dev_release_port(eth_dev) : 0;
+
 	if (eth_dev != NULL && eth_dev->state != RTE_ETH_DEV_UNUSED) {
 		ret = dpaa_eth_dev_close(eth_dev);
 		if (ret != 0)
-- 
2.25.1