Date: Fri, 28 Apr 2023 17:11:41 +0300 From: Christos Margiolis <christos@freebsd.org> To: status-updates@freebsdfoundation.org Cc: freebsd-dtrace@freebsd.org, markj@freebsd.org, jrm@freebsd.org Subject: [Development report #8] Improve the kinst DTrace provider Message-ID: <20230428141141.y74jra2tfvirwcdd@pleb>
next in thread | raw e-mail | index | archive | help
A brief overview of the things I've been working on since the last update: * The RISC-V port of kinst was crashing if we tried to access the address pointed to by `frame->tf_tp`. This was a result of TP being unitialized if accessed by DTrace in probe context. I wrote a patch [1] which saves the thread pointer in both userland and kernel mode exceptions. * Emulated all PC-relative RV64[IC] instructions. * Fixed various bugs in both the emulation and the double breakpoint mechanism. There is however still an unsolved bug which hangs the system if we try to trace a very specific block of instructions in vm_fault() (instructions 4436 to 4450). Once this is fixed, the RISC-V port will most likely be complete. * Added register bindings to RISC-V DTrace [2]. * Exposed dtrace_instr_size() to userland and implemented it for RISC-V [3]. Other architectures are meant to also implement this function as well and use it across DTrace code. * In report #5 I mentioned that cpu_switch() was safe to trace. It turns out it's not. * Submitted and working on various as-of-now open PRs related to DTrace and kinst. [1] https://cgit.freebsd.org/src/commit/?id=d95fbf4e1a12565908b04b442263fe60c9e890b4 [2] https://cgit.freebsd.org/src/commit/?id=1fef7abdc76bd6f0c93775bd0c1b62b523abd20d [3] https://cgit.freebsd.org/src/commit/?id=080e56a6c98c7e7a241a4260d4c19d3cad5b22af
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20230428141141.y74jra2tfvirwcdd>