Skip to contentSkip to conversation

Allow dyn MSI-X vector allocation of MANA

PATCH Superseded Shradha Gupta 2 participants 8 messages Jun 9, 2025 – Jun 11, 2025

Subsystems (7)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover Allow dyn MSI-X vector allocation of MANA

In this patchset we want to enable the MANA driver to be able to allocate MSI-X vectors in PCI dynamically.

The first patch exports pci_msix_prepare_desc() in PCI to be able to correctly prepare descriptors for dynamically added MSI-X vectors.

The second patch adds the support of dynamic vector allocation in pci-hyperv PCI controller by enabling the MSI_FLAG_PCI_MSIX_ALLOC_DYN flag and using the pci_msix_prepare_desc() exported in first patch.

The third patch adds a detailed description of the irq_setup(), to help understand the function design better.

The fourth patch is a preparation patch for mana changes to support dynamic IRQ allocation. It contains changes in irq_setup() to allow skipping first sibling CPU sets, in case certain IRQs are already affinitized to them.

The fifth patch has the changes in MANA driver to be able to allocate MSI-X vectors dynamically. If the support does not exist it defaults to older behavior.

Since this patchset has patches from PCI and net tree, I am not entirely sure what should be the target tree. Any suggestions/recommendations on the same are welcomed.

---
Changes in v5
* Added Yury as Author of the 3rd patch
* Fixed base commit information in the cover letter
* Correctly initialized start_irqs, so that it is cleaned properly
* rearranged the cpu_lock to minimize the critical section
---
Change in v4
* add a patch describing the functionality of irq_setup() through a comment
* In irq_setup(), avoid using a label next_cpumask:
* modify the changes in MANA patch about restructuring the error handling path in mana_gd_setup_dyn_irqs()
* modify the mana_gd_setup_irqs() to simplify handling around start_irq_index
* add warning if an invalid gic is returned
* place the xa_destroy() cleanup in mana_gd_remove
---
Changes in v3
* split the 3rd patch into preparation patch around irq_setup() and changes in mana driver to allow dynamic IRQ allocation
* Add arm64 support for dynamic MSI-X allocation in pci_hyperv controller
---
Changes in v2
* split the first patch into two(exporting the preapre_desc func and using the function and flag in pci-hyperv)
* replace 'pci vectors' by 'MSI-X vectors'
* Change the cover letter description to align with changes made
---

Shradha Gupta (5):
PCI/MSI: Export pci_msix_prepare_desc() for dynamic MSI-X allocations
PCI: hv: Allow dynamic MSI-X vector allocation
net: mana: explain irq_setup() algorithm
net: mana: Allow irq_setup() to skip cpus for affinity
net: mana: Allocate MSI-X vectors dynamically

 .../net/ethernet/microsoft/mana/gdma_main.c   | 363 ++++++++++++++----
 drivers/pci/controller/pci-hyperv.c           |   3 +-
 drivers/pci/msi/irqdomain.c                   |   5 +-
 include/linux/msi.h                           |   2 +
 include/net/mana/gdma.h                       |   8 +-
 5 files changed, 295 insertions(+), 86 deletions(-)

base-commit: 2c7e4a2663a1ab5a740c59c31991579b6b865a26

Signature · 1 line
2.34.1
#1 lore raw

PATCH v5 1/5 PCI/MSI: Export pci_msix_prepare_desc() for dynamic MSI-X allocations

For supporting dynamic MSI-X vector allocation by PCI controllers, enabling the flag MSI_FLAG_PCI_MSIX_ALLOC_DYN is not enough, msix_prepare_msi_desc() to prepare the MSI descriptor is also needed.

Export pci_msix_prepare_desc() to allow PCI controllers to support dynamic MSI-X vector allocation.

---
 drivers/pci/msi/irqdomain.c | 5 +++--
 include/linux/msi.h         | 2 ++
 2 files changed, 5 insertions(+), 2 deletions(-)
drivers/pci/msi/irqdomain.c +3 -2
index d7ba8795d60f..43129aa6d6c7 100644
@@ -222,13 +222,14 @@ static void pci_irq_unmask_msix(struct irq_data *data)
 	pci_msix_unmask(irq_data_get_msi_desc(data));
 }
 
-static void pci_msix_prepare_desc(struct irq_domain *domain, msi_alloc_info_t *arg,
-				  struct msi_desc *desc)
+void pci_msix_prepare_desc(struct irq_domain *domain, msi_alloc_info_t *arg,
+			   struct msi_desc *desc)
 {
 	/* Don't fiddle with preallocated MSI descriptors */
 	if (!desc->pci.mask_base)
 		msix_prepare_msi_desc(to_pci_dev(desc->dev), desc);
 }
+EXPORT_SYMBOL_GPL(pci_msix_prepare_desc);
 
 static const struct msi_domain_template pci_msix_template = {
 	.chip = {
include/linux/msi.h +2 -0
index 6863540f4b71..7f254bde5426 100644
@@ -706,6 +706,8 @@ struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
 					     struct irq_domain *parent);
 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev);
 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev);
+void pci_msix_prepare_desc(struct irq_domain *domain, msi_alloc_info_t *arg,
+			   struct msi_desc *desc);
 #else /* CONFIG_PCI_MSI */
 static inline struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
 {
Signature · 1 line
2.34.1
#2 lore raw

PATCH v5 2/5 PCI: hv: Allow dynamic MSI-X vector allocation

In reply to Shradha Gupta

Allow dynamic MSI-X vector allocation for pci_hyperv PCI controller by adding support for the flag MSI_FLAG_PCI_MSIX_ALLOC_DYN and using pci_msix_prepare_desc() to prepare the MSI-X descriptors.

Feature support added for both x86 and ARM64

---
Changes in v4:
* use the same prepare_desc() callback for arm and x86
---
Changes in v3:
* Add arm64 support
---
Changes in v2:
* split the patch to keep changes in PCI and pci_hyperv controller seperate
* replace strings "pci vectors" by "MSI-X vectors"

---
 drivers/pci/controller/pci-hyperv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/pci/controller/pci-hyperv.c +2 -1
index ef5d655a0052..86ca041bf74a 100644
@@ -2119,6 +2119,7 @@ static struct irq_chip hv_msi_irq_chip = {
 static struct msi_domain_ops hv_msi_ops = {
 	.msi_prepare	= hv_msi_prepare,
 	.msi_free	= hv_msi_free,
+	.prepare_desc	= pci_msix_prepare_desc,
 };
 
 /**
@@ -2140,7 +2141,7 @@ static int hv_pcie_init_irq_domain(struct hv_pcibus_device *hbus)
 	hbus->msi_info.ops = &hv_msi_ops;
 	hbus->msi_info.flags = (MSI_FLAG_USE_DEF_DOM_OPS |
 		MSI_FLAG_USE_DEF_CHIP_OPS | MSI_FLAG_MULTI_PCI_MSI |
-		MSI_FLAG_PCI_MSIX);
+		MSI_FLAG_PCI_MSIX | MSI_FLAG_PCI_MSIX_ALLOC_DYN);
 	hbus->msi_info.handler = FLOW_HANDLER;
 	hbus->msi_info.handler_name = FLOW_NAME;
 	hbus->msi_info.data = hbus;
Signature · 1 line
2.34.1
#3 lore raw

PATCH v5 3/5 net: mana: explain irq_setup() algorithm

In reply to Shradha Gupta

From: Yury Norov <[email protected]>

Commit 91bfe210e196 ("net: mana: add a function to spread IRQs per CPUs") added the irq_setup() function that distributes IRQs on CPUs according to a tricky heuristic. The corresponding commit message explains the heuristic.

Duplicate it in the source code to make available for readers without digging git in history. Also, add more detailed explanation about how the heuristics is implemented.

---
Changes in v5:
* Corrected the Authur of the patch

---
 .../net/ethernet/microsoft/mana/gdma_main.c   | 41 +++++++++++++++++++
 1 file changed, 41 insertions(+)
drivers/net/ethernet/microsoft/mana/gdma_main.c +41 -0
index 3504507477c6..6c4e143972a1 100644
@@ -1288,6 +1288,47 @@ void mana_gd_free_res_map(struct gdma_resource *r)
 	r->size = 0;
 }
 
+/*
+ * Spread on CPUs with the following heuristics:
+ *
+ * 1. No more than one IRQ per CPU, if possible;
+ * 2. NUMA locality is the second priority;
+ * 3. Sibling dislocality is the last priority.
+ *
+ * Let's consider this topology:
+ *
+ * Node            0               1
+ * Core        0       1       2       3
+ * CPU       0   1   2   3   4   5   6   7
+ *
+ * The most performant IRQ distribution based on the above topology
+ * and heuristics may look like this:
+ *
+ * IRQ     Nodes   Cores   CPUs
+ * 0       1       0       0-1
+ * 1       1       1       2-3
+ * 2       1       0       0-1
+ * 3       1       1       2-3
+ * 4       2       2       4-5
+ * 5       2       3       6-7
+ * 6       2       2       4-5
+ * 7       2       3       6-7
+ *
+ * The heuristics is implemented as follows.
+ *
+ * The outer for_each() loop resets the 'weight' to the actual number
+ * of CPUs in the hop. Then inner for_each() loop decrements it by the
+ * number of sibling groups (cores) while assigning first set of IRQs
+ * to each group. IRQs 0 and 1 above are distributed this way.
+ *
+ * Now, because NUMA locality is more important, we should walk the
+ * same set of siblings and assign 2nd set of IRQs (2 and 3), and it's
+ * implemented by the medium while() loop. We do like this unless the
+ * number of IRQs assigned on this hop will not become equal to number
+ * of CPUs in the hop (weight == 0). Then we switch to the next hop and
+ * do the same thing.
+ */
+
 static int irq_setup(unsigned int *irqs, unsigned int len, int node)
 {
 	const struct cpumask *next, *prev = cpu_none_mask;
Signature · 1 line
2.34.1
#4 lore raw

PATCH v5 4/5 net: mana: Allow irq_setup() to skip cpus for affinity

In reply to Shradha Gupta

In order to prepare the MANA driver to allocate the MSI-X IRQs dynamically, we need to enhance irq_setup() to allow skipping affinitizing IRQs to the first CPU sibling group.

This would be for cases when the number of IRQs is less than or equal to the number of online CPUs. In such cases for dynamically added IRQs the first CPU sibling group would already be affinitized with HWC IRQ.

---
Changes in v4
* fix commit description
* avoided using next_cpumask: label in the irq_setup()

---
 drivers/net/ethernet/microsoft/mana/gdma_main.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
drivers/net/ethernet/microsoft/mana/gdma_main.c +12 -4
index 6c4e143972a1..6e468c0f2c40 100644
@@ -1329,7 +1329,8 @@ void mana_gd_free_res_map(struct gdma_resource *r)
  * do the same thing.
  */
 
-static int irq_setup(unsigned int *irqs, unsigned int len, int node)
+static int irq_setup(unsigned int *irqs, unsigned int len, int node,
+		     bool skip_first_cpu)
 {
 	const struct cpumask *next, *prev = cpu_none_mask;
 	cpumask_var_t cpus __free(free_cpumask_var);
@@ -1344,11 +1345,18 @@ static int irq_setup(unsigned int *irqs, unsigned int len, int node)
 		while (weight > 0) {
 			cpumask_andnot(cpus, next, prev);
 			for_each_cpu(cpu, cpus) {
+				cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
+				--weight;
+
+				if (unlikely(skip_first_cpu)) {
+					skip_first_cpu = false;
+					continue;
+				}
+
 				if (len-- == 0)
 					goto done;
+
 				irq_set_affinity_and_hint(*irqs++, topology_sibling_cpumask(cpu));
-				cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
-				--weight;
 			}
 		}
 		prev = next;
@@ -1444,7 +1452,7 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
 		}
 	}
 
-	err = irq_setup(irqs, (nvec - start_irq_index), gc->numa_node);
+	err = irq_setup(irqs, nvec - start_irq_index, gc->numa_node, false);
 	if (err)
 		goto free_irq;
 
Signature · 1 line
2.34.1
#5 lore raw

PATCH v5 5/5 net: mana: Allocate MSI-X vectors dynamically

In reply to Shradha Gupta

Currently, the MANA driver allocates MSI-X vectors statically based on MANA_MAX_NUM_QUEUES and num_online_cpus() values and in some cases ends up allocating more vectors than it needs. This is because, by this time we do not have a HW channel and do not know how many IRQs should be allocated.

To avoid this, we allocate 1 MSI-X vector during the creation of HWC and after getting the value supported by hardware, dynamically add the remaining MSI-X vectors.

---
Changes in v5:
* Correctly initialized start_irqs, so that it is cleaned properly
* rearranged the cpu_lock to minimize the critical section
---
Changes in v4:
* added BUG_ON at appropriate places
* moved xa_destroy to mana_gd_remove()
* rearragned the cleanup logic in mana_gd_setup_dyn_irqs()
* simplified processing around start_irq_index in mana_gd_setup_irqs()
* return 0 instead of return err as appropriate
---
Changes in v3:
* implemented irq_contexts as xarrays rather than list
* split the patch to create a perparation patch around irq_setup()
* add log when IRQ allocation/setup for remaining IRQs fails
---
Changes in v2:
* Use string 'MSI-X vectors' instead of 'pci vectors'
* make skip-cpu a bool instead of int
* rearrange the comment arout skip_cpu variable appropriately
* update the capability bit for driver indicating dynamic IRQ allocation
* enforced max line length to 80
* enforced RCT convention
* initialized gic to NULL, for when there is a possibility of gic not being populated correctly

---
 .../net/ethernet/microsoft/mana/gdma_main.c   | 311 +++++++++++++-----
 include/net/mana/gdma.h                       |   8 +-
 2 files changed, 235 insertions(+), 84 deletions(-)
drivers/net/ethernet/microsoft/mana/gdma_main.c +229 -82
index 6e468c0f2c40..d0040c12b8a2 100644
@@ -6,6 +6,8 @@
 #include <linux/pci.h>
 #include <linux/utsname.h>
 #include <linux/version.h>
+#include <linux/msi.h>
+#include <linux/irqdomain.h>
 
 #include <net/mana/mana.h>
 
@@ -80,8 +82,15 @@ static int mana_gd_query_max_resources(struct pci_dev *pdev)
 		return err ? err : -EPROTO;
 	}
 
-	if (gc->num_msix_usable > resp.max_msix)
-		gc->num_msix_usable = resp.max_msix;
+	if (!pci_msix_can_alloc_dyn(pdev)) {
+		if (gc->num_msix_usable > resp.max_msix)
+			gc->num_msix_usable = resp.max_msix;
+	} else {
+		/* If dynamic allocation is enabled we have already allocated
+		 * hwc msi
+		 */
+		gc->num_msix_usable = min(resp.max_msix, num_online_cpus() + 1);
+	}
 
 	if (gc->num_msix_usable <= 1)
 		return -ENOSPC;
@@ -483,7 +492,9 @@ static int mana_gd_register_irq(struct gdma_queue *queue,
 	}
 
 	queue->eq.msix_index = msi_index;
-	gic = &gc->irq_contexts[msi_index];
+	gic = xa_load(&gc->irq_contexts, msi_index);
+	if (WARN_ON(!gic))
+		return -EINVAL;
 
 	spin_lock_irqsave(&gic->lock, flags);
 	list_add_rcu(&queue->entry, &gic->eq_list);
@@ -508,7 +519,10 @@ static void mana_gd_deregiser_irq(struct gdma_queue *queue)
 	if (WARN_ON(msix_index >= gc->num_msix_usable))
 		return;
 
-	gic = &gc->irq_contexts[msix_index];
+	gic = xa_load(&gc->irq_contexts, msix_index);
+	if (WARN_ON(!gic))
+		return;
+
 	spin_lock_irqsave(&gic->lock, flags);
 	list_for_each_entry_rcu(eq, &gic->eq_list, entry) {
 		if (queue == eq) {
@@ -1366,47 +1380,108 @@ static int irq_setup(unsigned int *irqs, unsigned int len, int node,
 	return 0;
 }
 
-static int mana_gd_setup_irqs(struct pci_dev *pdev)
+static int mana_gd_setup_dyn_irqs(struct pci_dev *pdev, int nvec)
 {
 	struct gdma_context *gc = pci_get_drvdata(pdev);
-	unsigned int max_queues_per_port;
 	struct gdma_irq_context *gic;
-	unsigned int max_irqs, cpu;
-	int start_irq_index = 1;
-	int nvec, *irqs, irq;
-	int err, i = 0, j;
+	bool skip_first_cpu = false;
+	int *irqs, irq, err, i;
 
-	cpus_read_lock();
-	max_queues_per_port = num_online_cpus();
-	if (max_queues_per_port > MANA_MAX_NUM_QUEUES)
-		max_queues_per_port = MANA_MAX_NUM_QUEUES;
+	irqs = kmalloc_array(nvec, sizeof(int), GFP_KERNEL);
+	if (!irqs)
+		return -ENOMEM;
+
+	/*
+	 * While processing the next pci irq vector, we start with index 1,
+	 * as IRQ vector at index 0 is already processed for HWC.
+	 * However, the population of irqs array starts with index 0, to be
+	 * further used in irq_setup()
+	 */
+	for (i = 1; i <= nvec; i++) {
+		gic = kzalloc(sizeof(*gic), GFP_KERNEL);
+		if (!gic) {
+			err = -ENOMEM;
+			goto free_irq;
+		}
+		gic->handler = mana_gd_process_eq_events;
+		INIT_LIST_HEAD(&gic->eq_list);
+		spin_lock_init(&gic->lock);
 
-	/* Need 1 interrupt for the Hardware communication Channel (HWC) */
-	max_irqs = max_queues_per_port + 1;
+		snprintf(gic->name, MANA_IRQ_NAME_SZ, "mana_q%d@pci:%s",
+			 i - 1, pci_name(pdev));
 
-	nvec = pci_alloc_irq_vectors(pdev, 2, max_irqs, PCI_IRQ_MSIX);
-	if (nvec < 0) {
-		cpus_read_unlock();
-		return nvec;
+		/* one pci vector is already allocated for HWC */
+		irqs[i - 1] = pci_irq_vector(pdev, i);
+		if (irqs[i - 1] < 0) {
+			err = irqs[i - 1];
+			goto free_current_gic;
+		}
+
+		err = request_irq(irqs[i - 1], mana_gd_intr, 0, gic->name, gic);
+		if (err)
+			goto free_current_gic;
+
+		xa_store(&gc->irq_contexts, i, gic, GFP_KERNEL);
 	}
-	if (nvec <= num_online_cpus())
-		start_irq_index = 0;
 
-	irqs = kmalloc_array((nvec - start_irq_index), sizeof(int), GFP_KERNEL);
-	if (!irqs) {
-		err = -ENOMEM;
-		goto free_irq_vector;
+	/*
+	 * When calling irq_setup() for dynamically added IRQs, if number of
+	 * CPUs is more than or equal to allocated MSI-X, we need to skip the
+	 * first CPU sibling group since they are already affinitized to HWC IRQ
+	 */
+	cpus_read_lock();
+	if (gc->num_msix_usable <= num_online_cpus())
+		skip_first_cpu = true;
+
+	err = irq_setup(irqs, nvec, gc->numa_node, skip_first_cpu);
+	if (err) {
+		cpus_read_unlock();
+		goto free_irq;
 	}
 
-	gc->irq_contexts = kcalloc(nvec, sizeof(struct gdma_irq_context),
-				   GFP_KERNEL);
-	if (!gc->irq_contexts) {
-		err = -ENOMEM;
-		goto free_irq_array;
+	cpus_read_unlock();
+	kfree(irqs);
+	return 0;
+
+free_current_gic:
+	kfree(gic);
+free_irq:
+	for (i -= 1; i > 0; i--) {
+		irq = pci_irq_vector(pdev, i);
+		gic = xa_load(&gc->irq_contexts, i);
+		if (WARN_ON(!gic))
+			continue;
+
+		irq_update_affinity_hint(irq, NULL);
+		free_irq(irq, gic);
+		xa_erase(&gc->irq_contexts, i);
+		kfree(gic);
 	}
+	kfree(irqs);
+	return err;
+}
+
+static int mana_gd_setup_irqs(struct pci_dev *pdev, int nvec)
+{
+	struct gdma_context *gc = pci_get_drvdata(pdev);
+	struct gdma_irq_context *gic;
+	int *irqs, *start_irqs, irq;
+	unsigned int cpu;
+	int err, i;
+
+	irqs = kmalloc_array(nvec, sizeof(int), GFP_KERNEL);
+	if (!irqs)
+		return -ENOMEM;
+
+	start_irqs = irqs;
 
 	for (i = 0; i < nvec; i++) {
-		gic = &gc->irq_contexts[i];
+		gic = kzalloc(sizeof(*gic), GFP_KERNEL);
+		if (!gic) {
+			err = -ENOMEM;
+			goto free_irq;
+		}
+
 		gic->handler = mana_gd_process_eq_events;
 		INIT_LIST_HEAD(&gic->eq_list);
 		spin_lock_init(&gic->lock);
@@ -1418,69 +1493,128 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
 			snprintf(gic->name, MANA_IRQ_NAME_SZ, "mana_q%d@pci:%s",
 				 i - 1, pci_name(pdev));
 
-		irq = pci_irq_vector(pdev, i);
-		if (irq < 0) {
-			err = irq;
-			goto free_irq;
+		irqs[i] = pci_irq_vector(pdev, i);
+		if (irqs[i] < 0) {
+			err = irqs[i];
+			goto free_current_gic;
 		}
 
-		if (!i) {
-			err = request_irq(irq, mana_gd_intr, 0, gic->name, gic);
-			if (err)
-				goto free_irq;
-
-			/* If number of IRQ is one extra than number of online CPUs,
-			 * then we need to assign IRQ0 (hwc irq) and IRQ1 to
-			 * same CPU.
-			 * Else we will use different CPUs for IRQ0 and IRQ1.
-			 * Also we are using cpumask_local_spread instead of
-			 * cpumask_first for the node, because the node can be
-			 * mem only.
-			 */
-			if (start_irq_index) {
-				cpu = cpumask_local_spread(i, gc->numa_node);
-				irq_set_affinity_and_hint(irq, cpumask_of(cpu));
-			} else {
-				irqs[start_irq_index] = irq;
-			}
-		} else {
-			irqs[i - start_irq_index] = irq;
-			err = request_irq(irqs[i - start_irq_index], mana_gd_intr, 0,
-					  gic->name, gic);
-			if (err)
-				goto free_irq;
-		}
+		err = request_irq(irqs[i], mana_gd_intr, 0, gic->name, gic);
+		if (err)
+			goto free_current_gic;
+
+		xa_store(&gc->irq_contexts, i, gic, GFP_KERNEL);
 	}
 
-	err = irq_setup(irqs, nvec - start_irq_index, gc->numa_node, false);
-	if (err)
+	/* If number of IRQ is one extra than number of online CPUs,
+	 * then we need to assign IRQ0 (hwc irq) and IRQ1 to
+	 * same CPU.
+	 * Else we will use different CPUs for IRQ0 and IRQ1.
+	 * Also we are using cpumask_local_spread instead of
+	 * cpumask_first for the node, because the node can be
+	 * mem only.
+	 */
+	cpus_read_lock();
+	if (nvec > num_online_cpus()) {
+		cpu = cpumask_local_spread(0, gc->numa_node);
+		irq_set_affinity_and_hint(irqs[0], cpumask_of(cpu));
+		irqs++;
+		nvec -= 1;
+	}
+
+	err = irq_setup(irqs, nvec, gc->numa_node, false);
+	if (err) {
+		cpus_read_unlock();
 		goto free_irq;
+	}
 
-	gc->max_num_msix = nvec;
-	gc->num_msix_usable = nvec;
 	cpus_read_unlock();
-	kfree(irqs);
+	kfree(start_irqs);
 	return 0;
 
+free_current_gic:
+	kfree(gic);
 free_irq:
-	for (j = i - 1; j >= 0; j--) {
-		irq = pci_irq_vector(pdev, j);
-		gic = &gc->irq_contexts[j];
+	for (i -= 1; i >= 0; i--) {
+		irq = pci_irq_vector(pdev, i);
+		gic = xa_load(&gc->irq_contexts, i);
+		if (WARN_ON(!gic))
+			continue;
 
 		irq_update_affinity_hint(irq, NULL);
 		free_irq(irq, gic);
+		xa_erase(&gc->irq_contexts, i);
+		kfree(gic);
 	}
 
-	kfree(gc->irq_contexts);
-	gc->irq_contexts = NULL;
-free_irq_array:
-	kfree(irqs);
-free_irq_vector:
-	cpus_read_unlock();
-	pci_free_irq_vectors(pdev);
+	kfree(start_irqs);
 	return err;
 }
 
+static int mana_gd_setup_hwc_irqs(struct pci_dev *pdev)
+{
+	struct gdma_context *gc = pci_get_drvdata(pdev);
+	unsigned int max_irqs, min_irqs;
+	int nvec, err;
+
+	if (pci_msix_can_alloc_dyn(pdev)) {
+		max_irqs = 1;
+		min_irqs = 1;
+	} else {
+		/* Need 1 interrupt for HWC */
+		max_irqs = min(num_online_cpus(), MANA_MAX_NUM_QUEUES) + 1;
+		min_irqs = 2;
+	}
+
+	nvec = pci_alloc_irq_vectors(pdev, min_irqs, max_irqs, PCI_IRQ_MSIX);
+	if (nvec < 0)
+		return nvec;
+
+	err = mana_gd_setup_irqs(pdev, nvec);
+	if (err) {
+		pci_free_irq_vectors(pdev);
+		return err;
+	}
+
+	gc->num_msix_usable = nvec;
+	gc->max_num_msix = nvec;
+
+	return 0;
+}
+
+static int mana_gd_setup_remaining_irqs(struct pci_dev *pdev)
+{
+	struct gdma_context *gc = pci_get_drvdata(pdev);
+	struct msi_map irq_map;
+	int max_irqs, i, err;
+
+	if (!pci_msix_can_alloc_dyn(pdev))
+		/* remain irqs are already allocated with HWC IRQ */
+		return 0;
+
+	/* allocate only remaining IRQs*/
+	max_irqs = gc->num_msix_usable - 1;
+
+	for (i = 1; i <= max_irqs; i++) {
+		irq_map = pci_msix_alloc_irq_at(pdev, i, NULL);
+		if (!irq_map.virq) {
+			err = irq_map.index;
+			/* caller will handle cleaning up all allocated
+			 * irqs, after HWC is destroyed
+			 */
+			return err;
+		}
+	}
+
+	err = mana_gd_setup_dyn_irqs(pdev, max_irqs);
+	if (err)
+		return err;
+
+	gc->max_num_msix = gc->max_num_msix + max_irqs;
+
+	return 0;
+}
+
 static void mana_gd_remove_irqs(struct pci_dev *pdev)
 {
 	struct gdma_context *gc = pci_get_drvdata(pdev);
@@ -1495,19 +1629,21 @@ static void mana_gd_remove_irqs(struct pci_dev *pdev)
 		if (irq < 0)
 			continue;
 
-		gic = &gc->irq_contexts[i];
+		gic = xa_load(&gc->irq_contexts, i);
+		if (WARN_ON(!gic))
+			continue;
 
 		/* Need to clear the hint before free_irq */
 		irq_update_affinity_hint(irq, NULL);
 		free_irq(irq, gic);
+		xa_erase(&gc->irq_contexts, i);
+		kfree(gic);
 	}
 
 	pci_free_irq_vectors(pdev);
 
 	gc->max_num_msix = 0;
 	gc->num_msix_usable = 0;
-	kfree(gc->irq_contexts);
-	gc->irq_contexts = NULL;
 }
 
 static int mana_gd_setup(struct pci_dev *pdev)
@@ -1522,9 +1658,10 @@ static int mana_gd_setup(struct pci_dev *pdev)
 	if (!gc->service_wq)
 		return -ENOMEM;
 
-	err = mana_gd_setup_irqs(pdev);
+	err = mana_gd_setup_hwc_irqs(pdev);
 	if (err) {
-		dev_err(gc->dev, "Failed to setup IRQs: %d\n", err);
+		dev_err(gc->dev, "Failed to setup IRQs for HWC creation: %d\n",
+			err);
 		goto free_workqueue;
 	}
 
@@ -1540,6 +1677,12 @@ static int mana_gd_setup(struct pci_dev *pdev)
 	if (err)
 		goto destroy_hwc;
 
+	err = mana_gd_setup_remaining_irqs(pdev);
+	if (err) {
+		dev_err(gc->dev, "Failed to setup remaining IRQs: %d", err);
+		goto destroy_hwc;
+	}
+
 	err = mana_gd_detect_devices(pdev);
 	if (err)
 		goto destroy_hwc;
@@ -1620,6 +1763,7 @@ static int mana_gd_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	gc->is_pf = mana_is_pf(pdev->device);
 	gc->bar0_va = bar0_va;
 	gc->dev = &pdev->dev;
+	xa_init(&gc->irq_contexts);
 
 	if (gc->is_pf)
 		gc->mana_pci_debugfs = debugfs_create_dir("0", mana_debugfs_root);
@@ -1654,6 +1798,7 @@ static int mana_gd_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	 */
 	debugfs_remove_recursive(gc->mana_pci_debugfs);
 	gc->mana_pci_debugfs = NULL;
+	xa_destroy(&gc->irq_contexts);
 	pci_iounmap(pdev, bar0_va);
 free_gc:
 	pci_set_drvdata(pdev, NULL);
@@ -1679,6 +1824,8 @@ static void mana_gd_remove(struct pci_dev *pdev)
 
 	gc->mana_pci_debugfs = NULL;
 
+	xa_destroy(&gc->irq_contexts);
+
 	pci_iounmap(pdev, gc->bar0_va);
 
 	vfree(gc);
include/net/mana/gdma.h +6 -2
index 3ce56a816425..87162ba96d91 100644
@@ -388,7 +388,7 @@ struct gdma_context {
 	unsigned int		max_num_queues;
 	unsigned int		max_num_msix;
 	unsigned int		num_msix_usable;
-	struct gdma_irq_context	*irq_contexts;
+	struct xarray		irq_contexts;
 
 	/* L2 MTU */
 	u16 adapter_mtu;
@@ -578,12 +578,16 @@ enum {
 /* Driver can handle holes (zeros) in the device list */
 #define GDMA_DRV_CAP_FLAG_1_DEV_LIST_HOLES_SUP BIT(11)
 
+/* Driver supports dynamic MSI-X vector allocation */
+#define GDMA_DRV_CAP_FLAG_1_DYNAMIC_IRQ_ALLOC_SUPPORT BIT(13)
+
 #define GDMA_DRV_CAP_FLAGS1 \
 	(GDMA_DRV_CAP_FLAG_1_EQ_SHARING_MULTI_VPORT | \
 	 GDMA_DRV_CAP_FLAG_1_NAPI_WKDONE_FIX | \
 	 GDMA_DRV_CAP_FLAG_1_HWC_TIMEOUT_RECONFIG | \
 	 GDMA_DRV_CAP_FLAG_1_VARIABLE_INDIRECTION_TABLE_SUPPORT | \
-	 GDMA_DRV_CAP_FLAG_1_DEV_LIST_HOLES_SUP)
+	 GDMA_DRV_CAP_FLAG_1_DEV_LIST_HOLES_SUP | \
+	 GDMA_DRV_CAP_FLAG_1_DYNAMIC_IRQ_ALLOC_SUPPORT)
 
 #define GDMA_DRV_CAP_FLAGS2 0
 
Signature · 1 line
2.34.1
maintainer
#6 lore raw
In reply to Shradha Gupta
Quoted 3 lines · On Mon, 9 Jun 2025 06:48:21 -0700 Shradha Gupta wrote:

Since this patchset has patches from PCI and net tree, I am not entirely sure what should be the target tree. Any suggestions/recommendations on the same are welcomed.

Could you put these on a branch based on v6.16-rc1 ? That why if someone else needs them before the next merge window they can pull that branch. The patches look fine to me.

#7 lore raw
Quoted 8 lines · Jakub Kicinski
Quoted 3 lines · On Mon, 9 Jun 2025 06:48:21 -0700 Shradha Gupta wrote:

Since this patchset has patches from PCI and net tree, I am not entirely sure what should be the target tree. Any suggestions/recommendations on the same are welcomed.

Could you put these on a branch based on v6.16-rc1 ? That why if someone else needs them before the next merge window they can pull that branch. The patches look fine to me.

Thanks Jakub. Sure I will send out an update rebasing the series to linux-next's v6.16-rc1.

Regards,
Shradha.