Re: [PATCH xenvif] Replace uses of MmAllocatePagesForMdlEx in __AllocatePages()...

[Jan Bakuwel <jan.bakuwel@xxxxxxxxx>]

Hi Paul,

On 18/06/20 3:02 am, Paul Durrant wrote:
> > -----Original Message-----
> > From: win-pv-devel <win-pv-devel-bounces@xxxxxxxxxxxxxxxxxxxx> On Behalf Of 
> > Paul Durrant
> > Sent: 16 June 2020 14:03
> > To: win-pv-devel@xxxxxxxxxxxxxxxxxxxx
> > Cc: Paul Durrant <pdurrant@xxxxxxxxxx>
> > Subject: [PATCH xenvif] Replace uses of MmAllocatePagesForMdlEx in 
> > __AllocatePages()...
> >
> > From: Paul Durrant <pdurrant@xxxxxxxxxx>
> >
> > ... again.
> >
> > Commit 4f85d004 "Replace uses of MmAllocatePagesForMdlEx in __AllocatePage"
> > modified __AllocatePage() (only) to use
> > MmAllocateContinguousMemorySpecifyCache() as its source of memory. As stated
> > in that commit:
> >
> > "Windows appears to have an edge case bug in which zeroing memory using
> > MmAllocatePAgesForMdlEx (which in Win 10 1803 happens even if you specify
> > MM_DONT_ZERO_ALLOCATION) can cause a BSOD 139 1e."
> >
> > That commit was reverted by e8fe14f6 since it was believed that the bug in
> > Windows had been fixed. Subsequently, however, the same symptoms have been
> > seen with recent updates of Server 2019. The stack is generally of the form:
> >
> > nt!KeBugCheckEx
> > nt!KiBugCheckDispatch+0x69
> > nt!KiFastFailDispatch+0xd0
> > nt!KiRaiseSecurityCheckFailure+0x30e
> > nt!KiAcquireThreadStateLock+0x11fa90
> > nt!KeSetIdealProcessorThreadEx+0xd0
> > nt!MiZeroInParallelWorker+0x115016
> > nt!MiZeroInParallel+0x11c
> > nt!MiInitializeMdlBatchPages+0x2ae
> > nt!MiAllocatePagesForMdl+0x192
> > nt!MmAllocatePartitionNodePagesForMdlEx+0xc9
> > nt!MmAllocatePagesForMdlEx+0x4d
> >
> > Hence, this patch re-instates the fix originally put in place by 4f85d004
> > but generalizes it to include AllocatePages() rather than just
> > AllocatePage().
> >
> > Reported-by: Jan Bakuwel <jan.bakuwel@xxxxxxxxx>
> > Signed-off-by: Paul Durrant <pdurrant@xxxxxxxxxx>
> > ---
> >
> > This fix will also be propogated to all other PV drivers.
> This is actually failing in xenvbd, trying to allocate 2 pages for the shared 
> ring, so I'm going to try the simpler alternative of
> continuing to use MmAllocatePagesForMdlEx() but without passing the "don't 
> zero" flag.
Many thanks. Will you apply the fix to both 8.2.2 and 9.0 drivers?

I've raised on issue in the xcp-ng/xcp github repo linking to the mail 
archive of this list, so hopefully the patches will be pulled in there 
as well. The XCP-NG driver (so far) is the only one I can install 
without issues on Windows 2019 Server.
Jan