kernel/linux.git/tools/testing/selftests/powerpc/math, branch v6.12.80

selftests/powerpc: make sub-folders buildable on their own

2024-04-29T13:54:42+00:00

Build breaks when executing make with run_tests for sub-folders under powerpc. This is because, CFLAGS and GIT_VERSION macros are defined in Makefile of toplevel powerpc folder. make: Entering directory '/home/maddy/linux/tools/testing/selftests/powerpc/mm' gcc hugetlb_vs_thp_test.c ../harness.c ../utils.c -o /home/maddy/selftest_output//hugetlb_vs_thp_test hugetlb_vs_thp_test.c:6:10: fatal error: utils.h: No such file or directory 6 | #include "utils.h" | ^~~~~~~~~ compilation terminated. Fix this by adding the flags.mk in each sub-folder Makefile. Also remove the CFLAGS and GIT_VERSION macros from powerpc/ folder Makefile since the same is definied in flags.mk Signed-off-by: Madhavan Srinivasan Signed-off-by: Michael Ellerman Link: https://msgid.link/20240229093711.581230-3-maddy@linux.ibm.com

selftests/powerpc: Fix fpu_signal failures

2024-03-01T11:15:30+00:00

My recent commit e5d00aaac651 ("selftests/powerpc: Check all FPRs in fpu_preempt") inadvertently broke the fpu_signal test. It needs to take into account that fpu_preempt now loads 32 FPRs, so enlarge darray. Also use the newly added randomise_darray() to properly randomise darray. Finally the checking done in signal_fpu_sig() needs to skip checking f30/f31, because they are used as scratch registers in check_all_fprs(), called by preempt_fpu(), and so could hold other values when the signal is taken. Fixes: e5d00aaac651 ("selftests/powerpc: Check all FPRs in fpu_preempt") Reported-by: Spoorthy Depends-on: 2ba107f6795d ("selftests/powerpc: Generate better bit patterns for FPU tests") Signed-off-by: Michael Ellerman Link: https://msgid.link/20240301101035.1230024-1-mpe@ellerman.id.au

selftests/powerpc: Check all FPRs in fpu_syscall test

2023-12-13T02:29:09+00:00

There is a selftest that checks if FPRs are corrupted across a fork, aka clone. It was added as part of the series that optimised the clone path to save the parent's FP state without "giving up" (turning off FP). See commit 8792468da5e1 ("powerpc: Add the ability to save FPU without giving it up"). The test encodes the assumption that FPRs 0-13 are volatile across the syscall, by only checking the volatile FPRs are not changed by the fork. There was also a comment in the fpu_preempt test alluding to that: The check_fpu function in asm only checks the non volatile registers as it is reused from the syscall test It is true that the function call ABI treats f0-f13 as volatile, however the syscall ABI has since been documented as *not* treating those registers as volatile. See commit 7b8845a2a2ec ("powerpc/64: Document the syscall ABI"). So change the test to check all FPRs are not corrupted by the syscall. Note that this currently fails, because save_fpu() etc. do not restore f0/vsr0. Signed-off-by: Michael Ellerman Link: https://msgid.link/20231128132748.1990179-5-mpe@ellerman.id.au

selftests/powerpc: Run fpu_preempt test for 60 seconds

2023-12-13T02:29:09+00:00

The FPU preempt test only runs for 20 seconds, which is not particularly long. Run it for 60 seconds to increase the chance of detecting corruption. Signed-off-by: Michael Ellerman Link: https://msgid.link/20231128132748.1990179-4-mpe@ellerman.id.au

selftests/powerpc: Generate better bit patterns for FPU tests

2023-12-13T02:29:08+00:00

The fpu_preempt test randomly initialises an array of doubles to try and detect FPU register corruption. However the values it generates do not occupy the full range of values possible in the 64-bit double, meaning some partial register corruption could go undetected. Without getting too carried away, add some better initialisation to generate values that occupy more bits. Sample values before: f0 902677510 (raw 0x41cae6e203000000) f1 325217596 (raw 0x41b3626d3c000000) f2 1856578300 (raw 0x41dbaa48bf000000) f3 1247189984 (raw 0x41d295a6f8000000) And after: f0 1.1078153481413311e-09 (raw 0x3e13083932805cc2) f1 1.0576648474801922e+17 (raw 0x43777c20eb88c261) f2 -6.6245033413594075e-10 (raw 0xbe06c2f989facae9) f3 3.0085988827156291e+18 (raw 0x43c4e0585f2df37b) Signed-off-by: Michael Ellerman Link: https://msgid.link/20231128132748.1990179-3-mpe@ellerman.id.au

selftests/powerpc: Check all FPRs in fpu_preempt

2023-12-13T02:29:08+00:00

There's a selftest that checks FPRs aren't corrupted by preemption, or just process scheduling. However it only checks the non-volatile FPRs, meaning corruption of the volatile FPRs could go undetected. The check_fpu function it calls is used by several other tests, so for now add a new routine to check all the FPRs. Increase the size of the array of FPRs to 32, and initialise them all with random values. Signed-off-by: Michael Ellerman Link: https://msgid.link/20231128132748.1990179-2-mpe@ellerman.id.au

selftests/powerpc: Fix error handling in FPU/VMX preemption tests

2023-12-13T02:29:08+00:00

The FPU & VMX preemption tests do not check for errors returned by the low-level asm routines, preempt_fpu() / preempt_vsx() respectively. That means any register corruption detected by the asm routines does not result in a test failure. Fix it by returning the return value of the asm routines from the pthread child routines. Fixes: e5ab8be68e44 ("selftests/powerpc: Test preservation of FPU and VMX regs across preemption") Signed-off-by: Michael Ellerman Link: https://msgid.link/20231128132748.1990179-1-mpe@ellerman.id.au

selftests/powerpc: Increase timeout for vsx_signal test

2023-03-14T13:52:09+00:00

On the max config P10 machine (1920 threads and 64TB) this test fails with a timeout: Sending signals to all threads 10 times...!! killing vmx_signal !! child died by signal 15 failure: vmx_signal The default timeout is 120sec so increase this 3x to 360sec. With this change the test passes on these large machines. Signed-off-by: Michael Neuling Signed-off-by: Michael Ellerman Link: https://msgid.link/20230307213614.2652059-1-mikey@neuling.org

selftests/powerpc: Fix matrix multiply assist test

2022-07-28T06:22:15+00:00

The ISA states: "when ACC[i] contains defined data, the contents of VSRs 4×i to 4×i+3 are undefined until either a VSX Move From ACC instruction is used to copy the contents of ACC[i] to VSRs 4×i to 4×i+3 or some other instruction directly writes to one of these VSRs." We aren't doing this. This test only works on Power10 because the hardware implementation happens to map ACC0 to VSRs 0-3, but will fail on any other implementation that doesn't do this. So add xxmfacc between writing to the accumulator and accessing the VSRs. Fixes: 3527e1ab9a79 ("selftests/powerpc: Add matrix multiply assist (MMA) test") Signed-off-by: Rashmica Gupta Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220617043935.428083-1-rashmica@linux.ibm.com

selftests/powerpc: Add missing files to .gitignores

2022-06-29T09:37:04+00:00

These were missed when the respective tests were added, add them now. Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220616070705.1941829-1-mpe@ellerman.id.au