cancellation: Unmask inside of invoke command #732
Conversation
There was a problem hiding this comment.
cancellation: Unmask inside of invoke command
The specification seems to indicate (see #731)
that the cancellation mask is reset at the start of every entry point.
As a result, the TEE_UnmakCancellation() should not be called at the end of
TA_CreateEntryPoint(), but at the start of TA_OpenSessionEntryPoint().
s/TA_CreateEntryPoint()/TA_OpenSessionEntryPoint()/
s/TA_OpenSessionEntryPoint()/TA_InvokeCommandEntryPoint()/
With that fixed:
Reviewed-by: Jerome Forissier <[email protected]>
Sure, if you don't mind, please provide such a test in a separate PR. I can take care of the secure side of things. |
|
Hi @c-auger, thanks for the patch!
That would be nice, please feel free to open a new PR. Then we can consider how to make the implementation compliant. |
|
@c-auger we need a |
There was a problem hiding this comment.
Commit header line should rather be "ta: os_test: Unmask cancellation from invoke command handler".
Acked-by: Etienne Carriere <[email protected]>
The specification seems to indicate (see OP-TEE#731) that the cancellation mask is reset at the start of every entry point. As a result, the TEE_UnmakCancellation() should not be called at the end of TA_OpenSessionEntryPoint(), but at the start of TA_InvokeCommandEntryPoint(). Signed-off-by: Cedric Auger <[email protected]> Reviewed-by: Jerome Forissier <[email protected]> Acked-by: Etienne Carriere <[email protected]>
|
@c-auger thanks! |
The specification seems to indicate (see #731)
that the cancellation mask is reset at the start of every entry point.
As a result, the TEE_UnmakCancellation() should not be called at the end of
TA_OpenSessionEntryPoint(), but at the start of TA_InvokeCommandEntryPoint().
Also, note that this patch should not break the current test.
Still, there is a test which could be written to test that unmasking cancellations
really does not affect subsequent entry points. There is no such test for now,
and if one were designed, then the current implementation would report a failure to this test,
until the implementation is fixed with respect to the Global Platform specification.
If you need such a test, I could consider writing one.