<html xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"> <head> <meta http-equiv="Content-Type" content="text/html; charset=Windows-1252"> <meta name="Generator" content="Microsoft Word 15 (filtered medium)"> <style></style> </head> <body lang="EN-US" link="#0563C1" vlink="#954F72" style="word-wrap:break-word"> <div class="WordSection1"> > On Fri, Jul 29, 2022 at 09:38:35 +0200, Michal Prívozník wrote:<o:p></o:p> > > On 7/28/22 14:41, Peter Krempa wrote:<o:p></o:p> > > > On Wed, Jul 27, 2022 at 12:34:57 +0200, Michal Privoznik wrote:<o:p></o:p> > > >> From: Lin Yang <lin.a.yang@intel.com><o:p></o:p> > > >><o:p></o:p> > > >> With NUMA config:<o:p></o:p> > > >><o:p></o:p> > > >> <devices><o:p></o:p> > > >> ...<o:p></o:p> > > >> <memory model='sgx-epc'><o:p></o:p> > > >> <source><o:p></o:p> > > >> <nodemask>0-1</nodemask><o:p></o:p> > > >> </source><o:p></o:p> > > >> <target><o:p></o:p> > > >> <size unit='KiB'>512</size><o:p></o:p> > > >> <node>0</node><o:p></o:p> > > >> </target><o:p></o:p> > > >> </memory><o:p></o:p> > > >> ...<o:p></o:p> > > >> </devices><o:p></o:p> > > >><o:p></o:p> > > >> Without NUMA config:<o:p></o:p> > > >><o:p></o:p> > > >> <devices><o:p></o:p> > > >> ...<o:p></o:p> > > >> <memory model='sgx-epc'><o:p></o:p> > > >> <target><o:p></o:p> > > >> <size unit='KiB'>512</size><o:p></o:p> > > >> </target><o:p></o:p> > > >> </memory><o:p></o:p> > > >> ...<o:p></o:p> > > >> </devices><o:p></o:p> > > >><o:p></o:p> > > >> Signed-off-by: Lin Yang <lin.a.yang@intel.com><o:p></o:p> > > >> Signed-off-by: Michal Privoznik <mprivozn@redhat.com><o:p></o:p> > > >> ---<o:p></o:p> > > > <o:p></o:p> > > > [...]<o:p></o:p> > > > <o:p></o:p> > > >> diff --git a/docs/formatdomain.rst b/docs/formatdomain.rst<o:p></o:p> > > >> index 1ed969ac3e..bdd0fcea8e 100644<o:p></o:p> > > >> --- a/docs/formatdomain.rst<o:p></o:p> > > >> +++ b/docs/formatdomain.rst<o:p></o:p> > > >> @@ -7940,6 +7940,20 @@ Example: usage of the memory devices<o:p></o:p> > > >> <current unit='KiB'>524288</current><o:p></o:p> > > >> </target><o:p></o:p> > > >> </memory><o:p></o:p> > > >> + <memory model='sgx-epc'><o:p></o:p> > > >> + <source><o:p></o:p> > > >> + <nodemask>0-1</nodemask><o:p></o:p> > > >> + </source><o:p></o:p> > > >> + <target><o:p></o:p> > > >> + <size unit='KiB'>16384</size><o:p></o:p> > > >> + <node>0</node><o:p></o:p> > > >> + </target><o:p></o:p> > > >> + </memory><o:p></o:p> > > >> + <memory model='sgx-epc'><o:p></o:p> > > >> + <target><o:p></o:p> > > >> + <size unit='KiB'>16384</size><o:p></o:p> > > >> + </target><o:p></o:p> > > >> + </memory><o:p></o:p> > > >> </devices><o:p></o:p> > > >> ...<o:p></o:p> > > >> <o:p></o:p> > > >> @@ -7948,7 +7962,9 @@ Example: usage of the memory devices<o:p></o:p> > > >> 1.2.14` Provide ``nvdimm`` model that adds a Non-Volatile DIMM module.<o:p></o:p> > > >> :since:`Since 3.2.0` Provide ``virtio-pmem`` model to add a paravirtualized<o:p></o:p> > > >> persistent memory device. :since:`Since 7.1.0` Provide ``virtio-mem`` model<o:p></o:p> > > >> - to add paravirtualized memory device. :since:`Since 7.9.0`<o:p></o:p> > > >> + to add paravirtualized memory device. :since:`Since 7.9.0` Provide<o:p></o:p> > > >> + ``sgx-epc`` model to add a SGX enclave page cache (EPC) memory to the guest.<o:p></o:p> > > >> + :since:`Since 8.7.0 and QEMU 6.2.0`<o:p></o:p> > > > <o:p></o:p> > > > I don't quite understand from this description whether this is real<o:p></o:p> > > > memory usable by the guest OS or something for internal use by the<o:p></o:p> > > > hypervisor.<o:p></o:p> > > > <o:p></o:p> > > > Specifically which leads me to questioning this is that the example<o:p></o:p> > > > sizes are very tiny compared to real memory sizing.<o:p></o:p> > > <o:p></o:p> > > I admit that I don't know all the details and let somebody from Intel to<o:p></o:p> > > provide them. But basically, SGX works by reserving a chunk of RAM (on<o:p></o:p> > > each NUMA node) which is then encrypted and the processor controls<o:p></o:p> > > access to it. This chunk is referred to as Processor Reserved Memory and<o:p></o:p> > > to my understanding is the maximum size of an enclave. Thus I can<o:p></o:p> > > understand why new <memory/> model was used.<o:p></o:p> > <o:p></o:p> > Even this interpretation would not sit well with the semantics of the<o:p></o:p> > <memory> element.<o:p></o:p> > <o:p></o:p> > > But I'm not sure how QEMU<o:p></o:p> > > accounts for this memory, whether say 16KiB worth of SGX is added to<o:p></o:p> > > whatever current guest has OR it's taken from an existing memory.<o:p></o:p> > <o:p></o:p> > Note that the name expands to "enclave page cache", so even the name<o:p></o:p> > doesn't really seem to be hinting that the memory declared like this is<o:p></o:p> > used by the guest OS.<o:p></o:p> > <o:p></o:p> > Preferrably the documentation added in this patch will clarify that,<o:p></o:p> > because if it is not clarified by the documentation, we will be getting<o:p></o:p> > clarification requests and it's very apparent that we can't answer them<o:p></o:p> > since neither of us actually knows what's going on.<o:p></o:p> > <o:p></o:p> > > <o:p></o:p> > > > <o:p></o:p> > > > Additionally in qemuDomainDefValidateMemoryHotplug you are changing the<o:p></o:p> > > > code to specifically exclude the sizing of the 'sgx-epc' memory devices<o:p></o:p> > > > from the total size of the memory, but this contrasts with the memory<o:p></o:p> > > > _not_ being excluded from the initial memory calculation in<o:p></o:p> > > > virDomainDefGetMemoryInitial which is used to format the initial memory<o:p></o:p> > > > argument ('-m size=...'). Thus at least one of them is wrong.<o:p></o:p> > > > <o:p></o:p> > > > If this is not guest usable memory, then the <memory> element should not<o:p></o:p> > > > be used to bolt this on, but rather add a new element similarly to what<o:p></o:p> > > > we have when AMD SEV is in use.<o:p></o:p> > > > <o:p></o:p> > > <o:p></o:p> > > Alright, I'm hold off reworking these patches per your review until we<o:p></o:p> > > are clear on this.<o:p></o:p> <o:p> </o:p> Basically sgx-epc memory created here is usable by guest OS. KVM SGX creates<o:p></o:p> One new misc device, and QEMU will open '/dev/sgx_vepc' device node to <o:p></o:p> mmap() host EPC memory to guest.<o:p></o:p> <o:p> </o:p> The sgx-epc memory is a separate memory range. Its size will not change the <o:p></o:p> guest OS size that is initialized by ‘-m size=…’. So it should be excluded from<o:p></o:p> calculation in virDomainDefGetMemoryInitial. Thanks for pointing this out.<o:p></o:p> <o:p> </o:p> Thanks,<o:p></o:p> Lin.<o:p></o:p> </div> </body> </html>