update kernel to 5.10.198

hc

2024-01-03 2f7c68cb55ecb7331f2381deb497c27155f32faf

 kernel/drivers/infiniband/hw/hfi1/chip.c
..
..
@@ -1097,7 +1097,7 @@
1097
1097
 static void handle_temp_err(struct hfi1_devdata *dd);
1098
1098
 static void dc_shutdown(struct hfi1_devdata *dd);
1099
1099
 static void dc_start(struct hfi1_devdata *dd);
1100
-static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
1100
+static int qos_rmt_entries(unsigned int n_krcv_queues, unsigned int *mp,
1101
1101
 			   unsigned int *np);
1102
1102
 static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd);
1103
1103
 static int wait_link_transfer_active(struct hfi1_devdata *dd, int wait_ms);
..
..
@@ -12348,6 +12348,7 @@
12348
12348
 
12349
12349
 	if (dd->synth_stats_timer.function)
12350
12350
 		del_timer_sync(&dd->synth_stats_timer);
12351
+	cancel_work_sync(&dd->update_cntr_work);
12351
12352
 	ppd = (struct hfi1_pportdata *)(dd + 1);
12352
12353
 	for (i = 0; i < dd->num_pports; i++, ppd++) {
12353
12354
 		kfree(ppd->cntrs);
..
..
@@ -13403,7 +13404,6 @@
13403
13404
 	int ret;
13404
13405
 	unsigned ngroups;
13405
13406
 	int rmt_count;
13406
-	int user_rmt_reduced;
13407
13407
 	u32 n_usr_ctxts;
13408
13408
 	u32 send_contexts = chip_send_contexts(dd);
13409
13409
 	u32 rcv_contexts = chip_rcv_contexts(dd);
..
..
@@ -13462,28 +13462,34 @@
13462
13462
 					 (num_kernel_contexts + n_usr_ctxts),
13463
13463
 					 &node_affinity.real_cpu_mask);
13464
13464
 	/*
13465
-	 * The RMT entries are currently allocated as shown below:
13466
-	 * 1. QOS (0 to 128 entries);
13467
-	 * 2. FECN (num_kernel_context - 1 + num_user_contexts +
13468
-	 *    num_netdev_contexts);
13469
-	 * 3. netdev (num_netdev_contexts).
13470
-	 * It should be noted that FECN oversubscribe num_netdev_contexts
13471
-	 * entries of RMT because both netdev and PSM could allocate any receive
13472
-	 * context between dd->first_dyn_alloc_text and dd->num_rcv_contexts,
13473
-	 * and PSM FECN must reserve an RMT entry for each possible PSM receive
13474
-	 * context.
13465
+	 * RMT entries are allocated as follows:
13466
+	 * 1. QOS (0 to 128 entries)
13467
+	 * 2. FECN (num_kernel_context - 1 [a] + num_user_contexts +
13468
+	 *          num_netdev_contexts [b])
13469
+	 * 3. netdev (NUM_NETDEV_MAP_ENTRIES)
13470
+	 *
13471
+	 * Notes:
13472
+	 * [a] Kernel contexts (except control) are included in FECN if kernel
13473
+	 *     TID_RDMA is active.
13474
+	 * [b] Netdev and user contexts are randomly allocated from the same
13475
+	 *     context pool, so FECN must cover all contexts in the pool.
13475
13476
 	 */
13476
-	rmt_count = qos_rmt_entries(dd, NULL, NULL) + (num_netdev_contexts * 2);
13477
-	if (HFI1_CAP_IS_KSET(TID_RDMA))
13478
-		rmt_count += num_kernel_contexts - 1;
13479
-	if (rmt_count + n_usr_ctxts > NUM_MAP_ENTRIES) {
13480
-		user_rmt_reduced = NUM_MAP_ENTRIES - rmt_count;
13481
-		dd_dev_err(dd,
13482
-			   "RMT size is reducing the number of user receive contexts from %u to %d\n",
13483
-			   n_usr_ctxts,
13484
-			   user_rmt_reduced);
13485
-		/* recalculate */
13486
-		n_usr_ctxts = user_rmt_reduced;
13477
+	rmt_count = qos_rmt_entries(num_kernel_contexts - 1, NULL, NULL)
13478
+		    + (HFI1_CAP_IS_KSET(TID_RDMA) ? num_kernel_contexts - 1
13479
+						  : 0)
13480
+		    + n_usr_ctxts
13481
+		    + num_netdev_contexts
13482
+		    + NUM_NETDEV_MAP_ENTRIES;
13483
+	if (rmt_count > NUM_MAP_ENTRIES) {
13484
+		int over = rmt_count - NUM_MAP_ENTRIES;
13485
+		/* try to squish user contexts, minimum of 1 */
13486
+		if (over >= n_usr_ctxts) {
13487
+			dd_dev_err(dd, "RMT overflow: reduce the requested number of contexts\n");
13488
+			return -EINVAL;
13489
+		}
13490
+		dd_dev_err(dd, "RMT overflow: reducing # user contexts from %u to %u\n",
13491
+			   n_usr_ctxts, n_usr_ctxts - over);
13492
+		n_usr_ctxts -= over;
13487
13493
 	}
13488
13494
 
13489
13495
 	/* the first N are kernel contexts, the rest are user/netdev contexts */
..
..
@@ -14340,15 +14346,15 @@
14340
14346
 }
14341
14347
 
14342
14348
 /* return the number of RSM map table entries that will be used for QOS */
14343
-static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
14349
+static int qos_rmt_entries(unsigned int n_krcv_queues, unsigned int *mp,
14344
14350
 			   unsigned int *np)
14345
14351
 {
14346
14352
 	int i;
14347
14353
 	unsigned int m, n;
14348
-	u8 max_by_vl = 0;
14354
+	uint max_by_vl = 0;
14349
14355
 
14350
14356
 	/* is QOS active at all? */
14351
-	if (dd->n_krcv_queues <= MIN_KERNEL_KCTXTS ||
14357
+	if (n_krcv_queues < MIN_KERNEL_KCTXTS ||
14352
14358
 	    num_vls == 1 ||
14353
14359
 	    krcvqsset <= 1)
14354
14360
 		goto no_qos;
..
..
@@ -14406,7 +14412,7 @@
14406
14412
 
14407
14413
 	if (!rmt)
14408
14414
 		goto bail;
14409
-	rmt_entries = qos_rmt_entries(dd, &m, &n);
14415
+	rmt_entries = qos_rmt_entries(dd->n_krcv_queues - 1, &m, &n);
14410
14416
 	if (rmt_entries == 0)
14411
14417
 		goto bail;
14412
14418
 	qpns_per_vl = 1 << m;