ktests wiki: breakpoints: basic description, compilation tips

Author: pvts-mat

kselftests/breakpoints/README.rst

@@ -0,0 +1,256 @@
+About
+=====
+
+This collection tests the ``ptrace()`` system call, specifically the
+ability of a tracer to set breakpoints (on functions) and watchpoints
+(on variables) in the tracee and whether they are correctly triggered
+when tracee hits them.
+
+The ``breakpoint_test`` test from the collection is chronologically the
+very first kernel selftest.
+
+Tests suite compilation and running
+===================================
+
+The suite generally contains two tests:
+
+#. ``breakpoint_test`` (on x86\* platforms) or ``breakpoint_test_arm64``
+   (on the arm64 platforms),
+#. ``step_after_suspend_test``.
+
+For other platforms than x86\* and arm64 the suite contains only
+``step_after_suspend_test``.
+
+``breakpoint_test``
+-------------------
+
+When compiling the ``breakpoints`` collection on *x86_64* beware not to
+set the ``ARCH`` variable explicitly, or the ``breakpoint_test`` test
+won't be produced. So, if your ``uname -m`` reports ``x86_64`` then this
+is ok:
+
+.. code:: shell
+
+   make -C tools/testing/selftests TARGETS=breakpoints
+
+and this is **not** ok:
+
+.. code:: shell
+
+   make ARCH=x86_64 -C tools/testing/selftests TARGETS=breakpoints
+
+It's obviously a bug in the
+``tools/testing/selftests/breakpoints/Makefile`` file, precisely in the
+way the ``ARCH`` variable is normalized with ``sed``, with this
+normalization being bound to setting default ``ARCH`` value instead of
+being general,
+
+.. code:: makefile
+
+   ARCH ?= $(shell echo $(uname_M) | sed -e s/i.86/x86/ -e s/x86_64/x86/)
+
+yet the conditional compilation of ``breakpoint_test`` assuming this
+normalization in all cases.
+
+If compiled successfully the test is expected to pass on all versions
+with a message similar to
+
+::
+
+   TAP version 13
+   1..1
+   # timeout set to 45
+   # selftests: breakpoints: breakpoint_test
+   # TAP version 13
+   # 1..110
+   # ok 1 Test breakpoint 0 with local: 0 global: 1
+   # ok 2 Test breakpoint 1 with local: 0 global: 1
+   # ok 3 Test breakpoint 2 with local: 0 global: 1
+   # ok 4 Test breakpoint 3 with local: 0 global: 1
+   # ok 5 Test breakpoint 0 with local: 1 global: 0
+   # ok 6 Test breakpoint 1 with local: 1 global: 0
+   …
+   # ok 104 Test read watchpoint 3 with len: 8 local: 1 global: 0
+   # ok 105 Test read watchpoint 0 with len: 8 local: 1 global: 1
+   # ok 106 Test read watchpoint 1 with len: 8 local: 1 global: 1
+   # ok 107 Test read watchpoint 2 with len: 8 local: 1 global: 1
+   # ok 108 Test read watchpoint 3 with len: 8 local: 1 global: 1
+   # ok 109 Test icebp
+   # ok 110 Test int 3 trap
+   # # Totals: pass:110 fail:0 xfail:0 xpass:0 skip:0 error:0
+   ok 1 selftests: breakpoints: breakpoint_test
+
+``breakpoint_test_arm64``
+-------------------------
+
+On versions ``ciqlts8_6``, ``ciqlts8_8`` the test contains a bug which
+prevents it from being compiled successfully for ``arm64``:
+
+::
+
+   gcc     breakpoint_test_arm64.c /mnt/build_files/kernel-src-tree-kselftests-ciqlts8_8/tools/testing/selftests/kselftest_harness.h /mnt/build_files/kernel-src-tree-kselftests-ciqlts8_8/tools/testing/selftests/kselftest.h  -o /mnt/build_files/kernel-src-tree-kselftests-ciqlts8_8/tools/testing/selftests/breakpoints/breakpoint_test_arm64
+   breakpoint_test_arm64.c: In function ‘main’:
+   breakpoint_test_arm64.c:226:14: warning: implicit declaration of function ‘run_test’; did you mean ‘arun_test’? [-Wimplicit-function-declaration]
+        result = run_test(size, MIN(size, 8), wr, wp);
+                 ^~~~~~~~
+                 arun_test
+   /tmp/cc5dMu7r.o: In function `main':
+   breakpoint_test_arm64.c:(.text+0xb08): undefined reference to `run_test'
+   breakpoint_test_arm64.c:(.text+0xc10): undefined reference to `run_test'
+   collect2: error: ld returned 1 exit status
+
+This bug was fixed in the mainline with the
+``5b06eeae52c02dd0d9bc8488275a1207d410870b`` commit. It's in the process
+of backporting to ``ciqlts8_6``, ``ciqlts8_8`` at the time of this
+writing.
+
+Additionally, when compiling the test on base cloud images for
+``ciqlts8_6`` and ``ciqlts8_8``\  [1]_, and using Mountain-provided
+repositories, the following error may be encountered:
+
+::
+
+   breakpoint_test_arm64.c: In function ‘run_test’:
+   breakpoint_test_arm64.c:188:25: error: ‘TRAP_HWBKPT’ undeclared (first use in this function); did you mean ‘TRAP_BRKPT’?
+     if (siginfo.si_code != TRAP_HWBKPT) {
+                            ^~~~~~~~~~~
+                            TRAP_BRKPT
+   breakpoint_test_arm64.c:188:25: note: each undeclared identifier is reported only once for each function it appears in
+
+The ``TRAP_BRKPT`` constant is searched for in the
+``/usr/include/bits/siginfo-consts.h`` file, which in versions ≥ 9.2
+contains
+
+.. code:: c
+
+   /* `si_code' values for SIGTRAP signal.  */
+   enum
+   {
+     TRAP_BRKPT = 1,       /* Process breakpoint.  */
+   #  define TRAP_BRKPT    TRAP_BRKPT
+     TRAP_TRACE,           /* Process trace trap.  */
+   #  define TRAP_TRACE    TRAP_TRACE
+     TRAP_BRANCH,          /* Process taken branch trap.  */
+   #  define TRAP_BRANCH   TRAP_BRANCH
+     TRAP_HWBKPT,          /* Hardware breakpoint/watchpoint.  */
+   #  define TRAP_HWBKPT   TRAP_HWBKPT
+     TRAP_UNK          /* Undiagnosed trap.  */
+   #  define TRAP_UNK  TRAP_UNK
+   };
+   # endif
+
+while in versions < 9.2 contains
+
+.. code:: c
+
+   /* `si_code' values for SIGTRAP signal.  */
+   enum
+   {
+     TRAP_BRKPT = 1,       /* Process breakpoint.  */
+   #  define TRAP_BRKPT    TRAP_BRKPT
+     TRAP_TRACE            /* Process trace trap.  */
+   #  define TRAP_TRACE    TRAP_TRACE
+   };
+   # endif
+
+If the ``TRAP_HWBKPT`` constant can't be included in
+``/usr/include/bits/siginfo-consts.h`` in some civilised manner then
+simply expanding the set of known ``TRAP_*`` codes in the system to
+match those on ≥ 9.2 should work:
+
+.. code:: shell
+
+   sudo sed -e '144 d' -e 's/#  define TRAP_TRACE\tTRAP_TRACE/  TRAP_TRACE,\n#  define TRAP_TRACE\tTRAP_TRACE\n  TRAP_BRANCH,\n#  define TRAP_BRANCH\tTRAP_BRANCH\n  TRAP_HWBKPT,\n#  define TRAP_HWBKPT\tTRAP_HWBKPT\n  TRAP_UNK\n#  define TRAP_UNK\tTRAP_UNK/g' -i /usr/include/bits/siginfo-consts.h
+
+No compilation problems for ``breakpoint_test_arm64`` were encountered
+on versions ``ciqlts9_2`` and ``ciqlts9_4``.
+
+If the test compiled successfully it's expected to pass with the message
+similar to the following
+
+::
+
+   # TAP version 13
+   # 1..213
+   # # child did not single-step
+   # ok 1 Test size = 1 write offset = 0 watchpoint offset = -1
+   # ok 2 Test size = 1 write offset = 0 watchpoint offset = 0
+   # # child did not single-step
+   # ok 3 Test size = 1 write offset = 0 watchpoint offset = 1
+   # # child did not single-step
+   # ok 4 Test size = 1 write offset = 1 watchpoint offset = 0
+   # ok 5 Test size = 1 write offset = 1 watchpoint offset = 1
+   …
+   # ok 206 Test size = 32 write offset = 32 watchpoint offset = 32
+   # # child did not single-step
+   # ok 207 Test size = 32 write offset = 32 watchpoint offset = 64
+   # ok 208 Test size = 1 write offset = -1 watchpoint offset = -8
+   # ok 209 Test size = 2 write offset = -2 watchpoint offset = -8
+   # ok 210 Test size = 4 write offset = -4 watchpoint offset = -8
+   # ok 211 Test size = 8 write offset = -8 watchpoint offset = -8
+   # ok 212 Test size = 16 write offset = -16 watchpoint offset = -8
+   # ok 213 Test size = 32 write offset = -32 watchpoint offset = -8
+   # # Totals: pass:213 fail:0 xfail:0 xpass:0 skip:0 error:0
+   ok 1 selftests: breakpoints: breakpoint_test_arm64
+
+``step_after_suspend_test``
+---------------------------
+
+No problems were encountered when compiling this test, on any of the
+versions ``ciqlts8_6``, ``ciqlts8_8``, ``ciqlts9_2``, ``ciqlts9_4``,
+archs ``x86_64``, ``aarch64``. However, for the test to run the system
+must provide the ability to suspend it, as it would be done with
+
+::
+
+   root@…# echo mem > /sys/power/state
+
+Can be checked with this command directly, that's what
+``step_after_suspend_test`` is doing internally. Unfortunately on the
+qemu-kvm virtual machines this fails with the ``virtio-pci`` device (the
+one connecting the machine with the hypervisor) refusing to cooperate:
+
+::
+
+   [root@ciqlts-9-2 pvts]# echo mem > /sys/power/state
+   [   24.434434] PM: suspend entry (s2idle)
+   [   24.498974] Filesystems sync: 0.062 seconds
+   [   24.501574] Freezing user space processes ... (elapsed 0.001 seconds) done.
+   [   24.507156] OOM killer disabled.
+   [   24.509062] Freezing remaining freezable tasks ... (elapsed 0.001 seconds) done.
+   [   24.514480] printk: Suspending console(s) (use no_console_suspend to debug)
+   [   24.540895] virtio-fs: suspend/resume not yet supported
+   [   24.540906] virtio-pci 0000:03:00.0: PM: pci_pm_suspend(): virtio_pci_freeze+0x0/0x50 returns -95
+   [   24.540970] virtio-pci 0000:03:00.0: PM: dpm_run_callback(): pci_pm_suspend+0x0/0x170 returns -95
+   [   24.541011] virtio-pci 0000:03:00.0: PM: failed to suspend async: error -95
+   [   24.554785] PM: Some devices failed to suspend, or early wake event detected
+   [   24.686487] OOM killer enabled.
+   [   24.687385] Restarting tasks ... done.
+   [   24.689834] PM: suspend exit
+   bash: echo: write error: Operation not supported
+   [root@ciqlts-9-2 pvts]# [   24.886811] ata3: SATA link down (SStatus 0 SControl 300)
+   [   24.889470] ata5: SATA link down (SStatus 0 SControl 300)
+   [   24.892095] ata2: SATA link down (SStatus 0 SControl 300)
+   [   24.894956] ata4: SATA link down (SStatus 0 SControl 300)
+   [   24.897717] ata6: SATA link down (SStatus 0 SControl 300)
+
+When running selftests in qemu-kvm it's therefore best to omit this
+test, or else it will keep scaring with failures like
+
+::
+
+   # selftests: breakpoints: step_after_suspend_test
+   # TAP version 13
+   # Bail out! Failed to enter Suspend state
+   # # Totals: pass:0 fail:0 xfail:0 xpass:0 skip:0 error:0
+   not ok 1 selftests: breakpoints: step_after_suspend_test # exit=1
+
+(Probably a better behavior would be for the test to ``# SKIP`` in this
+case.)
+
+.. [1]
+   Referring to the images available at
+   https://download.rockylinux.org/vault/rocky/8.6/images/Rocky-8-GenericCloud-8.6.20220702.0.aarch64.qcow2
+   for ``ciqlts8_6`` and at
+   https://download.rockylinux.org/vault/rocky/8.8/images/aarch64/Rocky-8-GenericCloud-Base-8.8-20230518.0.aarch64.qcow2
+   for ``ciqlts8_8``, to be precise.