[RFC] Another attempt at Provision TEE threads for system invocation

[jenswi-linaro]

An alternative to what is proposed in https://github.com/linaro-swg/linux/pull/109

@etienne-lms feel free to use these patches if they make sense to you.

[jenswi-linaro]

There is no fallback if an earlier execution stage left a suspended tee thread open. Worst case effect would be regular sessions will not reach OP-TEE, in case sys_thread_cnt == total_thread_cnt - lost_thread_cnt. Should it be addressed?

The worst case is a total deadlock in the secure world if threads are left suspended, that's basically a DOS by starvation of CPU.

A warning in optee_smc_do_call_with_arg()?

      if (res.a0 == OPTEE_SMC_RETURN_ETHREAD_LIMIT) {
+         WARN_ONCE(!w->sys_thread, "TEE thread count mismatch\n")

          /*
           * Out of threads in secure world, wait for a thread
           * become available.
           */
          optee_cq_wait_for_completion(&optee->call_queue, &w);
      } else if (OPTEE_SMC_RETURN_IS_RPC(res.a0)) {

Yes, it makes sense.

[jenswi-linaro]

Adding the WARN_ONCE

[etienne-lms]

I've posted the changes to LKML: system session patches v5. Note I removed the call to WARN_ONCE() I suggested. I found the backtrace information printed does not add much value.

[etienne-lms]

Patch v6 posted, see https://lore.kernel.org/lkml/20230505173012.881083 series.

[etienne-lms]

Closing. The feature is being addressed through the LKML: see latest post https://lore.kernel.org/lkml/20231030084812.905549-1-etienne.carriere@foss.st.com/.