LoongArch: BPF: Add struct ops support for trampoline

[RFC PATCH] LoongArch: BPF: Add struct ops support for trampoline

Posted by Tiezhu Yang 2 months, 3 weeks ago

Use BPF_TRAMP_F_INDIRECT flag to detect struct ops and emit proper
prologue and epilogue for this case.

With this patch, all of the struct_ops related testcases (except
struct_ops_multi_pages) passed on LoongArch.

The testcase struct_ops_multi_pages failed is because the actual
image_pages_cnt is 40 which is bigger than MAX_TRAMP_IMAGE_PAGES.

Before:

  $ sudo ./test_progs -t struct_ops -d struct_ops_multi_pages
  ...
  WATCHDOG: test case struct_ops_module/struct_ops_load executes for 10 seconds...

After:

  $ sudo ./test_progs -t struct_ops -d struct_ops_multi_pages
  ...
  #15      bad_struct_ops:OK
  ...
  #399     struct_ops_autocreate:OK
  ...
  #400     struct_ops_kptr_return:OK
  ...
  #401     struct_ops_maybe_null:OK
  ...
  #402     struct_ops_module:OK
  ...
  #404     struct_ops_no_cfi:OK
  ...
  #405     struct_ops_private_stack:SKIP
  ...
  #406     struct_ops_refcounted:OK
  Summary: 8/25 PASSED, 3 SKIPPED, 0 FAILED

Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
---

This is a RFC patch, based on 6.16-rc6 and the following series:

[PATCH v3 0/5] Support trampoline for LoongArch
https://lore.kernel.org/loongarch/20250709055029.723243-1-duanchenghao@kylinos.cn/

 arch/loongarch/net/bpf_jit.c | 71 ++++++++++++++++++++++++------------
 1 file changed, 47 insertions(+), 24 deletions(-)

diff --git a/arch/loongarch/net/bpf_jit.c b/arch/loongarch/net/bpf_jit.c
index 6820558afce5..7663b84a4ff1 100644
--- a/arch/loongarch/net/bpf_jit.c
+++ b/arch/loongarch/net/bpf_jit.c
@@ -1599,6 +1599,7 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
 	struct bpf_tramp_links *fentry = &tlinks[BPF_TRAMP_FENTRY];
 	struct bpf_tramp_links *fexit = &tlinks[BPF_TRAMP_FEXIT];
 	struct bpf_tramp_links *fmod_ret = &tlinks[BPF_TRAMP_MODIFY_RETURN];
+	bool is_struct_ops = flags & BPF_TRAMP_F_INDIRECT;
 	int ret, save_ret;
 	void *orig_call = func_addr;
 	u32 **branches = NULL;
@@ -1674,18 +1675,31 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
 
 	stack_size = round_up(stack_size, 16);
 
-	/* For the trampoline called from function entry */
-	/* RA and FP for parent function*/
-	emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -16);
-	emit_insn(ctx, std, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
-	emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
-	emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 16);
-
-	/* RA and FP for traced function*/
-	emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -stack_size);
-	emit_insn(ctx, std, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
-	emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
-	emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size);
+	if (!is_struct_ops) {
+		/*
+		 * For the trampoline called from function entry,
+		 * the frame of traced function and the frame of
+		 * trampoline need to be considered.
+		 */
+		emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -16);
+		emit_insn(ctx, std, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
+		emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
+		emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 16);
+
+		emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -stack_size);
+		emit_insn(ctx, std, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
+		emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
+		emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size);
+	} else {
+		/*
+		 * For the trampoline called directly, just handle
+		 * the frame of trampoline.
+		 */
+		emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -stack_size);
+		emit_insn(ctx, std, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, stack_size - 8);
+		emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
+		emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size);
+	}
 
 	/* callee saved register S1 to pass start time */
 	emit_insn(ctx, std, LOONGARCH_GPR_S1, LOONGARCH_GPR_FP, -sreg_off);
@@ -1772,21 +1786,30 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
 
 	emit_insn(ctx, ldd, LOONGARCH_GPR_S1, LOONGARCH_GPR_FP, -sreg_off);
 
-	/* trampoline called from function entry */
-	emit_insn(ctx, ldd, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
-	emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
-	emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, stack_size);
+	if (!is_struct_ops) {
+		/* trampoline called from function entry */
+		emit_insn(ctx, ldd, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
+		emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
+		emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, stack_size);
+
+		emit_insn(ctx, ldd, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
+		emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
+		emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, 16);
 
-	emit_insn(ctx, ldd, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
-	emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
-	emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, 16);
+		if (flags & BPF_TRAMP_F_SKIP_FRAME)
+			/* return to parent function */
+			emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_RA, 0);
+		else
+			/* return to traced function */
+			emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_T0, 0);
+	} else {
+		/* trampoline called directly */
+		emit_insn(ctx, ldd, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, stack_size - 8);
+		emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
+		emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, stack_size);
 
-	if (flags & BPF_TRAMP_F_SKIP_FRAME)
-		/* return to parent function */
 		emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_RA, 0);
-	else
-		/* return to traced function */
-		emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_T0, 0);
+	}
 
 	ret = ctx->idx;
 out:
-- 
2.42.0

Re: [RFC PATCH] LoongArch: BPF: Add struct ops support for trampoline

Posted by Huacai Chen 2 months, 3 weeks ago

Hi, Tiezhu,

I hope this patch can be squashed to V4 of chenghao's patchset, as a
co-developer.

Huacai

On Wed, Jul 16, 2025 at 12:39 PM Tiezhu Yang <yangtiezhu@loongson.cn> wrote:
>
> Use BPF_TRAMP_F_INDIRECT flag to detect struct ops and emit proper
> prologue and epilogue for this case.
>
> With this patch, all of the struct_ops related testcases (except
> struct_ops_multi_pages) passed on LoongArch.
>
> The testcase struct_ops_multi_pages failed is because the actual
> image_pages_cnt is 40 which is bigger than MAX_TRAMP_IMAGE_PAGES.
>
> Before:
>
>   $ sudo ./test_progs -t struct_ops -d struct_ops_multi_pages
>   ...
>   WATCHDOG: test case struct_ops_module/struct_ops_load executes for 10 seconds...
>
> After:
>
>   $ sudo ./test_progs -t struct_ops -d struct_ops_multi_pages
>   ...
>   #15      bad_struct_ops:OK
>   ...
>   #399     struct_ops_autocreate:OK
>   ...
>   #400     struct_ops_kptr_return:OK
>   ...
>   #401     struct_ops_maybe_null:OK
>   ...
>   #402     struct_ops_module:OK
>   ...
>   #404     struct_ops_no_cfi:OK
>   ...
>   #405     struct_ops_private_stack:SKIP
>   ...
>   #406     struct_ops_refcounted:OK
>   Summary: 8/25 PASSED, 3 SKIPPED, 0 FAILED
>
> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
> ---
>
> This is a RFC patch, based on 6.16-rc6 and the following series:
>
> [PATCH v3 0/5] Support trampoline for LoongArch
> https://lore.kernel.org/loongarch/20250709055029.723243-1-duanchenghao@kylinos.cn/
>
>  arch/loongarch/net/bpf_jit.c | 71 ++++++++++++++++++++++++------------
>  1 file changed, 47 insertions(+), 24 deletions(-)
>
> diff --git a/arch/loongarch/net/bpf_jit.c b/arch/loongarch/net/bpf_jit.c
> index 6820558afce5..7663b84a4ff1 100644
> --- a/arch/loongarch/net/bpf_jit.c
> +++ b/arch/loongarch/net/bpf_jit.c
> @@ -1599,6 +1599,7 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
>         struct bpf_tramp_links *fentry = &tlinks[BPF_TRAMP_FENTRY];
>         struct bpf_tramp_links *fexit = &tlinks[BPF_TRAMP_FEXIT];
>         struct bpf_tramp_links *fmod_ret = &tlinks[BPF_TRAMP_MODIFY_RETURN];
> +       bool is_struct_ops = flags & BPF_TRAMP_F_INDIRECT;
>         int ret, save_ret;
>         void *orig_call = func_addr;
>         u32 **branches = NULL;
> @@ -1674,18 +1675,31 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
>
>         stack_size = round_up(stack_size, 16);
>
> -       /* For the trampoline called from function entry */
> -       /* RA and FP for parent function*/
> -       emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -16);
> -       emit_insn(ctx, std, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
> -       emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
> -       emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 16);
> -
> -       /* RA and FP for traced function*/
> -       emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -stack_size);
> -       emit_insn(ctx, std, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
> -       emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
> -       emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size);
> +       if (!is_struct_ops) {
> +               /*
> +                * For the trampoline called from function entry,
> +                * the frame of traced function and the frame of
> +                * trampoline need to be considered.
> +                */
> +               emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -16);
> +               emit_insn(ctx, std, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
> +               emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
> +               emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 16);
> +
> +               emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -stack_size);
> +               emit_insn(ctx, std, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
> +               emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
> +               emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size);
> +       } else {
> +               /*
> +                * For the trampoline called directly, just handle
> +                * the frame of trampoline.
> +                */
> +               emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, -stack_size);
> +               emit_insn(ctx, std, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, stack_size - 8);
> +               emit_insn(ctx, std, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
> +               emit_insn(ctx, addid, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size);
> +       }
>
>         /* callee saved register S1 to pass start time */
>         emit_insn(ctx, std, LOONGARCH_GPR_S1, LOONGARCH_GPR_FP, -sreg_off);
> @@ -1772,21 +1786,30 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
>
>         emit_insn(ctx, ldd, LOONGARCH_GPR_S1, LOONGARCH_GPR_FP, -sreg_off);
>
> -       /* trampoline called from function entry */
> -       emit_insn(ctx, ldd, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
> -       emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
> -       emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, stack_size);
> +       if (!is_struct_ops) {
> +               /* trampoline called from function entry */
> +               emit_insn(ctx, ldd, LOONGARCH_GPR_T0, LOONGARCH_GPR_SP, stack_size - 8);
> +               emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
> +               emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, stack_size);
> +
> +               emit_insn(ctx, ldd, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
> +               emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
> +               emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, 16);
>
> -       emit_insn(ctx, ldd, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, 8);
> -       emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, 0);
> -       emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, 16);
> +               if (flags & BPF_TRAMP_F_SKIP_FRAME)
> +                       /* return to parent function */
> +                       emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_RA, 0);
> +               else
> +                       /* return to traced function */
> +                       emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_T0, 0);
> +       } else {
> +               /* trampoline called directly */
> +               emit_insn(ctx, ldd, LOONGARCH_GPR_RA, LOONGARCH_GPR_SP, stack_size - 8);
> +               emit_insn(ctx, ldd, LOONGARCH_GPR_FP, LOONGARCH_GPR_SP, stack_size - 16);
> +               emit_insn(ctx, addid, LOONGARCH_GPR_SP, LOONGARCH_GPR_SP, stack_size);
>
> -       if (flags & BPF_TRAMP_F_SKIP_FRAME)
> -               /* return to parent function */
>                 emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_RA, 0);
> -       else
> -               /* return to traced function */
> -               emit_insn(ctx, jirl, LOONGARCH_GPR_ZERO, LOONGARCH_GPR_T0, 0);
> +       }
>
>         ret = ctx->idx;
>  out:
> --
> 2.42.0
>

Re: [RFC PATCH] LoongArch: BPF: Add struct ops support for trampoline

Posted by Tiezhu Yang 2 months, 3 weeks ago

On 2025/7/16 下午12:42, Huacai Chen wrote:
> Hi, Tiezhu,
> 
> I hope this patch can be squashed to V4 of chenghao's patchset, as a
> co-developer.

No, do not squash, just add it as the last patch if necessary.

I think keep it as a single patch is proper due to it is an
additional independent feature and the log can be clear.

Thanks,
Tiezhu

Re: [RFC PATCH] LoongArch: BPF: Add struct ops support for trampoline

Posted by Huacai Chen 2 months, 3 weeks ago

On Wed, Jul 16, 2025 at 1:10 PM Tiezhu Yang <yangtiezhu@loongson.cn> wrote:
>
> On 2025/7/16 下午12:42, Huacai Chen wrote:
> > Hi, Tiezhu,
> >
> > I hope this patch can be squashed to V4 of chenghao's patchset, as a
> > co-developer.
>
> No, do not squash, just add it as the last patch if necessary.
>
> I think keep it as a single patch is proper due to it is an
> additional independent feature and the log can be clear.
OK, then put this as the last patch of chenghao's V4.

Huacai
>
> Thanks,
> Tiezhu
>
>