On 08/20/2010 12:12 PM, Wei Yongjun wrote: > Hi all: > > I have the following patch to SIDT emulation instruction, but it > does not work because we can not writeback more then 8 bytes > memory, the SIDT under PROTO64 is 10 bytes. > > I change the code to write twice, first time write limit, then write > address, and it does not work too. If I change the c->dst.bytes > to only writeback 8 bytes, it will writeback 8 bytes. > > Is this a bug of KVM or it is the IO limit? It's complicated... if you write to real memory, then you can do c->dst.* = ...; writeback(); c->dst.* = ...; writeback(); However that will not work for mmio, since writeback() only schedules mmio writes which are retired later on. As a temporary fix we can extend writeback() to support 2+N byte writes (and fail them on mmio). See also the sse-mmio branch on kvm.git on how to support >8 byte mmio. Longer term I'd like a queue based approach so we can have any number of reads and writes for an instruction, and have multiple writes possible for mmio. It would look something like this: - every read or write is assigned a position in the queue starting from 0 - the queue is maintained across invocations of x86_emulate_insn() - if an operation position is already in the queue, we read the value from the queue, and ignore writes (e.g. a replay after exit to userspace) - if an operation position is not already in the queue, we add it. If we cannot satisfy it, we exit to userspace - need to check read-after-write for collision in the queue. There's some beginning of that already (see struct read_cache). For now we can live without SIDT to mmio, that means you can do the simple fix to writeback() above. But you will only be able to test is with realmode.flat, not with emulator.flat since that requires mmio. -- error compiling committee.c: too many arguments to function -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
