[PATCH bpf-next v3 0/9] Introduce load-acquire and store-release BPF instructions

[PATCH bpf-next v3 0/9] Introduce load-acquire and store-release BPF instructions

[Peilin Ye]

Hi all!

This patchset adds kernel support for BPF load-acquire and store-release
instructions (for background, please see [1]), mainly including
core/verifier, arm64 JIT compiler, and Documentation/ changes.  x86-64
and riscv64 are also planned to be supported.  The corresponding LLVM
changes can be found at:

  [2] https://github.com/llvm/llvm-project/pull/108636

Tested using PLATFORM=aarch64 CROSS_COMPILE=aarch64-linux-gnu-
vmtest.sh, with llvm-17, llvm-18 and [2].

Please refer to the LLVM PR and individual kernel patches for details.
Feedback is much appreciated!

v2: https://lore.kernel.org/bpf/cover.1738888641.git.yepeilin@google.com/
v2..v3 notable changes:

  o (Alexei) change encoding to BPF_LOAD_ACQ=0x100, BPF_STORE_REL=0x110
  o add Acked-by: tags from Ilya and Eduard
  o make new selftests depend on:
    * __clang_major__ >= 18, and
    * ENABLE_ATOMICS_TESTS is defined (currently this means -mcpu=v3 or
      v4), and
    * JIT supports load_acq/store_rel (currenty only arm64)
  o work around llvm-17 CI job failure by conditionally define
    __arena_global variables as 64-bit if __clang_major__ < 18, to make
    sure .addr_space.1 has no holes
  o add Google copyright notice in new files

v1: https://lore.kernel.org/all/cover.1737763916.git.yepeilin@google.com/
v1..v2 notable changes:

  o (Eduard) for x86 and s390, make
             bpf_jit_supports_insn(..., /*in_arena=*/true) return false
	     for load_acq/store_rel
  o add Eduard's Acked-by: tag
  o (Eduard) extract LDX and non-ATOMIC STX handling into helpers, see
             PATCH v2 3/9
  o allow unpriv programs to store-release pointers to stack
  o (Alexei) make it clearer in the interpreter code (PATCH v2 4/9) that
             only W and DW are supported for atomic RMW
  o test misaligned load_acq/store_rel
  o (Eduard) other selftests/ changes:
    * test load_acq/store_rel with !atomic_ptr_type_ok() pointers:
      - PTR_TO_CTX, for is_ctx_reg()
      - PTR_TO_PACKET, for is_pkt_reg()
      - PTR_TO_FLOW_KEYS, for is_flow_key_reg()
      - PTR_TO_SOCKET, for is_sk_reg()
    * drop atomics/ tests
    * delete unnecessary 'pid' checks from arena_atomics/ tests
    * avoid depending on __BPF_FEATURE_LOAD_ACQ_STORE_REL, use
      __imm_insn() and inline asm macros instead

RFC v1: https://lore.kernel.org/all/cover.1734742802.git.yepeilin@google.com
RFC v1..v1 notable changes:

  o 1-2/8: minor verifier.c refactoring patches
  o   3/8: core/verifier changes
         * (Eduard) handle load-acquire properly in backtrack_insn()
         * (Eduard) avoid skipping checks (e.g.,
                    bpf_jit_supports_insn()) for load-acquires
         * track the value stored by store-releases, just like how
           non-atomic STX instructions are handled
         * (Eduard) add missing link in commit message
         * (Eduard) always print 'r' for disasm.c changes
  o   4/8: arm64/insn: avoid treating load_acq/store_rel as
           load_ex/store_ex
  o   5/8: arm64/insn: add load_acq/store_rel
         * (Xu) include Should-Be-One (SBO) bits in "mask" and "value",
                to avoid setting fixed bits during runtime (JIT-compile
                time)
  o   6/8: arm64 JIT compiler changes
         * (Xu) use emit_a64_add_i() for "pointer + offset" to optimize
                code emission
  o   7/8: selftests
         * (Eduard) avoid adding new tests to the 'test_verifier' runner
         * add more tests, e.g., checking mark_precise logic
  o   8/8: instruction-set.rst changes

[1] https://lore.kernel.org/all/20240729183246.4110549-1-yepeilin@google.com/

Thanks,
Peilin Ye (9):
  bpf/verifier: Factor out atomic_ptr_type_ok()
  bpf/verifier: Factor out check_atomic_rmw()
  bpf/verifier: Factor out check_load_mem() and check_store_reg()
  bpf: Introduce load-acquire and store-release instructions
  arm64: insn: Add BIT(23) to {load,store}_ex's mask
  arm64: insn: Add load-acquire and store-release instructions
  bpf, arm64: Support load-acquire and store-release instructions
  selftests/bpf: Add selftests for load-acquire and store-release
    instructions
  bpf, docs: Update instruction-set.rst for load-acquire and
    store-release instructions

 .../bpf/standardization/instruction-set.rst   |  78 ++++--
 arch/arm64/include/asm/insn.h                 |  12 +-
 arch/arm64/lib/insn.c                         |  29 ++
 arch/arm64/net/bpf_jit.h                      |  20 ++
 arch/arm64/net/bpf_jit_comp.c                 |  87 +++++-
 arch/s390/net/bpf_jit_comp.c                  |  14 +-
 arch/x86/net/bpf_jit_comp.c                   |   4 +
 include/linux/bpf.h                           |  15 +
 include/linux/filter.h                        |   2 +
 include/uapi/linux/bpf.h                      |   3 +
 kernel/bpf/core.c                             |  63 ++++-
 kernel/bpf/disasm.c                           |  12 +
 kernel/bpf/verifier.c                         | 234 +++++++++++-----
 tools/include/uapi/linux/bpf.h                |   3 +
 .../selftests/bpf/prog_tests/arena_atomics.c  |  66 ++++-
 .../selftests/bpf/prog_tests/verifier.c       |   4 +
 .../selftests/bpf/progs/arena_atomics.c       | 118 +++++++-
 .../bpf/progs/verifier_load_acquire.c         | 197 +++++++++++++
 .../selftests/bpf/progs/verifier_precision.c  |  48 ++++
 .../bpf/progs/verifier_store_release.c        | 264 ++++++++++++++++++
 20 files changed, 1165 insertions(+), 108 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/verifier_load_acquire.c
 create mode 100644 tools/testing/selftests/bpf/progs/verifier_store_release.c

-- 
2.48.1.601.g30ceb7b040-goog

[PATCH bpf-next v3 1/9] bpf/verifier: Factor out atomic_ptr_type_ok()

[Peilin Ye]

Factor out atomic_ptr_type_ok() as a helper function to be used later.

Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 kernel/bpf/verifier.c | 26 +++++++++++++++++++++-----
 1 file changed, 21 insertions(+), 5 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index e57b7c949860..21658bd5e6d8 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -6194,6 +6194,26 @@ static bool is_arena_reg(struct bpf_verifier_env *env, int regno)
 	return reg->type == PTR_TO_ARENA;
 }
 
+/* Return false if @regno contains a pointer whose type isn't supported for
+ * atomic instruction @insn.
+ */
+static bool atomic_ptr_type_ok(struct bpf_verifier_env *env, int regno,
+			       struct bpf_insn *insn)
+{
+	if (is_ctx_reg(env, regno))
+		return false;
+	if (is_pkt_reg(env, regno))
+		return false;
+	if (is_flow_key_reg(env, regno))
+		return false;
+	if (is_sk_reg(env, regno))
+		return false;
+	if (is_arena_reg(env, regno))
+		return bpf_jit_supports_insn(insn, true);
+
+	return true;
+}
+
 static u32 *reg2btf_ids[__BPF_REG_TYPE_MAX] = {
 #ifdef CONFIG_NET
 	[PTR_TO_SOCKET] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK],
@@ -7651,11 +7671,7 @@ static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_i
 		return -EACCES;
 	}
 
-	if (is_ctx_reg(env, insn->dst_reg) ||
-	    is_pkt_reg(env, insn->dst_reg) ||
-	    is_flow_key_reg(env, insn->dst_reg) ||
-	    is_sk_reg(env, insn->dst_reg) ||
-	    (is_arena_reg(env, insn->dst_reg) && !bpf_jit_supports_insn(insn, true))) {
+	if (!atomic_ptr_type_ok(env, insn->dst_reg, insn)) {
 		verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n",
 			insn->dst_reg,
 			reg_type_str(env, reg_state(env, insn->dst_reg)->type));
-- 
2.48.1.601.g30ceb7b040-goog

[PATCH bpf-next v3 2/9] bpf/verifier: Factor out check_atomic_rmw()

[Peilin Ye]

Currently, check_atomic() only handles atomic read-modify-write (RMW)
instructions.  Since we are planning to introduce other types of atomic
instructions (i.e., atomic load/store), extract the existing RMW
handling logic into its own function named check_atomic_rmw().

Remove the @insn_idx parameter as it is not really necessary.  Use
'env->insn_idx' instead, as in other places in verifier.c.

Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 kernel/bpf/verifier.c | 53 +++++++++++++++++++++++--------------------
 1 file changed, 29 insertions(+), 24 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 21658bd5e6d8..63d810bbc26e 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -7615,28 +7615,12 @@ static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
 static int save_aux_ptr_type(struct bpf_verifier_env *env, enum bpf_reg_type type,
 			     bool allow_trust_mismatch);
 
-static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn)
+static int check_atomic_rmw(struct bpf_verifier_env *env,
+			    struct bpf_insn *insn)
 {
 	int load_reg;
 	int err;
 
-	switch (insn->imm) {
-	case BPF_ADD:
-	case BPF_ADD | BPF_FETCH:
-	case BPF_AND:
-	case BPF_AND | BPF_FETCH:
-	case BPF_OR:
-	case BPF_OR | BPF_FETCH:
-	case BPF_XOR:
-	case BPF_XOR | BPF_FETCH:
-	case BPF_XCHG:
-	case BPF_CMPXCHG:
-		break;
-	default:
-		verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm);
-		return -EINVAL;
-	}
-
 	if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) {
 		verbose(env, "invalid atomic operand size\n");
 		return -EINVAL;
@@ -7698,12 +7682,12 @@ static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_i
 	/* Check whether we can read the memory, with second call for fetch
 	 * case to simulate the register fill.
 	 */
-	err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
+	err = check_mem_access(env, env->insn_idx, insn->dst_reg, insn->off,
 			       BPF_SIZE(insn->code), BPF_READ, -1, true, false);
 	if (!err && load_reg >= 0)
-		err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
-				       BPF_SIZE(insn->code), BPF_READ, load_reg,
-				       true, false);
+		err = check_mem_access(env, env->insn_idx, insn->dst_reg,
+				       insn->off, BPF_SIZE(insn->code),
+				       BPF_READ, load_reg, true, false);
 	if (err)
 		return err;
 
@@ -7713,13 +7697,34 @@ static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_i
 			return err;
 	}
 	/* Check whether we can write into the same memory. */
-	err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
+	err = check_mem_access(env, env->insn_idx, insn->dst_reg, insn->off,
 			       BPF_SIZE(insn->code), BPF_WRITE, -1, true, false);
 	if (err)
 		return err;
 	return 0;
 }
 
+static int check_atomic(struct bpf_verifier_env *env, struct bpf_insn *insn)
+{
+	switch (insn->imm) {
+	case BPF_ADD:
+	case BPF_ADD | BPF_FETCH:
+	case BPF_AND:
+	case BPF_AND | BPF_FETCH:
+	case BPF_OR:
+	case BPF_OR | BPF_FETCH:
+	case BPF_XOR:
+	case BPF_XOR | BPF_FETCH:
+	case BPF_XCHG:
+	case BPF_CMPXCHG:
+		return check_atomic_rmw(env, insn);
+	default:
+		verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n",
+			insn->imm);
+		return -EINVAL;
+	}
+}
+
 /* When register 'regno' is used to read the stack (either directly or through
  * a helper function) make sure that it's within stack boundary and, depending
  * on the access type and privileges, that all elements of the stack are
@@ -19187,7 +19192,7 @@ static int do_check(struct bpf_verifier_env *env)
 			enum bpf_reg_type dst_reg_type;
 
 			if (BPF_MODE(insn->code) == BPF_ATOMIC) {
-				err = check_atomic(env, env->insn_idx, insn);
+				err = check_atomic(env, insn);
 				if (err)
 					return err;
 				env->insn_idx++;
-- 
2.48.1.601.g30ceb7b040-goog

[PATCH bpf-next v3 3/9] bpf/verifier: Factor out check_load_mem() and check_store_reg()

[Peilin Ye]

Extract BPF_LDX and most non-ATOMIC BPF_STX instruction handling logic
in do_check() into helper functions to be used later.  While we are
here, make that comment about "reserved fields" more specific.

Suggested-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 kernel/bpf/verifier.c | 110 +++++++++++++++++++++++++-----------------
 1 file changed, 67 insertions(+), 43 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 63d810bbc26e..3b20bff1c5eb 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -7615,6 +7615,67 @@ static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
 static int save_aux_ptr_type(struct bpf_verifier_env *env, enum bpf_reg_type type,
 			     bool allow_trust_mismatch);
 
+static int check_load_mem(struct bpf_verifier_env *env, struct bpf_insn *insn,
+			  bool strict_alignment_once, bool is_ldsx,
+			  bool allow_trust_mismatch, const char *ctx)
+{
+	struct bpf_reg_state *regs = cur_regs(env);
+	enum bpf_reg_type src_reg_type;
+	int err;
+
+	/* check src operand */
+	err = check_reg_arg(env, insn->src_reg, SRC_OP);
+	if (err)
+		return err;
+
+	/* check dst operand */
+	err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
+	if (err)
+		return err;
+
+	src_reg_type = regs[insn->src_reg].type;
+
+	/* Check if (src_reg + off) is readable. The state of dst_reg will be
+	 * updated by this call.
+	 */
+	err = check_mem_access(env, env->insn_idx, insn->src_reg, insn->off,
+			       BPF_SIZE(insn->code), BPF_READ, insn->dst_reg,
+			       strict_alignment_once, is_ldsx);
+	err = err ?: save_aux_ptr_type(env, src_reg_type,
+				       allow_trust_mismatch);
+	err = err ?: reg_bounds_sanity_check(env, &regs[insn->dst_reg], ctx);
+
+	return err;
+}
+
+static int check_store_reg(struct bpf_verifier_env *env, struct bpf_insn *insn,
+			   bool strict_alignment_once)
+{
+	struct bpf_reg_state *regs = cur_regs(env);
+	enum bpf_reg_type dst_reg_type;
+	int err;
+
+	/* check src1 operand */
+	err = check_reg_arg(env, insn->src_reg, SRC_OP);
+	if (err)
+		return err;
+
+	/* check src2 operand */
+	err = check_reg_arg(env, insn->dst_reg, SRC_OP);
+	if (err)
+		return err;
+
+	dst_reg_type = regs[insn->dst_reg].type;
+
+	/* Check if (dst_reg + off) is writeable. */
+	err = check_mem_access(env, env->insn_idx, insn->dst_reg, insn->off,
+			       BPF_SIZE(insn->code), BPF_WRITE, insn->src_reg,
+			       strict_alignment_once, false);
+	err = err ?: save_aux_ptr_type(env, dst_reg_type, false);
+
+	return err;
+}
+
 static int check_atomic_rmw(struct bpf_verifier_env *env,
 			    struct bpf_insn *insn)
 {
@@ -19162,35 +19223,16 @@ static int do_check(struct bpf_verifier_env *env)
 				return err;
 
 		} else if (class == BPF_LDX) {
-			enum bpf_reg_type src_reg_type;
-
-			/* check for reserved fields is already done */
-
-			/* check src operand */
-			err = check_reg_arg(env, insn->src_reg, SRC_OP);
-			if (err)
-				return err;
+			bool is_ldsx = BPF_MODE(insn->code) == BPF_MEMSX;
 
-			err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
-			if (err)
-				return err;
-
-			src_reg_type = regs[insn->src_reg].type;
-
-			/* check that memory (src_reg + off) is readable,
-			 * the state of dst_reg will be updated by this func
+			/* Check for reserved fields is already done in
+			 * resolve_pseudo_ldimm64().
 			 */
-			err = check_mem_access(env, env->insn_idx, insn->src_reg,
-					       insn->off, BPF_SIZE(insn->code),
-					       BPF_READ, insn->dst_reg, false,
-					       BPF_MODE(insn->code) == BPF_MEMSX);
-			err = err ?: save_aux_ptr_type(env, src_reg_type, true);
-			err = err ?: reg_bounds_sanity_check(env, &regs[insn->dst_reg], "ldx");
+			err = check_load_mem(env, insn, false, is_ldsx, true,
+					     "ldx");
 			if (err)
 				return err;
 		} else if (class == BPF_STX) {
-			enum bpf_reg_type dst_reg_type;
-
 			if (BPF_MODE(insn->code) == BPF_ATOMIC) {
 				err = check_atomic(env, insn);
 				if (err)
@@ -19204,25 +19246,7 @@ static int do_check(struct bpf_verifier_env *env)
 				return -EINVAL;
 			}
 
-			/* check src1 operand */
-			err = check_reg_arg(env, insn->src_reg, SRC_OP);
-			if (err)
-				return err;
-			/* check src2 operand */
-			err = check_reg_arg(env, insn->dst_reg, SRC_OP);
-			if (err)
-				return err;
-
-			dst_reg_type = regs[insn->dst_reg].type;
-
-			/* check that memory (dst_reg + off) is writeable */
-			err = check_mem_access(env, env->insn_idx, insn->dst_reg,
-					       insn->off, BPF_SIZE(insn->code),
-					       BPF_WRITE, insn->src_reg, false, false);
-			if (err)
-				return err;
-
-			err = save_aux_ptr_type(env, dst_reg_type, false);
+			err = check_store_reg(env, insn, false);
 			if (err)
 				return err;
 		} else if (class == BPF_ST) {
-- 
2.48.1.601.g30ceb7b040-goog

[PATCH bpf-next v3 4/9] bpf: Introduce load-acquire and store-release instructions

[Peilin Ye]

Introduce BPF instructions with load-acquire and store-release
semantics, as discussed in [1].  Define 2 new flags:

  #define BPF_LOAD_ACQ    0x100
  #define BPF_STORE_REL   0x110

A "load-acquire" is a BPF_STX | BPF_ATOMIC instruction with the 'imm'
field set to BPF_LOAD_ACQ (0x100).

Similarly, a "store-release" is a BPF_STX | BPF_ATOMIC instruction with
the 'imm' field set to BPF_STORE_REL (0x110).

Unlike existing atomic read-modify-write operations that only support
BPF_W (32-bit) and BPF_DW (64-bit) size modifiers, load-acquires and
store-releases also support BPF_B (8-bit) and BPF_H (16-bit).  An 8- or
16-bit load-acquire zero-extends the value before writing it to a 32-bit
register, just like ARM64 instruction LDARH and friends.

Similar to existing atomic read-modify-write operations, misaligned
load-acquires/store-releases are not allowed (even if
BPF_F_ANY_ALIGNMENT is set).

As an example, consider the following 64-bit load-acquire BPF
instruction (assuming little-endian):

  db 10 00 00 00 01 00 00  r0 = load_acquire((u64 *)(r1 + 0x0))

  opcode (0xdb): BPF_ATOMIC | BPF_DW | BPF_STX
  imm (0x00000100): BPF_LOAD_ACQ

Similarly, a 16-bit BPF store-release:

  cb 21 00 00 10 01 00 00  store_release((u16 *)(r1 + 0x0), w2)

  opcode (0xcb): BPF_ATOMIC | BPF_H | BPF_STX
  imm (0x00000110): BPF_STORE_REL

In arch/{arm64,s390,x86}/net/bpf_jit_comp.c, have
bpf_jit_supports_insn(..., /*in_arena=*/true) return false for the new
instructions, until the corresponding JIT compiler supports them.

[1] https://lore.kernel.org/all/20240729183246.4110549-1-yepeilin@google.com/

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 arch/arm64/net/bpf_jit_comp.c  |  4 +++
 arch/s390/net/bpf_jit_comp.c   | 14 +++++---
 arch/x86/net/bpf_jit_comp.c    |  4 +++
 include/linux/bpf.h            | 15 ++++++++
 include/linux/filter.h         |  2 ++
 include/uapi/linux/bpf.h       |  3 ++
 kernel/bpf/core.c              | 63 ++++++++++++++++++++++++++++++----
 kernel/bpf/disasm.c            | 12 +++++++
 kernel/bpf/verifier.c          | 45 ++++++++++++++++++++++--
 tools/include/uapi/linux/bpf.h |  3 ++
 10 files changed, 152 insertions(+), 13 deletions(-)

diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
index 8446848edddb..8c3b47d9e441 100644
--- a/arch/arm64/net/bpf_jit_comp.c
+++ b/arch/arm64/net/bpf_jit_comp.c
@@ -2667,8 +2667,12 @@ bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
 	if (!in_arena)
 		return true;
 	switch (insn->code) {
+	case BPF_STX | BPF_ATOMIC | BPF_B:
+	case BPF_STX | BPF_ATOMIC | BPF_H:
 	case BPF_STX | BPF_ATOMIC | BPF_W:
 	case BPF_STX | BPF_ATOMIC | BPF_DW:
+		if (bpf_atomic_is_load_store(insn))
+			return false;
 		if (!cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
 			return false;
 	}
diff --git a/arch/s390/net/bpf_jit_comp.c b/arch/s390/net/bpf_jit_comp.c
index 9d440a0b729e..0776dfde2dba 100644
--- a/arch/s390/net/bpf_jit_comp.c
+++ b/arch/s390/net/bpf_jit_comp.c
@@ -2919,10 +2919,16 @@ bool bpf_jit_supports_arena(void)
 
 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
 {
-	/*
-	 * Currently the verifier uses this function only to check which
-	 * atomic stores to arena are supported, and they all are.
-	 */
+	if (!in_arena)
+		return true;
+	switch (insn->code) {
+	case BPF_STX | BPF_ATOMIC | BPF_B:
+	case BPF_STX | BPF_ATOMIC | BPF_H:
+	case BPF_STX | BPF_ATOMIC | BPF_W:
+	case BPF_STX | BPF_ATOMIC | BPF_DW:
+		if (bpf_atomic_is_load_store(insn))
+			return false;
+	}
 	return true;
 }
 
diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index a43fc5af973d..f0c31c940fb8 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -3771,8 +3771,12 @@ bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
 	if (!in_arena)
 		return true;
 	switch (insn->code) {
+	case BPF_STX | BPF_ATOMIC | BPF_B:
+	case BPF_STX | BPF_ATOMIC | BPF_H:
 	case BPF_STX | BPF_ATOMIC | BPF_W:
 	case BPF_STX | BPF_ATOMIC | BPF_DW:
+		if (bpf_atomic_is_load_store(insn))
+			return false;
 		if (insn->imm == (BPF_AND | BPF_FETCH) ||
 		    insn->imm == (BPF_OR | BPF_FETCH) ||
 		    insn->imm == (BPF_XOR | BPF_FETCH))
diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 15164787ce7f..9be9c586152f 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -991,6 +991,21 @@ static inline bool bpf_pseudo_func(const struct bpf_insn *insn)
 	return bpf_is_ldimm64(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
 }
 
+/* Given a BPF_ATOMIC instruction @atomic_insn, return true if it is an
+ * atomic load or store, and false if it is a read-modify-write instruction.
+ */
+static inline bool
+bpf_atomic_is_load_store(const struct bpf_insn *atomic_insn)
+{
+	switch (atomic_insn->imm) {
+	case BPF_LOAD_ACQ:
+	case BPF_STORE_REL:
+		return true;
+	default:
+		return false;
+	}
+}
+
 struct bpf_prog_ops {
 	int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr,
 			union bpf_attr __user *uattr);
diff --git a/include/linux/filter.h b/include/linux/filter.h
index a3ea46281595..e36812a5b01f 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -364,6 +364,8 @@ static inline bool insn_is_cast_user(const struct bpf_insn *insn)
  *   BPF_XOR | BPF_FETCH      src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
  *   BPF_XCHG                 src_reg = atomic_xchg(dst_reg + off16, src_reg)
  *   BPF_CMPXCHG              r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
+ *   BPF_LOAD_ACQ             dst_reg = smp_load_acquire(src_reg + off16)
+ *   BPF_STORE_REL            smp_store_release(dst_reg + off16, src_reg)
  */
 
 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF)			\
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index fff6cdb8d11a..c52946c27822 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -51,6 +51,9 @@
 #define BPF_XCHG	(0xe0 | BPF_FETCH)	/* atomic exchange */
 #define BPF_CMPXCHG	(0xf0 | BPF_FETCH)	/* atomic compare-and-write */
 
+#define BPF_LOAD_ACQ	0x100	/* load-acquire */
+#define BPF_STORE_REL	0x110	/* store-release */
+
 enum bpf_cond_pseudo_jmp {
 	BPF_MAY_GOTO = 0,
 };
diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index a0200fbbace9..323af18d7d49 100644
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c
@@ -1663,14 +1663,17 @@ EXPORT_SYMBOL_GPL(__bpf_call_base);
 	INSN_3(JMP, JSET, K),			\
 	INSN_2(JMP, JA),			\
 	INSN_2(JMP32, JA),			\
+	/* Atomic operations. */		\
+	INSN_3(STX, ATOMIC, B),			\
+	INSN_3(STX, ATOMIC, H),			\
+	INSN_3(STX, ATOMIC, W),			\
+	INSN_3(STX, ATOMIC, DW),		\
 	/* Store instructions. */		\
 	/*   Register based. */			\
 	INSN_3(STX, MEM,  B),			\
 	INSN_3(STX, MEM,  H),			\
 	INSN_3(STX, MEM,  W),			\
 	INSN_3(STX, MEM,  DW),			\
-	INSN_3(STX, ATOMIC, W),			\
-	INSN_3(STX, ATOMIC, DW),		\
 	/*   Immediate based. */		\
 	INSN_3(ST, MEM, B),			\
 	INSN_3(ST, MEM, H),			\
@@ -2152,24 +2155,33 @@ static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
 			if (BPF_SIZE(insn->code) == BPF_W)		\
 				atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
 					     (DST + insn->off));	\
-			else						\
+			else if (BPF_SIZE(insn->code) == BPF_DW)	\
 				atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
 					       (DST + insn->off));	\
+			else						\
+				goto default_label;			\
 			break;						\
 		case BOP | BPF_FETCH:					\
 			if (BPF_SIZE(insn->code) == BPF_W)		\
 				SRC = (u32) atomic_fetch_##KOP(		\
 					(u32) SRC,			\
 					(atomic_t *)(unsigned long) (DST + insn->off)); \
-			else						\
+			else if (BPF_SIZE(insn->code) == BPF_DW)	\
 				SRC = (u64) atomic64_fetch_##KOP(	\
 					(u64) SRC,			\
 					(atomic64_t *)(unsigned long) (DST + insn->off)); \
+			else						\
+				goto default_label;			\
 			break;
 
 	STX_ATOMIC_DW:
 	STX_ATOMIC_W:
+	STX_ATOMIC_H:
+	STX_ATOMIC_B:
 		switch (IMM) {
+		/* Atomic read-modify-write instructions support only W and DW
+		 * size modifiers.
+		 */
 		ATOMIC_ALU_OP(BPF_ADD, add)
 		ATOMIC_ALU_OP(BPF_AND, and)
 		ATOMIC_ALU_OP(BPF_OR, or)
@@ -2181,20 +2193,59 @@ static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
 				SRC = (u32) atomic_xchg(
 					(atomic_t *)(unsigned long) (DST + insn->off),
 					(u32) SRC);
-			else
+			else if (BPF_SIZE(insn->code) == BPF_DW)
 				SRC = (u64) atomic64_xchg(
 					(atomic64_t *)(unsigned long) (DST + insn->off),
 					(u64) SRC);
+			else
+				goto default_label;
 			break;
 		case BPF_CMPXCHG:
 			if (BPF_SIZE(insn->code) == BPF_W)
 				BPF_R0 = (u32) atomic_cmpxchg(
 					(atomic_t *)(unsigned long) (DST + insn->off),
 					(u32) BPF_R0, (u32) SRC);
-			else
+			else if (BPF_SIZE(insn->code) == BPF_DW)
 				BPF_R0 = (u64) atomic64_cmpxchg(
 					(atomic64_t *)(unsigned long) (DST + insn->off),
 					(u64) BPF_R0, (u64) SRC);
+			else
+				goto default_label;
+			break;
+		/* Atomic load and store instructions support all size
+		 * modifiers.
+		 */
+		case BPF_LOAD_ACQ:
+			switch (BPF_SIZE(insn->code)) {
+#define LOAD_ACQUIRE(SIZEOP, SIZE)				\
+			case BPF_##SIZEOP:			\
+				DST = (SIZE)smp_load_acquire(	\
+					(SIZE *)(unsigned long)(SRC + insn->off));	\
+				break;
+			LOAD_ACQUIRE(B,   u8)
+			LOAD_ACQUIRE(H,  u16)
+			LOAD_ACQUIRE(W,  u32)
+			LOAD_ACQUIRE(DW, u64)
+#undef LOAD_ACQUIRE
+			default:
+				goto default_label;
+			}
+			break;
+		case BPF_STORE_REL:
+			switch (BPF_SIZE(insn->code)) {
+#define STORE_RELEASE(SIZEOP, SIZE)			\
+			case BPF_##SIZEOP:		\
+				smp_store_release(	\
+					(SIZE *)(unsigned long)(DST + insn->off), (SIZE)SRC);	\
+				break;
+			STORE_RELEASE(B,   u8)
+			STORE_RELEASE(H,  u16)
+			STORE_RELEASE(W,  u32)
+			STORE_RELEASE(DW, u64)
+#undef STORE_RELEASE
+			default:
+				goto default_label;
+			}
 			break;
 
 		default:
diff --git a/kernel/bpf/disasm.c b/kernel/bpf/disasm.c
index 309c4aa1b026..974d172d6735 100644
--- a/kernel/bpf/disasm.c
+++ b/kernel/bpf/disasm.c
@@ -267,6 +267,18 @@ void print_bpf_insn(const struct bpf_insn_cbs *cbs,
 				BPF_SIZE(insn->code) == BPF_DW ? "64" : "",
 				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 				insn->dst_reg, insn->off, insn->src_reg);
+		} else if (BPF_MODE(insn->code) == BPF_ATOMIC &&
+			   insn->imm == BPF_LOAD_ACQ) {
+			verbose(cbs->private_data, "(%02x) r%d = load_acquire((%s *)(r%d %+d))\n",
+				insn->code, insn->dst_reg,
+				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
+				insn->src_reg, insn->off);
+		} else if (BPF_MODE(insn->code) == BPF_ATOMIC &&
+			   insn->imm == BPF_STORE_REL) {
+			verbose(cbs->private_data, "(%02x) store_release((%s *)(r%d %+d), r%d)\n",
+				insn->code,
+				bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
+				insn->dst_reg, insn->off, insn->src_reg);
 		} else {
 			verbose(cbs->private_data, "BUG_%02x\n", insn->code);
 		}
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 3b20bff1c5eb..89e06fce5ece 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -579,6 +579,13 @@ static bool is_cmpxchg_insn(const struct bpf_insn *insn)
 	       insn->imm == BPF_CMPXCHG;
 }
 
+static bool is_atomic_load_insn(const struct bpf_insn *insn)
+{
+	return BPF_CLASS(insn->code) == BPF_STX &&
+	       BPF_MODE(insn->code) == BPF_ATOMIC &&
+	       insn->imm == BPF_LOAD_ACQ;
+}
+
 static int __get_spi(s32 off)
 {
 	return (-off - 1) / BPF_REG_SIZE;
@@ -3550,7 +3557,7 @@ static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn,
 	}
 
 	if (class == BPF_STX) {
-		/* BPF_STX (including atomic variants) has multiple source
+		/* BPF_STX (including atomic variants) has one or more source
 		 * operands, one of which is a ptr. Check whether the caller is
 		 * asking about it.
 		 */
@@ -4164,7 +4171,7 @@ static int backtrack_insn(struct bpf_verifier_env *env, int idx, int subseq_idx,
 			   * dreg still needs precision before this insn
 			   */
 		}
-	} else if (class == BPF_LDX) {
+	} else if (class == BPF_LDX || is_atomic_load_insn(insn)) {
 		if (!bt_is_reg_set(bt, dreg))
 			return 0;
 		bt_clear_reg(bt, dreg);
@@ -7765,6 +7772,32 @@ static int check_atomic_rmw(struct bpf_verifier_env *env,
 	return 0;
 }
 
+static int check_atomic_load(struct bpf_verifier_env *env,
+			     struct bpf_insn *insn)
+{
+	if (!atomic_ptr_type_ok(env, insn->src_reg, insn)) {
+		verbose(env, "BPF_ATOMIC loads from R%d %s is not allowed\n",
+			insn->src_reg,
+			reg_type_str(env, reg_state(env, insn->src_reg)->type));
+		return -EACCES;
+	}
+
+	return check_load_mem(env, insn, true, false, false, "atomic_load");
+}
+
+static int check_atomic_store(struct bpf_verifier_env *env,
+			      struct bpf_insn *insn)
+{
+	if (!atomic_ptr_type_ok(env, insn->dst_reg, insn)) {
+		verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n",
+			insn->dst_reg,
+			reg_type_str(env, reg_state(env, insn->dst_reg)->type));
+		return -EACCES;
+	}
+
+	return check_store_reg(env, insn, true);
+}
+
 static int check_atomic(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
 	switch (insn->imm) {
@@ -7779,6 +7812,10 @@ static int check_atomic(struct bpf_verifier_env *env, struct bpf_insn *insn)
 	case BPF_XCHG:
 	case BPF_CMPXCHG:
 		return check_atomic_rmw(env, insn);
+	case BPF_LOAD_ACQ:
+		return check_atomic_load(env, insn);
+	case BPF_STORE_REL:
+		return check_atomic_store(env, insn);
 	default:
 		verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n",
 			insn->imm);
@@ -20558,7 +20595,9 @@ static int convert_ctx_accesses(struct bpf_verifier_env *env)
 			   insn->code == (BPF_ST | BPF_MEM | BPF_W) ||
 			   insn->code == (BPF_ST | BPF_MEM | BPF_DW)) {
 			type = BPF_WRITE;
-		} else if ((insn->code == (BPF_STX | BPF_ATOMIC | BPF_W) ||
+		} else if ((insn->code == (BPF_STX | BPF_ATOMIC | BPF_B) ||
+			    insn->code == (BPF_STX | BPF_ATOMIC | BPF_H) ||
+			    insn->code == (BPF_STX | BPF_ATOMIC | BPF_W) ||
 			    insn->code == (BPF_STX | BPF_ATOMIC | BPF_DW)) &&
 			   env->insn_aux_data[i + delta].ptr_type == PTR_TO_ARENA) {
 			insn->code = BPF_STX | BPF_PROBE_ATOMIC | BPF_SIZE(insn->code);
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index fff6cdb8d11a..c52946c27822 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -51,6 +51,9 @@
 #define BPF_XCHG	(0xe0 | BPF_FETCH)	/* atomic exchange */
 #define BPF_CMPXCHG	(0xf0 | BPF_FETCH)	/* atomic compare-and-write */
 
+#define BPF_LOAD_ACQ	0x100	/* load-acquire */
+#define BPF_STORE_REL	0x110	/* store-release */
+
 enum bpf_cond_pseudo_jmp {
 	BPF_MAY_GOTO = 0,
 };
-- 
2.48.1.601.g30ceb7b040-goog

Re: [PATCH bpf-next v3 4/9] bpf: Introduce load-acquire and store-release instructions

[Alexei Starovoitov]

On Wed, Feb 19, 2025 at 5:21 PM Peilin Ye <yepeilin@google.com> wrote:
>
>
> In arch/{arm64,s390,x86}/net/bpf_jit_comp.c, have
> bpf_jit_supports_insn(..., /*in_arena=*/true) return false for the new
> instructions, until the corresponding JIT compiler supports them.

...

> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> index a43fc5af973d..f0c31c940fb8 100644
> --- a/arch/x86/net/bpf_jit_comp.c
> +++ b/arch/x86/net/bpf_jit_comp.c
> @@ -3771,8 +3771,12 @@ bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
>         if (!in_arena)
>                 return true;
>         switch (insn->code) {
> +       case BPF_STX | BPF_ATOMIC | BPF_B:
> +       case BPF_STX | BPF_ATOMIC | BPF_H:
>         case BPF_STX | BPF_ATOMIC | BPF_W:
>         case BPF_STX | BPF_ATOMIC | BPF_DW:
> +               if (bpf_atomic_is_load_store(insn))
> +                       return false;
>                 if (insn->imm == (BPF_AND | BPF_FETCH) ||
>                     insn->imm == (BPF_OR | BPF_FETCH) ||
>                     insn->imm == (BPF_XOR | BPF_FETCH))

It's border line ok-ish to delay arena ld_acq/st_rel on x86 for a follow up,
but non-arena on x86 should be done in this patch.
It's trivial to add support on x86-64 and
limiting the selftest to:
.. defined(__TARGET_ARCH_arm64)
is just not right.

Even arena ld_acq/st_rel can be done in this patch too.
It is such a minor addition compared to the rest, so I don't see
a reason not to do it right away.

pw-bot: cr

Re: [PATCH bpf-next v3 4/9] bpf: Introduce load-acquire and store-release instructions

[Peilin Ye]

On Wed, Feb 26, 2025 at 06:18:47PM -0800, Alexei Starovoitov wrote:
> It's border line ok-ish to delay arena ld_acq/st_rel on x86 for a follow up,
> but non-arena on x86 should be done in this patch.
> It's trivial to add support on x86-64 and
> limiting the selftest to:
> .. defined(__TARGET_ARCH_arm64)
> is just not right.
> 
> Even arena ld_acq/st_rel can be done in this patch too.
> It is such a minor addition compared to the rest, so I don't see
> a reason not to do it right away.

Sure!  I'll add x86-64 support (including arena) in v4.

Thanks,
Peilin Ye

[PATCH bpf-next v3 5/9] arm64: insn: Add BIT(23) to {load,store}_ex's mask

[Peilin Ye]

We are planning to add load-acquire (LDAR{,B,H}) and store-release
(STLR{,B,H}) instructions to insn.{c,h}; add BIT(23) to mask of load_ex
and store_ex to prevent aarch64_insn_is_{load,store}_ex() from returning
false-positives for load-acquire and store-release instructions.

Reference: Arm Architecture Reference Manual (ARM DDI 0487K.a,
           ID032224),

  * C6.2.228 LDXR
  * C6.2.165 LDAXR
  * C6.2.161 LDAR
  * C6.2.393 STXR
  * C6.2.360 STLXR
  * C6.2.353 STLR

Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 arch/arm64/include/asm/insn.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/include/asm/insn.h b/arch/arm64/include/asm/insn.h
index e390c432f546..2d8316b3abaf 100644
--- a/arch/arm64/include/asm/insn.h
+++ b/arch/arm64/include/asm/insn.h
@@ -351,8 +351,8 @@ __AARCH64_INSN_FUNCS(ldr_imm,	0x3FC00000, 0x39400000)
 __AARCH64_INSN_FUNCS(ldr_lit,	0xBF000000, 0x18000000)
 __AARCH64_INSN_FUNCS(ldrsw_lit,	0xFF000000, 0x98000000)
 __AARCH64_INSN_FUNCS(exclusive,	0x3F800000, 0x08000000)
-__AARCH64_INSN_FUNCS(load_ex,	0x3F400000, 0x08400000)
-__AARCH64_INSN_FUNCS(store_ex,	0x3F400000, 0x08000000)
+__AARCH64_INSN_FUNCS(load_ex,	0x3FC00000, 0x08400000)
+__AARCH64_INSN_FUNCS(store_ex,	0x3FC00000, 0x08000000)
 __AARCH64_INSN_FUNCS(mops,	0x3B200C00, 0x19000400)
 __AARCH64_INSN_FUNCS(stp,	0x7FC00000, 0x29000000)
 __AARCH64_INSN_FUNCS(ldp,	0x7FC00000, 0x29400000)
-- 
2.48.1.601.g30ceb7b040-goog

Re: [PATCH bpf-next v3 5/9] arm64: insn: Add BIT(23) to {load,store}_ex's mask

[Xu Kuohai]

On 2/20/2025 9:21 AM, Peilin Ye wrote:
> We are planning to add load-acquire (LDAR{,B,H}) and store-release
> (STLR{,B,H}) instructions to insn.{c,h}; add BIT(23) to mask of load_ex
> and store_ex to prevent aarch64_insn_is_{load,store}_ex() from returning
> false-positives for load-acquire and store-release instructions.
> 
> Reference: Arm Architecture Reference Manual (ARM DDI 0487K.a,
>             ID032224),
> 
>    * C6.2.228 LDXR
>    * C6.2.165 LDAXR
>    * C6.2.161 LDAR
>    * C6.2.393 STXR
>    * C6.2.360 STLXR
>    * C6.2.353 STLR
> 
> Signed-off-by: Peilin Ye <yepeilin@google.com>
> ---
>   arch/arm64/include/asm/insn.h | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/insn.h b/arch/arm64/include/asm/insn.h
> index e390c432f546..2d8316b3abaf 100644
> --- a/arch/arm64/include/asm/insn.h
> +++ b/arch/arm64/include/asm/insn.h
> @@ -351,8 +351,8 @@ __AARCH64_INSN_FUNCS(ldr_imm,	0x3FC00000, 0x39400000)
>   __AARCH64_INSN_FUNCS(ldr_lit,	0xBF000000, 0x18000000)
>   __AARCH64_INSN_FUNCS(ldrsw_lit,	0xFF000000, 0x98000000)
>   __AARCH64_INSN_FUNCS(exclusive,	0x3F800000, 0x08000000)
> -__AARCH64_INSN_FUNCS(load_ex,	0x3F400000, 0x08400000)
> -__AARCH64_INSN_FUNCS(store_ex,	0x3F400000, 0x08000000)
> +__AARCH64_INSN_FUNCS(load_ex,	0x3FC00000, 0x08400000)
> +__AARCH64_INSN_FUNCS(store_ex,	0x3FC00000, 0x08000000)
>   __AARCH64_INSN_FUNCS(mops,	0x3B200C00, 0x19000400)
>   __AARCH64_INSN_FUNCS(stp,	0x7FC00000, 0x29000000)
>   __AARCH64_INSN_FUNCS(ldp,	0x7FC00000, 0x29400000)


Looks good to me

Acked-by: Xu Kuohai <xukuohai@huawei.com>

[PATCH bpf-next v3 6/9] arm64: insn: Add load-acquire and store-release instructions

[Peilin Ye]

Add load-acquire ("load_acq", LDAR{,B,H}) and store-release
("store_rel", STLR{,B,H}) instructions.  Breakdown of encoding:

                                size        L   (Rs)  o0 (Rt2) Rn    Rt
             mask (0x3fdffc00): 00 111111 1 1 0 11111 1  11111 00000 00000
  value, load_acq (0x08dffc00): 00 001000 1 1 0 11111 1  11111 00000 00000
 value, store_rel (0x089ffc00): 00 001000 1 0 0 11111 1  11111 00000 00000

As suggested by Xu [1], include all Should-Be-One (SBO) bits ("Rs" and
"Rt2" fields) in the "mask" and "value" numbers.

It is worth noting that we are adding the "no offset" variant of STLR
instead of the "pre-index" variant, which has a different encoding.

Reference: Arm Architecture Reference Manual (ARM DDI 0487K.a,
           ID032224),

  * C6.2.161 LDAR
  * C6.2.353 STLR

[1] https://lore.kernel.org/bpf/4e6641ce-3f1e-4251-8daf-4dd4b77d08c4@huaweicloud.com/

Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 arch/arm64/include/asm/insn.h |  8 ++++++++
 arch/arm64/lib/insn.c         | 29 +++++++++++++++++++++++++++++
 2 files changed, 37 insertions(+)

diff --git a/arch/arm64/include/asm/insn.h b/arch/arm64/include/asm/insn.h
index 2d8316b3abaf..39577f1d079a 100644
--- a/arch/arm64/include/asm/insn.h
+++ b/arch/arm64/include/asm/insn.h
@@ -188,8 +188,10 @@ enum aarch64_insn_ldst_type {
 	AARCH64_INSN_LDST_STORE_PAIR_PRE_INDEX,
 	AARCH64_INSN_LDST_LOAD_PAIR_POST_INDEX,
 	AARCH64_INSN_LDST_STORE_PAIR_POST_INDEX,
+	AARCH64_INSN_LDST_LOAD_ACQ,
 	AARCH64_INSN_LDST_LOAD_EX,
 	AARCH64_INSN_LDST_LOAD_ACQ_EX,
+	AARCH64_INSN_LDST_STORE_REL,
 	AARCH64_INSN_LDST_STORE_EX,
 	AARCH64_INSN_LDST_STORE_REL_EX,
 	AARCH64_INSN_LDST_SIGNED_LOAD_IMM_OFFSET,
@@ -351,6 +353,8 @@ __AARCH64_INSN_FUNCS(ldr_imm,	0x3FC00000, 0x39400000)
 __AARCH64_INSN_FUNCS(ldr_lit,	0xBF000000, 0x18000000)
 __AARCH64_INSN_FUNCS(ldrsw_lit,	0xFF000000, 0x98000000)
 __AARCH64_INSN_FUNCS(exclusive,	0x3F800000, 0x08000000)
+__AARCH64_INSN_FUNCS(load_acq,  0x3FDFFC00, 0x08DFFC00)
+__AARCH64_INSN_FUNCS(store_rel, 0x3FDFFC00, 0x089FFC00)
 __AARCH64_INSN_FUNCS(load_ex,	0x3FC00000, 0x08400000)
 __AARCH64_INSN_FUNCS(store_ex,	0x3FC00000, 0x08000000)
 __AARCH64_INSN_FUNCS(mops,	0x3B200C00, 0x19000400)
@@ -602,6 +606,10 @@ u32 aarch64_insn_gen_load_store_pair(enum aarch64_insn_register reg1,
 				     int offset,
 				     enum aarch64_insn_variant variant,
 				     enum aarch64_insn_ldst_type type);
+u32 aarch64_insn_gen_load_acq_store_rel(enum aarch64_insn_register reg,
+					enum aarch64_insn_register base,
+					enum aarch64_insn_size_type size,
+					enum aarch64_insn_ldst_type type);
 u32 aarch64_insn_gen_load_store_ex(enum aarch64_insn_register reg,
 				   enum aarch64_insn_register base,
 				   enum aarch64_insn_register state,
diff --git a/arch/arm64/lib/insn.c b/arch/arm64/lib/insn.c
index b008a9b46a7f..9bef696e2230 100644
--- a/arch/arm64/lib/insn.c
+++ b/arch/arm64/lib/insn.c
@@ -540,6 +540,35 @@ u32 aarch64_insn_gen_load_store_pair(enum aarch64_insn_register reg1,
 					     offset >> shift);
 }
 
+u32 aarch64_insn_gen_load_acq_store_rel(enum aarch64_insn_register reg,
+					enum aarch64_insn_register base,
+					enum aarch64_insn_size_type size,
+					enum aarch64_insn_ldst_type type)
+{
+	u32 insn;
+
+	switch (type) {
+	case AARCH64_INSN_LDST_LOAD_ACQ:
+		insn = aarch64_insn_get_load_acq_value();
+		break;
+	case AARCH64_INSN_LDST_STORE_REL:
+		insn = aarch64_insn_get_store_rel_value();
+		break;
+	default:
+		pr_err("%s: unknown load-acquire/store-release encoding %d\n",
+		       __func__, type);
+		return AARCH64_BREAK_FAULT;
+	}
+
+	insn = aarch64_insn_encode_ldst_size(size, insn);
+
+	insn = aarch64_insn_encode_register(AARCH64_INSN_REGTYPE_RT, insn,
+					    reg);
+
+	return aarch64_insn_encode_register(AARCH64_INSN_REGTYPE_RN, insn,
+					    base);
+}
+
 u32 aarch64_insn_gen_load_store_ex(enum aarch64_insn_register reg,
 				   enum aarch64_insn_register base,
 				   enum aarch64_insn_register state,
-- 
2.48.1.601.g30ceb7b040-goog

Re: [PATCH bpf-next v3 6/9] arm64: insn: Add load-acquire and store-release instructions

[Alexei Starovoitov]

On Wed, Feb 19, 2025 at 5:21 PM Peilin Ye <yepeilin@google.com> wrote:
>
> Add load-acquire ("load_acq", LDAR{,B,H}) and store-release
> ("store_rel", STLR{,B,H}) instructions.  Breakdown of encoding:
>
>                                 size        L   (Rs)  o0 (Rt2) Rn    Rt
>              mask (0x3fdffc00): 00 111111 1 1 0 11111 1  11111 00000 00000
>   value, load_acq (0x08dffc00): 00 001000 1 1 0 11111 1  11111 00000 00000
>  value, store_rel (0x089ffc00): 00 001000 1 0 0 11111 1  11111 00000 00000
>
> As suggested by Xu [1], include all Should-Be-One (SBO) bits ("Rs" and
> "Rt2" fields) in the "mask" and "value" numbers.
>
> It is worth noting that we are adding the "no offset" variant of STLR
> instead of the "pre-index" variant, which has a different encoding.
>
> Reference: Arm Architecture Reference Manual (ARM DDI 0487K.a,
>            ID032224),
>
>   * C6.2.161 LDAR
>   * C6.2.353 STLR
>
> [1] https://lore.kernel.org/bpf/4e6641ce-3f1e-4251-8daf-4dd4b77d08c4@huaweicloud.com/
>
> Signed-off-by: Peilin Ye <yepeilin@google.com>
> ---
>  arch/arm64/include/asm/insn.h |  8 ++++++++
>  arch/arm64/lib/insn.c         | 29 +++++++++++++++++++++++++++++
>  2 files changed, 37 insertions(+)
>
> diff --git a/arch/arm64/include/asm/insn.h b/arch/arm64/include/asm/insn.h
> index 2d8316b3abaf..39577f1d079a 100644
> --- a/arch/arm64/include/asm/insn.h
> +++ b/arch/arm64/include/asm/insn.h
> @@ -188,8 +188,10 @@ enum aarch64_insn_ldst_type {
>         AARCH64_INSN_LDST_STORE_PAIR_PRE_INDEX,
>         AARCH64_INSN_LDST_LOAD_PAIR_POST_INDEX,
>         AARCH64_INSN_LDST_STORE_PAIR_POST_INDEX,
> +       AARCH64_INSN_LDST_LOAD_ACQ,
>         AARCH64_INSN_LDST_LOAD_EX,
>         AARCH64_INSN_LDST_LOAD_ACQ_EX,
> +       AARCH64_INSN_LDST_STORE_REL,
>         AARCH64_INSN_LDST_STORE_EX,
>         AARCH64_INSN_LDST_STORE_REL_EX,
>         AARCH64_INSN_LDST_SIGNED_LOAD_IMM_OFFSET,

Xu, Puranjay, other arm experts,

Please help review these patches.

Thanks!

Re: [PATCH bpf-next v3 6/9] arm64: insn: Add load-acquire and store-release instructions

[Xu Kuohai]

On 2/20/2025 9:21 AM, Peilin Ye wrote:
> Add load-acquire ("load_acq", LDAR{,B,H}) and store-release
> ("store_rel", STLR{,B,H}) instructions.  Breakdown of encoding:
> 
>                                  size        L   (Rs)  o0 (Rt2) Rn    Rt
>               mask (0x3fdffc00): 00 111111 1 1 0 11111 1  11111 00000 00000
>    value, load_acq (0x08dffc00): 00 001000 1 1 0 11111 1  11111 00000 00000
>   value, store_rel (0x089ffc00): 00 001000 1 0 0 11111 1  11111 00000 00000
> 
> As suggested by Xu [1], include all Should-Be-One (SBO) bits ("Rs" and
> "Rt2" fields) in the "mask" and "value" numbers.
> 
> It is worth noting that we are adding the "no offset" variant of STLR
> instead of the "pre-index" variant, which has a different encoding.
> 
> Reference: Arm Architecture Reference Manual (ARM DDI 0487K.a,
>             ID032224),
> 
>    * C6.2.161 LDAR
>    * C6.2.353 STLR
> 
> [1] https://lore.kernel.org/bpf/4e6641ce-3f1e-4251-8daf-4dd4b77d08c4@huaweicloud.com/
> 
> Signed-off-by: Peilin Ye <yepeilin@google.com>
> ---
>   arch/arm64/include/asm/insn.h |  8 ++++++++
>   arch/arm64/lib/insn.c         | 29 +++++++++++++++++++++++++++++
>   2 files changed, 37 insertions(+)
> 
> diff --git a/arch/arm64/include/asm/insn.h b/arch/arm64/include/asm/insn.h
> index 2d8316b3abaf..39577f1d079a 100644
> --- a/arch/arm64/include/asm/insn.h
> +++ b/arch/arm64/include/asm/insn.h
> @@ -188,8 +188,10 @@ enum aarch64_insn_ldst_type {
>   	AARCH64_INSN_LDST_STORE_PAIR_PRE_INDEX,
>   	AARCH64_INSN_LDST_LOAD_PAIR_POST_INDEX,
>   	AARCH64_INSN_LDST_STORE_PAIR_POST_INDEX,
> +	AARCH64_INSN_LDST_LOAD_ACQ,
>   	AARCH64_INSN_LDST_LOAD_EX,
>   	AARCH64_INSN_LDST_LOAD_ACQ_EX,
> +	AARCH64_INSN_LDST_STORE_REL,
>   	AARCH64_INSN_LDST_STORE_EX,
>   	AARCH64_INSN_LDST_STORE_REL_EX,
>   	AARCH64_INSN_LDST_SIGNED_LOAD_IMM_OFFSET,
> @@ -351,6 +353,8 @@ __AARCH64_INSN_FUNCS(ldr_imm,	0x3FC00000, 0x39400000)
>   __AARCH64_INSN_FUNCS(ldr_lit,	0xBF000000, 0x18000000)
>   __AARCH64_INSN_FUNCS(ldrsw_lit,	0xFF000000, 0x98000000)
>   __AARCH64_INSN_FUNCS(exclusive,	0x3F800000, 0x08000000)
> +__AARCH64_INSN_FUNCS(load_acq,  0x3FDFFC00, 0x08DFFC00)
> +__AARCH64_INSN_FUNCS(store_rel, 0x3FDFFC00, 0x089FFC00)
>   __AARCH64_INSN_FUNCS(load_ex,	0x3FC00000, 0x08400000)
>   __AARCH64_INSN_FUNCS(store_ex,	0x3FC00000, 0x08000000)
>   __AARCH64_INSN_FUNCS(mops,	0x3B200C00, 0x19000400)
> @@ -602,6 +606,10 @@ u32 aarch64_insn_gen_load_store_pair(enum aarch64_insn_register reg1,
>   				     int offset,
>   				     enum aarch64_insn_variant variant,
>   				     enum aarch64_insn_ldst_type type);
> +u32 aarch64_insn_gen_load_acq_store_rel(enum aarch64_insn_register reg,
> +					enum aarch64_insn_register base,
> +					enum aarch64_insn_size_type size,
> +					enum aarch64_insn_ldst_type type);
>   u32 aarch64_insn_gen_load_store_ex(enum aarch64_insn_register reg,
>   				   enum aarch64_insn_register base,
>   				   enum aarch64_insn_register state,
> diff --git a/arch/arm64/lib/insn.c b/arch/arm64/lib/insn.c
> index b008a9b46a7f..9bef696e2230 100644
> --- a/arch/arm64/lib/insn.c
> +++ b/arch/arm64/lib/insn.c
> @@ -540,6 +540,35 @@ u32 aarch64_insn_gen_load_store_pair(enum aarch64_insn_register reg1,
>   					     offset >> shift);
>   }
>   
> +u32 aarch64_insn_gen_load_acq_store_rel(enum aarch64_insn_register reg,
> +					enum aarch64_insn_register base,
> +					enum aarch64_insn_size_type size,
> +					enum aarch64_insn_ldst_type type)
> +{
> +	u32 insn;
> +
> +	switch (type) {
> +	case AARCH64_INSN_LDST_LOAD_ACQ:
> +		insn = aarch64_insn_get_load_acq_value();
> +		break;
> +	case AARCH64_INSN_LDST_STORE_REL:
> +		insn = aarch64_insn_get_store_rel_value();
> +		break;
> +	default:
> +		pr_err("%s: unknown load-acquire/store-release encoding %d\n",
> +		       __func__, type);
> +		return AARCH64_BREAK_FAULT;
> +	}
> +
> +	insn = aarch64_insn_encode_ldst_size(size, insn);
> +
> +	insn = aarch64_insn_encode_register(AARCH64_INSN_REGTYPE_RT, insn,
> +					    reg);
> +
> +	return aarch64_insn_encode_register(AARCH64_INSN_REGTYPE_RN, insn,
> +					    base);
> +}
> +
>   u32 aarch64_insn_gen_load_store_ex(enum aarch64_insn_register reg,
>   				   enum aarch64_insn_register base,
>   				   enum aarch64_insn_register state,

Looks good to me

Acked-by: Xu Kuohai <xukuohai@huawei.com>

[PATCH bpf-next v3 7/9] bpf, arm64: Support load-acquire and store-release instructions

[Peilin Ye]

Support BPF load-acquire (BPF_LOAD_ACQ) and store-release
(BPF_STORE_REL) instructions in the arm64 JIT compiler.  For example
(assuming little-endian):

  db 10 00 00 00 01 00 00  r0 = load_acquire((u64 *)(r1 + 0x0))
  95 00 00 00 00 00 00 00  exit

  opcode (0xdb): BPF_ATOMIC | BPF_DW | BPF_STX
  imm (0x00000100): BPF_LOAD_ACQ

The JIT compiler would emit an LDAR instruction for the above, e.g.:

  ldar  x7, [x0]

Similarly, consider the following 16-bit store-release:

  cb 21 00 00 10 01 00 00  store_release((u16 *)(r1 + 0x0), w2)
  95 00 00 00 00 00 00 00  exit

  opcode (0xcb): BPF_ATOMIC | BPF_H | BPF_STX
  imm (0x00000110): BPF_STORE_REL

An STLRH instruction would be emitted, e.g.:

  stlrh  w1, [x0]

For a complete mapping:

  load-acquire     8-bit  LDARB
 (BPF_LOAD_ACQ)   16-bit  LDARH
                  32-bit  LDAR (32-bit)
                  64-bit  LDAR (64-bit)
  store-release    8-bit  STLRB
 (BPF_STORE_REL)  16-bit  STLRH
                  32-bit  STLR (32-bit)
                  64-bit  STLR (64-bit)

Arena accesses are supported.
bpf_jit_supports_insn(..., /*in_arena=*/true) always returns true for
BPF_LOAD_ACQ and BPF_STORE_REL instructions, as they don't depend on
ARM64_HAS_LSE_ATOMICS.

Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 arch/arm64/net/bpf_jit.h      | 20 ++++++++
 arch/arm64/net/bpf_jit_comp.c | 91 ++++++++++++++++++++++++++++++++---
 2 files changed, 105 insertions(+), 6 deletions(-)

diff --git a/arch/arm64/net/bpf_jit.h b/arch/arm64/net/bpf_jit.h
index b22ab2f97a30..a3b0e693a125 100644
--- a/arch/arm64/net/bpf_jit.h
+++ b/arch/arm64/net/bpf_jit.h
@@ -119,6 +119,26 @@
 	aarch64_insn_gen_load_store_ex(Rt, Rn, Rs, A64_SIZE(sf), \
 				       AARCH64_INSN_LDST_STORE_REL_EX)
 
+/* Load-acquire & store-release */
+#define A64_LDAR(Rt, Rn, size)  \
+	aarch64_insn_gen_load_acq_store_rel(Rt, Rn, AARCH64_INSN_SIZE_##size, \
+					    AARCH64_INSN_LDST_LOAD_ACQ)
+#define A64_STLR(Rt, Rn, size)  \
+	aarch64_insn_gen_load_acq_store_rel(Rt, Rn, AARCH64_INSN_SIZE_##size, \
+					    AARCH64_INSN_LDST_STORE_REL)
+
+/* Rt = [Rn] (load acquire) */
+#define A64_LDARB(Wt, Xn)	A64_LDAR(Wt, Xn, 8)
+#define A64_LDARH(Wt, Xn)	A64_LDAR(Wt, Xn, 16)
+#define A64_LDAR32(Wt, Xn)	A64_LDAR(Wt, Xn, 32)
+#define A64_LDAR64(Xt, Xn)	A64_LDAR(Xt, Xn, 64)
+
+/* [Rn] = Rt (store release) */
+#define A64_STLRB(Wt, Xn)	A64_STLR(Wt, Xn, 8)
+#define A64_STLRH(Wt, Xn)	A64_STLR(Wt, Xn, 16)
+#define A64_STLR32(Wt, Xn)	A64_STLR(Wt, Xn, 32)
+#define A64_STLR64(Xt, Xn)	A64_STLR(Xt, Xn, 64)
+
 /*
  * LSE atomics
  *
diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
index 8c3b47d9e441..25562bdb8eb5 100644
--- a/arch/arm64/net/bpf_jit_comp.c
+++ b/arch/arm64/net/bpf_jit_comp.c
@@ -647,6 +647,82 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx)
 	return 0;
 }
 
+static int emit_atomic_load_store(const struct bpf_insn *insn,
+				  struct jit_ctx *ctx)
+{
+	const s32 imm = insn->imm;
+	const s16 off = insn->off;
+	const u8 code = insn->code;
+	const bool arena = BPF_MODE(code) == BPF_PROBE_ATOMIC;
+	const u8 arena_vm_base = bpf2a64[ARENA_VM_START];
+	const u8 dst = bpf2a64[insn->dst_reg];
+	const u8 src = bpf2a64[insn->src_reg];
+	const u8 tmp = bpf2a64[TMP_REG_1];
+	u8 reg;
+
+	switch (imm) {
+	case BPF_LOAD_ACQ:
+		reg = src;
+		break;
+	case BPF_STORE_REL:
+		reg = dst;
+		break;
+	default:
+		pr_err_once("unknown atomic load/store op code %02x\n", imm);
+		return -EINVAL;
+	}
+
+	if (off) {
+		emit_a64_add_i(1, tmp, reg, tmp, off, ctx);
+		reg = tmp;
+	}
+	if (arena) {
+		emit(A64_ADD(1, tmp, reg, arena_vm_base), ctx);
+		reg = tmp;
+	}
+
+	switch (imm) {
+	case BPF_LOAD_ACQ:
+		switch (BPF_SIZE(code)) {
+		case BPF_B:
+			emit(A64_LDARB(dst, reg), ctx);
+			break;
+		case BPF_H:
+			emit(A64_LDARH(dst, reg), ctx);
+			break;
+		case BPF_W:
+			emit(A64_LDAR32(dst, reg), ctx);
+			break;
+		case BPF_DW:
+			emit(A64_LDAR64(dst, reg), ctx);
+			break;
+		}
+		break;
+	case BPF_STORE_REL:
+		switch (BPF_SIZE(code)) {
+		case BPF_B:
+			emit(A64_STLRB(src, reg), ctx);
+			break;
+		case BPF_H:
+			emit(A64_STLRH(src, reg), ctx);
+			break;
+		case BPF_W:
+			emit(A64_STLR32(src, reg), ctx);
+			break;
+		case BPF_DW:
+			emit(A64_STLR64(src, reg), ctx);
+			break;
+		}
+		break;
+	default:
+		pr_err_once("unexpected atomic load/store op code %02x\n",
+			    imm);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 #ifdef CONFIG_ARM64_LSE_ATOMICS
 static int emit_lse_atomic(const struct bpf_insn *insn, struct jit_ctx *ctx)
 {
@@ -1641,11 +1717,17 @@ static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 			return ret;
 		break;
 
+	case BPF_STX | BPF_ATOMIC | BPF_B:
+	case BPF_STX | BPF_ATOMIC | BPF_H:
 	case BPF_STX | BPF_ATOMIC | BPF_W:
 	case BPF_STX | BPF_ATOMIC | BPF_DW:
+	case BPF_STX | BPF_PROBE_ATOMIC | BPF_B:
+	case BPF_STX | BPF_PROBE_ATOMIC | BPF_H:
 	case BPF_STX | BPF_PROBE_ATOMIC | BPF_W:
 	case BPF_STX | BPF_PROBE_ATOMIC | BPF_DW:
-		if (cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
+		if (bpf_atomic_is_load_store(insn))
+			ret = emit_atomic_load_store(insn, ctx);
+		else if (cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
 			ret = emit_lse_atomic(insn, ctx);
 		else
 			ret = emit_ll_sc_atomic(insn, ctx);
@@ -2667,13 +2749,10 @@ bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
 	if (!in_arena)
 		return true;
 	switch (insn->code) {
-	case BPF_STX | BPF_ATOMIC | BPF_B:
-	case BPF_STX | BPF_ATOMIC | BPF_H:
 	case BPF_STX | BPF_ATOMIC | BPF_W:
 	case BPF_STX | BPF_ATOMIC | BPF_DW:
-		if (bpf_atomic_is_load_store(insn))
-			return false;
-		if (!cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
+		if (!bpf_atomic_is_load_store(insn) &&
+		    !cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
 			return false;
 	}
 	return true;
-- 
2.48.1.601.g30ceb7b040-goog

Re: [PATCH bpf-next v3 7/9] bpf, arm64: Support load-acquire and store-release instructions

[Xu Kuohai]

On 2/20/2025 9:21 AM, Peilin Ye wrote:
> Support BPF load-acquire (BPF_LOAD_ACQ) and store-release
> (BPF_STORE_REL) instructions in the arm64 JIT compiler.  For example
> (assuming little-endian):
> 
>    db 10 00 00 00 01 00 00  r0 = load_acquire((u64 *)(r1 + 0x0))
>    95 00 00 00 00 00 00 00  exit
> 
>    opcode (0xdb): BPF_ATOMIC | BPF_DW | BPF_STX
>    imm (0x00000100): BPF_LOAD_ACQ
> 
> The JIT compiler would emit an LDAR instruction for the above, e.g.:
> 
>    ldar  x7, [x0]
> 
> Similarly, consider the following 16-bit store-release:
> 
>    cb 21 00 00 10 01 00 00  store_release((u16 *)(r1 + 0x0), w2)
>    95 00 00 00 00 00 00 00  exit
> 
>    opcode (0xcb): BPF_ATOMIC | BPF_H | BPF_STX
>    imm (0x00000110): BPF_STORE_REL
> 
> An STLRH instruction would be emitted, e.g.:
> 
>    stlrh  w1, [x0]
> 
> For a complete mapping:
> 
>    load-acquire     8-bit  LDARB
>   (BPF_LOAD_ACQ)   16-bit  LDARH
>                    32-bit  LDAR (32-bit)
>                    64-bit  LDAR (64-bit)
>    store-release    8-bit  STLRB
>   (BPF_STORE_REL)  16-bit  STLRH
>                    32-bit  STLR (32-bit)
>                    64-bit  STLR (64-bit)
> 
> Arena accesses are supported.
> bpf_jit_supports_insn(..., /*in_arena=*/true) always returns true for
> BPF_LOAD_ACQ and BPF_STORE_REL instructions, as they don't depend on
> ARM64_HAS_LSE_ATOMICS.
> 
> Signed-off-by: Peilin Ye <yepeilin@google.com>
> ---
>   arch/arm64/net/bpf_jit.h      | 20 ++++++++
>   arch/arm64/net/bpf_jit_comp.c | 91 ++++++++++++++++++++++++++++++++---
>   2 files changed, 105 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/arm64/net/bpf_jit.h b/arch/arm64/net/bpf_jit.h
> index b22ab2f97a30..a3b0e693a125 100644
> --- a/arch/arm64/net/bpf_jit.h
> +++ b/arch/arm64/net/bpf_jit.h
> @@ -119,6 +119,26 @@
>   	aarch64_insn_gen_load_store_ex(Rt, Rn, Rs, A64_SIZE(sf), \
>   				       AARCH64_INSN_LDST_STORE_REL_EX)
>   
> +/* Load-acquire & store-release */
> +#define A64_LDAR(Rt, Rn, size)  \
> +	aarch64_insn_gen_load_acq_store_rel(Rt, Rn, AARCH64_INSN_SIZE_##size, \
> +					    AARCH64_INSN_LDST_LOAD_ACQ)
> +#define A64_STLR(Rt, Rn, size)  \
> +	aarch64_insn_gen_load_acq_store_rel(Rt, Rn, AARCH64_INSN_SIZE_##size, \
> +					    AARCH64_INSN_LDST_STORE_REL)
> +
> +/* Rt = [Rn] (load acquire) */
> +#define A64_LDARB(Wt, Xn)	A64_LDAR(Wt, Xn, 8)
> +#define A64_LDARH(Wt, Xn)	A64_LDAR(Wt, Xn, 16)
> +#define A64_LDAR32(Wt, Xn)	A64_LDAR(Wt, Xn, 32)
> +#define A64_LDAR64(Xt, Xn)	A64_LDAR(Xt, Xn, 64)
> +
> +/* [Rn] = Rt (store release) */
> +#define A64_STLRB(Wt, Xn)	A64_STLR(Wt, Xn, 8)
> +#define A64_STLRH(Wt, Xn)	A64_STLR(Wt, Xn, 16)
> +#define A64_STLR32(Wt, Xn)	A64_STLR(Wt, Xn, 32)
> +#define A64_STLR64(Xt, Xn)	A64_STLR(Xt, Xn, 64)
> +
>   /*
>    * LSE atomics
>    *
> diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
> index 8c3b47d9e441..25562bdb8eb5 100644
> --- a/arch/arm64/net/bpf_jit_comp.c
> +++ b/arch/arm64/net/bpf_jit_comp.c
> @@ -647,6 +647,82 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx)
>   	return 0;
>   }
>   
> +static int emit_atomic_load_store(const struct bpf_insn *insn,
> +				  struct jit_ctx *ctx)
> +{
> +	const s32 imm = insn->imm;
> +	const s16 off = insn->off;
> +	const u8 code = insn->code;
> +	const bool arena = BPF_MODE(code) == BPF_PROBE_ATOMIC;
> +	const u8 arena_vm_base = bpf2a64[ARENA_VM_START];
> +	const u8 dst = bpf2a64[insn->dst_reg];
> +	const u8 src = bpf2a64[insn->src_reg];
> +	const u8 tmp = bpf2a64[TMP_REG_1];
> +	u8 reg;
> +
> +	switch (imm) {
> +	case BPF_LOAD_ACQ:
> +		reg = src;
> +		break;
> +	case BPF_STORE_REL:
> +		reg = dst;
> +		break;
> +	default:
> +		pr_err_once("unknown atomic load/store op code %02x\n", imm);
> +		return -EINVAL;
> +	}
> +
> +	if (off) {
> +		emit_a64_add_i(1, tmp, reg, tmp, off, ctx);
> +		reg = tmp;
> +	}
> +	if (arena) {
> +		emit(A64_ADD(1, tmp, reg, arena_vm_base), ctx);
> +		reg = tmp;
> +	}
> +
> +	switch (imm) {
> +	case BPF_LOAD_ACQ:
> +		switch (BPF_SIZE(code)) {
> +		case BPF_B:
> +			emit(A64_LDARB(dst, reg), ctx);
> +			break;
> +		case BPF_H:
> +			emit(A64_LDARH(dst, reg), ctx);
> +			break;
> +		case BPF_W:
> +			emit(A64_LDAR32(dst, reg), ctx);
> +			break;
> +		case BPF_DW:
> +			emit(A64_LDAR64(dst, reg), ctx);
> +			break;
> +		}
> +		break;
> +	case BPF_STORE_REL:
> +		switch (BPF_SIZE(code)) {
> +		case BPF_B:
> +			emit(A64_STLRB(src, reg), ctx);
> +			break;
> +		case BPF_H:
> +			emit(A64_STLRH(src, reg), ctx);
> +			break;
> +		case BPF_W:
> +			emit(A64_STLR32(src, reg), ctx);
> +			break;
> +		case BPF_DW:
> +			emit(A64_STLR64(src, reg), ctx);
> +			break;
> +		}
> +		break;
> +	default:
> +		pr_err_once("unexpected atomic load/store op code %02x\n",
> +			    imm);
> +		return -EINVAL;
> +	}
> +
> +	return 0;
> +}
> +
>   #ifdef CONFIG_ARM64_LSE_ATOMICS
>   static int emit_lse_atomic(const struct bpf_insn *insn, struct jit_ctx *ctx)
>   {
> @@ -1641,11 +1717,17 @@ static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
>   			return ret;
>   		break;
>   
> +	case BPF_STX | BPF_ATOMIC | BPF_B:
> +	case BPF_STX | BPF_ATOMIC | BPF_H:
>   	case BPF_STX | BPF_ATOMIC | BPF_W:
>   	case BPF_STX | BPF_ATOMIC | BPF_DW:
> +	case BPF_STX | BPF_PROBE_ATOMIC | BPF_B:
> +	case BPF_STX | BPF_PROBE_ATOMIC | BPF_H:
>   	case BPF_STX | BPF_PROBE_ATOMIC | BPF_W:
>   	case BPF_STX | BPF_PROBE_ATOMIC | BPF_DW:
> -		if (cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
> +		if (bpf_atomic_is_load_store(insn))
> +			ret = emit_atomic_load_store(insn, ctx);
> +		else if (cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
>   			ret = emit_lse_atomic(insn, ctx);
>   		else
>   			ret = emit_ll_sc_atomic(insn, ctx);
> @@ -2667,13 +2749,10 @@ bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
>   	if (!in_arena)
>   		return true;
>   	switch (insn->code) {
> -	case BPF_STX | BPF_ATOMIC | BPF_B:
> -	case BPF_STX | BPF_ATOMIC | BPF_H:
>   	case BPF_STX | BPF_ATOMIC | BPF_W:
>   	case BPF_STX | BPF_ATOMIC | BPF_DW:
> -		if (bpf_atomic_is_load_store(insn))
> -			return false;
> -		if (!cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
> +		if (!bpf_atomic_is_load_store(insn) &&
> +		    !cpus_have_cap(ARM64_HAS_LSE_ATOMICS))
>   			return false;
>   	}
>   	return true;


Acked-by: Xu Kuohai <xukuohai@huawei.com>

[PATCH bpf-next v3 8/9] selftests/bpf: Add selftests for load-acquire and store-release instructions

[Peilin Ye]

Add several ./test_progs tests:

  - arena_atomics/load_acquire
  - arena_atomics/store_release
  - verifier_load_acquire/*
  - verifier_store_release/*
  - verifier_precision/bpf_load_acquire
  - verifier_precision/bpf_store_release

The last two tests are added to check if backtrack_insn() handles the
new instructions correctly.

Additionally, the last test also makes sure that the verifier
"remembers" the value (in src_reg) we store-release into e.g. a stack
slot.  For example, if we take a look at the test program:

    #0:  r1 = 8;
      /* store_release((u64 *)(r10 - 8), r1); */
    #1:  .8byte %[store_release];
    #2:  r1 = *(u64 *)(r10 - 8);
    #3:  r2 = r10;
    #4:  r2 += r1;
    #5:  r0 = 0;
    #6:  exit;

At #1, if the verifier doesn't remember that we wrote 8 to the stack,
then later at #4 we would be adding an unbounded scalar value to the
stack pointer, which would cause the program to be rejected:

  VERIFIER LOG:
  =============
...
  math between fp pointer and register with unbounded min value is not allowed

For easier CI integration, instead of using built-ins like
__atomic_{load,store}_n() which depend on the new
__BPF_FEATURE_LOAD_ACQ_STORE_REL pre-defined macro, manually craft
load-acquire/store-release instructions using __imm_insn(), as suggested
by Eduard.

All new tests depend on:

  (1) Clang major version >= 18, and
  (2) ENABLE_ATOMICS_TESTS is defined (currently implies -mcpu=v3 or
      v4), and
  (3) JIT supports load-acquire/store-release (currently only arm64)

In .../progs/arena_atomics.c:

  /* 8-byte-aligned */
  __u8 __arena_global load_acquire8_value = 0x12;
  /* 1-byte hole */
  __u16 __arena_global load_acquire16_value = 0x1234;

That 1-byte hole in the .addr_space.1 ELF section caused clang-17 to
crash:

  fatal error: error in backend: unable to write nop sequence of 1 bytes

To work around such llvm-17 CI job failures, conditionally define
__arena_global variables as 64-bit if __clang_major__ < 18, to make sure
.addr_space.1 has no holes.  Ideally we should avoid compiling this file
using clang-17 at all (arena tests depend on
__BPF_FEATURE_ADDR_SPACE_CAST, and are skipped for llvm-17 anyway), but
that is a separate topic.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 .../selftests/bpf/prog_tests/arena_atomics.c  |  66 ++++-
 .../selftests/bpf/prog_tests/verifier.c       |   4 +
 .../selftests/bpf/progs/arena_atomics.c       | 118 +++++++-
 .../bpf/progs/verifier_load_acquire.c         | 197 +++++++++++++
 .../selftests/bpf/progs/verifier_precision.c  |  48 ++++
 .../bpf/progs/verifier_store_release.c        | 264 ++++++++++++++++++
 6 files changed, 694 insertions(+), 3 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/verifier_load_acquire.c
 create mode 100644 tools/testing/selftests/bpf/progs/verifier_store_release.c

diff --git a/tools/testing/selftests/bpf/prog_tests/arena_atomics.c b/tools/testing/selftests/bpf/prog_tests/arena_atomics.c
index 26e7c06c6cb4..d98577a6babc 100644
--- a/tools/testing/selftests/bpf/prog_tests/arena_atomics.c
+++ b/tools/testing/selftests/bpf/prog_tests/arena_atomics.c
@@ -162,6 +162,66 @@ static void test_uaf(struct arena_atomics *skel)
 	ASSERT_EQ(skel->arena->uaf_recovery_fails, 0, "uaf_recovery_fails");
 }
 
+static void test_load_acquire(struct arena_atomics *skel)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, topts);
+	int err, prog_fd;
+
+	if (skel->data->skip_lacq_srel_tests) {
+		printf("%s:SKIP: ENABLE_ATOMICS_TESTS not defined, Clang doesn't support addr_space_cast, and/or JIT doesn't support load-acquire\n",
+		       __func__);
+		test__skip();
+		return;
+	}
+
+	/* No need to attach it, just run it directly */
+	prog_fd = bpf_program__fd(skel->progs.load_acquire);
+	err = bpf_prog_test_run_opts(prog_fd, &topts);
+	if (!ASSERT_OK(err, "test_run_opts err"))
+		return;
+	if (!ASSERT_OK(topts.retval, "test_run_opts retval"))
+		return;
+
+	ASSERT_EQ(skel->arena->load_acquire8_result, 0x12,
+		  "load_acquire8_result");
+	ASSERT_EQ(skel->arena->load_acquire16_result, 0x1234,
+		  "load_acquire16_result");
+	ASSERT_EQ(skel->arena->load_acquire32_result, 0x12345678,
+		  "load_acquire32_result");
+	ASSERT_EQ(skel->arena->load_acquire64_result, 0x1234567890abcdef,
+		  "load_acquire64_result");
+}
+
+static void test_store_release(struct arena_atomics *skel)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, topts);
+	int err, prog_fd;
+
+	if (skel->data->skip_lacq_srel_tests) {
+		printf("%s:SKIP: ENABLE_ATOMICS_TESTS not defined, Clang doesn't support addr_space_cast, and/or JIT doesn't support store-release\n",
+		       __func__);
+		test__skip();
+		return;
+	}
+
+	/* No need to attach it, just run it directly */
+	prog_fd = bpf_program__fd(skel->progs.store_release);
+	err = bpf_prog_test_run_opts(prog_fd, &topts);
+	if (!ASSERT_OK(err, "test_run_opts err"))
+		return;
+	if (!ASSERT_OK(topts.retval, "test_run_opts retval"))
+		return;
+
+	ASSERT_EQ(skel->arena->store_release8_result, 0x12,
+		  "store_release8_result");
+	ASSERT_EQ(skel->arena->store_release16_result, 0x1234,
+		  "store_release16_result");
+	ASSERT_EQ(skel->arena->store_release32_result, 0x12345678,
+		  "store_release32_result");
+	ASSERT_EQ(skel->arena->store_release64_result, 0x1234567890abcdef,
+		  "store_release64_result");
+}
+
 void test_arena_atomics(void)
 {
 	struct arena_atomics *skel;
@@ -171,7 +231,7 @@ void test_arena_atomics(void)
 	if (!ASSERT_OK_PTR(skel, "arena atomics skeleton open"))
 		return;
 
-	if (skel->data->skip_tests) {
+	if (skel->data->skip_all_tests) {
 		printf("%s:SKIP:no ENABLE_ATOMICS_TESTS or no addr_space_cast support in clang",
 		       __func__);
 		test__skip();
@@ -198,6 +258,10 @@ void test_arena_atomics(void)
 		test_xchg(skel);
 	if (test__start_subtest("uaf"))
 		test_uaf(skel);
+	if (test__start_subtest("load_acquire"))
+		test_load_acquire(skel);
+	if (test__start_subtest("store_release"))
+		test_store_release(skel);
 
 cleanup:
 	arena_atomics__destroy(skel);
diff --git a/tools/testing/selftests/bpf/prog_tests/verifier.c b/tools/testing/selftests/bpf/prog_tests/verifier.c
index 8a0e1ff8a2dc..cfe47b529e01 100644
--- a/tools/testing/selftests/bpf/prog_tests/verifier.c
+++ b/tools/testing/selftests/bpf/prog_tests/verifier.c
@@ -45,6 +45,7 @@
 #include "verifier_ldsx.skel.h"
 #include "verifier_leak_ptr.skel.h"
 #include "verifier_linked_scalars.skel.h"
+#include "verifier_load_acquire.skel.h"
 #include "verifier_loops1.skel.h"
 #include "verifier_lwt.skel.h"
 #include "verifier_map_in_map.skel.h"
@@ -80,6 +81,7 @@
 #include "verifier_spill_fill.skel.h"
 #include "verifier_spin_lock.skel.h"
 #include "verifier_stack_ptr.skel.h"
+#include "verifier_store_release.skel.h"
 #include "verifier_subprog_precision.skel.h"
 #include "verifier_subreg.skel.h"
 #include "verifier_tailcall_jit.skel.h"
@@ -173,6 +175,7 @@ void test_verifier_int_ptr(void)              { RUN(verifier_int_ptr); }
 void test_verifier_iterating_callbacks(void)  { RUN(verifier_iterating_callbacks); }
 void test_verifier_jeq_infer_not_null(void)   { RUN(verifier_jeq_infer_not_null); }
 void test_verifier_jit_convergence(void)      { RUN(verifier_jit_convergence); }
+void test_verifier_load_acquire(void)         { RUN(verifier_load_acquire); }
 void test_verifier_ld_ind(void)               { RUN(verifier_ld_ind); }
 void test_verifier_ldsx(void)                  { RUN(verifier_ldsx); }
 void test_verifier_leak_ptr(void)             { RUN(verifier_leak_ptr); }
@@ -211,6 +214,7 @@ void test_verifier_sockmap_mutate(void)       { RUN(verifier_sockmap_mutate); }
 void test_verifier_spill_fill(void)           { RUN(verifier_spill_fill); }
 void test_verifier_spin_lock(void)            { RUN(verifier_spin_lock); }
 void test_verifier_stack_ptr(void)            { RUN(verifier_stack_ptr); }
+void test_verifier_store_release(void)        { RUN(verifier_store_release); }
 void test_verifier_subprog_precision(void)    { RUN(verifier_subprog_precision); }
 void test_verifier_subreg(void)               { RUN(verifier_subreg); }
 void test_verifier_tailcall_jit(void)         { RUN(verifier_tailcall_jit); }
diff --git a/tools/testing/selftests/bpf/progs/arena_atomics.c b/tools/testing/selftests/bpf/progs/arena_atomics.c
index 40dd57fca5cc..12de414156e1 100644
--- a/tools/testing/selftests/bpf/progs/arena_atomics.c
+++ b/tools/testing/selftests/bpf/progs/arena_atomics.c
@@ -6,6 +6,8 @@
 #include <stdbool.h>
 #include <stdatomic.h>
 #include "bpf_arena_common.h"
+#include "../../../include/linux/filter.h"
+#include "bpf_misc.h"
 
 struct {
 	__uint(type, BPF_MAP_TYPE_ARENA);
@@ -19,9 +21,16 @@ struct {
 } arena SEC(".maps");
 
 #if defined(ENABLE_ATOMICS_TESTS) && defined(__BPF_FEATURE_ADDR_SPACE_CAST)
-bool skip_tests __attribute((__section__(".data"))) = false;
+bool skip_all_tests __attribute((__section__(".data"))) = false;
 #else
-bool skip_tests = true;
+bool skip_all_tests = true;
+#endif
+
+#if defined(ENABLE_ATOMICS_TESTS) && \
+	defined(__BPF_FEATURE_ADDR_SPACE_CAST) && defined(__TARGET_ARCH_arm64)
+bool skip_lacq_srel_tests __attribute((__section__(".data"))) = false;
+#else
+bool skip_lacq_srel_tests = true;
 #endif
 
 __u32 pid = 0;
@@ -274,4 +283,109 @@ int uaf(const void *ctx)
 	return 0;
 }
 
+#if __clang_major__ >= 18
+__u8 __arena_global load_acquire8_value = 0x12;
+__u16 __arena_global load_acquire16_value = 0x1234;
+__u32 __arena_global load_acquire32_value = 0x12345678;
+__u64 __arena_global load_acquire64_value = 0x1234567890abcdef;
+
+__u8 __arena_global load_acquire8_result = 0;
+__u16 __arena_global load_acquire16_result = 0;
+__u32 __arena_global load_acquire32_result = 0;
+__u64 __arena_global load_acquire64_result = 0;
+#else
+/* clang-17 crashes if the .addr_space.1 ELF section has holes. Work around
+ * this issue by defining the below variables as 64-bit.
+ */
+__u64 __arena_global load_acquire8_value;
+__u64 __arena_global load_acquire16_value;
+__u64 __arena_global load_acquire32_value;
+__u64 __arena_global load_acquire64_value;
+
+__u64 __arena_global load_acquire8_result;
+__u64 __arena_global load_acquire16_result;
+__u64 __arena_global load_acquire32_result;
+__u64 __arena_global load_acquire64_result;
+#endif
+
+SEC("raw_tp/sys_enter")
+int load_acquire(const void *ctx)
+{
+#if defined(ENABLE_ATOMICS_TESTS) && \
+	defined(__BPF_FEATURE_ADDR_SPACE_CAST) && defined(__TARGET_ARCH_arm64)
+
+#define LOAD_ACQUIRE_ARENA(SIZEOP, SIZE, SRC, DST)	\
+	{ asm volatile (				\
+	"r1 = %[" #SRC "] ll;"				\
+	"r1 = addr_space_cast(r1, 0x0, 0x1);"		\
+	".8byte %[load_acquire_insn];"			\
+	"r3 = %[" #DST "] ll;"				\
+	"r3 = addr_space_cast(r3, 0x0, 0x1);"		\
+	"*(" #SIZE " *)(r3 + 0) = r2;"			\
+	:						\
+	: __imm_addr(SRC),				\
+	  __imm_insn(load_acquire_insn,			\
+		     BPF_ATOMIC_OP(BPF_##SIZEOP, BPF_LOAD_ACQ,	\
+				   BPF_REG_2, BPF_REG_1, 0)),	\
+	  __imm_addr(DST)				\
+	: __clobber_all); }				\
+
+	LOAD_ACQUIRE_ARENA(B, u8, load_acquire8_value, load_acquire8_result)
+	LOAD_ACQUIRE_ARENA(H, u16, load_acquire16_value,
+			   load_acquire16_result)
+	LOAD_ACQUIRE_ARENA(W, u32, load_acquire32_value,
+			   load_acquire32_result)
+	LOAD_ACQUIRE_ARENA(DW, u64, load_acquire64_value,
+			   load_acquire64_result)
+#undef LOAD_ACQUIRE_ARENA
+
+#endif
+	return 0;
+}
+
+#if __clang_major__ >= 18
+__u8 __arena_global store_release8_result = 0;
+__u16 __arena_global store_release16_result = 0;
+__u32 __arena_global store_release32_result = 0;
+__u64 __arena_global store_release64_result = 0;
+#else
+/* clang-17 crashes if the .addr_space.1 ELF section has holes. Work around
+ * this issue by defining the below variables as 64-bit.
+ */
+__u64 __arena_global store_release8_result;
+__u64 __arena_global store_release16_result;
+__u64 __arena_global store_release32_result;
+__u64 __arena_global store_release64_result;
+#endif
+
+SEC("raw_tp/sys_enter")
+int store_release(const void *ctx)
+{
+#if defined(ENABLE_ATOMICS_TESTS) && \
+	defined(__BPF_FEATURE_ADDR_SPACE_CAST) && defined(__TARGET_ARCH_arm64)
+
+#define STORE_RELEASE_ARENA(SIZEOP, DST, VAL)	\
+	{ asm volatile (			\
+	"r1 = " VAL ";"				\
+	"r2 = %[" #DST "] ll;"			\
+	"r2 = addr_space_cast(r2, 0x0, 0x1);"	\
+	".8byte %[store_release_insn];"		\
+	:					\
+	: __imm_addr(DST),			\
+	  __imm_insn(store_release_insn,	\
+		     BPF_ATOMIC_OP(BPF_##SIZEOP, BPF_STORE_REL,	\
+				   BPF_REG_2, BPF_REG_1, 0))	\
+	: __clobber_all); }			\
+
+	STORE_RELEASE_ARENA(B, store_release8_result, "0x12")
+	STORE_RELEASE_ARENA(H, store_release16_result, "0x1234")
+	STORE_RELEASE_ARENA(W, store_release32_result, "0x12345678")
+	STORE_RELEASE_ARENA(DW, store_release64_result,
+			    "0x1234567890abcdef ll")
+#undef STORE_RELEASE_ARENA
+
+#endif
+	return 0;
+}
+
 char _license[] SEC("license") = "GPL";
diff --git a/tools/testing/selftests/bpf/progs/verifier_load_acquire.c b/tools/testing/selftests/bpf/progs/verifier_load_acquire.c
new file mode 100644
index 000000000000..529756501f10
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/verifier_load_acquire.c
@@ -0,0 +1,197 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2025 Google LLC. */
+
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include "../../../include/linux/filter.h"
+#include "bpf_misc.h"
+
+#if __clang_major__ >= 18 && defined(ENABLE_ATOMICS_TESTS) && \
+	defined(__TARGET_ARCH_arm64)
+
+SEC("socket")
+__description("load-acquire, 8-bit")
+__success __success_unpriv __retval(0x12)
+__naked void load_acquire_8(void)
+{
+	asm volatile (
+	"w1 = 0x12;"
+	"*(u8 *)(r10 - 1) = w1;"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u8 *)(r10 - 1));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_10, -1))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("load-acquire, 16-bit")
+__success __success_unpriv __retval(0x1234)
+__naked void load_acquire_16(void)
+{
+	asm volatile (
+	"w1 = 0x1234;"
+	"*(u16 *)(r10 - 2) = w1;"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u16 *)(r10 - 2));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_H, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_10, -2))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("load-acquire, 32-bit")
+__success __success_unpriv __retval(0x12345678)
+__naked void load_acquire_32(void)
+{
+	asm volatile (
+	"w1 = 0x12345678;"
+	"*(u32 *)(r10 - 4) = w1;"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u32 *)(r10 - 4));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_W, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_10, -4))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("load-acquire, 64-bit")
+__success __success_unpriv __retval(0x1234567890abcdef)
+__naked void load_acquire_64(void)
+{
+	asm volatile (
+	"r1 = 0x1234567890abcdef ll;"
+	"*(u64 *)(r10 - 8) = r1;"
+	".8byte %[load_acquire_insn];" // r0 = load_acquire((u64 *)(r10 - 8));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_10, -8))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("load-acquire with uninitialized src_reg")
+__failure __failure_unpriv __msg("R2 !read_ok")
+__naked void load_acquire_with_uninitialized_src_reg(void)
+{
+	asm volatile (
+	".8byte %[load_acquire_insn];" // r0 = load_acquire((u64 *)(r2 + 0));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_2, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("load-acquire with non-pointer src_reg")
+__failure __failure_unpriv __msg("R1 invalid mem access 'scalar'")
+__naked void load_acquire_with_non_pointer_src_reg(void)
+{
+	asm volatile (
+	"r1 = 0;"
+	".8byte %[load_acquire_insn];" // r0 = load_acquire((u64 *)(r1 + 0));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_1, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("misaligned load-acquire")
+__failure __failure_unpriv __msg("misaligned stack access off")
+__flag(BPF_F_ANY_ALIGNMENT)
+__naked void load_acquire_misaligned(void)
+{
+	asm volatile (
+	"r1 = 0;"
+	"*(u64 *)(r10 - 8) = r1;"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u32 *)(r10 - 5));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_W, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_10, -5))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("load-acquire from ctx pointer")
+__failure __failure_unpriv __msg("BPF_ATOMIC loads from R1 ctx is not allowed")
+__naked void load_acquire_from_ctx_pointer(void)
+{
+	asm volatile (
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u8 *)(r1 + 0));
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_1, 0))
+	: __clobber_all);
+}
+
+SEC("xdp")
+__description("load-acquire from pkt pointer")
+__failure __msg("BPF_ATOMIC loads from R2 pkt is not allowed")
+__naked void load_acquire_from_pkt_pointer(void)
+{
+	asm volatile (
+	"r2 = *(u32 *)(r1 + %[xdp_md_data]);"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u8 *)(r2 + 0));
+	"exit;"
+	:
+	: __imm_const(xdp_md_data, offsetof(struct xdp_md, data)),
+	  __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_2, 0))
+	: __clobber_all);
+}
+
+SEC("flow_dissector")
+__description("load-acquire from flow_keys pointer")
+__failure __msg("BPF_ATOMIC loads from R2 flow_keys is not allowed")
+__naked void load_acquire_from_flow_keys_pointer(void)
+{
+	asm volatile (
+	"r2 = *(u64 *)(r1 + %[__sk_buff_flow_keys]);"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u8 *)(r2 + 0));
+	"exit;"
+	:
+	: __imm_const(__sk_buff_flow_keys,
+		      offsetof(struct __sk_buff, flow_keys)),
+	  __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_2, 0))
+	: __clobber_all);
+}
+
+SEC("sk_reuseport")
+__description("load-acquire from sock pointer")
+__failure __msg("BPF_ATOMIC loads from R2 sock is not allowed")
+__naked void load_acquire_from_sock_pointer(void)
+{
+	asm volatile (
+	"r2 = *(u64 *)(r1 + %[sk_reuseport_md_sk]);"
+	".8byte %[load_acquire_insn];" // w0 = load_acquire((u8 *)(r2 + 0));
+	"exit;"
+	:
+	: __imm_const(sk_reuseport_md_sk, offsetof(struct sk_reuseport_md, sk)),
+	  __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_LOAD_ACQ, BPF_REG_0, BPF_REG_2, 0))
+	: __clobber_all);
+}
+
+#else
+
+SEC("socket")
+__description("Clang version < 18, ENABLE_ATOMICS_TESTS not defined, and/or JIT doesn't support load-acquire, use a dummy test")
+__success
+int dummy_test(void)
+{
+	return 0;
+}
+
+#endif
+
+char _license[] SEC("license") = "GPL";
diff --git a/tools/testing/selftests/bpf/progs/verifier_precision.c b/tools/testing/selftests/bpf/progs/verifier_precision.c
index 6b564d4c0986..793fac62d63b 100644
--- a/tools/testing/selftests/bpf/progs/verifier_precision.c
+++ b/tools/testing/selftests/bpf/progs/verifier_precision.c
@@ -2,6 +2,7 @@
 /* Copyright (C) 2023 SUSE LLC */
 #include <linux/bpf.h>
 #include <bpf/bpf_helpers.h>
+#include "../../../include/linux/filter.h"
 #include "bpf_misc.h"
 
 SEC("?raw_tp")
@@ -90,6 +91,53 @@ __naked int bpf_end_bswap(void)
 		::: __clobber_all);
 }
 
+#if defined(ENABLE_ATOMICS_TESTS) && defined(__TARGET_ARCH_arm64)
+
+SEC("?raw_tp")
+__success __log_level(2)
+__msg("mark_precise: frame0: regs=r2 stack= before 3: (bf) r3 = r10")
+__msg("mark_precise: frame0: regs=r2 stack= before 2: (db) r2 = load_acquire((u64 *)(r10 -8))")
+__msg("mark_precise: frame0: regs= stack=-8 before 1: (7b) *(u64 *)(r10 -8) = r1")
+__msg("mark_precise: frame0: regs=r1 stack= before 0: (b7) r1 = 8")
+__naked int bpf_load_acquire(void)
+{
+	asm volatile (
+	"r1 = 8;"
+	"*(u64 *)(r10 - 8) = r1;"
+	".8byte %[load_acquire_insn];" /* r2 = load_acquire((u64 *)(r10 - 8)); */
+	"r3 = r10;"
+	"r3 += r2;" /* mark_precise */
+	"r0 = 0;"
+	"exit;"
+	:
+	: __imm_insn(load_acquire_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_LOAD_ACQ, BPF_REG_2, BPF_REG_10, -8))
+	: __clobber_all);
+}
+
+SEC("?raw_tp")
+__success __log_level(2)
+__msg("mark_precise: frame0: regs=r1 stack= before 3: (bf) r2 = r10")
+__msg("mark_precise: frame0: regs=r1 stack= before 2: (79) r1 = *(u64 *)(r10 -8)")
+__msg("mark_precise: frame0: regs= stack=-8 before 1: (db) store_release((u64 *)(r10 -8), r1)")
+__msg("mark_precise: frame0: regs=r1 stack= before 0: (b7) r1 = 8")
+__naked int bpf_store_release(void)
+{
+	asm volatile (
+	"r1 = 8;"
+	".8byte %[store_release_insn];" /* store_release((u64 *)(r10 - 8), r1); */
+	"r1 = *(u64 *)(r10 - 8);"
+	"r2 = r10;"
+	"r2 += r1;" /* mark_precise */
+	"r0 = 0;"
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_10, BPF_REG_1, -8))
+	: __clobber_all);
+}
+
+#endif /* load-acquire, store-release */
 #endif /* v4 instruction */
 
 SEC("?raw_tp")
diff --git a/tools/testing/selftests/bpf/progs/verifier_store_release.c b/tools/testing/selftests/bpf/progs/verifier_store_release.c
new file mode 100644
index 000000000000..fd0aded3479e
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/verifier_store_release.c
@@ -0,0 +1,264 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2025 Google LLC. */
+
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include "../../../include/linux/filter.h"
+#include "bpf_misc.h"
+
+#if __clang_major__ >= 18 && defined(ENABLE_ATOMICS_TESTS) && \
+	defined(__TARGET_ARCH_arm64)
+
+SEC("socket")
+__description("store-release, 8-bit")
+__success __success_unpriv __retval(0x12)
+__naked void store_release_8(void)
+{
+	asm volatile (
+	"w1 = 0x12;"
+	".8byte %[store_release_insn];" // store_release((u8 *)(r10 - 1), w1);
+	"w0 = *(u8 *)(r10 - 1);"
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_STORE_REL, BPF_REG_10, BPF_REG_1, -1))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release, 16-bit")
+__success __success_unpriv __retval(0x1234)
+__naked void store_release_16(void)
+{
+	asm volatile (
+	"w1 = 0x1234;"
+	".8byte %[store_release_insn];" // store_release((u16 *)(r10 - 2), w1);
+	"w0 = *(u16 *)(r10 - 2);"
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_H, BPF_STORE_REL, BPF_REG_10, BPF_REG_1, -2))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release, 32-bit")
+__success __success_unpriv __retval(0x12345678)
+__naked void store_release_32(void)
+{
+	asm volatile (
+	"w1 = 0x12345678;"
+	".8byte %[store_release_insn];" // store_release((u32 *)(r10 - 4), w1);
+	"w0 = *(u32 *)(r10 - 4);"
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_W, BPF_STORE_REL, BPF_REG_10, BPF_REG_1, -4))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release, 64-bit")
+__success __success_unpriv __retval(0x1234567890abcdef)
+__naked void store_release_64(void)
+{
+	asm volatile (
+	"r1 = 0x1234567890abcdef ll;"
+	".8byte %[store_release_insn];" // store_release((u64 *)(r10 - 8), r1);
+	"r0 = *(u64 *)(r10 - 8);"
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_10, BPF_REG_1, -8))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release with uninitialized src_reg")
+__failure __failure_unpriv __msg("R2 !read_ok")
+__naked void store_release_with_uninitialized_src_reg(void)
+{
+	asm volatile (
+	".8byte %[store_release_insn];" // store_release((u64 *)(r10 - 8), r2);
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_10, BPF_REG_2, -8))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release with uninitialized dst_reg")
+__failure __failure_unpriv __msg("R2 !read_ok")
+__naked void store_release_with_uninitialized_dst_reg(void)
+{
+	asm volatile (
+	"r1 = 0;"
+	".8byte %[store_release_insn];" // store_release((u64 *)(r2 - 8), r1);
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_2, BPF_REG_1, -8))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release with non-pointer dst_reg")
+__failure __failure_unpriv __msg("R1 invalid mem access 'scalar'")
+__naked void store_release_with_non_pointer_dst_reg(void)
+{
+	asm volatile (
+	"r1 = 0;"
+	".8byte %[store_release_insn];" // store_release((u64 *)(r1 + 0), r1);
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_1, BPF_REG_1, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("misaligned store-release")
+__failure __failure_unpriv __msg("misaligned stack access off")
+__flag(BPF_F_ANY_ALIGNMENT)
+__naked void store_release_misaligned(void)
+{
+	asm volatile (
+	"w0 = 0;"
+	".8byte %[store_release_insn];" // store_release((u32 *)(r10 - 5), w0);
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_W, BPF_STORE_REL, BPF_REG_10, BPF_REG_0, -5))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release to ctx pointer")
+__failure __failure_unpriv __msg("BPF_ATOMIC stores into R1 ctx is not allowed")
+__naked void store_release_to_ctx_pointer(void)
+{
+	asm volatile (
+	"w0 = 0;"
+	".8byte %[store_release_insn];" // store_release((u8 *)(r1 + 0), w0);
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_STORE_REL, BPF_REG_1, BPF_REG_0, 0))
+	: __clobber_all);
+}
+
+SEC("xdp")
+__description("store-release to pkt pointer")
+__failure __msg("BPF_ATOMIC stores into R2 pkt is not allowed")
+__naked void store_release_to_pkt_pointer(void)
+{
+	asm volatile (
+	"w0 = 0;"
+	"r2 = *(u32 *)(r1 + %[xdp_md_data]);"
+	".8byte %[store_release_insn];" // store_release((u8 *)(r2 + 0), w0);
+	"exit;"
+	:
+	: __imm_const(xdp_md_data, offsetof(struct xdp_md, data)),
+	  __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_STORE_REL, BPF_REG_2, BPF_REG_0, 0))
+	: __clobber_all);
+}
+
+SEC("flow_dissector")
+__description("store-release to flow_keys pointer")
+__failure __msg("BPF_ATOMIC stores into R2 flow_keys is not allowed")
+__naked void store_release_to_flow_keys_pointer(void)
+{
+	asm volatile (
+	"w0 = 0;"
+	"r2 = *(u64 *)(r1 + %[__sk_buff_flow_keys]);"
+	".8byte %[store_release_insn];" // store_release((u8 *)(r2 + 0), w0);
+	"exit;"
+	:
+	: __imm_const(__sk_buff_flow_keys,
+		      offsetof(struct __sk_buff, flow_keys)),
+	  __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_STORE_REL, BPF_REG_2, BPF_REG_0, 0))
+	: __clobber_all);
+}
+
+SEC("sk_reuseport")
+__description("store-release to sock pointer")
+__failure __msg("BPF_ATOMIC stores into R2 sock is not allowed")
+__naked void store_release_to_sock_pointer(void)
+{
+	asm volatile (
+	"w0 = 0;"
+	"r2 = *(u64 *)(r1 + %[sk_reuseport_md_sk]);"
+	".8byte %[store_release_insn];" // store_release((u8 *)(r2 + 0), w0);
+	"exit;"
+	:
+	: __imm_const(sk_reuseport_md_sk, offsetof(struct sk_reuseport_md, sk)),
+	  __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_B, BPF_STORE_REL, BPF_REG_2, BPF_REG_0, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("store-release, leak pointer to stack")
+__success __success_unpriv __retval(0)
+__naked void store_release_leak_pointer_to_stack(void)
+{
+	asm volatile (
+	".8byte %[store_release_insn];" // store_release((u64 *)(r10 - 8), r1);
+	"r0 = 0;"
+	"exit;"
+	:
+	: __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_10, BPF_REG_1, -8))
+	: __clobber_all);
+}
+
+struct {
+	__uint(type, BPF_MAP_TYPE_HASH);
+	__uint(max_entries, 1);
+	__type(key, long long);
+	__type(value, long long);
+} map_hash_8b SEC(".maps");
+
+SEC("socket")
+__description("store-release, leak pointer to map")
+__success __retval(0)
+__failure_unpriv __msg_unpriv("R6 leaks addr into map")
+__naked void store_release_leak_pointer_to_map(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	"r1 = %[map_hash_8b] ll;"
+	"r2 = 0;"
+	"*(u64 *)(r10 - 8) = r2;"
+	"r2 = r10;"
+	"r2 += -8;"
+	"call %[bpf_map_lookup_elem];"
+	"if r0 == 0 goto l0_%=;"
+	".8byte %[store_release_insn];" // store_release((u64 *)(r0 + 0), r6);
+"l0_%=:"
+	"r0 = 0;"
+	"exit;"
+	:
+	: __imm_addr(map_hash_8b),
+	  __imm(bpf_map_lookup_elem),
+	  __imm_insn(store_release_insn,
+		     BPF_ATOMIC_OP(BPF_DW, BPF_STORE_REL, BPF_REG_0, BPF_REG_6, 0))
+	: __clobber_all);
+}
+
+#else
+
+SEC("socket")
+__description("Clang version < 18, ENABLE_ATOMICS_TESTS not defined, and/or JIT doesn't support store-release, use a dummy test")
+__success
+int dummy_test(void)
+{
+	return 0;
+}
+
+#endif
+
+char _license[] SEC("license") = "GPL";
-- 
2.48.1.601.g30ceb7b040-goog

[PATCH bpf-next v3 9/9] bpf, docs: Update instruction-set.rst for load-acquire and store-release instructions

[Peilin Ye]

Update documentation for the new load-acquire and store-release
instructions.  Rename existing atomic operations as "atomic
read-modify-write (RMW) operations".

Following RFC 9669, section 7.3. "Adding Instructions", create new
conformance groups "atomic32v2" and "atomic64v2", where:

  * atomic32v2: includes all instructions in "atomic32", plus the new
                8-bit, 16-bit and 32-bit atomic load-acquire and
                store-release instructions

  * atomic64v2: includes all instructions in "atomic64" and
                "atomic32v2", plus the new 64-bit atomic load-acquire
                and store-release instructions

Cc: bpf@ietf.org
Signed-off-by: Peilin Ye <yepeilin@google.com>
---
 .../bpf/standardization/instruction-set.rst   | 78 +++++++++++++++----
 1 file changed, 62 insertions(+), 16 deletions(-)

diff --git a/Documentation/bpf/standardization/instruction-set.rst b/Documentation/bpf/standardization/instruction-set.rst
index ab820d565052..6bd62e79c4ed 100644
--- a/Documentation/bpf/standardization/instruction-set.rst
+++ b/Documentation/bpf/standardization/instruction-set.rst
@@ -139,8 +139,14 @@ This document defines the following conformance groups:
   specification unless otherwise noted.
 * base64: includes base32, plus instructions explicitly noted
   as being in the base64 conformance group.
-* atomic32: includes 32-bit atomic operation instructions (see `Atomic operations`_).
-* atomic64: includes atomic32, plus 64-bit atomic operation instructions.
+* atomic32: includes 32-bit atomic read-modify-write instructions (see
+  `Atomic operations`_).
+* atomic32v2: includes atomic32, plus 8-bit, 16-bit and 32-bit atomic
+  load-acquire and store-release instructions.
+* atomic64: includes atomic32, plus 64-bit atomic read-modify-write
+  instructions.
+* atomic64v2: unifies atomic32v2 and atomic64, plus 64-bit atomic load-acquire
+  and store-release instructions.
 * divmul32: includes 32-bit division, multiplication, and modulo instructions.
 * divmul64: includes divmul32, plus 64-bit division, multiplication,
   and modulo instructions.
@@ -653,20 +659,29 @@ Atomic operations are operations that operate on memory and can not be
 interrupted or corrupted by other access to the same memory region
 by other BPF programs or means outside of this specification.
 
-All atomic operations supported by BPF are encoded as store operations
-that use the ``ATOMIC`` mode modifier as follows:
+All atomic operations supported by BPF are encoded as ``STX`` instructions
+that use the ``ATOMIC`` mode modifier, with the 'imm' field encoding the
+actual atomic operation.  These operations fall into two categories, as
+described in the following sections:
 
-* ``{ATOMIC, W, STX}`` for 32-bit operations, which are
+* `Atomic read-modify-write operations`_
+* `Atomic load and store operations`_
+
+Atomic read-modify-write operations
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The atomic read-modify-write (RMW) operations are encoded as follows:
+
+* ``{ATOMIC, W, STX}`` for 32-bit RMW operations, which are
   part of the "atomic32" conformance group.
-* ``{ATOMIC, DW, STX}`` for 64-bit operations, which are
+* ``{ATOMIC, DW, STX}`` for 64-bit RMW operations, which are
   part of the "atomic64" conformance group.
-* 8-bit and 16-bit wide atomic operations are not supported.
+* 8-bit and 16-bit wide atomic RMW operations are not supported.
 
-The 'imm' field is used to encode the actual atomic operation.
-Simple atomic operation use a subset of the values defined to encode
-arithmetic operations in the 'imm' field to encode the atomic operation:
+Simple atomic RMW operation use a subset of the values defined to encode
+arithmetic operations in the 'imm' field to encode the atomic RMW operation:
 
-.. table:: Simple atomic operations
+.. table:: Simple atomic read-modify-write operations
 
   ========  =====  ===========
   imm       value  description
@@ -686,10 +701,10 @@ arithmetic operations in the 'imm' field to encode the atomic operation:
 
   *(u64 *)(dst + offset) += src
 
-In addition to the simple atomic operations, there also is a modifier and
-two complex atomic operations:
+In addition to the simple atomic RMW operations, there also is a modifier and
+two complex atomic RMW operations:
 
-.. table:: Complex atomic operations
+.. table:: Complex atomic read-modify-write operations
 
   ===========  ================  ===========================
   imm          value             description
@@ -699,8 +714,8 @@ two complex atomic operations:
   CMPXCHG      0xf0 | FETCH      atomic compare and exchange
   ===========  ================  ===========================
 
-The ``FETCH`` modifier is optional for simple atomic operations, and
-always set for the complex atomic operations.  If the ``FETCH`` flag
+The ``FETCH`` modifier is optional for simple atomic RMW operations, and
+always set for the complex atomic RMW operations.  If the ``FETCH`` flag
 is set, then the operation also overwrites ``src`` with the value that
 was in memory before it was modified.
 
@@ -713,6 +728,37 @@ The ``CMPXCHG`` operation atomically compares the value addressed by
 value that was at ``dst + offset`` before the operation is zero-extended
 and loaded back to ``R0``.
 
+Atomic load and store operations
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+To encode an atomic load or store operation, the 'imm' field is one of:
+
+.. table:: Atomic load and store operations
+
+  ========= =====  ====================
+  imm       value  description
+  ========= =====  ====================
+  LOAD_ACQ  0x100  atomic load-acquire
+  STORE_REL 0x110  atomic store-release
+  ========= =====  ====================
+
+``{ATOMIC, <size>, STX}`` with 'imm' = LOAD_ACQ means::
+
+  dst = load_acquire((unsigned size *)(src + offset))
+
+``{ATOMIC, <size>, STX}`` with 'imm' = STORE_REL means::
+
+  store_release((unsigned size *)(dst + offset), src)
+
+Where '<size>' is one of: ``B``, ``H``, ``W``, or ``DW``, and 'unsigned size'
+is one of: u8, u16, u32, or u64.
+
+8-bit, 16-bit and 32-bit atomic load-acquire and store-release instructions
+are part of the "atomic32v2" conformance group.
+
+64-bit atomic load-acquire and store-release instructions are part of the
+"atomic64v2" conformance group.
+
 64-bit immediate instructions
 -----------------------------
 
-- 
2.48.1.601.g30ceb7b040-goog