diff options
| author | Alan Wu <[email protected]> | 2021-08-26 14:58:13 -0400 |
|---|---|---|
| committer | Alan Wu <[email protected]> | 2021-10-20 18:19:39 -0400 |
| commit | 54db64f7a59f1f14b280a80491336a128f792b42 (patch) | |
| tree | 10c897bde6a2d600a869e3982f2716f6062f1d21 /yjit_codegen.c | |
| parent | 4b815abb373801b94d2471997f1467421620c1e5 (diff) | |
filter out internal events. add comments. reorder
Diffstat (limited to 'yjit_codegen.c')
| -rw-r--r-- | yjit_codegen.c | 11 |
1 files changed, 8 insertions, 3 deletions
diff --git a/yjit_codegen.c b/yjit_codegen.c index 41670189b7..db0694523e 100644 --- a/yjit_codegen.c +++ b/yjit_codegen.c @@ -3842,12 +3842,17 @@ static void invalidate_all_blocks_for_tracing(const rb_iseq_t *iseq); // they are waiting for a return from a C routine. For every routine call, we // patch in an exit after the body of the containing VM instruction. This makes // it so all the invalidated code exit as soon as execution logically reaches -// the next VM instruction. +// the next VM instruction. The interpreter takes care of firing the tracing +// event if it so happens that the next VM instruction has one attached. +// // The c_return event needs special handling as our codegen never outputs code // that contains tracing logic. If we let the normal output code run until the // start of the next VM instruction by relying on the patching scheme above, we -// would fail to fire the c_return event. To handle it, we patch in the full -// logic at the return address. See full_cfunc_return(). +// would fail to fire the c_return event. The interpreter doesn't fire the +// event at an instruction boundary, so simply exiting to the interpreter isn't +// enough. To handle it, we patch in the full logic at the return address. See +// full_cfunc_return(). +// // In addition to patching, we prevent future entries into invalidated code by // removing all live blocks from their iseq. void |