On Wed, Jul 13, 2011 at 1:56 PM, Pekka Enberg <[email protected]> wrote: > On Wed, Jul 13, 2011 at 1:26 PM, Sasha Levin <[email protected]> wrote: >> The problem is that if we received a signal during the read notification >> the write and before receiving the read, we have to go back to >> userspace. >> >> This means that userspace will see same read request twice (once in the >> socket and once in the MMIO exit). > > So the problem is only in ioeventfd_read() if socket_write() succeeds > but socket_read() fails? If so, can we do the socket_read() somewhere > else in the code which is able to restart it? > > AFAICT, both ioeventfd_read() and ioeventfd_write() should -ENOSUPP if > socket_write() fails. If socket_write() succeeds, we should return > -EINTR and teach vcpu_mmio_read() and kernel_pio() to KVM_EXIT_INTR in > those cases. We'll can just restart the read, no? The only > complication I can see is that ioeventfd_read() needs to keep track if > it did the socket_write() already or not.
Maybe struct _ioeventfd could hold a flag that tells ioevenfd_read() that the write part was already successful? -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
