search

seL4 / sel4test

Test suite for seL4.
http://sel4.systems
Other
24 stars 60 forks source link

Hardware debug API: Extend to aarch64 and add single stepping test #111

Closed alwin-joshy closed 7 months ago

alwin-joshy commented 7 months ago

seL4's hardware debug API is currently only implemented for ARMv7 and x86. I am currently implementing it for AARCH64 and as a part of this, have ported the software breakpoint test to AARCH64.

I have also implemented a new test to check that single stepping works on ARM. In doing so, I believe I have found a bug in the implementation which I will make another pull request for soon. This test is simpler than the existing x86 equivalent but I think it should be sufficient. If not, I can port the existing x86 test for single stepping to ARM.

Test with: https://github.com/seL4/seL4/pull/1162

Indanz commented 7 months ago

With 1162 in mind, I think it would be good to test whether code actually gets executed instead of only checking if the instruction pointer advances, so instead of nops just add a value to a register and check if it's correct at the second label.

alwin-joshy commented 7 months ago

With 1162 in mind, I think it would be good to test whether code actually gets executed instead of only checking if the instruction pointer advances, so instead of nops just add a value to a register and check if it's correct at the second label.

Good point. In that case, I'll port the equivalent x86 test, which seems to be more robust.

Indanz commented 7 months ago

With 1162 in mind, I think it would be good to test whether code actually gets executed instead of only checking if the instruction pointer advances, so instead of nops just add a value to a register and check if it's correct at the second label.

Good point. In that case, I'll port the equivalent x86 test, which seems to be more robust.

At first glance I don't really see anything x86 specific in there, perhaps make that test cross-platform with some arch specific helper functions or defines to paper over the differences? That definitely seems to be the current intention. You'd need to move the test to some common place and make sure it still gets executed on x86.

alwin-joshy commented 7 months ago

With 1162 in mind, I think it would be good to test whether code actually gets executed instead of only checking if the instruction pointer advances, so instead of nops just add a value to a register and check if it's correct at the second label.

Good point. In that case, I'll port the equivalent x86 test, which seems to be more robust.

At first glance I don't really see anything x86 specific in there, perhaps make that test cross-platform with some arch specific helper functions or defines to paper over the differences? That definitely seems to be the current intention. You'd need to move the test to some common place and make sure it still gets executed on x86.

Yeah, should be simple. I do recall seeing a comment somewhere in the code about something being missing or not ready in ARM, but I'll see if I run into any issues. Didn't spot anything that immediately stood out.

alwin-joshy commented 7 months ago

@Indanz I moved the x86 single-stepping test into the arch-independent breakpoints file and it just worked.

lsf37 commented 7 months ago

It looks like the hw-test trigger isn't working properly on this repo. I'll investigate..

Edit: it doesn't seem to be working on the seL4 repo any more either. Looks like something changed in the infrastructure.

Edit 2: looking more closely, it did actually work. It's just not showing them in the PR summary panel. It is showing them in the "Check" tab of the PR.

Indanz commented 7 months ago

The Sabre clang fails with:

   All is well in the universe

  console_run returned 0

  0.05user 0.02system 2:48.06elapsed 0%CPU (0avgtext+0avgdata 7936maxresident)k
  0inputs+336outputs (0major+8920minor)pagefaults 0swaps
  +++ python3 ../projects/seL4_libs/libsel4test/tools/extract_results.py -q results.xml parsed_results.xml
  No log to check boot failure on.
  +++ mq.sh sem -signal sabre2 -k seL4/sel4test-seL4Test HW-7619487774-hw-run-5
  Error parsing parsed_results.xml
  -----------[ end test SABRE_debug_MCS_clang_32 ]-----------
SABRE_debug_MCS_clang_32 FAILED
Indanz commented 7 months ago

I verified that it still runs and passes on x86 and some 32 bit ARMs, it doesn't yet run on Aarch64 though. I think we should remove the lines around https://github.com/seL4/seL4/blob/cc3205ea486f6ce3da3a64d8a99e166f047b8419/src/arch/arm/config.cmake#L226.

Indanz commented 7 months ago

So do we want to remove KernelHardwareDebugAPIUnsupported from aarch64 before merging this, as the aarch64 code isn't tested yet, or merge this first and remove the define later?

@alwin-joshy, did you test this on aarch64 yet?

alwin-joshy commented 7 months ago

I verified that it still runs and passes on x86 and some 32 bit ARMs, it doesn't yet run on Aarch64 though. I think we should remove the lines around https://github.com/seL4/seL4/blob/cc3205ea486f6ce3da3a64d8a99e166f047b8419/src/arch/arm/config.cmake#L226.

Yep, that will be necessary for getting the test to run on AARCH64 eventually but there are other substantial changes required for this. I've been working on this and am close to done (https://github.com/alwin-joshy/seL4/pull/7) but there are still a few problems related to ARMv8 32-bit and ARM hyp mode that I want to iron out before submitting a proper PR.

alwin-joshy commented 7 months ago

So do we want to remove KernelHardwareDebugAPIUnsupported from aarch64 before merging this, as the aarch64 code isn't tested yet, or merge this first and remove the define later?

@alwin-joshy, did you test this on aarch64 yet?

I think to merge this first and remove the KernelHardwareDebugAPIUnsupported in the PR when I have the hardware debug API fully ported to aarch64