forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-10-09 244b2c5ca8b14627e4a17755e5922221e121c771 
 kernel/drivers/scsi/storvsc_drv.c
....@@ -1,18 +1,6 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  * Copyright (c) 2009, Microsoft Corporation.
3
- *
4
- * This program is free software; you can redistribute it and/or modify it
5
- * under the terms and conditions of the GNU General Public License,
6
- * version 2, as published by the Free Software Foundation.
7
- *
8
- * This program is distributed in the hope it will be useful, but WITHOUT
9
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11
- * more details.
12
- *
13
- * You should have received a copy of the GNU General Public License along with
14
- * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15
- * Place - Suite 330, Boston, MA 02111-1307 USA.
164  *
175  * Authors:
186  *   Haiyang Zhang <haiyangz@microsoft.com>
....@@ -368,16 +356,21 @@
368356 };
369357 
370358 /*
371
- * SRB status codes and masks; a subset of the codes used here.
359
+ * SRB status codes and masks. In the 8-bit field, the two high order bits
360
+ * are flags, while the remaining 6 bits are an integer status code.  The
361
+ * definitions here include only the subset of the integer status codes that
362
+ * are tested for in this driver.
372363  */
373
-
374364 #define SRB_STATUS_AUTOSENSE_VALID	0x80
375365 #define SRB_STATUS_QUEUE_FROZEN		0x40
376
-#define SRB_STATUS_INVALID_LUN	0x20
377
-#define SRB_STATUS_SUCCESS	0x01
378
-#define SRB_STATUS_ABORTED	0x02
379
-#define SRB_STATUS_ERROR	0x04
380
-#define SRB_STATUS_DATA_OVERRUN	0x12
366
+
367
+/* SRB status integer codes */
368
+#define SRB_STATUS_SUCCESS		0x01
369
+#define SRB_STATUS_ABORTED		0x02
370
+#define SRB_STATUS_ERROR		0x04
371
+#define SRB_STATUS_INVALID_REQUEST	0x06
372
+#define SRB_STATUS_DATA_OVERRUN		0x12
373
+#define SRB_STATUS_INVALID_LUN		0x20
381374 
382375 #define SRB_STATUS(status) \
383376 	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
....@@ -385,8 +378,9 @@
385378  * This is the end of Protocol specific defines.
386379  */
387380 
388
-static int storvsc_ringbuffer_size = (256 * PAGE_SIZE);
381
+static int storvsc_ringbuffer_size = (128 * 1024);
389382 static u32 max_outstanding_req_per_channel;
383
+static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
390384 
391385 static int storvsc_vcpus_per_sub_channel = 4;
392386 
....@@ -446,7 +440,6 @@
446440 
447441 	bool	 destroy;
448442 	bool	 drain_notify;
449
-	bool	 open_sub_channel;
450443 	atomic_t num_outstanding_req;
451444 	struct Scsi_Host *host;
452445 
....@@ -474,6 +467,11 @@
474467 	 * Mask of CPUs bound to subchannels.
475468 	 */
476469 	struct cpumask alloced_cpus;
470
+	/*
471
+	 * Serializes modifications of stor_chns[] from storvsc_do_io()
472
+	 * and storvsc_change_target_cpu().
473
+	 */
474
+	spinlock_t lock;
477475 	/* Used for vsc/vsp channel reset process */
478476 	struct storvsc_cmd_request init_request;
479477 	struct storvsc_cmd_request reset_request;
....@@ -633,36 +631,101 @@
633631 
634632 }
635633 
634
+static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
635
+				      u32 new)
636
+{
637
+	struct storvsc_device *stor_device;
638
+	struct vmbus_channel *cur_chn;
639
+	bool old_is_alloced = false;
640
+	struct hv_device *device;
641
+	unsigned long flags;
642
+	int cpu;
643
+
644
+	device = channel->primary_channel ?
645
+			channel->primary_channel->device_obj
646
+				: channel->device_obj;
647
+	stor_device = get_out_stor_device(device);
648
+	if (!stor_device)
649
+		return;
650
+
651
+	/* See storvsc_do_io() -> get_og_chn(). */
652
+	spin_lock_irqsave(&stor_device->lock, flags);
653
+
654
+	/*
655
+	 * Determines if the storvsc device has other channels assigned to
656
+	 * the "old" CPU to update the alloced_cpus mask and the stor_chns
657
+	 * array.
658
+	 */
659
+	if (device->channel != channel && device->channel->target_cpu == old) {
660
+		cur_chn = device->channel;
661
+		old_is_alloced = true;
662
+		goto old_is_alloced;
663
+	}
664
+	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
665
+		if (cur_chn == channel)
666
+			continue;
667
+		if (cur_chn->target_cpu == old) {
668
+			old_is_alloced = true;
669
+			goto old_is_alloced;
670
+		}
671
+	}
672
+
673
+old_is_alloced:
674
+	if (old_is_alloced)
675
+		WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
676
+	else
677
+		cpumask_clear_cpu(old, &stor_device->alloced_cpus);
678
+
679
+	/* "Flush" the stor_chns array. */
680
+	for_each_possible_cpu(cpu) {
681
+		if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
682
+					cpu, &stor_device->alloced_cpus))
683
+			WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
684
+	}
685
+
686
+	WRITE_ONCE(stor_device->stor_chns[new], channel);
687
+	cpumask_set_cpu(new, &stor_device->alloced_cpus);
688
+
689
+	spin_unlock_irqrestore(&stor_device->lock, flags);
690
+}
691
+
636692 static void handle_sc_creation(struct vmbus_channel *new_sc)
637693 {
638694 	struct hv_device *device = new_sc->primary_channel->device_obj;
695
+	struct device *dev = &device->device;
639696 	struct storvsc_device *stor_device;
640697 	struct vmstorage_channel_properties props;
698
+	int ret;
641699 
642700 	stor_device = get_out_stor_device(device);
643701 	if (!stor_device)
644702 		return;
645703 
646
-	if (stor_device->open_sub_channel == false)
647
-		return;
648
-
649704 	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
650705 
651
-	vmbus_open(new_sc,
652
-		   storvsc_ringbuffer_size,
653
-		   storvsc_ringbuffer_size,
654
-		   (void *)&props,
655
-		   sizeof(struct vmstorage_channel_properties),
656
-		   storvsc_on_channel_callback, new_sc);
706
+	ret = vmbus_open(new_sc,
707
+			 storvsc_ringbuffer_size,
708
+			 storvsc_ringbuffer_size,
709
+			 (void *)&props,
710
+			 sizeof(struct vmstorage_channel_properties),
711
+			 storvsc_on_channel_callback, new_sc);
657712 
658
-	if (new_sc->state == CHANNEL_OPENED_STATE) {
659
-		stor_device->stor_chns[new_sc->target_cpu] = new_sc;
660
-		cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
713
+	/* In case vmbus_open() fails, we don't use the sub-channel. */
714
+	if (ret != 0) {
715
+		dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
716
+		return;
661717 	}
718
+
719
+	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
720
+
721
+	/* Add the sub-channel to the array of available channels. */
722
+	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
723
+	cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
662724 }
663725 
664726 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
665727 {
728
+	struct device *dev = &device->device;
666729 	struct storvsc_device *stor_device;
667730 	int num_sc;
668731 	struct storvsc_cmd_request *request;
....@@ -688,21 +751,11 @@
688751 	request = &stor_device->init_request;
689752 	vstor_packet = &request->vstor_packet;
690753 
691
-	stor_device->open_sub_channel = true;
692754 	/*
693755 	 * Establish a handler for dealing with subchannels.
694756 	 */
695757 	vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
696758 
697
-	/*
698
-	 * Check to see if sub-channels have already been created. This
699
-	 * can happen when this driver is re-loaded after unloading.
700
-	 */
701
-
702
-	if (vmbus_are_subchannels_present(device->channel))
703
-		return;
704
-
705
-	stor_device->open_sub_channel = false;
706759 	/*
707760 	 * Request the host to create sub-channels.
708761 	 */
....@@ -719,23 +772,29 @@
719772 			       VM_PKT_DATA_INBAND,
720773 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
721774 
722
-	if (ret != 0)
775
+	if (ret != 0) {
776
+		dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
723777 		return;
778
+	}
724779 
725780 	t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
726
-	if (t == 0)
781
+	if (t == 0) {
782
+		dev_err(dev, "Failed to create sub-channel: timed out\n");
727783 		return;
784
+	}
728785 
729786 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
730
-	    vstor_packet->status != 0)
787
+	    vstor_packet->status != 0) {
788
+		dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
789
+			vstor_packet->operation, vstor_packet->status);
731790 		return;
791
+	}
732792 
733793 	/*
734
-	 * Now that we created the sub-channels, invoke the check; this
735
-	 * may trigger the callback.
794
+	 * We need to do nothing here, because vmbus_process_offer()
795
+	 * invokes channel->sc_creation_callback, which will open and use
796
+	 * the sub-channel(s).
736797 	 */
737
-	stor_device->open_sub_channel = true;
738
-	vmbus_are_subchannels_present(device->channel);
739798 }
740799 
741800 static void cache_wwn(struct storvsc_device *stor_device,
....@@ -887,6 +946,8 @@
887946 	if (stor_device->stor_chns == NULL)
888947 		return -ENOMEM;
889948 
949
+	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
950
+
890951 	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
891952 	cpumask_set_cpu(device->channel->target_cpu,
892953 			&stor_device->alloced_cpus);
....@@ -938,17 +999,43 @@
938999 	struct storvsc_scan_work *wrk;
9391000 	void (*process_err_fn)(struct work_struct *work);
9401001 	struct hv_host_device *host_dev = shost_priv(host);
941
-	bool do_work = false;
9421002 
9431003 	switch (SRB_STATUS(vm_srb->srb_status)) {
9441004 	case SRB_STATUS_ERROR:
945
-		/*
946
-		 * Let upper layer deal with error when
947
-		 * sense message is present.
948
-		 */
1005
+	case SRB_STATUS_ABORTED:
1006
+	case SRB_STATUS_INVALID_REQUEST:
1007
+		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
1008
+			/* Check for capacity change */
1009
+			if ((asc == 0x2a) && (ascq == 0x9)) {
1010
+				process_err_fn = storvsc_device_scan;
1011
+				/* Retry the I/O that triggered this. */
1012
+				set_host_byte(scmnd, DID_REQUEUE);
1013
+				goto do_work;
1014
+			}
9491015 
950
-		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
951
-			break;
1016
+			/*
1017
+			 * Check for "Operating parameters have changed"
1018
+			 * due to Hyper-V changing the VHD/VHDX BlockSize
1019
+			 * when adding/removing a differencing disk. This
1020
+			 * causes discard_granularity to change, so do a
1021
+			 * rescan to pick up the new granularity. We don't
1022
+			 * want scsi_report_sense() to output a message
1023
+			 * that a sysadmin wouldn't know what to do with.
1024
+			 */
1025
+			if ((asc == 0x3f) && (ascq != 0x03) &&
1026
+					(ascq != 0x0e)) {
1027
+				process_err_fn = storvsc_device_scan;
1028
+				set_host_byte(scmnd, DID_REQUEUE);
1029
+				goto do_work;
1030
+			}
1031
+
1032
+			/*
1033
+			 * Otherwise, let upper layer deal with the
1034
+			 * error when sense message is present
1035
+			 */
1036
+			return;
1037
+		}
1038
+
9521039 		/*
9531040 		 * If there is an error; offline the device since all
9541041 		 * error recovery strategies would have already been
....@@ -961,37 +1048,26 @@
9611048 			set_host_byte(scmnd, DID_PASSTHROUGH);
9621049 			break;
9631050 		/*
964
-		 * On Some Windows hosts TEST_UNIT_READY command can return
965
-		 * SRB_STATUS_ERROR, let the upper level code deal with it
966
-		 * based on the sense information.
1051
+		 * On some Hyper-V hosts TEST_UNIT_READY command can
1052
+		 * return SRB_STATUS_ERROR. Let the upper level code
1053
+		 * deal with it based on the sense information.
9671054 		 */
9681055 		case TEST_UNIT_READY:
9691056 			break;
9701057 		default:
9711058 			set_host_byte(scmnd, DID_ERROR);
9721059 		}
973
-		break;
974
-	case SRB_STATUS_INVALID_LUN:
975
-		set_host_byte(scmnd, DID_NO_CONNECT);
976
-		do_work = true;
977
-		process_err_fn = storvsc_remove_lun;
978
-		break;
979
-	case SRB_STATUS_ABORTED:
980
-		if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
981
-		    (asc == 0x2a) && (ascq == 0x9)) {
982
-			do_work = true;
983
-			process_err_fn = storvsc_device_scan;
984
-			/*
985
-			 * Retry the I/O that trigerred this.
986
-			 */
987
-			set_host_byte(scmnd, DID_REQUEUE);
988
-		}
989
-		break;
990
-	}
991
-
992
-	if (!do_work)
9931060 		return;
9941061 
1062
+	case SRB_STATUS_INVALID_LUN:
1063
+		set_host_byte(scmnd, DID_NO_CONNECT);
1064
+		process_err_fn = storvsc_remove_lun;
1065
+		goto do_work;
1066
+
1067
+	}
1068
+	return;
1069
+
1070
+do_work:
9951071 	/*
9961072 	 * We need to schedule work to process this error; schedule it.
9971073 	 */
....@@ -1049,6 +1125,10 @@
10491125 			data_transfer_length = 0;
10501126 	}
10511127 
1128
+	/* Validate data_transfer_length (from Hyper-V) */
1129
+	if (data_transfer_length > cmd_request->payload->range.len)
1130
+		data_transfer_length = cmd_request->payload->range.len;
1131
+
10521132 	scsi_set_resid(scmnd,
10531133 		cmd_request->payload->range.len - data_transfer_length);
10541134 
....@@ -1089,6 +1169,11 @@
10891169 	/* Copy over the status...etc */
10901170 	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
10911171 	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1172
+
1173
+	/* Validate sense_info_length (from Hyper-V) */
1174
+	if (vstor_packet->vm_srb.sense_info_length > sense_buffer_size)
1175
+		vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1176
+
10921177 	stor_pkt->vm_srb.sense_info_length =
10931178 	vstor_packet->vm_srb.sense_info_length;
10941179 
....@@ -1259,8 +1344,10 @@
12591344 	const struct cpumask *node_mask;
12601345 	int num_channels, tgt_cpu;
12611346 
1262
-	if (stor_device->num_sc == 0)
1347
+	if (stor_device->num_sc == 0) {
1348
+		stor_device->stor_chns[q_num] = stor_device->device->channel;
12631349 		return stor_device->device->channel;
1350
+	}
12641351 
12651352 	/*
12661353 	 * Our channel array is sparsley populated and we
....@@ -1269,7 +1356,6 @@
12691356 	 * The strategy is simple:
12701357 	 * I. Ensure NUMA locality
12711358 	 * II. Distribute evenly (best effort)
1272
-	 * III. Mapping is persistent.
12731359 	 */
12741360 
12751361 	node_mask = cpumask_of_node(cpu_to_node(q_num));
....@@ -1279,8 +1365,10 @@
12791365 		if (cpumask_test_cpu(tgt_cpu, node_mask))
12801366 			num_channels++;
12811367 	}
1282
-	if (num_channels == 0)
1368
+	if (num_channels == 0) {
1369
+		stor_device->stor_chns[q_num] = stor_device->device->channel;
12831370 		return stor_device->device->channel;
1371
+	}
12841372 
12851373 	hash_qnum = q_num;
12861374 	while (hash_qnum >= num_channels)
....@@ -1306,6 +1394,7 @@
13061394 	struct storvsc_device *stor_device;
13071395 	struct vstor_packet *vstor_packet;
13081396 	struct vmbus_channel *outgoing_channel, *channel;
1397
+	unsigned long flags;
13091398 	int ret = 0;
13101399 	const struct cpumask *node_mask;
13111400 	int tgt_cpu;
....@@ -1319,10 +1408,11 @@
13191408 
13201409 	request->device  = device;
13211410 	/*
1322
-	 * Select an an appropriate channel to send the request out.
1411
+	 * Select an appropriate channel to send the request out.
13231412 	 */
1324
-	if (stor_device->stor_chns[q_num] != NULL) {
1325
-		outgoing_channel = stor_device->stor_chns[q_num];
1413
+	/* See storvsc_change_target_cpu(). */
1414
+	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1415
+	if (outgoing_channel != NULL) {
13261416 		if (outgoing_channel->target_cpu == q_num) {
13271417 			/*
13281418 			 * Ideally, we want to pick a different channel if
....@@ -1335,7 +1425,10 @@
13351425 					continue;
13361426 				if (tgt_cpu == q_num)
13371427 					continue;
1338
-				channel = stor_device->stor_chns[tgt_cpu];
1428
+				channel = READ_ONCE(
1429
+					stor_device->stor_chns[tgt_cpu]);
1430
+				if (channel == NULL)
1431
+					continue;
13391432 				if (hv_get_avail_to_write_percent(
13401433 							&channel->outbound)
13411434 						> ring_avail_percent_lowater) {
....@@ -1361,7 +1454,10 @@
13611454 			for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
13621455 				if (cpumask_test_cpu(tgt_cpu, node_mask))
13631456 					continue;
1364
-				channel = stor_device->stor_chns[tgt_cpu];
1457
+				channel = READ_ONCE(
1458
+					stor_device->stor_chns[tgt_cpu]);
1459
+				if (channel == NULL)
1460
+					continue;
13651461 				if (hv_get_avail_to_write_percent(
13661462 							&channel->outbound)
13671463 						> ring_avail_percent_lowater) {
....@@ -1371,7 +1467,14 @@
13711467 			}
13721468 		}
13731469 	} else {
1470
+		spin_lock_irqsave(&stor_device->lock, flags);
1471
+		outgoing_channel = stor_device->stor_chns[q_num];
1472
+		if (outgoing_channel != NULL) {
1473
+			spin_unlock_irqrestore(&stor_device->lock, flags);
1474
+			goto found_channel;
1475
+		}
13741476 		outgoing_channel = get_og_chn(stor_device, q_num);
1477
+		spin_unlock_irqrestore(&stor_device->lock, flags);
13751478 	}
13761479 
13771480 found_channel:
....@@ -1434,9 +1537,8 @@
14341537 {
14351538 	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
14361539 
1437
-	/* Ensure there are no gaps in presented sgls */
1438
-	blk_queue_virt_boundary(sdevice->request_queue, PAGE_SIZE - 1);
1439
-
1540
+	/* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1541
+	sdevice->no_report_opcodes = 1;
14401542 	sdevice->no_write_same = 1;
14411543 
14421544 	/*
....@@ -1499,6 +1601,7 @@
14991601 
15001602 	request = &stor_device->reset_request;
15011603 	vstor_packet = &request->vstor_packet;
1604
+	memset(vstor_packet, 0, sizeof(struct vstor_packet));
15021605 
15031606 	init_completion(&request->wait_event);
15041607 
....@@ -1540,10 +1643,6 @@
15401643  */
15411644 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
15421645 {
1543
-#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1544
-	if (scmnd->device->host->transportt == fc_transport_template)
1545
-		return fc_eh_timed_out(scmnd);
1546
-#endif
15471646 	return BLK_EH_RESET_TIMER;
15481647 }
15491648 
....@@ -1602,6 +1701,7 @@
16021701 	/* Setup the cmd request */
16031702 	cmd_request->cmd = scmnd;
16041703 
1704
+	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
16051705 	vm_srb = &cmd_request->vstor_packet.vm_srb;
16061706 	vm_srb->win8_extension.time_out_value = 60;
16071707 
....@@ -1654,26 +1754,68 @@
16541754 
16551755 	length = scsi_bufflen(scmnd);
16561756 	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1657
-	payload_sz = sizeof(cmd_request->mpb);
1757
+	payload_sz = 0;
16581758 
16591759 	if (sg_count) {
1660
-		if (sg_count > MAX_PAGE_BUFFER_COUNT) {
1760
+		unsigned int hvpgoff = 0;
1761
+		unsigned long offset_in_hvpg = sgl->offset & ~HV_HYP_PAGE_MASK;
1762
+		unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1763
+		u64 hvpfn;
16611764 
1662
-			payload_sz = (sg_count * sizeof(u64) +
1663
-				      sizeof(struct vmbus_packet_mpb_array));
1765
+		payload_sz = (hvpg_count * sizeof(u64) +
1766
+			      sizeof(struct vmbus_packet_mpb_array));
1767
+
1768
+		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
16641769 			payload = kzalloc(payload_sz, GFP_ATOMIC);
16651770 			if (!payload)
16661771 				return SCSI_MLQUEUE_DEVICE_BUSY;
16671772 		}
16681773 
1774
+		/*
1775
+		 * sgl is a list of PAGEs, and payload->range.pfn_array
1776
+		 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
1777
+		 * page size that Hyper-V uses, so here we need to divide PAGEs
1778
+		 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
1779
+		 * Besides, payload->range.offset should be the offset in one
1780
+		 * HV_HYP_PAGE.
1781
+		 */
16691782 		payload->range.len = length;
1670
-		payload->range.offset = sgl[0].offset;
1783
+		payload->range.offset = offset_in_hvpg;
1784
+		hvpgoff = sgl->offset >> HV_HYP_PAGE_SHIFT;
16711785 
16721786 		cur_sgl = sgl;
1673
-		for (i = 0; i < sg_count; i++) {
1674
-			payload->range.pfn_array[i] =
1675
-				page_to_pfn(sg_page((cur_sgl)));
1676
-			cur_sgl = sg_next(cur_sgl);
1787
+		for (i = 0; i < hvpg_count; i++) {
1788
+			/*
1789
+			 * 'i' is the index of hv pages in the payload and
1790
+			 * 'hvpgoff' is the offset (in hv pages) of the first
1791
+			 * hv page in the the first page. The relationship
1792
+			 * between the sum of 'i' and 'hvpgoff' and the offset
1793
+			 * (in hv pages) in a payload page ('hvpgoff_in_page')
1794
+			 * is as follow:
1795
+			 *
1796
+			 * |------------------ PAGE -------------------|
1797
+			 * |   NR_HV_HYP_PAGES_IN_PAGE hvpgs in total  |
1798
+			 * |hvpg|hvpg| ...              |hvpg|... |hvpg|
1799
+			 * ^         ^                                 ^                 ^
1800
+			 * +-hvpgoff-+                                 +-hvpgoff_in_page-+
1801
+			 *           ^                                                   |
1802
+			 *           +--------------------- i ---------------------------+
1803
+			 */
1804
+			unsigned int hvpgoff_in_page =
1805
+				(i + hvpgoff) % NR_HV_HYP_PAGES_IN_PAGE;
1806
+
1807
+			/*
1808
+			 * Two cases that we need to fetch a page:
1809
+			 * 1) i == 0, the first step or
1810
+			 * 2) hvpgoff_in_page == 0, when we reach the boundary
1811
+			 *    of a page.
1812
+			 */
1813
+			if (hvpgoff_in_page == 0 || i == 0) {
1814
+				hvpfn = page_to_hvpfn(sg_page(cur_sgl));
1815
+				cur_sgl = sg_next(cur_sgl);
1816
+			}
1817
+
1818
+			payload->range.pfn_array[i] = hvpfn + hvpgoff_in_page;
16771819 		}
16781820 	}
16791821 
....@@ -1707,11 +1849,13 @@
17071849 	.slave_configure =	storvsc_device_configure,
17081850 	.cmd_per_lun =		2048,
17091851 	.this_id =		-1,
1710
-	.use_clustering =	ENABLE_CLUSTERING,
17111852 	/* Make sure we dont get a sg segment crosses a page boundary */
17121853 	.dma_boundary =		PAGE_SIZE-1,
1854
+	/* Ensure there are no gaps in presented sgls */
1855
+	.virt_boundary_mask =	PAGE_SIZE-1,
17131856 	.no_write_same =	1,
17141857 	.track_queue_depth =	1,
1858
+	.change_queue_depth =	storvsc_change_queue_depth,
17151859 };
17161860 
17171861 enum {
....@@ -1738,6 +1882,13 @@
17381882 };
17391883 
17401884 MODULE_DEVICE_TABLE(vmbus, id_table);
1885
+
1886
+static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1887
+
1888
+static bool hv_dev_is_fc(struct hv_device *hv_dev)
1889
+{
1890
+	return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1891
+}
17411892 
17421893 static int storvsc_probe(struct hv_device *device,
17431894 			const struct hv_vmbus_device_id *dev_id)
....@@ -1803,10 +1954,10 @@
18031954 	}
18041955 
18051956 	stor_device->destroy = false;
1806
-	stor_device->open_sub_channel = false;
18071957 	init_waitqueue_head(&stor_device->waiting_to_drain);
18081958 	stor_device->device = device;
18091959 	stor_device->host = host;
1960
+	spin_lock_init(&stor_device->lock);
18101961 	hv_set_drvdata(device, stor_device);
18111962 
18121963 	stor_device->port_number = host->host_no;
....@@ -1848,20 +1999,23 @@
18481999 	 */
18492000 	host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
18502001 	/*
2002
+	 * For non-IDE disks, the host supports multiple channels.
18512003 	 * Set the number of HW queues we are supporting.
18522004 	 */
1853
-	if (stor_device->num_sc != 0)
1854
-		host->nr_hw_queues = stor_device->num_sc + 1;
2005
+	if (!dev_is_ide)
2006
+		host->nr_hw_queues = num_present_cpus();
18552007 
18562008 	/*
18572009 	 * Set the error handler work queue.
18582010 	 */
18592011 	host_dev->handle_error_wq =
18602012 			alloc_ordered_workqueue("storvsc_error_wq_%d",
1861
-						WQ_MEM_RECLAIM,
2013
+						0,
18622014 						host->host_no);
1863
-	if (!host_dev->handle_error_wq)
2015
+	if (!host_dev->handle_error_wq) {
2016
+		ret = -ENOMEM;
18642017 		goto err_out2;
2018
+	}
18652019 	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
18662020 	/* Register the HBA and start the scsi bus scan */
18672021 	ret = scsi_add_host(host, &device->device);
....@@ -1919,6 +2073,15 @@
19192073 	return ret;
19202074 }
19212075 
2076
+/* Change a scsi target's queue depth */
2077
+static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2078
+{
2079
+	if (queue_depth > scsi_driver.can_queue)
2080
+		queue_depth = scsi_driver.can_queue;
2081
+
2082
+	return scsi_change_queue_depth(sdev, queue_depth);
2083
+}
2084
+
19222085 static int storvsc_remove(struct hv_device *dev)
19232086 {
19242087 	struct storvsc_device *stor_device = hv_get_drvdata(dev);
....@@ -1939,11 +2102,42 @@
19392102 	return 0;
19402103 }
19412104 
2105
+static int storvsc_suspend(struct hv_device *hv_dev)
2106
+{
2107
+	struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2108
+	struct Scsi_Host *host = stor_device->host;
2109
+	struct hv_host_device *host_dev = shost_priv(host);
2110
+
2111
+	storvsc_wait_to_drain(stor_device);
2112
+
2113
+	drain_workqueue(host_dev->handle_error_wq);
2114
+
2115
+	vmbus_close(hv_dev->channel);
2116
+
2117
+	kfree(stor_device->stor_chns);
2118
+	stor_device->stor_chns = NULL;
2119
+
2120
+	cpumask_clear(&stor_device->alloced_cpus);
2121
+
2122
+	return 0;
2123
+}
2124
+
2125
+static int storvsc_resume(struct hv_device *hv_dev)
2126
+{
2127
+	int ret;
2128
+
2129
+	ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2130
+				     hv_dev_is_fc(hv_dev));
2131
+	return ret;
2132
+}
2133
+
19422134 static struct hv_driver storvsc_drv = {
19432135 	.name = KBUILD_MODNAME,
19442136 	.id_table = id_table,
19452137 	.probe = storvsc_probe,
19462138 	.remove = storvsc_remove,
2139
+	.suspend = storvsc_suspend,
2140
+	.resume = storvsc_resume,
19472141 	.driver = {
19482142 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
19492143 	},