Re: [PATCH v2] scsi: mpt3sas: avoid out-of-bounds cpumask_of_node() call in _base_assign_reply_queues()

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but incorrect subtraction in the fall_back path causes unassigned CPUs in the msix table and a potential divide by zero if nr_msix evaluates to zero.
--

commit 9e90c3b985462523aebc40d2af67de713d3be849
Author: Ivy Lopez <[email protected]>

scsi: mpt3sas: avoid out-of-bounds cpumask_of_node() call in _base_assign_reply_queues()

dev_to_node() can return NUMA_NO_NODE (-1) on systems without NUMA
topology information for the PCI device, such as single-socket
boards that don't expose device-to-node affinity. Fall back to
cpu_online_mask when no NUMA node is available, rather than assuming
dev_to_node() always returns a valid node index.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=221294

> diff --git a/drivers/scsi/mpt3sas/mpt3sas_base.c b/drivers/scsi/mpt3sas/mpt3sas_base.c
> index fed7aeffec58d..1af25a22611a1 100644
> --- a/drivers/scsi/mpt3sas/mpt3sas_base.c
> +++ b/drivers/scsi/mpt3sas/mpt3sas_base.c
> @@ -3238,7 +3238,10 @@ _base_assign_reply_queues(struct MPT3SAS_ADAPTER *ioc)
>  		 * corresponding to high iops queues.
>  		 */
>  		if (ioc->high_iops_queues) {
> -			mask = cpumask_of_node(dev_to_node(&ioc->pdev->dev));
> +			int node = dev_to_node(&ioc->pdev->dev);
> +
> +			mask = (node == NUMA_NO_NODE) ?
> +				cpu_online_mask : cpumask_of_node(node);
>  			for (index = 0; index < ioc->high_iops_queues;
>  			    index++) {
>  				irq = pci_irq_vector(ioc->pdev, index);

[Severity: High]
This isn't a bug introduced by this patch, but incorrect subtraction in the
fall_back path causes unassigned CPUs in the msix table and a potential
divide by zero.

In _base_assign_reply_queues(), the calculation evaluates incorrectly:

drivers/scsi/mpt3sas/mpt3sas_base.c:_base_assign_reply_queues() {
...
fall_back:
	cpu = cpumask_first(cpu_online_mask);
	nr_msix -= (ioc->high_iops_queues - iopoll_q_count);
	index = 0;

	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
		unsigned int i, group = nr_cpus / nr_msix;
...
}

If iopoll_q_count > 0, the calculation nr_msix -= (ioc->high_iops_queues -
iopoll_q_count) evaluates to nr_msix - high_iops_queues + iopoll_q_count.

This mathematically adds iopoll_q_count instead of subtracting it, inflating
nr_msix. This causes group = nr_cpus / nr_msix to calculate a smaller group
size, leaving trailing CPUs mapped to vector 0.

Furthermore, if PCI MSI-X allocation limits nr_msix to exactly
high_iops_queues (e.g. system interrupt limits), nr_msix evaluates to 0.

This would trigger a hardware divide-by-zero exception when evaluating
nr_cpus / nr_msix.

Is it possible to rewrite this subtraction to avoid inflating nr_msix and
prevent a potential divide-by-zero?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1