[PATCH bpf-next v6 1/3] riscv, bpf: Add 12-argument support for RV64 bpf trampoline

[Puranjay Mohan]

Pu Lehui <pulehui at huaweicloud.com> writes:

> From: Pu Lehui <pulehui at huawei.com>
>
> This patch adds 12 function arguments support for riscv64 bpf
> trampoline. The current bpf trampoline supports <= sizeof(u64) bytes
> scalar arguments [0] and <= 16 bytes struct arguments [1]. Therefore, we
> focus on the situation where scalars are at most XLEN bits and
> aggregates whose total size does not exceed 2×XLEN bits in the riscv
> calling convention [2].
>
> Link: https://elixir.bootlin.com/linux/v6.8/source/kernel/bpf/btf.c#L6184 [0]
> Link: https://elixir.bootlin.com/linux/v6.8/source/kernel/bpf/btf.c#L6769 [1]
> Link: https://github.com/riscv-non-isa/riscv-elf-psabi-doc/releases/download/draft-20230929-e5c800e661a53efe3c2678d71a306323b60eb13b/riscv-abi.pdf [2]
> Signed-off-by: Pu Lehui <pulehui at huawei.com>
> Acked-by: Björn Töpel <bjorn at kernel.org>
> Reviewed-by: Björn Töpel <bjorn at rivosinc.com>
> ---
>  arch/riscv/net/bpf_jit_comp64.c | 66 +++++++++++++++++++++++----------
>  1 file changed, 47 insertions(+), 19 deletions(-)
>
> diff --git a/arch/riscv/net/bpf_jit_comp64.c b/arch/riscv/net/bpf_jit_comp64.c
> index 351e1484205e..685c7389ae7e 100644
> --- a/arch/riscv/net/bpf_jit_comp64.c
> +++ b/arch/riscv/net/bpf_jit_comp64.c
> @@ -15,6 +15,7 @@
>  #include <asm/percpu.h>
>  #include "bpf_jit.h"
>  
> +#define RV_MAX_REG_ARGS 8
>  #define RV_FENTRY_NINSNS 2
>  /* imm that allows emit_imm to emit max count insns */
>  #define RV_MAX_COUNT_IMM 0x7FFF7FF7FF7FF7FF
> @@ -692,26 +693,45 @@ int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type poke_type,
>  	return ret;
>  }
>  
> -static void store_args(int nregs, int args_off, struct rv_jit_context *ctx)
> +static void store_args(int nr_arg_slots, int args_off, struct rv_jit_context *ctx)
>  {
>  	int i;
>  
> -	for (i = 0; i < nregs; i++) {
> -		emit_sd(RV_REG_FP, -args_off, RV_REG_A0 + i, ctx);
> +	for (i = 0; i < nr_arg_slots; i++) {
> +		if (i < RV_MAX_REG_ARGS) {
> +			emit_sd(RV_REG_FP, -args_off, RV_REG_A0 + i, ctx);
> +		} else {
> +			/* skip slots for T0 and FP of traced function */
> +			emit_ld(RV_REG_T1, 16 + (i - RV_MAX_REG_ARGS) * 8, RV_REG_FP, ctx);
> +			emit_sd(RV_REG_FP, -args_off, RV_REG_T1, ctx);
> +		}
>  		args_off -= 8;
>  	}
>  }
>  
> -static void restore_args(int nregs, int args_off, struct rv_jit_context *ctx)
> +static void restore_args(int nr_reg_args, int args_off, struct rv_jit_context *ctx)
>  {
>  	int i;
>  
> -	for (i = 0; i < nregs; i++) {
> +	for (i = 0; i < nr_reg_args; i++) {
>  		emit_ld(RV_REG_A0 + i, -args_off, RV_REG_FP, ctx);
>  		args_off -= 8;
>  	}
>  }
>  
> +static void restore_stack_args(int nr_stack_args, int args_off, int stk_arg_off,
> +			       struct rv_jit_context *ctx)
> +{
> +	int i;
> +
> +	for (i = 0; i < nr_stack_args; i++) {
> +		emit_ld(RV_REG_T1, -(args_off - RV_MAX_REG_ARGS * 8), RV_REG_FP, ctx);
> +		emit_sd(RV_REG_FP, -stk_arg_off, RV_REG_T1, ctx);
> +		args_off -= 8;
> +		stk_arg_off -= 8;
> +	}
> +}
> +
>  static int invoke_bpf_prog(struct bpf_tramp_link *l, int args_off, int retval_off,
>  			   int run_ctx_off, bool save_ret, struct rv_jit_context *ctx)
>  {
> @@ -784,8 +804,8 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
>  {
>  	int i, ret, offset;
>  	int *branches_off = NULL;
> -	int stack_size = 0, nregs = m->nr_args;
> -	int retval_off, args_off, nregs_off, ip_off, run_ctx_off, sreg_off;
> +	int stack_size = 0, nr_arg_slots = 0;
> +	int retval_off, args_off, nregs_off, ip_off, run_ctx_off, sreg_off, stk_arg_off;
>  	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];
> @@ -831,20 +851,21 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
>  	 * FP - sreg_off    [ callee saved reg	]
>  	 *
>  	 *		    [ pads              ] pads for 16 bytes alignment
> +	 *
> +	 *		    [ stack_argN        ]
> +	 *		    [ ...               ]
> +	 * FP - stk_arg_off [ stack_arg1        ] BPF_TRAMP_F_CALL_ORIG
>  	 */
>  
>  	if (flags & (BPF_TRAMP_F_ORIG_STACK | BPF_TRAMP_F_SHARE_IPMODIFY))
>  		return -ENOTSUPP;
>  
> -	/* extra regiters for struct arguments */
> -	for (i = 0; i < m->nr_args; i++)
> -		if (m->arg_flags[i] & BTF_FMODEL_STRUCT_ARG)
> -			nregs += round_up(m->arg_size[i], 8) / 8 - 1;
> -
> -	/* 8 arguments passed by registers */
> -	if (nregs > 8)
> +	if (m->nr_args > MAX_BPF_FUNC_ARGS)
>  		return -ENOTSUPP;
>  
> +	for (i = 0; i < m->nr_args; i++)
> +		nr_arg_slots += round_up(m->arg_size[i], 8) / 8;
> +
>  	/* room of trampoline frame to store return address and frame pointer */
>  	stack_size += 16;
>  
> @@ -854,7 +875,7 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
>  		retval_off = stack_size;
>  	}
>  
> -	stack_size += nregs * 8;
> +	stack_size += nr_arg_slots * 8;
>  	args_off = stack_size;
>  
>  	stack_size += 8;
> @@ -871,8 +892,14 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
>  	stack_size += 8;
>  	sreg_off = stack_size;
>  
> +	if (nr_arg_slots - RV_MAX_REG_ARGS > 0)
> +		stack_size += (nr_arg_slots - RV_MAX_REG_ARGS) * 8;

Hi Pu,
Although this is merged now, while working on this for arm64 I realised
that the above doesn't check for BPF_TRAMP_F_CALL_ORIG and can waste
some stack space, we should change this to:

if ((flags & BPF_TRAMP_F_CALL_ORIG) && (nr_arg_slots - RV_MAX_REG_ARGS > 0))
        stack_size += (nr_arg_slots - RV_MAX_REG_ARGS) * 8;

It will save some stack space when BPF_TRAMP_F_CALL_ORIG is not set?

I can send a patch if you think this is worth fixing.


Thanks,
Puranjay
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 255 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-riscv/attachments/20240705/f4a16d6f/attachment.sig>