TinyEMU Host I/O Flow

The run loop multiplexes host events and CPU slicesEach iteration computes sleep time, polls fds, handles ready host events, refreshes SDL, then executes guest instructions.

virt_machine_run(machine)
  delay = machine sleep duration
  FD_SET terminal if console queue can receive
  net->select_fill(...)
  fs_net_set_fdset(...)
  select(...)
  net->select_poll(...)
  terminal bytes -> virtio_console_write_data()
  sdl_refresh()
  virt_machine_interp(MAX_EXEC_CYCLE)

temu.c:537 run loop temu.c:604 terminal to console

Local block devices are synchronous callbacksThe block backend exposes sector count, read, and write functions to VirtIO block.

Read-only and read-write modes operate directly on the disk image. Snapshot mode overlays writes in `sector_table`, so guest writes are visible to later reads without modifying the original image.

virtio block request
  -> bs->read_async / write_async
  -> local file fseek/fread/fwrite
  -> snapshot writes copy sectors into sector_table
  -> status byte returned through VirtIO descriptor

temu.c:232 local read temu.c:268 local write temu.c:307 block init

HTTP block devices cache split image chunksNetwork block data is fetched block-by-block or by prefetch group, with per-cluster write overlay support.

block_device_init_http(url)
  -> load blk.txt metadata
  -> read request enters bf_rw_async1()
  -> if cluster overlay exists: read modified cluster
  -> else find cached block
       loaded  -> copy from cache
       loading -> return async wait
       missing -> fs_wget(blkNNN.bin)
  -> onload updates cache and resumes request

block_net.c:176 start load block block_net.c:262 rw core block_net.c:370 HTTP read block_net.c:404 HTTP init

Network packets cross an `EthernetDevice` bridgeTap uses a host fd; slirp routes through the user-mode network stack.

Guest TX starts in VirtIO net and calls `write_packet`. Host RX is polled through `select_fill`/`select_poll`; when a packet arrives, the backend calls `device_write_packet`, which copies it into the guest receive queue.

guest TX:
  virtio_net_recv_request(queue 1)
  -> EthernetDevice.write_packet()
  -> tap write(fd) or slirp_input()

host RX:
  select says backend readable
  -> backend reads packet
  -> EthernetDevice.device_write_packet()
  -> virtio_net_write_packet(queue 0)

temu.c:366 tap write temu.c:509 slirp open virtio.c:1153 guest TX virtio.c:1190 host RX

Local 9P maps guest FIDs to host paths and file descriptorsThe VirtIO 9P layer owns protocol framing; `fs_disk.c` owns filesystem operations.

guest 9P request
  -> virtio_9p_recv_request()
  -> fid lookup or create
  -> FSDevice method
       attach -> root path
       walk   -> compose child path
       open   -> opendir/open
       read   -> readdir/read
       write  -> write
  -> marshal 9P reply
  -> virtio_consume_desc()

fs_disk.c:172 attach fs_disk.c:190 walk fs_disk.c:233 open fs_disk.c:349 read fs_disk.c:363 write

Host I/O supplies the outside world