arm: virtualization: don't allow hypervisor to issue std calls

There is standing issue with having two versions of OP-TEE binary:
with virtualization enabled and without it. Correct variant needs to
be present on board before booting rest of the system.

If non-virtualized variant is present and user tries to boot a
hypervisor, hypervisor can (and should) detect that OP-TEE does not
provide OPTEE_SMC_SEC_CAP_VIRTUALIZATION capability and fail
gracefully.

On other hand, when virtualized variant of OP-TEE is booted, but user
then boots directly into Linux (or any other OS), OP-TEE crashes:

E/TC:0 0 0 Core data-abort at address 0xffffffffffffffa0 (translation fault)
E/TC:0 0 0  esr 0x96000044  ttbr0 0x4418d000   ttbr1 0x00000000   cidr 0x0
E/TC:0 0 0  cpu #0          cpsr 0x00000184
E/TC:0 0 0  x0  0000000032000004 x1  0000000000000004
E/TC:0 0 0  x2  000000008183c000 x3  0000000000000000
E/TC:0 0 0  x4  0000000000000000 x5  0000000000000000
E/TC:0 0 0  x6  0000000000000000 x7  0000000000000000
E/TC:0 0 0  x8  0000000000000000 x9  0000000000000000
E/TC:0 0 0  x10 0000000000000000 x11 0000000000000000
E/TC:0 0 0  x12 0000000000000000 x13 0000000000000000
E/TC:0 0 0  x14 0000000000000000 x15 0000000000000000
E/TC:0 0 0  x16 0000000000000000 x17 0000000000000000
E/TC:0 0 0  x18 0000000000000000 x19 0000000000000000
E/TC:0 0 0  x20 0000000000000000 x21 0000000000000000
E/TC:0 0 0  x22 0000000000000000 x23 0000000000000000
E/TC:0 0 0  x24 0000000000000000 x25 0000000000000000
E/TC:0 0 0  x26 0000000000000000 x27 0000000000000000
E/TC:0 0 0  x28 0000000000000000 x29 0000000000000000
E/TC:0 0 0  x30 0000000044103ce4 elr 0000000044106314
E/TC:0 0 0  sp_el0 0000000000000000
E/TC:0 0 0 TEE load address @ 0x44100000
E/TC:0 0 0 Call stack:
E/TC:0 0 0  0x44106314 thread_handle_std_smc at core/arch/arm/kernel/thread_optee_smc.c:62
E/TC:0 0 0 Panic 'unhandled pageable abort' at core/arch/arm/kernel/abort.c:584 <abort_handler>
E/TC:0 0 0 TEE load address @ 0x44100000
E/TC:0 0 0 Call stack:
E/TC:0 0 0  0x44107e14 print_kernel_stack at core/arch/arm/kernel/unwind_arm64.c:89
E/TC:0 0 0  0x44114ffc __do_panic at core/kernel/panic.c:73
E/TC:0 0 0  0x44107050 get_fault_type at core/arch/arm/kernel/abort.c:500

This crash happens because virtualization code has special case for
guest_id == HYP_CLNT_ID. This case is needed to allow hypervisor to
call fast SMCs, so it can check OP-TEE version, capabilities and ask
OP-TEE to create/destroy guest partitions. Problem is that
thread_handle_std_smc() assumes that virt_set_guest() really sets the
guest partition, which does not happen in this special case.

This patch removes this special case from virt_set_guest(). Instead
thread_handle_fast_smc() function checks for HYP_CLNT_ID explicitly.

If hypervisor really want to be able to issue STD calls, it should
create a partition for itself using OPTEE_SMC_VM_CREATED call.

With this patch applied, virtualized variant of OP-TEE does not crash
anymore when users tries to boot into a baremetal setup.

Signed-off-by: Volodymyr Babchuk <volodymyr_babchuk@epam.com>
Reviewed-by: Jens Wiklander <jens.wiklander@linaro.org>

core/arch/arm/kernel/thread_optee_smc.c

@@ -35,7 +35,7 @@ void thread_handle_fast_smc(struct thread_smc_args *args)
 	thread_check_canaries();
 
 	if (IS_ENABLED(CFG_NS_VIRTUALIZATION) &&
-	    virt_set_guest(args->a7)) {
+	    (virt_set_guest(args->a7) && args->a7 != HYP_CLNT_ID)) {
 		args->a0 = OPTEE_SMC_RETURN_ENOTAVAIL;
 		goto out;
 	}

core/arch/arm/kernel/virtualization.c

@@ -382,8 +382,6 @@ TEE_Result virt_set_guest(uint16_t guest_id)
 	}
 	cpu_spin_unlock_xrestore(&prtn_list_lock, exceptions);
 
-	if (guest_id == HYP_CLNT_ID)
-		return TEE_SUCCESS;
 	return TEE_ERROR_ITEM_NOT_FOUND;
 }