Age | Commit message (Collapse) | Author | Files | Lines |
|
`disassemble` shows the whole function which is usually too much.
Instead just show the nearest instructions for some context.
|
|
update gdbserver.py; tolerance value of MemorySampleTest()
|
|
The control sequences (^[[?2004h and ^[[?2004l) occur after the
gdb.command, which results in Exception fault. This commit removes the
control sequences and strips out the blank lines (^M).
|
|
|
|
|
|
|
|
|
|
These can change at any time and that is OK.
|
|
|
|
|
|
I don't have time to fix the bug right now, and nobody has run into the
bug yet in any case.
|
|
|
|
Turn semihosting_fileio on for every hart. This test still fails if it
ends up running on hart 1 instead of 0, but at least it's closer to
passing. Feels like the remaining problem is in OpenOCD.
|
|
|
|
* Another pylint upgrade.
Lots of format string changes, which are more readable.
More files to come...
* Satisfy pylint for two more files.
|
|
The source of benchmarks' Makefile does not use fragments anymore. This commit removes the mention of them from the readme.
Signed-off-by: Mehmet Oguz Derin <mehmetoguzderin@mehmetoguzderin.com>
|
|
Assume that misaligned loads work correctly.
Passes `make -C isa run` against spike (both with and without misaligned
support).
|
|
Test misaligned stores
|
|
In the original test, confirm that stdout data ends up in the OpenOCD
log.
In the new test, with `arm semihosting_fileio` enabled, confirm that
stdout data ends up in gdb's CLI.
This test requires https://github.com/riscv/riscv-openocd/pull/699.
|
|
Cover lh, lw, and ld (only on rv64).
|
|
Tests that might cause a trap during their code need TESTNUM (gp) set so
the trap handler can correctly identify which test is running, and also
report that to the user in case the test fails.
Fix up shamt.S and csr.S to handle the new behavior.
|
|
They lead to two problems:
1. The change to set TESTNUM before the test instead of after broke a
few tests that relied on the old behavior.
2. Some tests in the v variant do something in the exception handler
such that when they get an unaligned access exception, they end up
stuck in a loop of unaligned access exceptions and the exception
handler set up in the test is never called.
|
|
Identical to the lw change in #391.
|
|
The test passes if either the target register contains the correct
value, or the target raises a misaligned load exception.
|
|
Tests that might cause a trap during their code need TESTNUM (gp) set so
the trap handler can correctly identify which test is running, and also
report that to the user in case the test fails.
|
|
I'm running a newer version of pylint, and thus there are new warnings
to be fixed. All very minor.
|
|
regions (#388)
|
|
|
|
Adjust test to work with that.
|
|
* Add EbreakTest.
Confirm correct behavior when somebody bakes an ebreak instruction into
their code.
* Forgot to commit ebreak.c
|
|
That way it can go into flash.
|
|
Otherwise, "make run" doesn't work.
|
|
It includes the name in quotes:
```* 2 Thread 1 "Current Execution" (Name: Current Execution) 0x10000100 in main```
Just ignore that part.
|
|
Now it will give slightly more helpful output if it fails.
|
|
(#369)
|
|
This feature lets you easily interact with the gdb after the test has
run to a certain point.
|
|
They used to set U, A, D, in intermediate page table entries which is no
longer allowed.
|
|
The tests don't confirm that the order actually changes, but at least
the code that does the work now is executed during the tests.
|
|
Disconnects from gdb, and then reconnects, making sure that didn't
change any of the registers.
This test will start passing when
https://github.com/riscv/riscv-openocd/pull/661 merges.
|
|
Useful for estimating interactive performance.
|
|
It's not an argument to spike anymore.
Also switch testing the vector unit from multi-gdb to `-rtos hwthread`.
This exposes a bug in OpenOCD (which is already fixed).
|
|
Thanks to @pdonahue-ventana for pointing this out
|
|
...since not all implementations will support it
|
|
Fixes #350
|
|
|
|
|
|
|
|
1. Don't run all tests in multi-spike. Extra coverage is negligible, and
it just takes too long.
2. Increase a few timeouts.
|
|
There are two reasons that writing the cycle counter might trap:
- Because it's a read-only CSR
- Because mcounteren.CY=0 or scounteren.CY=0
We want to make sure we're testing the first property, so set up
the other bits accordingly.
|
|
|