]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
bpf,s390: add fsession support for trampolines
authorMenglong Dong <menglong8.dong@gmail.com>
Tue, 24 Feb 2026 09:22:07 +0000 (17:22 +0800)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 3 Mar 2026 16:36:26 +0000 (08:36 -0800)
Implement BPF_TRACE_FSESSION support for s390. The logic here is similar
to what we did in x86_64.

In order to simply the logic, we factor out the function invoke_bpf() for
fentry and fexit.

Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
Link: https://lore.kernel.org/r/20260224092208.1395085-3-dongml2@chinatelecom.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
arch/s390/net/bpf_jit_comp.c

index d72d682ad220422dee24484109f4e1e41436c3a7..1f9a6b728beb34e26fdefccc6f7d9743f422ddab 100644 (file)
@@ -2487,8 +2487,8 @@ struct bpf_tramp_jit {
        int ip_off;             /* For bpf_get_func_ip(), has to be at
                                 * (ctx - 16)
                                 */
-       int arg_cnt_off;        /* For bpf_get_func_arg_cnt(), has to be at
-                                * (ctx - 8)
+       int func_meta_off;      /* For bpf_get_func_arg_cnt()/fsession, has
+                                * to be at (ctx - 8)
                                 */
        int bpf_args_off;       /* Offset of BPF_PROG context, which consists
                                 * of BPF arguments followed by return value
@@ -2592,6 +2592,28 @@ static int invoke_bpf_prog(struct bpf_tramp_jit *tjit,
        return 0;
 }
 
+static int invoke_bpf(struct bpf_tramp_jit *tjit,
+                     const struct btf_func_model *m,
+                     struct bpf_tramp_links *tl, bool save_ret,
+                     u64 func_meta, int cookie_off)
+{
+       int i, cur_cookie = (tjit->bpf_args_off - cookie_off) / sizeof(u64);
+       struct bpf_jit *jit = &tjit->common;
+
+       for (i = 0; i < tl->nr_links; i++) {
+               if (bpf_prog_calls_session_cookie(tl->links[i])) {
+                       u64 meta = func_meta | ((u64)cur_cookie << BPF_TRAMP_COOKIE_INDEX_SHIFT);
+
+                       emit_store_stack_imm64(jit, REG_0, tjit->func_meta_off, meta);
+                       cur_cookie--;
+               }
+               if (invoke_bpf_prog(tjit, m, tl->links[i], save_ret))
+                       return -EINVAL;
+       }
+
+       return 0;
+}
+
 static int alloc_stack(struct bpf_tramp_jit *tjit, size_t size)
 {
        int stack_offset = tjit->stack_size;
@@ -2621,8 +2643,10 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
        struct bpf_tramp_links *fentry = &tlinks[BPF_TRAMP_FENTRY];
        struct bpf_tramp_links *fexit = &tlinks[BPF_TRAMP_FEXIT];
        int nr_bpf_args, nr_reg_args, nr_stack_args;
+       int cookie_cnt, cookie_off, fsession_cnt;
        struct bpf_jit *jit = &tjit->common;
        int arg, bpf_arg_off;
+       u64 func_meta;
        int i, j;
 
        /* Support as many stack arguments as "mvc" instruction can handle. */
@@ -2654,6 +2678,9 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
                        return -ENOTSUPP;
        }
 
+       cookie_cnt = bpf_fsession_cookie_cnt(tlinks);
+       fsession_cnt = bpf_fsession_cnt(tlinks);
+
        /*
         * Calculate the stack layout.
         */
@@ -2666,8 +2693,9 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
        tjit->backchain_off = tjit->stack_size - sizeof(u64);
        tjit->stack_args_off = alloc_stack(tjit, nr_stack_args * sizeof(u64));
        tjit->reg_args_off = alloc_stack(tjit, nr_reg_args * sizeof(u64));
+       cookie_off = alloc_stack(tjit, cookie_cnt * sizeof(u64));
        tjit->ip_off = alloc_stack(tjit, sizeof(u64));
-       tjit->arg_cnt_off = alloc_stack(tjit, sizeof(u64));
+       tjit->func_meta_off = alloc_stack(tjit, sizeof(u64));
        tjit->bpf_args_off = alloc_stack(tjit, nr_bpf_args * sizeof(u64));
        tjit->retval_off = alloc_stack(tjit, sizeof(u64));
        tjit->r7_r8_off = alloc_stack(tjit, 2 * sizeof(u64));
@@ -2756,11 +2784,12 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
 
        if (flags & BPF_TRAMP_F_IP_ARG)
                emit_store_stack_imm64(jit, REG_0, tjit->ip_off, (u64)func_addr);
-       /* lghi %r0,nr_bpf_args */
-       EMIT4_IMM(0xa7090000, REG_0, nr_bpf_args);
-       /* stg %r0,arg_cnt_off(%r15) */
+       func_meta = nr_bpf_args;
+       /* lghi %r0,func_meta */
+       EMIT4_IMM(0xa7090000, REG_0, func_meta);
+       /* stg %r0,func_meta_off(%r15) */
        EMIT6_DISP_LH(0xe3000000, 0x0024, REG_0, REG_0, REG_15,
-                     tjit->arg_cnt_off);
+                     tjit->func_meta_off);
 
        if (flags & BPF_TRAMP_F_CALL_ORIG) {
                /*
@@ -2773,10 +2802,17 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
                EMIT6_PCREL_RILB_PTR(0xc0050000, REG_14, __bpf_tramp_enter);
        }
 
-       for (i = 0; i < fentry->nr_links; i++)
-               if (invoke_bpf_prog(tjit, m, fentry->links[i],
-                                   flags & BPF_TRAMP_F_RET_FENTRY_RET))
-                       return -EINVAL;
+       if (fsession_cnt) {
+               /* Clear all the session cookies' value. */
+               for (i = 0; i < cookie_cnt; i++)
+                       emit_store_stack_imm64(jit, REG_0, cookie_off + 8 * i, 0);
+               /* Clear the return value to make sure fentry always gets 0. */
+               emit_store_stack_imm64(jit, REG_0, tjit->retval_off, 0);
+       }
+
+       if (invoke_bpf(tjit, m, fentry, flags & BPF_TRAMP_F_RET_FENTRY_RET,
+                      func_meta, cookie_off))
+               return -EINVAL;
 
        if (fmod_ret->nr_links) {
                /*
@@ -2853,11 +2889,16 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
                EMIT6_PCREL_RILC(0xc0040000, 0, (u64)im->ip_epilogue);
        }
 
+       /* Set the "is_return" flag for fsession. */
+       func_meta |= (1ULL << BPF_TRAMP_IS_RETURN_SHIFT);
+       if (fsession_cnt)
+               emit_store_stack_imm64(jit, REG_W0, tjit->func_meta_off,
+                                      func_meta);
+
        /* do_fexit: */
        tjit->do_fexit = jit->prg;
-       for (i = 0; i < fexit->nr_links; i++)
-               if (invoke_bpf_prog(tjit, m, fexit->links[i], false))
-                       return -EINVAL;
+       if (invoke_bpf(tjit, m, fexit, false, func_meta, cookie_off))
+               return -EINVAL;
 
        if (flags & BPF_TRAMP_F_CALL_ORIG) {
                im->ip_epilogue = jit->prg_buf + jit->prg;
@@ -2962,6 +3003,11 @@ bool bpf_jit_supports_arena(void)
        return true;
 }
 
+bool bpf_jit_supports_fsession(void)
+{
+       return true;
+}
+
 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena)
 {
        if (!in_arena)