diff mbox series

[06/15] bpf: Add bpf_ksym_tree tree

Message ID 20200302143154.258569-7-jolsa@kernel.org
State Changes Requested
Delegated to: BPF Maintainers
Headers show
Series bpf: Add trampoline and dispatcher to /proc/kallsyms | expand

Commit Message

Jiri Olsa March 2, 2020, 2:31 p.m. UTC
The bpf_tree is used both for kallsyms iterations and searching
for exception tables of bpf programs, which is needed only for
bpf programs.

Adding bpf_ksym_tree that will hold symbols for all bpf_prog
bpf_trampoline and bpf_dispatcher objects and keeping bpf_tree
only for bpf_prog objects to keep it fast.

Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 include/linux/bpf.h |  1 +
 kernel/bpf/core.c   | 56 ++++++++++++++++++++++++++++++++++++++++-----
 2 files changed, 51 insertions(+), 6 deletions(-)

Comments

Alexei Starovoitov March 3, 2020, 6:03 p.m. UTC | #1
On Mon, Mar 02, 2020 at 03:31:45PM +0100, Jiri Olsa wrote:
> The bpf_tree is used both for kallsyms iterations and searching
> for exception tables of bpf programs, which is needed only for
> bpf programs.
> 
> Adding bpf_ksym_tree that will hold symbols for all bpf_prog
> bpf_trampoline and bpf_dispatcher objects and keeping bpf_tree
> only for bpf_prog objects to keep it fast.

...

>  static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
> @@ -616,6 +650,7 @@ static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
>  	WARN_ON_ONCE(!list_empty(&aux->ksym.lnode));
>  	list_add_tail_rcu(&aux->ksym.lnode, &bpf_kallsyms);
>  	latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
> +	latch_tree_insert(&aux->ksym.tnode, &bpf_ksym_tree, &bpf_ksym_tree_ops);
>  }
>  
>  static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
> @@ -624,6 +659,7 @@ static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
>  		return;
>  
>  	latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
> +	latch_tree_erase(&aux->ksym.tnode, &bpf_ksym_tree, &bpf_ksym_tree_ops);

I have to agree with Daniel here.
Having bpf prog in two latch trees is unnecessary.
Especially looking at the patch 7 that moves update to the other tree.
The whole thing becomes assymetrical and harder to follow.
Consider that walking extable is slow anyway. It's a page fault.
Having trampoline and dispatch in the same tree will not be measurable
on the speed of search_bpf_extables->bpf_prog_kallsyms_find.
So please consolidate.

Also I don't see a hunk that deletes tnode from 'struct bpf_image'.
These patches suppose to generalize it too, no?
And at the end kernel_text_address() suppose to call
is_bpf_text_address() only, right?
Instead of is_bpf_text_address() || is_bpf_image_address() ?
That _will_ actually speed up backtrace collection.
Jiri Olsa March 3, 2020, 8:12 p.m. UTC | #2
On Tue, Mar 03, 2020 at 10:03:19AM -0800, Alexei Starovoitov wrote:
> On Mon, Mar 02, 2020 at 03:31:45PM +0100, Jiri Olsa wrote:
> > The bpf_tree is used both for kallsyms iterations and searching
> > for exception tables of bpf programs, which is needed only for
> > bpf programs.
> > 
> > Adding bpf_ksym_tree that will hold symbols for all bpf_prog
> > bpf_trampoline and bpf_dispatcher objects and keeping bpf_tree
> > only for bpf_prog objects to keep it fast.
> 
> ...
> 
> >  static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
> > @@ -616,6 +650,7 @@ static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
> >  	WARN_ON_ONCE(!list_empty(&aux->ksym.lnode));
> >  	list_add_tail_rcu(&aux->ksym.lnode, &bpf_kallsyms);
> >  	latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
> > +	latch_tree_insert(&aux->ksym.tnode, &bpf_ksym_tree, &bpf_ksym_tree_ops);
> >  }
> >  
> >  static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
> > @@ -624,6 +659,7 @@ static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
> >  		return;
> >  
> >  	latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
> > +	latch_tree_erase(&aux->ksym.tnode, &bpf_ksym_tree, &bpf_ksym_tree_ops);
> 
> I have to agree with Daniel here.
> Having bpf prog in two latch trees is unnecessary.
> Especially looking at the patch 7 that moves update to the other tree.
> The whole thing becomes assymetrical and harder to follow.
> Consider that walking extable is slow anyway. It's a page fault.
> Having trampoline and dispatch in the same tree will not be measurable
> on the speed of search_bpf_extables->bpf_prog_kallsyms_find.
> So please consolidate.

ok

> 
> Also I don't see a hunk that deletes tnode from 'struct bpf_image'.
> These patches suppose to generalize it too, no?

__bpf_ksym_del function added in patch:

    bpf: Separate kallsyms add/del functions

> And at the end kernel_text_address() suppose to call
> is_bpf_text_address() only, right?
> Instead of is_bpf_text_address() || is_bpf_image_address() ?
> That _will_ actually speed up backtrace collection.

right, this one could have already used just the ksym tree

will send new version.. meanwhile I was checking struct_ops,
so will include kallsyms support them as well

thanks,
jirka
diff mbox series

Patch

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index f1174d24c185..5d6649cdc3df 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -468,6 +468,7 @@  struct bpf_ksym {
 	unsigned long		 end;
 	char			 name[KSYM_NAME_LEN];
 	struct list_head	 lnode;
+	struct latch_tree_node	 tnode;
 };
 
 enum bpf_tramp_prog_type {
diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index 084abfbc3362..c36558c44637 100644
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c
@@ -607,8 +607,42 @@  static const struct latch_tree_ops bpf_tree_ops = {
 	.comp	= bpf_tree_comp,
 };
 
+static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
+{
+	return container_of(n, struct bpf_ksym, tnode)->start;
+}
+
+static bool
+bpf_ksym_tree_less(struct latch_tree_node *a,
+		   struct latch_tree_node *b)
+{
+	return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
+}
+
+static int
+bpf_ksym_tree_comp(void *key, struct latch_tree_node *n)
+{
+	unsigned long val = (unsigned long)key;
+	const struct bpf_ksym *ksym;
+
+	ksym = container_of(n, struct bpf_ksym, tnode);
+
+	if (val < ksym->start)
+		return -1;
+	if (val >= ksym->end)
+		return  1;
+
+	return 0;
+}
+
+static const struct latch_tree_ops bpf_ksym_tree_ops = {
+	.less	= bpf_ksym_tree_less,
+	.comp	= bpf_ksym_tree_comp,
+};
+
 static DEFINE_SPINLOCK(bpf_lock);
 static LIST_HEAD(bpf_kallsyms);
+static struct latch_tree_root bpf_ksym_tree __cacheline_aligned;
 static struct latch_tree_root bpf_tree __cacheline_aligned;
 
 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
@@ -616,6 +650,7 @@  static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
 	WARN_ON_ONCE(!list_empty(&aux->ksym.lnode));
 	list_add_tail_rcu(&aux->ksym.lnode, &bpf_kallsyms);
 	latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
+	latch_tree_insert(&aux->ksym.tnode, &bpf_ksym_tree, &bpf_ksym_tree_ops);
 }
 
 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
@@ -624,6 +659,7 @@  static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
 		return;
 
 	latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
+	latch_tree_erase(&aux->ksym.tnode, &bpf_ksym_tree, &bpf_ksym_tree_ops);
 	list_del_rcu(&aux->ksym.lnode);
 }
 
@@ -672,19 +708,27 @@  static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
 	       NULL;
 }
 
+static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
+{
+	struct latch_tree_node *n;
+
+	n = latch_tree_find((void *)addr, &bpf_ksym_tree, &bpf_ksym_tree_ops);
+	return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
+}
+
 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
 				 unsigned long *off, char *sym)
 {
-	struct bpf_prog *prog;
+	struct bpf_ksym *ksym;
 	char *ret = NULL;
 
 	rcu_read_lock();
-	prog = bpf_prog_kallsyms_find(addr);
-	if (prog) {
-		unsigned long symbol_start = prog->aux->ksym.start;
-		unsigned long symbol_end = prog->aux->ksym.end;
+	ksym = bpf_ksym_find(addr);
+	if (ksym) {
+		unsigned long symbol_start = ksym->start;
+		unsigned long symbol_end = ksym->end;
 
-		strncpy(sym, prog->aux->ksym.name, KSYM_NAME_LEN);
+		strncpy(sym, ksym->name, KSYM_NAME_LEN);
 
 		ret = sym;
 		if (size)