[lldb][AArch64] Fix tagged watch test on Graviton 3

[DavidSpickett]

Summary

During __do_global_dtors_aux glibc sets a flag that is right
next to the global variable. This is done using a store byte.

On QEMU the watchpoints are handled with a finer granularity
than real hardware, so this wasn't a problem. On Graviton 3
(and Mountain Jade, though this test won't run there) watchpoints
look at larger chunks of memory.

This means that the final continue actually stops in __do_global_dtors_aux
instead of exiting.

We could fix this by padding the global to be away from the flag,
but that is fiddly and it is easier just to remove the watchpoint
before the final continue. We have already verified it worked by that
point.

[jasonmolenda]

Interesting. I remember reading in the recent arm arm docs about how there may be false watchpoint notifications when a processor is in SVE Streaming Mode and a write happens near a watchpoint, but I didn't realize that could happen outside of that context. I'm working on using power-of-2 mask watchpoints on Darwin and one of the problems I'll need to handle is similar, where a user watches a 48 byte object, I need to watch 64 bytes and ignore writes to the other 16. My current thinking is to add a new type of watchpoint in addition to "read" and "write" -- "modify", where lldb will compare the bytes of the watched object and silently continue if they are unmodified. I suspect that's what most people using write watchpoints would actually want, outside of this issue of false watchpoint hits.

[DavidSpickett]

I suspect that's what most people using write watchpoints would actually want, outside of this issue of false watchpoint hits.

Seems like the sensible route.

The other option is understanding every instruction's behaviour, which is in theory possible but I wouldn't like to do it for SVE scatter stores for example.

lldb/test/API/commands/watchpoints/watch_tagged_addr/TestWatchTaggedAddress.py

def verify_watch_hits(self):
         # We should be stopped again due to the watchpoint (read_write type).
         # The stop reason of the thread should be watchpoint.
         self.expect(
             "thread backtrace",
             STOPPED_DUE_TO_WATCHPOINT,
             substrs=["stop reason = watchpoint"],
         )
 
+        # Use the '-v' option to do verbose listing of the watchpoint.
+        # The hit count should now be 2.
+        self.expect("watchpoint list -v", substrs=["hit_count = 2"])
+
+        # On some hardware, during __do_global_dtors_aux a flag is set near
+        # the global which can trigger the watchpoint. So we must remove it.
+        self.runCmd("watchpoint delete 1")
         self.runCmd("process continue")
 
         # There should be no more watchpoint hit and the process status should
         # be 'exited'.
         self.expect("process status", substrs=["exited"])
-
-        # Use the '-v' option to do verbose listing of the watchpoint.
-        # The hit count should now be 2.
-        self.expect("watchpoint list -v", substrs=["hit_count = 2"])