On Tue, May 27, 2025 at 08:59:03AM -0700, Shradha Gupta wrote:
> 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.
>
> Signed-off-by: Shradha Gupta <shradhagu...@linux.microsoft.com>
> Reviewed-by: Haiyang Zhang <haiya...@microsoft.com>
> ---
> 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 | 306 +++++++++++++-----
> include/net/mana/gdma.h | 8 +-
> 2 files changed, 235 insertions(+), 79 deletions(-)
>
> diff --git a/drivers/net/ethernet/microsoft/mana/gdma_main.c
> b/drivers/net/ethernet/microsoft/mana/gdma_main.c
> index 763a548c4a2b..98ebecbec9a7 100644
> --- a/drivers/net/ethernet/microsoft/mana/gdma_main.c
> +++ b/drivers/net/ethernet/microsoft/mana/gdma_main.c
> @@ -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;
> @@ -482,7 +491,9 @@ 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);
> @@ -507,7 +518,10 @@ 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 +1380,113 @@ 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;
>
> - /* Need 1 interrupt for the Hardware communication Channel (HWC) */
> - max_irqs = max_queues_per_port + 1;
> -
> - nvec = pci_alloc_irq_vectors(pdev, 2, max_irqs, PCI_IRQ_MSIX);
> - if (nvec < 0) {
> - cpus_read_unlock();
> - return nvec;
> - }
> - if (nvec <= num_online_cpus())
> - start_irq_index = 0;
> -
> - irqs = kmalloc_array((nvec - start_irq_index), sizeof(int), GFP_KERNEL);
> + irqs = kmalloc_array(nvec, sizeof(int), GFP_KERNEL);
> if (!irqs) {
> err = -ENOMEM;
> goto free_irq_vector;
> }
>
> - gc->irq_contexts = kcalloc(nvec, sizeof(struct gdma_irq_context),
> - GFP_KERNEL);
> - if (!gc->irq_contexts) {
> + /*
> + * 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);
> +
> + snprintf(gic->name, MANA_IRQ_NAME_SZ, "mana_q%d@pci:%s",
> + i - 1, pci_name(pdev));
> +
> + /* 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);
> + }
> +
> + /*
> + * 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
> + */
> + 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)
> + goto free_irq;
> +
> + 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);
> +free_irq_vector:
> + cpus_read_unlock();
> + 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;
> +
> + cpus_read_lock();
> +
> + irqs = kmalloc_array(nvec, sizeof(int), GFP_KERNEL);
> + if (!irqs) {
> err = -ENOMEM;
> - goto free_irq_array;
> + goto free_irq_vector;
> }
>
> 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 +1498,128 @@ 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 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.
> + */
> + start_irqs = irqs;
> + 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)
> 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);
> + kfree(start_irqs);
There is a case when start_irqs can be used here uninitialized.
- Saurabh
