add new system file

hc

2024-02-20 102a0743326a03cd1a1202ceda21e175b7d3575c

 kernel/drivers/pci/controller/pci-hyperv.c
..
..
@@ -520,19 +520,10 @@
520
520
 	struct hv_pcidev_description func[];
521
521
 };
522
522
 
523
-enum hv_pcichild_state {
524
-	hv_pcichild_init = 0,
525
-	hv_pcichild_requirements,
526
-	hv_pcichild_resourced,
527
-	hv_pcichild_ejecting,
528
-	hv_pcichild_maximum
529
-};
530
-
531
523
 struct hv_pci_dev {
532
524
 	/* List protected by pci_rescan_remove_lock */
533
525
 	struct list_head list_entry;
534
526
 	refcount_t refs;
535
-	enum hv_pcichild_state state;
536
527
 	struct pci_slot *pci_slot;
537
528
 	struct hv_pcidev_description desc;
538
529
 	bool reported_missing;
..
..
@@ -1237,6 +1228,11 @@
1237
1228
 	pbus = pdev->bus;
1238
1229
 	hbus = container_of(pbus->sysdata, struct hv_pcibus_device, sysdata);
1239
1230
 	int_desc = data->chip_data;
1231
+	if (!int_desc) {
1232
+		dev_warn(&hbus->hdev->device, "%s() can not unmask irq %u\n",
1233
+			 __func__, data->irq);
1234
+		return;
1235
+	}
1240
1236
 
1241
1237
 	spin_lock_irqsave(&hbus->retarget_msi_interrupt_lock, flags);
1242
1238
 
..
..
@@ -1522,7 +1518,7 @@
1522
1518
 	 * Prevents hv_pci_onchannelcallback() from running concurrently
1523
1519
 	 * in the tasklet.
1524
1520
 	 */
1525
-	tasklet_disable_in_atomic(&channel->callback_event);
1521
+	tasklet_disable(&channel->callback_event);
1526
1522
 
1527
1523
 	/*
1528
1524
 	 * Since this function is called with IRQ locks held, can't
..
..
@@ -1552,12 +1548,6 @@
1552
1548
 		}
1553
1549
 		hv_pci_onchannelcallback(hbus);
1554
1550
 		spin_unlock_irqrestore(&channel->sched_lock, flags);
1555
-
1556
-		if (hpdev->state == hv_pcichild_ejecting) {
1557
-			dev_err_once(&hbus->hdev->device,
1558
-				     "the device is being ejected\n");
1559
-			goto enable_tasklet;
1560
-		}
1561
1551
 
1562
1552
 		udelay(100);
1563
1553
 	}
..
..
@@ -2378,8 +2368,6 @@
2378
2368
 	hpdev = container_of(work, struct hv_pci_dev, wrk);
2379
2369
 	hbus = hpdev->hbus;
2380
2370
 
2381
-	WARN_ON(hpdev->state != hv_pcichild_ejecting);
2382
-
2383
2371
 	/*
2384
2372
 	 * Ejection can come before or after the PCI bus has been set up, so
2385
2373
 	 * attempt to find it and tear down the bus state, if it exists.  This
..
..
@@ -2438,7 +2426,6 @@
2438
2426
 		return;
2439
2427
 	}
2440
2428
 
2441
-	hpdev->state = hv_pcichild_ejecting;
2442
2429
 	get_pcichild(hpdev);
2443
2430
 	INIT_WORK(&hpdev->wrk, hv_eject_device_work);
2444
2431
 	get_hvpcibus(hbus);
..
..
@@ -2842,8 +2829,10 @@
2842
2829
 	struct pci_bus_d0_entry *d0_entry;
2843
2830
 	struct hv_pci_compl comp_pkt;
2844
2831
 	struct pci_packet *pkt;
2832
+	bool retry = true;
2845
2833
 	int ret;
2846
2834
 
2835
+enter_d0_retry:
2847
2836
 	/*
2848
2837
 	 * Tell the host that the bus is ready to use, and moved into the
2849
2838
 	 * powered-on state.  This includes telling the host which region
..
..
@@ -2869,6 +2858,38 @@
2869
2858
 
2870
2859
 	if (ret)
2871
2860
 		goto exit;
2861
+
2862
+	/*
2863
+	 * In certain case (Kdump) the pci device of interest was
2864
+	 * not cleanly shut down and resource is still held on host
2865
+	 * side, the host could return invalid device status.
2866
+	 * We need to explicitly request host to release the resource
2867
+	 * and try to enter D0 again.
2868
+	 */
2869
+	if (comp_pkt.completion_status < 0 && retry) {
2870
+		retry = false;
2871
+
2872
+		dev_err(&hdev->device, "Retrying D0 Entry\n");
2873
+
2874
+		/*
2875
+		 * Hv_pci_bus_exit() calls hv_send_resource_released()
2876
+		 * to free up resources of its child devices.
2877
+		 * In the kdump kernel we need to set the
2878
+		 * wslot_res_allocated to 255 so it scans all child
2879
+		 * devices to release resources allocated in the
2880
+		 * normal kernel before panic happened.
2881
+		 */
2882
+		hbus->wslot_res_allocated = 255;
2883
+
2884
+		ret = hv_pci_bus_exit(hdev, true);
2885
+
2886
+		if (ret == 0) {
2887
+			kfree(pkt);
2888
+			goto enter_d0_retry;
2889
+		}
2890
+		dev_err(&hdev->device,
2891
+			"Retrying D0 failed with ret %d\n", ret);
2892
+	}
2872
2893
 
2873
2894
 	if (comp_pkt.completion_status < 0) {
2874
2895
 		dev_err(&hdev->device,
..
..
@@ -2911,6 +2932,24 @@
2911
2932
 			       0, VM_PKT_DATA_INBAND, 0);
2912
2933
 	if (!ret)
2913
2934
 		ret = wait_for_response(hdev, &comp);
2935
+
2936
+	/*
2937
+	 * In the case of fast device addition/removal, it's possible that
2938
+	 * vmbus_sendpacket() or wait_for_response() returns -ENODEV but we
2939
+	 * already got a PCI_BUS_RELATIONS* message from the host and the
2940
+	 * channel callback already scheduled a work to hbus->wq, which can be
2941
+	 * running pci_devices_present_work() -> survey_child_resources() ->
2942
+	 * complete(&hbus->survey_event), even after hv_pci_query_relations()
2943
+	 * exits and the stack variable 'comp' is no longer valid; as a result,
2944
+	 * a hang or a page fault may happen when the complete() calls
2945
+	 * raw_spin_lock_irqsave(). Flush hbus->wq before we exit from
2946
+	 * hv_pci_query_relations() to avoid the issues. Note: if 'ret' is
2947
+	 * -ENODEV, there can't be any more work item scheduled to hbus->wq
2948
+	 * after the flush_workqueue(): see vmbus_onoffer_rescind() ->
2949
+	 * vmbus_reset_channel_cb(), vmbus_rescind_cleanup() ->
2950
+	 * channel->rescind = true.
2951
+	 */
2952
+	flush_workqueue(hbus->wq);
2914
2953
 
2915
2954
 	return ret;
2916
2955
 }
..
..
@@ -3107,7 +3146,6 @@
3107
3146
 	struct hv_pcibus_device *hbus;
3108
3147
 	u16 dom_req, dom;
3109
3148
 	char *name;
3110
-	bool enter_d0_retry = true;
3111
3149
 	int ret;
3112
3150
 
3113
3151
 	/*
..
..
@@ -3228,47 +3266,11 @@
3228
3266
 	if (ret)
3229
3267
 		goto free_fwnode;
3230
3268
 
3231
-retry:
3232
3269
 	ret = hv_pci_query_relations(hdev);
3233
3270
 	if (ret)
3234
3271
 		goto free_irq_domain;
3235
3272
 
3236
3273
 	ret = hv_pci_enter_d0(hdev);
3237
-	/*
3238
-	 * In certain case (Kdump) the pci device of interest was
3239
-	 * not cleanly shut down and resource is still held on host
3240
-	 * side, the host could return invalid device status.
3241
-	 * We need to explicitly request host to release the resource
3242
-	 * and try to enter D0 again.
3243
-	 * Since the hv_pci_bus_exit() call releases structures
3244
-	 * of all its child devices, we need to start the retry from
3245
-	 * hv_pci_query_relations() call, requesting host to send
3246
-	 * the synchronous child device relations message before this
3247
-	 * information is needed in hv_send_resources_allocated()
3248
-	 * call later.
3249
-	 */
3250
-	if (ret == -EPROTO && enter_d0_retry) {
3251
-		enter_d0_retry = false;
3252
-
3253
-		dev_err(&hdev->device, "Retrying D0 Entry\n");
3254
-
3255
-		/*
3256
-		 * Hv_pci_bus_exit() calls hv_send_resources_released()
3257
-		 * to free up resources of its child devices.
3258
-		 * In the kdump kernel we need to set the
3259
-		 * wslot_res_allocated to 255 so it scans all child
3260
-		 * devices to release resources allocated in the
3261
-		 * normal kernel before panic happened.
3262
-		 */
3263
-		hbus->wslot_res_allocated = 255;
3264
-		ret = hv_pci_bus_exit(hdev, true);
3265
-
3266
-		if (ret == 0)
3267
-			goto retry;
3268
-
3269
-		dev_err(&hdev->device,
3270
-			"Retrying D0 failed with ret %d\n", ret);
3271
-	}
3272
3274
 	if (ret)
3273
3275
 		goto free_irq_domain;
3274
3276