Back to home page

OSCL-LXR

 
 

    


0001 ==================
0002 BPF Selftest Notes
0003 ==================
0004 General instructions on running selftests can be found in
0005 `Documentation/bpf/bpf_devel_QA.rst`__.
0006 
0007 __ /Documentation/bpf/bpf_devel_QA.rst#q-how-to-run-bpf-selftests
0008 
0009 =========================
0010 Running Selftests in a VM
0011 =========================
0012 
0013 It's now possible to run the selftests using ``tools/testing/selftests/bpf/vmtest.sh``.
0014 The script tries to ensure that the tests are run with the same environment as they
0015 would be run post-submit in the CI used by the Maintainers.
0016 
0017 This script downloads a suitable Kconfig and VM userspace image from the system used by
0018 the CI. It builds the kernel (without overwriting your existing Kconfig), recompiles the
0019 bpf selftests, runs them (by default ``tools/testing/selftests/bpf/test_progs``) and
0020 saves the resulting output (by default in ``~/.bpf_selftests``).
0021 
0022 Script dependencies:
0023 - clang (preferably built from sources, https://github.com/llvm/llvm-project);
0024 - pahole (preferably built from sources, https://git.kernel.org/pub/scm/devel/pahole/pahole.git/);
0025 - qemu;
0026 - docutils (for ``rst2man``);
0027 - libcap-devel.
0028 
0029 For more information on about using the script, run:
0030 
0031 .. code-block:: console
0032 
0033   $ tools/testing/selftests/bpf/vmtest.sh -h
0034 
0035 In case of linker errors when running selftests, try using static linking:
0036 
0037 .. code-block:: console
0038 
0039   $ LDLIBS=-static vmtest.sh
0040 
0041 .. note:: Some distros may not support static linking.
0042 
0043 .. note:: The script uses pahole and clang based on host environment setting.
0044           If you want to change pahole and llvm, you can change `PATH` environment
0045           variable in the beginning of script.
0046 
0047 .. note:: The script currently only supports x86_64 and s390x architectures.
0048 
0049 Additional information about selftest failures are
0050 documented here.
0051 
0052 profiler[23] test failures with clang/llvm <12.0.0
0053 ==================================================
0054 
0055 With clang/llvm <12.0.0, the profiler[23] test may fail.
0056 The symptom looks like
0057 
0058 .. code-block:: c
0059 
0060   // r9 is a pointer to map_value
0061   // r7 is a scalar
0062   17:       bf 96 00 00 00 00 00 00 r6 = r9
0063   18:       0f 76 00 00 00 00 00 00 r6 += r7
0064   math between map_value pointer and register with unbounded min value is not allowed
0065 
0066   // the instructions below will not be seen in the verifier log
0067   19:       a5 07 01 00 01 01 00 00 if r7 < 257 goto +1
0068   20:       bf 96 00 00 00 00 00 00 r6 = r9
0069   // r6 is used here
0070 
0071 The verifier will reject such code with above error.
0072 At insn 18 the r7 is indeed unbounded. The later insn 19 checks the bounds and
0073 the insn 20 undoes map_value addition. It is currently impossible for the
0074 verifier to understand such speculative pointer arithmetic.
0075 Hence `this patch`__ addresses it on the compiler side. It was committed on llvm 12.
0076 
0077 __ https://reviews.llvm.org/D85570
0078 
0079 The corresponding C code
0080 
0081 .. code-block:: c
0082 
0083   for (int i = 0; i < MAX_CGROUPS_PATH_DEPTH; i++) {
0084           filepart_length = bpf_probe_read_str(payload, ...);
0085           if (filepart_length <= MAX_PATH) {
0086                   barrier_var(filepart_length); // workaround
0087                   payload += filepart_length;
0088           }
0089   }
0090 
0091 bpf_iter test failures with clang/llvm 10.0.0
0092 =============================================
0093 
0094 With clang/llvm 10.0.0, the following two bpf_iter tests failed:
0095   * ``bpf_iter/ipv6_route``
0096   * ``bpf_iter/netlink``
0097 
0098 The symptom for ``bpf_iter/ipv6_route`` looks like
0099 
0100 .. code-block:: c
0101 
0102   2: (79) r8 = *(u64 *)(r1 +8)
0103   ...
0104   14: (bf) r2 = r8
0105   15: (0f) r2 += r1
0106   ; BPF_SEQ_PRINTF(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
0107   16: (7b) *(u64 *)(r8 +64) = r2
0108   only read is supported
0109 
0110 The symptom for ``bpf_iter/netlink`` looks like
0111 
0112 .. code-block:: c
0113 
0114   ; struct netlink_sock *nlk = ctx->sk;
0115   2: (79) r7 = *(u64 *)(r1 +8)
0116   ...
0117   15: (bf) r2 = r7
0118   16: (0f) r2 += r1
0119   ; BPF_SEQ_PRINTF(seq, "%pK %-3d ", s, s->sk_protocol);
0120   17: (7b) *(u64 *)(r7 +0) = r2
0121   only read is supported
0122 
0123 This is due to a llvm BPF backend bug. `The fix`__
0124 has been pushed to llvm 10.x release branch and will be
0125 available in 10.0.1. The patch is available in llvm 11.0.0 trunk.
0126 
0127 __  https://reviews.llvm.org/D78466
0128 
0129 bpf_verif_scale/loop6.o test failure with Clang 12
0130 ==================================================
0131 
0132 With Clang 12, the following bpf_verif_scale test failed:
0133   * ``bpf_verif_scale/loop6.o``
0134 
0135 The verifier output looks like
0136 
0137 .. code-block:: c
0138 
0139   R1 type=ctx expected=fp
0140   The sequence of 8193 jumps is too complex.
0141 
0142 The reason is compiler generating the following code
0143 
0144 .. code-block:: c
0145 
0146   ;       for (i = 0; (i < VIRTIO_MAX_SGS) && (i < num); i++) {
0147       14:       16 05 40 00 00 00 00 00 if w5 == 0 goto +64 <LBB0_6>
0148       15:       bc 51 00 00 00 00 00 00 w1 = w5
0149       16:       04 01 00 00 ff ff ff ff w1 += -1
0150       17:       67 05 00 00 20 00 00 00 r5 <<= 32
0151       18:       77 05 00 00 20 00 00 00 r5 >>= 32
0152       19:       a6 01 01 00 05 00 00 00 if w1 < 5 goto +1 <LBB0_4>
0153       20:       b7 05 00 00 06 00 00 00 r5 = 6
0154   00000000000000a8 <LBB0_4>:
0155       21:       b7 02 00 00 00 00 00 00 r2 = 0
0156       22:       b7 01 00 00 00 00 00 00 r1 = 0
0157   ;       for (i = 0; (i < VIRTIO_MAX_SGS) && (i < num); i++) {
0158       23:       7b 1a e0 ff 00 00 00 00 *(u64 *)(r10 - 32) = r1
0159       24:       7b 5a c0 ff 00 00 00 00 *(u64 *)(r10 - 64) = r5
0160 
0161 Note that insn #15 has w1 = w5 and w1 is refined later but
0162 r5(w5) is eventually saved on stack at insn #24 for later use.
0163 This cause later verifier failure. The bug has been `fixed`__ in
0164 Clang 13.
0165 
0166 __  https://reviews.llvm.org/D97479
0167 
0168 BPF CO-RE-based tests and Clang version
0169 =======================================
0170 
0171 A set of selftests use BPF target-specific built-ins, which might require
0172 bleeding-edge Clang versions (Clang 12 nightly at this time).
0173 
0174 Few sub-tests of core_reloc test suit (part of test_progs test runner) require
0175 the following built-ins, listed with corresponding Clang diffs introducing
0176 them to Clang/LLVM. These sub-tests are going to be skipped if Clang is too
0177 old to support them, they shouldn't cause build failures or runtime test
0178 failures:
0179 
0180 - __builtin_btf_type_id() [0_, 1_, 2_];
0181 - __builtin_preserve_type_info(), __builtin_preserve_enum_value() [3_, 4_].
0182 
0183 .. _0: https://reviews.llvm.org/D74572
0184 .. _1: https://reviews.llvm.org/D74668
0185 .. _2: https://reviews.llvm.org/D85174
0186 .. _3: https://reviews.llvm.org/D83878
0187 .. _4: https://reviews.llvm.org/D83242
0188 
0189 Floating-point tests and Clang version
0190 ======================================
0191 
0192 Certain selftests, e.g. core_reloc, require support for the floating-point
0193 types, which was introduced in `Clang 13`__. The older Clang versions will
0194 either crash when compiling these tests, or generate an incorrect BTF.
0195 
0196 __  https://reviews.llvm.org/D83289
0197 
0198 Kernel function call test and Clang version
0199 ===========================================
0200 
0201 Some selftests (e.g. kfunc_call and bpf_tcp_ca) require a LLVM support
0202 to generate extern function in BTF.  It was introduced in `Clang 13`__.
0203 
0204 Without it, the error from compiling bpf selftests looks like:
0205 
0206 .. code-block:: console
0207 
0208   libbpf: failed to find BTF for extern 'tcp_slow_start' [25] section: -2
0209 
0210 __ https://reviews.llvm.org/D93563
0211 
0212 btf_tag test and Clang version
0213 ==============================
0214 
0215 The btf_tag selftest requires LLVM support to recognize the btf_decl_tag and
0216 btf_type_tag attributes. They are introduced in `Clang 14` [0_, 1_].
0217 The subtests ``btf_type_tag_user_{mod1, mod2, vmlinux}`` also requires
0218 pahole version ``1.23``.
0219 
0220 Without them, the btf_tag selftest will be skipped and you will observe:
0221 
0222 .. code-block:: console
0223 
0224   #<test_num> btf_tag:SKIP
0225 
0226 .. _0: https://reviews.llvm.org/D111588
0227 .. _1: https://reviews.llvm.org/D111199
0228 
0229 Clang dependencies for static linking tests
0230 ===========================================
0231 
0232 linked_vars, linked_maps, and linked_funcs tests depend on `Clang fix`__ to
0233 generate valid BTF information for weak variables. Please make sure you use
0234 Clang that contains the fix.
0235 
0236 __ https://reviews.llvm.org/D100362
0237 
0238 Clang relocation changes
0239 ========================
0240 
0241 Clang 13 patch `clang reloc patch`_  made some changes on relocations such
0242 that existing relocation types are broken into more types and
0243 each new type corresponds to only one way to resolve relocation.
0244 See `kernel llvm reloc`_ for more explanation and some examples.
0245 Using clang 13 to compile old libbpf which has static linker support,
0246 there will be a compilation failure::
0247 
0248   libbpf: ELF relo #0 in section #6 has unexpected type 2 in .../bpf_tcp_nogpl.o
0249 
0250 Here, ``type 2`` refers to new relocation type ``R_BPF_64_ABS64``.
0251 To fix this issue, user newer libbpf.
0252 
0253 .. Links
0254 .. _clang reloc patch: https://reviews.llvm.org/D102712
0255 .. _kernel llvm reloc: /Documentation/bpf/llvm_reloc.rst
0256 
0257 Clang dependencies for the u32 spill test (xdpwall)
0258 ===================================================
0259 The xdpwall selftest requires a change in `Clang 14`__.
0260 
0261 Without it, the xdpwall selftest will fail and the error message
0262 from running test_progs will look like:
0263 
0264 .. code-block:: console
0265 
0266   test_xdpwall:FAIL:Does LLVM have https://reviews.llvm.org/D109073? unexpected error: -4007
0267 
0268 __ https://reviews.llvm.org/D109073