Skip to contentSkip to conversation

BPF kselftest cross-build/RISC-V fixes

PATCH In review Björn Töpel 2 participants 5 messages Jul 5, 2023

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

cover BPF kselftest cross-build/RISC-V fixes

From: Björn Töpel <[email protected]>

This series has two minor fixes, found when cross-compiling for the RISC-V architecture.

Some RISC-V systems do not define HAVE_EFFICIENT_UNALIGNED_ACCESS, which made some of tests bail out. Fix the failing tests by adding F_NEEDS_EFFICIENT_UNALIGNED_ACCESS.

...and some RISC-V systems *do* define HAVE_EFFICIENT_UNALIGNED_ACCESS. In this case the autoconf.h was not correctly picked up by the build system.

Cheers,
Björn

Björn Töpel (2):
selftests/bpf: Add F_NEEDS_EFFICIENT_UNALIGNED_ACCESS to some tests selftests/bpf: Honor $(O) when figuring out paths

 tools/testing/selftests/bpf/Makefile                  | 4 ++++
 tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c | 1 +
 tools/testing/selftests/bpf/verifier/ctx_skb.c        | 2 ++
 tools/testing/selftests/bpf/verifier/jmp32.c          | 8 ++++++++
 tools/testing/selftests/bpf/verifier/map_kptr.c       | 2 ++
 tools/testing/selftests/bpf/verifier/precise.c        | 2 +-
 6 files changed, 18 insertions(+), 1 deletion(-)

base-commit: a94098d490e17d652770f2309fcb9b46bc4cf864

Signature · 1 line
2.39.2
#1 lore raw

PATCH 1/2 selftests/bpf: Add F_NEEDS_EFFICIENT_UNALIGNED_ACCESS to some tests

From: Björn Töpel <[email protected]>

Some verifier tests were missing F_NEEDS_EFFICIENT_UNALIGNED_ACCESS, which made the test fail. Add the flag where needed.

---
 tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c | 1 +
 tools/testing/selftests/bpf/verifier/ctx_skb.c        | 2 ++
 tools/testing/selftests/bpf/verifier/jmp32.c          | 8 ++++++++
 tools/testing/selftests/bpf/verifier/map_kptr.c       | 2 ++
 tools/testing/selftests/bpf/verifier/precise.c        | 2 +-
 5 files changed, 14 insertions(+), 1 deletion(-)
tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c +1 -0
index b39665f33524..319337bdcfc8 100644
@@ -242,4 +242,5 @@
 	.result = REJECT,
 	.errstr = "R0 invalid mem access",
 	.errstr_unpriv = "R10 partial copy of pointer",
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
tools/testing/selftests/bpf/verifier/ctx_skb.c +2 -0
index 83cecfbd6739..0b394a7f7a2d 100644
@@ -1169,6 +1169,7 @@
        },
        .result = ACCEPT,
        .prog_type = BPF_PROG_TYPE_SK_SKB,
+       .flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
        "pkt_end < pkt taken check",
@@ -1190,4 +1191,5 @@
        },
        .result = ACCEPT,
        .prog_type = BPF_PROG_TYPE_SK_SKB,
+       .flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
tools/testing/selftests/bpf/verifier/jmp32.c +8 -0
index 1a27a6210554..43776f6f92f4 100644
@@ -290,6 +290,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jgt32: BPF_K",
@@ -360,6 +361,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jle32: BPF_K",
@@ -430,6 +432,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jlt32: BPF_K",
@@ -500,6 +503,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jsge32: BPF_K",
@@ -570,6 +574,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jsgt32: BPF_K",
@@ -640,6 +645,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jsle32: BPF_K",
@@ -710,6 +716,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jslt32: BPF_K",
@@ -780,6 +787,7 @@
 	.result_unpriv = REJECT,
 	.result = ACCEPT,
 	.retval = 2,
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"jgt32: range bound deduction, reg op imm",
tools/testing/selftests/bpf/verifier/map_kptr.c +2 -0
index a0cfc06d75bc..d25c3e9605f1 100644
@@ -68,6 +68,7 @@
 	.fixup_map_kptr = { 1 },
 	.result = REJECT,
 	.errstr = "kptr access cannot have variable offset",
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"map_kptr: bpf_kptr_xchg non-const var_off",
@@ -121,6 +122,7 @@
 	.fixup_map_kptr = { 1 },
 	.result = REJECT,
 	.errstr = "kptr access misaligned expected=0 off=7",
+	.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 },
 {
 	"map_kptr: reject var_off != 0",
tools/testing/selftests/bpf/verifier/precise.c +1 -1
index 99272bb890da..0d84dd1f38b6 100644
@@ -216,7 +216,7 @@
 	},
 	.fixup_map_ringbuf = { 1 },
 	.prog_type = BPF_PROG_TYPE_XDP,
-	.flags = BPF_F_TEST_STATE_FREQ,
+	.flags = BPF_F_TEST_STATE_FREQ | F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
 	.errstr = "invalid access to memory, mem_size=1 off=42 size=8",
 	.result = REJECT,
 },
Signature · 1 line
2.39.2
#2 lore raw

PATCH 2/2 selftests/bpf: Honor $(O) when figuring out paths

In reply to Björn Töpel

From: Björn Töpel <[email protected]>

When building the kselftests out-of-tree, e.g.

  | make ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- \
  |   O=/tmp/kselftest headers
  | make ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- \
  |  O=/tmp/kselftest HOSTCC=gcc FORMAT= \
  |  SKIP_TARGETS="arm64 ia64 powerpc sparc64 x86 sgx" \
  |  -C tools/testing/selftests gen_tar

the kselftest build would not pick up the correct GENDIR path, and therefore not including autoconf.h.

Correct that by taking $(O) into consideration when figuring out the GENDIR path.

---
 tools/testing/selftests/bpf/Makefile | 4 ++++
 1 file changed, 4 insertions(+)
tools/testing/selftests/bpf/Makefile +4 -0
index ad6b585e0d7c..daccc1b8573a 100644
@@ -12,7 +12,11 @@ BPFDIR := $(LIBDIR)/bpf
 TOOLSINCDIR := $(TOOLSDIR)/include
 BPFTOOLDIR := $(TOOLSDIR)/bpf/bpftool
 APIDIR := $(TOOLSINCDIR)/uapi
+ifneq ($(O),)
+GENDIR := $(O)/include/generated
+else
 GENDIR := $(abspath ../../../../include/generated)
+endif
 GENHDR := $(GENDIR)/autoconf.h
 HOSTPKG_CONFIG := pkg-config
 
Signature · 1 line
2.39.2
maintainer
#3 lore raw
In reply to Björn Töpel
Quoted 5 lines · Björn Töpel

From: Björn Töpel <[email protected]>

This series has two minor fixes, found when cross-compiling for the RISC-V architecture.

Quoted 7 lines

Some RISC-V systems do not define HAVE_EFFICIENT_UNALIGNED_ACCESS, which made some of tests bail out. Fix the failing tests by adding F_NEEDS_EFFICIENT_UNALIGNED_ACCESS.

...and some RISC-V systems *do* define HAVE_EFFICIENT_UNALIGNED_ACCESS. In this case the autoconf.h was not correctly picked up by the build system.

Looks good, applied thanks! Any plans on working towards integrating riscv into upstream BPF CI? Would love to see that happening. :)

Thanks,
Daniel

#4 lore raw
Quoted 10 lines · Daniel Borkmann
Quoted 9 lines · Björn Töpel

From: Björn Töpel <[email protected]>

This series has two minor fixes, found when cross-compiling for the RISC-V architecture.

Some RISC-V systems do not define HAVE_EFFICIENT_UNALIGNED_ACCESS, which made some of tests bail out. Fix the failing tests by adding F_NEEDS_EFFICIENT_UNALIGNED_ACCESS.

Quoted 5 lines
Quoted 3 lines

...and some RISC-V systems *do* define HAVE_EFFICIENT_UNALIGNED_ACCESS. In this case the autoconf.h was not correctly picked up by the build system.

Looks good, applied thanks!

Thank you!

Quoted 2 lines

Any plans on working towards integrating riscv into upstream BPF CI? Would love to see that happening. :)

Yes! I started hacking a bit on that some time back:

https://github.com/libbpf/ci/pull/87
https://github.com/kernel-patches/vmtest/pull/194

(TL;DR: I'll continuing that work at some point.)

RISC-V still needs cross-compilation, and testing on qemu/TCG (on typically x86-hosts), which puts some constraints on the rootfs/cross-compilation host; For RISC-V Debian Bullseye is way too old (a lot packages are missing/broken). Typically for BPF it would be Ubuntu Kinetic (or later), or some Debian Sid snapshot.

The rootfs, the host, and the host foreign arch would need to be the same for "no-hassle cross-compilation on Debian derivatives" -- and at least younger than "Ubuntu Kinetic"-age.

AFAIU, there are some issues with rootfs version and build host versioning for other archs as well: https://github.com/libbpf/ci/pull/83

Björn