Google traces: LOAD without data

32 views
Skip to first unread message

Heiner Litz

unread,
Mar 28, 2025, 3:19:28 PMMar 28
to DynamoRIO Users
Hi,
In Google v2 traces, I see category=LOAD instructions with no attached data. I think this could be:
1. REP with count 0
2. CMOV
3. LEA? (counts as load?)
4. other?

Is there a way to determine what these instructions are (REP info seems to have been stripped)?

Assuming REP/CMOV, the presence of a load address would be dynamic and could not be inferred from the instruction decoding (and would require checking the next trace element), correct?

Heiner

--

Enrico Deiana

unread,
Apr 1, 2025, 4:46:07 PMApr 1
to DynamoRIO Users
Hi Heiner,

We did find some miscategorized instructions (e.g., LEA should have had category "math" instead of "load").
This has been fixed by: https://github.com/DynamoRIO/dynamorio/pull/7411.
We might update the public traces and document the change in CHANGELOG.txt in the cloud bucket where the traces are stored.
Stay tuned!

As for the possible cases for the instruction you observed: "REP with count 0" could be a possibility, but "CMOV" is not as it always does the load (https://github.com/DynamoRIO/dynamorio/issues/2577#issuecomment-2767768900).

> Is there a way to determine what these instructions are

No. Preventing that was one of the objectives when releasing these traces (but you're welcome to speculate :) ).
To protect Google's intellectual property we designed a synthetic ISA (DR_ISA_REGDEPS) that these traces have been converted to, and it's a lossy conversion on purpose.
We believe we preserved enough information (register dependencies is a big one) to make these traces useful, but we are always open to suggestions for the next version!

> the presence of a load address would be dynamic and could not be inferred from the instruction decoding (and would require checking the next trace element), correct?

That's right, it requires checking the TRACE_TYPE_READ and TRACE_TYPE_WRITE records after the instruction.
Reply all
Reply to author
Forward
0 new messages