<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/62840>62840</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            lldb's frame diagnose cannot handle a DW_AT_frame_base that uses the CFA
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            new issue
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          khuey
      </td>
    </tr>
</table>

<pre>
    From D143463:

Based on code inspection what I think is happening is that we're getting to this point https://github.com/llvm/llvm-project/blob/d9610b4a56c532614545eef5995362e99b776535/lldb/source/Expression/DWARFExpression.cpp#L2676. DWARFExpression::MatchesOperand takes the operand from a crashing instruction (e.g. [rbp + 42]) and tries to match that up with the locations of the various variables provided in the DWARF. It does this symbolically, looking for exactly two forms of DWARF expressions in variable locations, DW_OP_regN/x and DW_OP_fbreg <offset> where DW_AT_frame_base is itself DW_OP_regN/x. Because I changed DW_AT_frame_base to be DW_OP_call_frame_cfa DW_OP_consts <offset> DW_OP_plus, nothing on the stack is recognized anymore. There's no code here that knows how to deal with DW_OP_call_frame_cfa.

I don't see any easy way to fix this. The CFA is only available at this point in value form (via StackFrame's StackID m_cfa). Despite the comments, StackFrame::GetFrameBaseExpression really does return the DW_AT_frame_base expression, not the CFA (the author of the comment describing the function seems not to have understood that the frame base and the canonical frame address are not the same thing). With the symbolic form of the CFA available we could recognize the sequence DW_AT_location = DW_OP_fbreg <offset1>, DW_AT_frame_base = DW_OP_call_frame_cfa DW_OP_consts <offset2> DW_OP_plus, CFA = rXX + offset3 and match it to e.g. mov [%rXX + offset4], %rax where offset4 = offset1 + offset2 + offset3, but that would require a bunch of work in lldb to plumb the symbolic form of the CFA up out of the unwinding layer to somewhere where it's available to CommandObjectFrameDiagnose.
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJyUVU2P4zYM_TXKhagRyx-JDznMTDbFAG23aBfYvQ1ki47VsSWvJCeT_vqCtLPJfqBoL5OxKJGPj3ykCsEcLeJOFI-i2K_UFDvnd6_dhJdV7fRld_BugH2aZ3mZiexBrPdivfx9VAE1OAuN0wjGhhGbaJyFc6ciPEPsjH0FE6BT44jW2CN9RDKeUciNRzhijHQeHd0OMDpjI3QxjoGCyYOQh6OJ3VQnjRuEPPT96frz0-jdX9hEIQ9172ohD7oq03Wdq6JsikyWaV7kBWJbVFWRlRKrqt5syiIr2IGmF8FNvkEhD-_eRo8hGGeFPOw_Pvxxd5I04yhk9ossN2UC3xgJZvbwq4pNh-H9iF5ZDVG9ImWK4JaTlmhU0HgVOibChuinmS4ht5gcExDFo69HEPIRcimKvZAVsDdvyJuDgaLMBE4jnE3sOEbvGkWOAriWD07KGzcF_lV1jwFG705GowZj-QYnkcBzBO0YqQkQLkPtetOovr8I-QS9c68EtXUe8E01sb9APDv6HjgUOwH8QkUg79eYN1Tka__x5f3vLx6Pvwl5eOOs5qO29ngEkT25tg0YRfYOzh16Qvjy8OGl9WrAl1oFpNYxMWDffuMsgUds1BQQnqHplD2i_v5xdFDj8pASXGxNq66HzoYYvgYyW8Z-4hSsi1w5NzMYomq4uT027mjN36hB2cvgPCbwgVIQchPAulkdnBNX7tW6c4DOnQmURtXPhfwRtuRebc-gqTk3EQIihQJU4QJndSFHrXnjKnJseDo8EDRn-wuokzI9V0TFe5FxrfoJuZ7Ugyej4E_K6kDxGT1_Pu9hIDRCVgnsMYwmIlPQuGFAG5mdu4csiJ8x8hfNiJtYwCN119x0HuPkr-34Tb3wTo5MPV-jtITc0r_zoLo2_IIENIbGm5onSofQTnZWWEAcwuzGQadOCJPV6EN0Ts9V4esUHzg-y44cK-ssSWIxKq0JFyiPX1AFMnBvMEMfr7K86mkmeEFKKdwqciboU69vTTS_xM8T2ubKy1VJILL9j2WTiuzdIrOveby9-E9dL79ve-Y824P_9Ikn03wxY4bmeWSYVZ5ggzvRFBOy-Pp6ztPsCcig3haFLyb2vqRx90TeR6O39RSX3bEw9nkyHkFBPdmmI37Pzr9SW9NsJ0hjPw31v5diGsFN8Xoy2bOxmrqnVxf05CO4AWe4818TWRi3EkYHT24YlNXva9pG3PV7o47WBUxWepfpKqvUCndpuc3LqsyyfNXtUqVlUdVls0m36XZd4lrnebXNi0ZiUaTtyuzkWmbrQqZplco0T9JW622Tt8VG5c26qES-xkGZPqFdmDh_XJkQJtyVcpuvV72qsQ-81qW0eAY2Ckl7ZeV3vD_r6RhEvu5NiOHmJZrY427ej5uw9L1eEiJBUN93ymqaKD-YtLydwrL-ng4Pq8n3u_-90RluEPLA6fwTAAD__xm08Hw">