From: Florian Westphal <fw@strlen.de>
To: <netfilter-devel@vger.kernel.org>
Cc: Florian Westphal <fw@strlen.de>, Eric Dumazet <eric.dumazet@gmail.com>
Subject: [PATCH nf-next 1/3] netfilter: nf_tables: add static key to skip retpoline workarounds
Date: Tue, 3 Jan 2023 13:47:15 +0100 [thread overview]
Message-ID: <20230103124717.8178-2-fw@strlen.de> (raw)
In-Reply-To: <20230103124717.8178-1-fw@strlen.de>
If CONFIG_RETPOLINE is enabled nf_tables avoids indirect calls for
builtin expressions.
On newer cpus indirect calls do not go through the retpoline thunk
anymore, even for RETPOLINE=y builds.
Just like with the new tc retpoline wrappers:
Add a static key to skip the if / else if cascade if the cpu
does not require retpolines.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
---
net/netfilter/nf_tables_core.c | 30 +++++++++++++++++++++++++++++-
1 file changed, 29 insertions(+), 1 deletion(-)
diff --git a/net/netfilter/nf_tables_core.c b/net/netfilter/nf_tables_core.c
index 709a736c301c..0f26d002d8b3 100644
--- a/net/netfilter/nf_tables_core.c
+++ b/net/netfilter/nf_tables_core.c
@@ -21,6 +21,26 @@
#include <net/netfilter/nf_log.h>
#include <net/netfilter/nft_meta.h>
+#if defined(CONFIG_RETPOLINE) && defined(CONFIG_X86)
+
+static struct static_key_false nf_tables_skip_direct_calls;
+
+static bool nf_skip_indirect_calls(void)
+{
+ return static_branch_likely(&nf_tables_skip_direct_calls);
+}
+
+static void __init nf_skip_indirect_calls_enable(void)
+{
+ if (!cpu_feature_enabled(X86_FEATURE_RETPOLINE))
+ static_branch_enable(&nf_tables_skip_direct_calls);
+}
+#else
+static inline bool nf_skip_indirect_calls(void) { return false; }
+
+static inline void nf_skip_indirect_calls_enable(void) { }
+#endif
+
static noinline void __nft_trace_packet(struct nft_traceinfo *info,
const struct nft_chain *chain,
enum nft_trace_types type)
@@ -193,7 +213,12 @@ static void expr_call_ops_eval(const struct nft_expr *expr,
struct nft_pktinfo *pkt)
{
#ifdef CONFIG_RETPOLINE
- unsigned long e = (unsigned long)expr->ops->eval;
+ unsigned long e;
+
+ if (nf_skip_indirect_calls())
+ goto indirect_call;
+
+ e = (unsigned long)expr->ops->eval;
#define X(e, fun) \
do { if ((e) == (unsigned long)(fun)) \
return fun(expr, regs, pkt); } while (0)
@@ -210,6 +235,7 @@ static void expr_call_ops_eval(const struct nft_expr *expr,
X(e, nft_rt_get_eval);
X(e, nft_bitwise_eval);
#undef X
+indirect_call:
#endif /* CONFIG_RETPOLINE */
expr->ops->eval(expr, regs, pkt);
}
@@ -369,6 +395,8 @@ int __init nf_tables_core_module_init(void)
goto err;
}
+ nf_skip_indirect_calls_enable();
+
return 0;
err:
--
2.38.2
next prev parent reply other threads:[~2023-01-03 12:47 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-03 12:47 [PATCH nf-next 0/3] netfilter: nf_tables: extend retpoline workarounds Florian Westphal
2023-01-03 12:47 ` Florian Westphal [this message]
2023-01-03 12:47 ` [PATCH nf-next 2/3] netfilter: nf_tables: avoid retpoline overhead for objref calls Florian Westphal
2023-01-03 12:47 ` [PATCH nf-next 3/3] netfilter: nf_tables: avoid retpoline overhead for some ct expression calls Florian Westphal
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230103124717.8178-2-fw@strlen.de \
--to=fw@strlen.de \
--cc=eric.dumazet@gmail.com \
--cc=netfilter-devel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).