The per-CPU variables used during bpf_prog_run_xdp() invocation and
later during xdp_do_redirect() rely on disabled BH for their protection.
Without locking in local_bh_disable() on PREEMPT_RT these data structure
require explicit locking.

This is a follow-up on the previous change which introduced
bpf_run_lock.redirect_lock and uses it now within drivers.

The simple way is to acquire the lock before bpf_prog_run_xdp() is
invoked and hold it until the end of function.
This does not always work because some drivers (cpsw, atlantic) invoke
xdp_do_flush() in the same context.
Acquiring the lock in bpf_prog_run_xdp() and dropping in
xdp_do_redirect() (without touching drivers) does not work because not
all driver, which use bpf_prog_run_xdp(), do support XDP_REDIRECT (and
invoke xdp_do_redirect()).

Ideally the minimal locking scope would be bpf_prog_run_xdp() +
xdp_do_redirect() and everything else (error recovery, DMA unmapping,
free/ alloc of memory, …) would happen outside of the locked section.

Cc: Alexandre Torgue <alexandre.torgue@foss.st.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Ariel Elior <redacted>
Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Cc: Jassi Brar <redacted>
Cc: Jesper Dangaard Brouer <hawk@kernel.org>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jose Abreu <redacted>
Cc: Manish Chopra <manishc@marvell.com>
Cc: Maxime Coquelin <mcoquelin.stm32@gmail.com>
Cc: Ravi Gunasekaran <redacted>
Cc: Roger Quadros <rogerq@kernel.org>
Cc: Siddharth Vadapalli <s-vadapalli@ti.com>
Cc: bpf@vger.kernel.org
Cc: linux-omap@vger.kernel.org
Cc: linux-stm32@st-md-mailman.stormreply.com
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/net/ethernet/qlogic/qede/qede_fp.c        |  1 +
 drivers/net/ethernet/socionext/netsec.c           |  1 +
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c |  1 +
 drivers/net/ethernet/ti/cpsw_priv.c               | 15 +++++++++------
 4 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/qlogic/qede/qede_fp.c b/drivers/net/ethernet/qlogic/qede/qede_fp.c
index cb1746bc0e0c5..ce5af094fb817 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_fp.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_fp.c

@@ -1091,6 +1091,7 @@ static bool qede_rx_xdp(struct qede_dev *edev,
 	xdp_prepare_buff(&xdp, page_address(bd->data), *data_offset,
 			 *len, false);
 
+	guard(local_lock_nested_bh)(&bpf_run_lock.redirect_lock);
 	act = bpf_prog_run_xdp(prog, &xdp);
 
 	/* Recalculate, as XDP might have changed the headers */

diff --git a/drivers/net/ethernet/socionext/netsec.c b/drivers/net/ethernet/socionext/netsec.c
index 0891e9e49ecb5..47e314338f3f3 100644
--- a/drivers/net/ethernet/socionext/netsec.c
+++ b/drivers/net/ethernet/socionext/netsec.c

@@ -905,6 +905,7 @@ static u32 netsec_run_xdp(struct netsec_priv *priv, struct bpf_prog *prog,
 	int err;
 	u32 act;
 
+	guard(local_lock_nested_bh)(&bpf_run_lock.redirect_lock);
 	act = bpf_prog_run_xdp(prog, xdp);
 
 	/* Due xdp_adjust_tail: DMA sync for_device cover max len CPU touch */

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index 37e64283f9107..9e92affc8c22c 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c

@@ -4893,6 +4893,7 @@ static int __stmmac_xdp_run_prog(struct stmmac_priv *priv,
 	u32 act;
 	int res;
 
+	guard(local_lock_nested_bh)(&bpf_run_lock.redirect_lock);
 	act = bpf_prog_run_xdp(prog, xdp);
 	switch (act) {
 	case XDP_PASS:

diff --git a/drivers/net/ethernet/ti/cpsw_priv.c b/drivers/net/ethernet/ti/cpsw_priv.c
index 764ed298b5708..f38c49f9fab35 100644
--- a/drivers/net/ethernet/ti/cpsw_priv.c
+++ b/drivers/net/ethernet/ti/cpsw_priv.c

@@ -1335,9 +1335,15 @@ int cpsw_run_xdp(struct cpsw_priv *priv, int ch, struct xdp_buff *xdp,
 	if (!prog)
 		return CPSW_XDP_PASS;
 
-	act = bpf_prog_run_xdp(prog, xdp);
-	/* XDP prog might have changed packet data and boundaries */
-	*len = xdp->data_end - xdp->data;
+	scoped_guard(local_lock_nested_bh, &bpf_run_lock.redirect_lock) {
+		act = bpf_prog_run_xdp(prog, xdp);
+		/* XDP prog might have changed packet data and boundaries */
+		*len = xdp->data_end - xdp->data;
+		if (act == XDP_REDIRECT) {
+			if (xdp_do_redirect(ndev, xdp, prog))
+				goto drop;
+		}
+	}
 
 	switch (act) {
 	case XDP_PASS:

@@ -1352,9 +1358,6 @@ int cpsw_run_xdp(struct cpsw_priv *priv, int ch, struct xdp_buff *xdp,
 			xdp_return_frame_rx_napi(xdpf);
 		break;
 	case XDP_REDIRECT:
-		if (xdp_do_redirect(ndev, xdp, prog))
-			goto drop;
-
 		/*  Have to flush here, per packet, instead of doing it in bulk
 		 *  at the end of the napi handler. The RX devices on this
 		 *  particular hardware is sharing a common queue, so the

-- 
2.43.0