From: Martin K. Petersen <martin.petersen@oracle.com>
Date: Sat, 4 Jan 2014 00:16:56 +0000 (-0500)
Subject: mpt3sas: Rework the MSI-X grouping code
X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=91b265bf0b5733a6c4865e809c93a2a812de46e8;p=GitHub%2FLineageOS%2FG12%2Fandroid_kernel_amlogic_linux-4.9.git

mpt3sas: Rework the MSI-X grouping code

On systems with a non power-of-two CPU count the existing MSI-X grouping
code failed to distribute interrupts correctly. Rework the code to
handle arbitrary processor counts.

Also remove the hardcoded upper limit on the number of processors so we
can boot on large systems.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Sreekanth Reddy <Sreekanth.reddy@avagotech.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---

diff --git a/drivers/scsi/mpt3sas/mpt3sas_base.c b/drivers/scsi/mpt3sas/mpt3sas_base.c
index 0cf4f7000f94..c608a616941a 100644
--- a/drivers/scsi/mpt3sas/mpt3sas_base.c
+++ b/drivers/scsi/mpt3sas/mpt3sas_base.c
@@ -1624,66 +1624,35 @@ _base_request_irq(struct MPT3SAS_ADAPTER *ioc, u8 index, u32 vector)
 static void
 _base_assign_reply_queues(struct MPT3SAS_ADAPTER *ioc)
 {
-	struct adapter_reply_queue *reply_q;
-	int cpu_id;
-	int cpu_grouping, loop, grouping, grouping_mod;
-	int reply_queue;
+	unsigned int cpu, nr_cpus, nr_msix, index = 0;
 
 	if (!_base_is_controller_msix_enabled(ioc))
 		return;
 
 	memset(ioc->cpu_msix_table, 0, ioc->cpu_msix_table_sz);
 
-	/* NUMA Hardware bug workaround - drop to less reply queues */
-	if (ioc->reply_queue_count > ioc->facts.MaxMSIxVectors) {
-		ioc->reply_queue_count = ioc->facts.MaxMSIxVectors;
-		reply_queue = 0;
-		list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
-			reply_q->msix_index = reply_queue;
-			if (++reply_queue == ioc->reply_queue_count)
-				reply_queue = 0;
-		}
-	}
+	nr_cpus = num_online_cpus();
+	nr_msix = ioc->reply_queue_count = min(ioc->reply_queue_count,
+					       ioc->facts.MaxMSIxVectors);
+	if (!nr_msix)
+		return;
 
-	/* when there are more cpus than available msix vectors,
-	 * then group cpus togeather on same irq
-	 */
-	if (ioc->cpu_count > ioc->msix_vector_count) {
-		grouping = ioc->cpu_count / ioc->msix_vector_count;
-		grouping_mod = ioc->cpu_count % ioc->msix_vector_count;
-		if (grouping < 2 || (grouping == 2 && !grouping_mod))
-			cpu_grouping = 2;
-		else if (grouping < 4 || (grouping == 4 && !grouping_mod))
-			cpu_grouping = 4;
-		else if (grouping < 8 || (grouping == 8 && !grouping_mod))
-			cpu_grouping = 8;
-		else
-			cpu_grouping = 16;
-	} else
-		cpu_grouping = 0;
-
-	loop = 0;
-	reply_q = list_entry(ioc->reply_queue_list.next,
-	     struct adapter_reply_queue, list);
-	for_each_online_cpu(cpu_id) {
-		if (!cpu_grouping) {
-			ioc->cpu_msix_table[cpu_id] = reply_q->msix_index;
-			reply_q = list_entry(reply_q->list.next,
-			    struct adapter_reply_queue, list);
-		} else {
-			if (loop < cpu_grouping) {
-				ioc->cpu_msix_table[cpu_id] =
-				    reply_q->msix_index;
-				loop++;
-			} else {
-				reply_q = list_entry(reply_q->list.next,
-				    struct adapter_reply_queue, list);
-				ioc->cpu_msix_table[cpu_id] =
-				    reply_q->msix_index;
-				loop = 1;
-			}
+	cpu = cpumask_first(cpu_online_mask);
+
+	do {
+		unsigned int i, group = nr_cpus / nr_msix;
+
+		if (index < nr_cpus % nr_msix)
+			group++;
+
+		for (i = 0 ; i < group ; i++) {
+			ioc->cpu_msix_table[cpu] = index;
+			cpu = cpumask_next(cpu, cpu_online_mask);
 		}
-	}
+
+		index++;
+
+	} while (cpu < nr_cpus);
 }
 
 /**