FreeCalypso > hg > fc-magnetite
diff src/g23m-gsm/l1/l1_pei.c @ 110:49254db81128
implemented fix for the L1/ETM boot time race condition bug
author | Mychaela Falconia <falcon@freecalypso.org> |
---|---|
date | Thu, 06 Oct 2016 00:13:12 +0000 |
parents | 267cefdca4b6 |
children |
line wrap: on
line diff
--- a/src/g23m-gsm/l1/l1_pei.c Wed Oct 05 23:44:54 2016 +0000 +++ b/src/g23m-gsm/l1/l1_pei.c Thu Oct 06 00:13:12 2016 +0000 @@ -183,6 +183,8 @@ T_PEI_RETURN pei_init (void) #endif { + extern int etm_is_running; /* FreeCalypso addition */ + #if defined (NEW_FRAME) L1_Handle = Handle; #endif @@ -266,6 +268,22 @@ #endif /* + * FreeCalypso addition: TI's original code contains a boot time + * race condition: if the L1A task starts running before ETM over + * in RiViera land has initialized itself, the etm_register() call + * at the beginning of l1a_task() that registers L1TM causes a + * boot time crash. It appears that this race condition has always + * been there, but it has been unmasked when we made a port of R2D + * to the Motorola C139 LCD. + * + * Our solution: we hold off on L1 GPF task startup until ETM + * has been initialized, just like we hold off until the message + * queues L1 needs to send to have been created. + */ + if (!etm_is_running) + return PEI_ERROR; + + /* * Register VSI_CALLER as generic caller entity. */ _osx_open (VSI_CALLER 0, 0);