vioscsi: add logic to make reset mechanism configurable

We have seen many cases where we see vioscsi 129 reset events leading to a complete guest hang. While overall reset logic added in #684 likely mitigates the problem, we still see sporadic requests from customer where they see guest OSes hangs. This change will make it possible configure system to bugcheck and get a memory dump and the state of the VQs. By default, the behavior will remain intact - complete pending SRBs, but can be changed to either to do nothing or bugcheck the guest by setting the registry value: Path: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\vioscsi\Parameters Name: VioscsiActionOnReset Type: REG_DWORD Values: 0x0 - Complete all pending SRBs (default) 0x1 - Do nothing (all unserved SRBs will be stored in memory) 0xdeaddead - Bugcheck on the reset event Signed-off-by: Sergey Bykov <[email protected]>
virtio-win · Aug 25, 2023 · eff7b43 · eff7b43
1 parent 322f1a4
commit eff7b43
Show file tree

Hide file tree

Showing 2 changed files with 29 additions and 7 deletions.
diff --git a/vioscsi/vioscsi.c b/vioscsi/vioscsi.c
@@ -487,7 +487,7 @@ ENTER_FN();
          * [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\vioscsi\Parameters\Device]
          * "PhysicalBreaks"={dword value here}
          */
-        VioScsiReadRegistryParameter(DeviceExtension, MAX_PH_BREAKS, FIELD_OFFSET(ADAPTER_EXTENSION, max_physical_breaks));
+        VioScsiReadRegistryParameter(DeviceExtension, REGISTRY_MAX_PH_BREAKS, FIELD_OFFSET(ADAPTER_EXTENSION, max_physical_breaks));
         adaptExt->max_physical_breaks = min(
                                     max(SCSI_MINIMUM_PHYSICAL_BREAKS, adaptExt->max_physical_breaks),
                                     MAX_PHYS_SEGMENTS);
@@ -520,6 +520,9 @@ ENTER_FN();
         adaptExt->num_queues = min(adaptExt->num_queues, (USHORT)num_cpus);
     }
 
+    adaptExt->action_on_reset = VioscsiResetCompleteRequests;
+    VioScsiReadRegistryParameter(DeviceExtension, REGISTRY_ACTION_ON_RESET, FIELD_OFFSET(ADAPTER_EXTENSION, action_on_reset));
+
     RhelDbgPrint(TRACE_LEVEL_INFORMATION, " Queues %d CPUs %d\n", adaptExt->num_queues, num_cpus);
 
     /* Figure out the maximum number of queues we will ever need to set up. Note that this may
@@ -1493,10 +1496,21 @@ ENTER_FN_SRB();
         case SRB_FUNCTION_RESET_DEVICE:
         case SRB_FUNCTION_RESET_LOGICAL_UNIT:
             RhelDbgPrint(TRACE_LEVEL_INFORMATION, " <--> SRB_FUNCTION_RESET_LOGICAL_UNIT Target (%d::%d::%d), SRB 0x%p\n", SRB_PATH_ID(Srb), SRB_TARGET_ID(Srb), SRB_LUN(Srb), Srb);
-            CompletePendingRequests(DeviceExtension);
-            SRB_SET_SRB_STATUS(Srb, SRB_STATUS_SUCCESS);
-            return TRUE;
-
+            switch (adaptExt->action_on_reset) {
+                case VioscsiResetCompleteRequests:
+                    RhelDbgPrint(TRACE_LEVEL_INFORMATION, " Completing all pending SRBs\n");
+                    CompletePendingRequests(DeviceExtension);
+                    SRB_SET_SRB_STATUS(Srb, SRB_STATUS_SUCCESS);
+                    return TRUE;
+                case VioscsiResetDoNothing:
+                    RhelDbgPrint(TRACE_LEVEL_INFORMATION, " Doing nothing with all pending SRBs\n");
+                    SRB_SET_SRB_STATUS(Srb, SRB_STATUS_SUCCESS);
+                    return TRUE;
+                case VioscsiResetBugCheck:
+                    RhelDbgPrint(TRACE_LEVEL_INFORMATION, " Let's bugcheck due to this reset event\n");
+                    KeBugCheckEx(0xDEADDEAD, (ULONG_PTR)Srb, SRB_PATH_ID(Srb), SRB_TARGET_ID(Srb), SRB_LUN(Srb));
+                    return TRUE;
+            }
         case SRB_FUNCTION_WMI:
             VioScsiWmiSrb(DeviceExtension, Srb);
             return TRUE;

diff --git a/vioscsi/vioscsi.h b/vioscsi/vioscsi.h
@@ -58,7 +58,8 @@ typedef struct VirtIOBufferDescriptor VIO_SG, *PVIO_SG;
 #define IO_PORT_LENGTH          0x40
 #define MAX_CPU                 256
 
-#define MAX_PH_BREAKS           "PhysicalBreaks"
+#define REGISTRY_MAX_PH_BREAKS           "PhysicalBreaks"
+#define REGISTRY_ACTION_ON_RESET         "VioscsiActionOnReset"
 
 
 /* Feature Bits */
@@ -271,6 +272,12 @@ typedef struct virtio_bar {
     BOOLEAN           bPortSpace;
 } VIRTIO_BAR, *PVIRTIO_BAR;
 
+typedef enum ACTION_ON_RESET {
+    VioscsiResetCompleteRequests,
+    VioscsiResetDoNothing,
+    VioscsiResetBugCheck = 0xDEADDEAD,
+} ACTION_ON_RESET;
+
 typedef struct _ADAPTER_EXTENSION {
     VirtIODevice          vdev;
 
@@ -326,10 +333,11 @@ typedef struct _ADAPTER_EXTENSION {
     UCHAR                 prod_id[16 + 1];
     UCHAR                 rev_id[4 + 1];
     BOOLEAN               reset_in_progress;
+    ACTION_ON_RESET       action_on_reset;
 #if (NTDDI_VERSION >= NTDDI_WINTHRESHOLD)
     ULONGLONG             fw_ver;
 #endif
-}ADAPTER_EXTENSION, * PADAPTER_EXTENSION;
+} ADAPTER_EXTENSION, * PADAPTER_EXTENSION;
 
 #ifndef PCIX_TABLE_POINTER
 typedef struct {