openhwgroup
HPTW accesses not checked by PMA/P checkers
Each memory access in a HPTWalk needs to be potentially flushed if the PMA/P checkers generate an access fault. Specially the store on UDPATE_PTE needs to check for access violation. // I think the solution is to do 1 of the following // 1. Allow the HPTW to generate exceptions and stop walking immediately. // 2. If the store would generate an exception don't store to dcache but still write the TLB. When we go back // to LEAF then the PMA/P. Wait this does not work. The PMA/P won't be looking a the address in the table, but // rather than physical address of the translated instruction/data. So we must generate the exception.
Line 272 of hptw.sv
Logic bug not caught by regression tests.
This was fixed a while ago, but there isn't a test to show that is works.