Re: [PATCHv2 bpf-next 3/3] bpf, x86: Add support for jit dry run

["Alexei Starovoitov" <alexei.starovoitov@gmail.com>]

On Fri Sep 4, 2026 at 12:45 AM PDT, Jiri Olsa wrote:
> On Thu, Sep 03, 2026 at 09:40:46PM -0700, Alexei Starovoitov wrote:
>> On Thu Sep 3, 2026 at 2:20 AM PDT, Jiri Olsa wrote:
>> > Adding support to run jit code generation in dry_run mode that won't
>> > store any code and only returns the jir code size.
>> >
>> > The dry_run is enabled when __arch_prepare_bpf_trampoline is called
>> > with rw_image argument as NULL.
>> >
>> > It's used in arch_bpf_trampoline_size where it allows to skip the
>> > image allocation, that gives speed up for tracing_multi attachment.
>> >
>> > With current code:
>> >
>> >   # ./test_progs -t tracing_multi_bench_attach -v
>> >   ...
>> >   serial_test_tracing_multi_bench_attach: found 55227 functions
>> >   serial_test_tracing_multi_bench_attach: attached in   1.563s
>> >   serial_test_tracing_multi_bench_attach: detached in   0.256s
>> >
>> > With the fix:
>> >
>> >   # ./test_progs -t tracing_multi_bench_attach -v
>> >   ...
>> >   serial_test_tracing_multi_bench_attach: found 55235 functions
>> >   serial_test_tracing_multi_bench_attach: attached in   0.798s
>> >   serial_test_tracing_multi_bench_attach: detached in   0.258s
>> >
>> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
>> > ---
>> >  arch/x86/net/bpf_jit_comp.c | 42 ++++++++++++++++++-------------------
>> >  1 file changed, 20 insertions(+), 22 deletions(-)
>> >
>> > diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
>> > index ee2ddeba3de1..d1af7dc5c5ce 100644
>> > --- a/arch/x86/net/bpf_jit_comp.c
>> > +++ b/arch/x86/net/bpf_jit_comp.c
>> > @@ -25,6 +25,7 @@ static bool all_callee_regs_used[4] = {true, true, true, true};
>> >  
>> >  struct jit_emit_context {
>> >  	u8 *prog;
>> > +	bool dry_run;
>> >  };
>> >  
>> >  static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
>> > @@ -42,7 +43,10 @@ static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
>> >  
>> >  static void emit_code_jit(struct jit_emit_context *jit, u32 bytes, unsigned int len)
>> >  {
>> > -	jit->prog = emit_code(jit->prog, bytes, len);
>> > +	if (jit->dry_run)
>> > +		jit->prog += len;
>> > +	else
>> > +		jit->prog = emit_code(jit->prog, bytes, len);
>> 
>> Sorry, I don't believe that skipping emit makes that much
>> of runtime difference.
>> I suspect jit_alloc + jit_free are costly.
>> In such case the earlier patches are not needed.
>> Point emit logic to some scratch area and discard it.
>> Same effect with half of the changes.
>> 
>> pw-bot: cr
>> 
>
> yes, I got same speed up with the change below

Great :)

> jirka
>
>
> ---
> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> index 48429fae0641..1bba19a9a17f 100644
> --- a/arch/x86/net/bpf_jit_comp.c
> +++ b/arch/x86/net/bpf_jit_comp.c
> @@ -9,6 +9,7 @@
>  #include <linux/filter.h>
>  #include <linux/if_vlan.h>
>  #include <linux/bitfield.h>
> +#include <linux/init.h>
>  #include <linux/bpf.h>
>  #include <linux/bpf_verifier.h>
>  #include <linux/memory.h>
> @@ -23,6 +24,19 @@
>  
>  static bool all_callee_regs_used[4] = {true, true, true, true};
>  
> +/*
> + * Reuse a writable image in the BPF execmem range for size calculation.
> + * Its contents do not affect size calculation.
> + */
> +static void *trampoline_size_image;
> +
> +static int __init init_trampoline_size_image(void)
> +{
> +	trampoline_size_image = bpf_jit_alloc_exec_rw(PAGE_SIZE);

I think you could have kvmalloc()-ed that memory instead.

> +	return trampoline_size_image ? 0 : -ENOMEM;
> +}
> +late_initcall(init_trampoline_size_image);
> +
>  static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
>  {
>  	if (len == 1)
> @@ -3819,23 +3833,14 @@ int arch_bpf_trampoline_size(const struct btf_func_model *m, u32 flags,
>  			     struct bpf_tramp_nodes *tnodes, void *func_addr)
>  {
>  	struct bpf_tramp_image im;
> -	void *image;
> -	int ret;
>  
> -	/* Allocate a temporary buffer for __arch_prepare_bpf_trampoline().
> -	 *
> -	 * We cannot use kvmalloc here, because we need image to be in
> -	 * module memory range.
> -	 * Since it must be writable use bpf_jit_alloc_exec_rw().
> -	 */
> -	image = bpf_jit_alloc_exec_rw(PAGE_SIZE);
> -	if (!image)
> +	if (!trampoline_size_image)
>  		return -ENOMEM;
>  
> -	ret = __arch_prepare_bpf_trampoline(&im, image, image + PAGE_SIZE, image,
> -					    m, flags, tnodes, func_addr);
> -	bpf_jit_free_exec(image);
> -	return ret;
> +	return __arch_prepare_bpf_trampoline(&im, trampoline_size_image,
> +					     trampoline_size_image + PAGE_SIZE,
> +					     trampoline_size_image, m, flags,
> +					     tnodes, func_addr);
>  }
>  
>  static int emit_bpf_dispatcher(u8 **pprog, int a, int b, s64 *progs, u8 *image, u8 *buf)