mirror of
https://github.com/torvalds/linux.git
synced 2026-06-04 20:46:48 +02:00
a2aa5dcc63
The definition of GNU_SOURCE was recently centralised in an upper layer
kselftest Makefile, so the definition in the arm64 signal tests Makefile
is no longer needed. To make things worse, since both definitions are
not strictly identical, the compiler warns about it:
<command-line>: warning: "_GNU_SOURCE" redefined
<command-line>: note: this is the location of the previous definition
Drop the definition in the arm64/signal Makefile.
Fixes:
|
||
|---|---|---|
| .. | ||
| testcases | ||
| .gitignore | ||
| Makefile | ||
| README | ||
| signals.S | ||
| sve_helpers.c | ||
| sve_helpers.h | ||
| test_signals_utils.c | ||
| test_signals_utils.h | ||
| test_signals.c | ||
| test_signals.h | ||
KSelfTest arm64/signal/
=======================
Signals Tests
+++++++++++++
- Tests are built around a common main compilation unit: such shared main
enforces a standard sequence of operations needed to perform a single
signal-test (setup/trigger/run/result/cleanup)
- The above mentioned ops are configurable on a test-by-test basis: each test
is described (and configured) using the descriptor signals.h::struct tdescr
- Each signal testcase is compiled into its own executable: a separate
executable is used for each test since many tests complete successfully
by receiving some kind of fatal signal from the Kernel, so it's safer
to run each test unit in its own standalone process, so as to start each
test from a clean slate.
- New tests can be simply defined in testcases/ dir providing a proper struct
tdescr overriding all the defaults we wish to change (as of now providing a
custom run method is mandatory though)
- Signals' test-cases hereafter defined belong currently to two
principal families:
- 'mangle_' tests: a real signal (SIGUSR1) is raised and used as a trigger
and then the test case code modifies the signal frame from inside the
signal handler itself.
- 'fake_sigreturn_' tests: a brand new custom artificial sigframe structure
is placed on the stack and a sigreturn syscall is called to simulate a
real signal return. This kind of tests does not use a trigger usually and
they are just fired using some simple included assembly trampoline code.
- Most of these tests are successfully passing if the process gets killed by
some fatal signal: usually SIGSEGV or SIGBUS. Since while writing this
kind of tests it is extremely easy in fact to end-up injecting other
unrelated SEGV bugs in the testcases, it becomes extremely tricky to
be really sure that the tests are really addressing what they are meant
to address and they are not instead falling apart due to unplanned bugs
in the test code.
In order to alleviate the misery of the life of such test-developer, a few
helpers are provided:
- a couple of ASSERT_BAD/GOOD_CONTEXT() macros to easily parse a ucontext_t
and verify if it is indeed GOOD or BAD (depending on what we were
expecting), using the same logic/perspective as in the arm64 Kernel signals
routines.
- a sanity mechanism to be used in 'fake_sigreturn_'-alike tests: enabled by
default it takes care to verify that the test-execution had at least
successfully progressed up to the stage of triggering the fake sigreturn
call.
In both cases test results are expected in terms of:
- some fatal signal sent by the Kernel to the test process
or
- analyzing some final regs state