Using the -p x flag to add a delay between requests also means that hitting ctrl+c takes x seconds to exit,

As discussed in the discord channel. Having a whitelist/blacklist approach where you can tell to (not)instrument specific libraries in QEMU mode. For example by passing library names in an env variable. Something like INSTRUMENT_QEMU_BLACKLIST="libc.so,another.so".
Could be done by replacing all the checks if (cur_loc < afl_code_start || cur_loc >= afl_code_end) with the traverse of a list (in

grep unknown opentitan.fasm  | wc
    557    6127   52651

557 bits isn't too many. Most of these are likely related to the DSP, as 1 DSP is being used:

DSP_L_X66Y110.DSP48.DSP_0.AREG_0
DSP_L_X66Y110.DSP48.DSP_0.BREG_0
DSP_L_X66Y110.DSP48.DSP_0.MASK[45:0] = 46'b1111111111111111111111111111111111111111111111
DSP_L_X66Y110.DSP48.DSP_0.ZADREG[0]
DSP_L_X66Y110.DSP48.DSP_0.ZAL

Bug Report

After updating to the latest test-tls13-obsolete-curves.py there were ~100 failures due to a new check that extensions are rejected due to a mismatch of the curve with the key share. While that's a good test, it is unrelated to the purpose of this test file. As such one updating to the new version that does not want to comply with the newly introduced requirement, would have to dis

GRR uses GRANARY_ASSERT, which is kind of a weird holdover from writing DBTs that can't depend on system libraries, and it's also very C-like. Now that we've made GRR start using cxx-common, we could easily upgrade it to use Google Log's CHECK macro.