Re: [Gems-users] Abort Trap in LogTM

[Polina Dudnik <pdudnik@xxxxxxxxxxx>]

Can anyone please answer these questions?

Polina

arrvindh shriraman wrote:
> I am trying to understand the abort handler sequence in In version 2.1 
> of logtm
> 1) Simics v9_exception_type doesn't seem to specify anything for trap 
> 0x122.
>     When you set the s.exception_t in SimicsProcessor.C what exactly is 
> the control flow ?
>     Do you get to the user level handler or is this just to faithfully 
> model the performance
> 2) I would also like to understand what this sequence of instructions in 
> XACT_MGR::traptoHandler does...
>   // We temporarily flip the LSB of PSTATE so that simics can access the 
> program       global registers instead of the alternate globals. Note 
> that we are currently  in a system 
> trap.                                                                                                                                                        
>
>   int pstate_rn_no = SIMICS_get_register_number(logical_proc_no, 
> "pstate");                                                                                                   
>
>   uint64 pstate_val = SIMICS_read_register(logical_proc_no, 
> pstate_rn_no);                                                                                                    
>
>   SIMICS_write_register(logical_proc_no, pstate_rn_no, (pstate_val ^ 
> 0x1));                                                                                                   
>
>   int tid_rn_no = SIMICS_get_register_number(logical_proc_no, 
> "g2");                                                                                                          
>
>   SIMICS_write_register(logical_proc_no, tid_rn_no, 
> m_tid[thread]);                                                                                                           
>
>   SIMICS_write_register(logical_proc_no, pstate_rn_no, 
> pstate_val);                                                                                                           
>
>                                                                                                                                                                 &n 
> bsp;            
>   3) and finally the transaction_manager_stub(), tm_trap_handler()
>      How do you actually invoke these given that you don't seem to 
> register either of these function pointers with the simulator. I am 
> assuming RUBY_MAGIC_CALL(60) is the call used to register handler addresses.
>  4) The ATMTP folks seem to be directly modifying the PC and NPC 
> registers to invoke their handlers is their any reason why you are doing 
> it differently ?
> Thanks                                        
>
> Arrvindh Shriraman
> Conputer Science Department
> University of Rochester
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.