Description of problem:

There are patches submitted upstream to fix RSS configuration and TX path race condition.
Those patches need additional work to become part of the code base.

https://github.com/YanVugenfirer/kvm-guest-drivers-windows/pull/83

--->>>
There is a race between ParaNdis_SetupRSSQueueMap and
ParaNdis6_SendNetBufferLists due to the lack of r/w sync:

Processor 1 call stack and code:
NDIS!ndisMInvokeOidRequest calls
ParaNdis6_OidRequest calls (through OidsDB callback table)
OIDENTRYPROC(OID_GEN_RECEIVE_SCALE_PARAMETERS, 0,0,0, ohfSet
| ohfSetMoreOK, RSSSetParameters),
RSSSetParameters calls
ParaNdis_SetupRSSQueueMap

if (pContext->RSS2QueueLength && 
    pContext->RSS2QueueLength < rssTableSize)
{
    DPrintf(0, ("[%s] Freeing RSS2Queue Map\n", __FUNCTION__));
--> NdisFreeMemoryWithTagPriority(pContext->MiniportHandle,
pContext->RSS2QueueMap, PARANDIS_MEMORY_TAG);
pContext->RSS2QueueLength = 0;
}

if (!pContext->RSS2QueueLength)
{
    pContext->RSS2QueueLength = USHORT(rssTableSize);
    pContext->RSS2QueueMap = (CPUPathesBundle **)
        NdisAllocateMemoryWithTagPriority(pContext->MiniportHandle,
        rssTableSize * sizeof(*pContext->RSS2QueueMap),
        PARANDIS_MEMORY_TAG, NormalPoolPriority);
    if (pContext->RSS2QueueMap == nullptr)
--> {
DPrintf(0, ("[%s] - Allocating RSS to queue mapping failed\n",
FUNCTION));
NdisFreeMemoryWithTagPriority(pContext->MiniportHandle,
cpuIndexTable, PARANDIS_MEMORY_TAG);
return NDIS_STATUS_RESOURCES;
}

--> NdisZeroMemory(pContext->RSS2QueueMap,
sizeof(*pContext->RSS2QueueMap) * pContext->RSS2QueueLength);
}

Processor 2 call stack and code:
NDIS!NdisSendNetBufferLists calls
netkvm!ParaNdis6_SendNetBufferLists

--> pContext->RSS2QueueMap[indirectionIndex]->txPath.Send(pNBL);

Another race is indirectionIndex calculation via RSSHashMask.
Mask can be updated before map update. So there is a need to check
that index fits into bounds. (We suppose that there is NO_PROBLEM
if packet will be sent not exactly in the proper queue by its hash value).

This patch set introduces r/w spinlock as well as index inbound validation.

Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info: