On Wed, 15 Oct 2025 16:32:37 +0800
Menglong Dong <[email protected]> wrote:

> For now, fgraph is used for the fprobe, even if we need trace the entry
> only. However, the performance of ftrace is better than fgraph, and we
> can use ftrace_ops for this case.
> 
> Then performance of kprobe-multi increases from 54M to 69M. Before this
> commit:
> 
>   $ ./benchs/run_bench_trigger.sh kprobe-multi
>   kprobe-multi   :   54.663 ± 0.493M/s
> 
> After this commit:
> 
>   $ ./benchs/run_bench_trigger.sh kprobe-multi
>   kprobe-multi   :   69.447 ± 0.143M/s
> 
> Mitigation is disable during the bench testing above.
> 

Looks good to me. Thanks!

> Signed-off-by: Menglong Dong <[email protected]>
> ---
> v4:
> - fallback to fgraph if FTRACE_OPS_FL_SAVE_REGS not supported
> 
> v3:
> - add some comment to the rcu_read_lock() in fprobe_ftrace_entry()
> 
> v2:
> - add some document for fprobe_fgraph_entry as Masami suggested
> - merge the rename of fprobe_entry into current patch
> - use ftrace_test_recursion_trylock() in fprobe_ftrace_entry()
> ---
>  kernel/trace/fprobe.c | 128 +++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 119 insertions(+), 9 deletions(-)
> 
> diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c
> index 99d83c08b9e2..ecd623eef68b 100644
> --- a/kernel/trace/fprobe.c
> +++ b/kernel/trace/fprobe.c
> @@ -254,8 +254,106 @@ static inline int __fprobe_kprobe_handler(unsigned long 
> ip, unsigned long parent
>       return ret;
>  }
>  
> -static int fprobe_entry(struct ftrace_graph_ent *trace, struct fgraph_ops 
> *gops,
> -                     struct ftrace_regs *fregs)
> +#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
> +/* ftrace_ops callback, this processes fprobes which have only 
> entry_handler. */
> +static void fprobe_ftrace_entry(unsigned long ip, unsigned long parent_ip,
> +     struct ftrace_ops *ops, struct ftrace_regs *fregs)
> +{
> +     struct fprobe_hlist_node *node;
> +     struct rhlist_head *head, *pos;
> +     struct fprobe *fp;
> +     int bit;
> +
> +     bit = ftrace_test_recursion_trylock(ip, parent_ip);
> +     if (bit < 0)
> +             return;
> +
> +     /*
> +      * ftrace_test_recursion_trylock() disables preemption, but
> +      * rhltable_lookup() checks whether rcu_read_lcok is held.
> +      * So we take rcu_read_lock() here.
> +      */
> +     rcu_read_lock();
> +     head = rhltable_lookup(&fprobe_ip_table, &ip, fprobe_rht_params);
> +
> +     rhl_for_each_entry_rcu(node, pos, head, hlist) {
> +             if (node->addr != ip)
> +                     break;
> +             fp = READ_ONCE(node->fp);
> +             if (unlikely(!fp || fprobe_disabled(fp) || fp->exit_handler))
> +                     continue;
> +
> +             if (fprobe_shared_with_kprobes(fp))
> +                     __fprobe_kprobe_handler(ip, parent_ip, fp, fregs, NULL);
> +             else
> +                     __fprobe_handler(ip, parent_ip, fp, fregs, NULL);
> +     }
> +     rcu_read_unlock();
> +     ftrace_test_recursion_unlock(bit);
> +}
> +NOKPROBE_SYMBOL(fprobe_ftrace_entry);
> +
> +static struct ftrace_ops fprobe_ftrace_ops = {
> +     .func   = fprobe_ftrace_entry,
> +     .flags  = FTRACE_OPS_FL_SAVE_REGS,
> +};
> +static int fprobe_ftrace_active;
> +
> +static int fprobe_ftrace_add_ips(unsigned long *addrs, int num)
> +{
> +     int ret;
> +
> +     lockdep_assert_held(&fprobe_mutex);
> +
> +     ret = ftrace_set_filter_ips(&fprobe_ftrace_ops, addrs, num, 0, 0);
> +     if (ret)
> +             return ret;
> +
> +     if (!fprobe_ftrace_active) {
> +             ret = register_ftrace_function(&fprobe_ftrace_ops);
> +             if (ret) {
> +                     ftrace_free_filter(&fprobe_ftrace_ops);
> +                     return ret;
> +             }
> +     }
> +     fprobe_ftrace_active++;
> +     return 0;
> +}
> +
> +static void fprobe_ftrace_remove_ips(unsigned long *addrs, int num)
> +{
> +     lockdep_assert_held(&fprobe_mutex);
> +
> +     fprobe_ftrace_active--;
> +     if (!fprobe_ftrace_active)
> +             unregister_ftrace_function(&fprobe_ftrace_ops);
> +     if (num)
> +             ftrace_set_filter_ips(&fprobe_ftrace_ops, addrs, num, 1, 0);
> +}
> +
> +static bool fprobe_is_ftrace(struct fprobe *fp)
> +{
> +     return !fp->exit_handler;
> +}
> +#else
> +static int fprobe_ftrace_add_ips(unsigned long *addrs, int num)
> +{
> +     return -ENOENT;
> +}
> +
> +static void fprobe_ftrace_remove_ips(unsigned long *addrs, int num)
> +{
> +}
> +
> +static bool fprobe_is_ftrace(struct fprobe *fp)
> +{
> +     return false;
> +}
> +#endif
> +
> +/* fgraph_ops callback, this processes fprobes which have exit_handler. */
> +static int fprobe_fgraph_entry(struct ftrace_graph_ent *trace, struct 
> fgraph_ops *gops,
> +                            struct ftrace_regs *fregs)
>  {
>       unsigned long *fgraph_data = NULL;
>       unsigned long func = trace->func;
> @@ -292,7 +390,7 @@ static int fprobe_entry(struct ftrace_graph_ent *trace, 
> struct fgraph_ops *gops,
>                               if (node->addr != func)
>                                       continue;
>                               fp = READ_ONCE(node->fp);
> -                             if (fp && !fprobe_disabled(fp))
> +                             if (fp && !fprobe_disabled(fp) && 
> !fprobe_is_ftrace(fp))
>                                       fp->nmissed++;
>                       }
>                       return 0;
> @@ -312,7 +410,7 @@ static int fprobe_entry(struct ftrace_graph_ent *trace, 
> struct fgraph_ops *gops,
>               if (node->addr != func)
>                       continue;
>               fp = READ_ONCE(node->fp);
> -             if (!fp || fprobe_disabled(fp))
> +             if (unlikely(!fp || fprobe_disabled(fp) || 
> fprobe_is_ftrace(fp)))
>                       continue;
>  
>               data_size = fp->entry_data_size;
> @@ -340,7 +438,7 @@ static int fprobe_entry(struct ftrace_graph_ent *trace, 
> struct fgraph_ops *gops,
>       /* If any exit_handler is set, data must be used. */
>       return used != 0;
>  }
> -NOKPROBE_SYMBOL(fprobe_entry);
> +NOKPROBE_SYMBOL(fprobe_fgraph_entry);
>  
>  static void fprobe_return(struct ftrace_graph_ret *trace,
>                         struct fgraph_ops *gops,
> @@ -379,7 +477,7 @@ static void fprobe_return(struct ftrace_graph_ret *trace,
>  NOKPROBE_SYMBOL(fprobe_return);
>  
>  static struct fgraph_ops fprobe_graph_ops = {
> -     .entryfunc      = fprobe_entry,
> +     .entryfunc      = fprobe_fgraph_entry,
>       .retfunc        = fprobe_return,
>  };
>  static int fprobe_graph_active;
> @@ -498,9 +596,14 @@ static int fprobe_module_callback(struct notifier_block 
> *nb,
>       } while (node == ERR_PTR(-EAGAIN));
>       rhashtable_walk_exit(&iter);
>  
> -     if (alist.index > 0)
> +     if (alist.index > 0) {
>               ftrace_set_filter_ips(&fprobe_graph_ops.ops,
>                                     alist.addrs, alist.index, 1, 0);
> +#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
> +             ftrace_set_filter_ips(&fprobe_ftrace_ops,
> +                                   alist.addrs, alist.index, 1, 0);
> +#endif
> +     }
>       mutex_unlock(&fprobe_mutex);
>  
>       kfree(alist.addrs);
> @@ -733,7 +836,11 @@ int register_fprobe_ips(struct fprobe *fp, unsigned long 
> *addrs, int num)
>       mutex_lock(&fprobe_mutex);
>  
>       hlist_array = fp->hlist_array;
> -     ret = fprobe_graph_add_ips(addrs, num);
> +     if (fprobe_is_ftrace(fp))
> +             ret = fprobe_ftrace_add_ips(addrs, num);
> +     else
> +             ret = fprobe_graph_add_ips(addrs, num);
> +
>       if (!ret) {
>               add_fprobe_hash(fp);
>               for (i = 0; i < hlist_array->size; i++) {
> @@ -829,7 +936,10 @@ int unregister_fprobe(struct fprobe *fp)
>       }
>       del_fprobe_hash(fp);
>  
> -     fprobe_graph_remove_ips(addrs, count);
> +     if (fprobe_is_ftrace(fp))
> +             fprobe_ftrace_remove_ips(addrs, count);
> +     else
> +             fprobe_graph_remove_ips(addrs, count);
>  
>       kfree_rcu(hlist_array, rcu);
>       fp->hlist_array = NULL;
> -- 
> 2.51.0
> 


-- 
Masami Hiramatsu (Google) <[email protected]>

Reply via email to