From patchwork Thu Aug 1 04:57:24 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Yonghong Song X-Patchwork-Id: 13749522 X-Patchwork-Delegate: bpf@iogearbox.net Received: from 69-171-232-180.mail-mxout.facebook.com (69-171-232-180.mail-mxout.facebook.com [69.171.232.180]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 83A0013D899 for ; Thu, 1 Aug 2024 04:57:41 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=69.171.232.180 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1722488264; cv=none; b=jvmV+C9wUJ2gdy0HSaTBp152k9c4AHQGVicvclj8vqUiuQI7SQBdxGlbZXGcwLuuVabyXz4KWER89lqjNlaC43wzNDeGVAgepBXlB43YRZbQ6kK+av+U8KwS1Czd1Gv81eLikz2ce98XW3prCwO/vK+Uc13jCudpEifzdnj6SQQ= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1722488264; c=relaxed/simple; bh=+2A/BTgraciyW+z5GBjPU5j/ICW5DzaWHM0Riv/8/+g=; h=From:To:Cc:Subject:Date:Message-ID:MIME-Version; b=gesW8Cw0JZ9e7V+Dynv8M8t1Z/n2HEkM6wpbHTvA8lwx7e9hvZxZy2JrmSyHIFjguLUGc9m88u/Xq5PxRI0dsQLEtTQ59TzQk5id0ZFm3Td75RiAGWdlVdfxj14INg8Csm4afKCx48mosYo0ZVPRms5J/thaChKyYkbQGCL5rQI= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=fail (p=none dis=none) header.from=linux.dev; spf=fail smtp.mailfrom=linux.dev; arc=none smtp.client-ip=69.171.232.180 Authentication-Results: smtp.subspace.kernel.org; dmarc=fail (p=none dis=none) header.from=linux.dev Authentication-Results: smtp.subspace.kernel.org; spf=fail smtp.mailfrom=linux.dev Received: by devbig309.ftw3.facebook.com (Postfix, from userid 128203) id 8786A739B554; Wed, 31 Jul 2024 21:57:24 -0700 (PDT) From: Yonghong Song To: bpf@vger.kernel.org Cc: Alexei Starovoitov , Andrii Nakryiko , Daniel Borkmann , kernel-team@fb.com, Martin KaFai Lau Subject: [PATCH bpf-next] selftests/bpf: Fix arena_atomics selftest failure due to llvm change Date: Wed, 31 Jul 2024 21:57:24 -0700 Message-ID: <20240801045724.1259104-1-yonghong.song@linux.dev> X-Mailer: git-send-email 2.43.0 Precedence: bulk X-Mailing-List: bpf@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-Patchwork-Delegate: bpf@iogearbox.net Peilen Ye reported an issue ([1]) where for __sync_fetch_and_add(...) without return value like __sync_fetch_and_add(&foo, 1); llvm BPF backend generates locked insn e.g. lock *(u32 *)(r1 + 0) += r2 If __sync_fetch_and_add(...) returns a value like res = __sync_fetch_and_add(&foo, 1); llvm BPF backend generates like r2 = atomic_fetch_add((u32 *)(r1 + 0), r2) But 'lock *(u32 *)(r1 + 0) += r2' caused a problem in jit since proper barrier is not inserted based on __sync_fetch_and_add() semantics. The above discrepancy is due to commit [2] where it tries to maintain backward compatability since before commit [2], __sync_fetch_and_add(...) generates lock insn in BPF backend. Based on discussion in [1], now it is time to fix the above discrepancy so we can have proper barrier support in jit. llvm patch [3] made sure that __sync_fetch_and_add(...) always generates atomic_fetch_add(...) insns. Now 'lock *(u32 *)(r1 + 0) += r2' can only be generated by inline asm. The same for __sync_fetch_and_and(), __sync_fetch_and_or() and __sync_fetch_and_xor(). But the change in [3] caused arena_atomics selftest failure. test_arena_atomics:PASS:arena atomics skeleton open 0 nsec libbpf: prog 'and': BPF program load failed: Permission denied libbpf: prog 'and': -- BEGIN PROG LOAD LOG -- arg#0 reference type('UNKNOWN ') size cannot be determined: -22 0: R1=ctx() R10=fp0 ; if (pid != (bpf_get_current_pid_tgid() >> 32)) @ arena_atomics.c:87 0: (18) r1 = 0xffffc90000064000 ; R1_w=map_value(map=arena_at.bss,ks=4,vs=4) 2: (61) r6 = *(u32 *)(r1 +0) ; R1_w=map_value(map=arena_at.bss,ks=4,vs=4) R6_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) 3: (85) call bpf_get_current_pid_tgid#14 ; R0_w=scalar() 4: (77) r0 >>= 32 ; R0_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) 5: (5d) if r0 != r6 goto pc+11 ; R0_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R6_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0x) ; __sync_fetch_and_and(&and64_value, 0x011ull << 32); @ arena_atomics.c:91 6: (18) r1 = 0x100000000060 ; R1_w=scalar() 8: (bf) r1 = addr_space_cast(r1, 0, 1) ; R1_w=arena 9: (18) r2 = 0x1100000000 ; R2_w=0x1100000000 11: (db) r2 = atomic64_fetch_and((u64 *)(r1 +0), r2) BPF_ATOMIC stores into R1 arena is not allowed processed 9 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0 -- END PROG LOAD LOG -- libbpf: prog 'and': failed to load: -13 libbpf: failed to load object 'arena_atomics' libbpf: failed to load BPF skeleton 'arena_atomics': -13 test_arena_atomics:FAIL:arena atomics skeleton load unexpected error: -13 (errno 13) #3 arena_atomics:FAIL The reason of the failure is due to [4] where atomic{64,}_fetch_{and,or,xor}() are not allowed by arena addresses. Without llvm patch [3], the compiler will generate 'lock ...' insn and everything will work fine. This patch fixed the problem by using inline asms. Instead of __sync_fetch_and_{and,or,xor}() functions, the inline asm with 'lock' insn is used and it will work with or without [3]. [1] https://lore.kernel.org/bpf/ZqqiQQWRnz7H93Hc@google.com/T/#mb68d67bc8f39e35a0c3db52468b9de59b79f021f [2] https://github.com/llvm/llvm-project/commit/286daafd65129228e08a1d07aa4ca74488615744 [3] https://github.com/llvm/llvm-project/pull/101428 [4] d503a04f8bc0 ("bpf: Add support for certain atomics in bpf_arena to x86 JIT") Signed-off-by: Yonghong Song --- .../selftests/bpf/progs/arena_atomics.c | 57 +++++++++++++++++-- 1 file changed, 51 insertions(+), 6 deletions(-) diff --git a/tools/testing/selftests/bpf/progs/arena_atomics.c b/tools/testing/selftests/bpf/progs/arena_atomics.c index bb0acd79d28a..611983446427 100644 --- a/tools/testing/selftests/bpf/progs/arena_atomics.c +++ b/tools/testing/selftests/bpf/progs/arena_atomics.c @@ -5,6 +5,7 @@ #include #include #include "bpf_arena_common.h" +#include "bpf_misc.h" struct { __uint(type, BPF_MAP_TYPE_ARENA); @@ -87,8 +88,22 @@ int and(const void *ctx) return 0; #ifdef ENABLE_ATOMICS_TESTS - __sync_fetch_and_and(&and64_value, 0x011ull << 32); - __sync_fetch_and_and(&and32_value, 0x011); + asm volatile( + "r1 = addr_space_cast(%[and64_value], 0, 1);" + "lock *(u64 *)(r1 + 0) &= %[val]" + : + : __imm_ptr(and64_value), + [val]"r"(0x011ull << 32) + : "r1" + ); + asm volatile( + "r1 = addr_space_cast(%[and32_value], 0, 1);" + "lock *(u32 *)(r1 + 0) &= %[val]" + : + : __imm_ptr(and32_value), + [val]"w"(0x011) + : "r1" + ); #endif return 0; @@ -103,8 +118,23 @@ int or(const void *ctx) if (pid != (bpf_get_current_pid_tgid() >> 32)) return 0; #ifdef ENABLE_ATOMICS_TESTS - __sync_fetch_and_or(&or64_value, 0x011ull << 32); - __sync_fetch_and_or(&or32_value, 0x011); + + asm volatile( + "r1 = addr_space_cast(%[or64_value], 0, 1);" + "lock *(u64 *)(r1 + 0) |= %[val]" + : + : __imm_ptr(or64_value), + [val]"r"(0x011ull << 32) + : "r1" + ); + asm volatile( + "r1 = addr_space_cast(%[or32_value], 0, 1);" + "lock *(u32 *)(r1 + 0) |= %[val]" + : + : __imm_ptr(or32_value), + [val]"w"(0x011) + : "r1" + ); #endif return 0; @@ -119,8 +149,23 @@ int xor(const void *ctx) if (pid != (bpf_get_current_pid_tgid() >> 32)) return 0; #ifdef ENABLE_ATOMICS_TESTS - __sync_fetch_and_xor(&xor64_value, 0x011ull << 32); - __sync_fetch_and_xor(&xor32_value, 0x011); + + asm volatile( + "r1 = addr_space_cast(%[xor64_value], 0, 1);" + "lock *(u64 *)(r1 + 0) ^= %[val]" + : + : __imm_ptr(xor64_value), + [val]"r"(0x011ull << 32) + : "r1" + ); + asm volatile( + "r1 = addr_space_cast(%[xor32_value], 0, 1);" + "lock *(u32 *)(r1 + 0) ^= %[val]" + : + : __imm_ptr(xor32_value), + [val]"w"(0x011) + : "r1" + ); #endif return 0;