On 02/20/2012 04:30 PM, Michael S. Tsirkin wrote:
On Wed, Dec 14, 2011 at 08:43:20AM -0500, Stefan Berger wrote:+ +static void *tpm_passthrough_thread(void *d) +{ + TPMPassthruThreadParams *thr_parms = d; + TPMPassthruState *tpm_pt = thr_parms->tb->s.tpm_pt; + uint32_t in_len; + + dprintf("tpm_passthrough: THREAD IS STARTING\n"); + + /* start command processing */ + while (!tpm_pt->thread_terminate) {IMO you want to checvk thread_terminate undet a lock/
The way the termination flag is used along with the condition it isn't necessary to call it with a lock held. It may run into the condition but then be woken up with a signal and then again check the flag and terminate.
+ /* receive and handle commands */ + in_len = 0; + do { + dprintf("tpm_passthrough: waiting for commands...\n"); + + if (tpm_pt->thread_terminate) { + break; + } +This will have to be duplicate in each backend, no?
There is this sequence of code around the condition that helps us waiting for the front-end to send a command that is indeed repeated in every backend. I isolated that now but would like to keep the general thread more flexible and inside the individual backends. There is quite some code that is different in every backend.
So I would say make this logic part of backend and expose some api.
Ok, a single function for now: void tpm_backend_wait_for_command(TPMState *tpm_state);
You can then call it from backend or invoke backend callbacks from frontend.
I was only going to invoke it from within the backend thread. Stefan
