Source-Changes-HG archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
[src/trunk]: src/sys/arch/arm/sunxi Just skipping sunxi_can_rx_intr() if the ...
details: https://anonhg.NetBSD.org/src/rev/ec41d5283827
branches: trunk
changeset: 370627:ec41d5283827
user: bouyer <bouyer%NetBSD.org@localhost>
date: Wed Sep 21 20:21:16 2022 +0000
description:
Just skipping sunxi_can_rx_intr() if the DATA_OR flag is set isn't enough
to properly recover from overrrun in all case. So go the linux way and reset
the hardware.
Don't write SUNXI_CAN_INT_RX_FLAG to SUNXI_CAN_INT_REG, this could race
with hardware and clear the interrupt while there are new packets received.
SUNXI_CAN_INT_RX_FLAG clears automatically when all pending packets have been
read, so when no more packets are pending just read SUNXI_CAN_INT_REG again
and process other interrupts, if any (or RX if there are new packets pending).
With this change it seems I get overruns less often in my use case.
diffstat:
sys/arch/arm/sunxi/sunxi_can.c | 26 +++++++++++++++++++-------
1 files changed, 19 insertions(+), 7 deletions(-)
diffs (74 lines):
diff -r 53646cf61318 -r ec41d5283827 sys/arch/arm/sunxi/sunxi_can.c
--- a/sys/arch/arm/sunxi/sunxi_can.c Wed Sep 21 18:55:21 2022 +0000
+++ b/sys/arch/arm/sunxi/sunxi_can.c Wed Sep 21 20:21:16 2022 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: sunxi_can.c,v 1.10 2022/09/19 11:21:36 bouyer Exp $ */
+/* $NetBSD: sunxi_can.c,v 1.11 2022/09/21 20:21:16 bouyer Exp $ */
/*-
* Copyright (c) 2017,2018 The NetBSD Foundation, Inc.
@@ -36,7 +36,7 @@
#include <sys/cdefs.h>
-__KERNEL_RCSID(1, "$NetBSD: sunxi_can.c,v 1.10 2022/09/19 11:21:36 bouyer Exp $");
+__KERNEL_RCSID(1, "$NetBSD: sunxi_can.c,v 1.11 2022/09/21 20:21:16 bouyer Exp $");
#include <sys/param.h>
#include <sys/bus.h>
@@ -101,6 +101,8 @@
static void sunxi_can_enter_reset(struct sunxi_can_softc *);
static void sunxi_can_exit_reset(struct sunxi_can_softc *);
+static void sunxi_can_ifdown(struct sunxi_can_softc * const);
+static int sunxi_can_ifup(struct sunxi_can_softc * const);
CFATTACH_DECL_NEW(sunxi_can, sizeof(struct sunxi_can_softc),
sunxi_can_match, sunxi_can_attach, NULL, NULL);
@@ -344,7 +346,9 @@
if (irq & SUNXI_CAN_INT_DATA_OR) {
if_statinc(ifp, if_ierrors);
+ sunxi_can_ifdown(sc);
sunxi_can_write(sc, SUNXI_CAN_CMD_REG, SUNXI_CAN_CMD_CLR_OR);
+ sunxi_can_ifup(sc);
}
if (irq & SUNXI_CAN_INT_ERR) {
reg = sunxi_can_read(sc, SUNXI_CAN_REC_REG);
@@ -383,23 +387,31 @@
while ((irq = sunxi_can_read(sc, SUNXI_CAN_INT_REG)) != 0) {
uint32_t sts = sunxi_can_read(sc, SUNXI_CAN_STA_REG);
rv = 1;
+ rnd_add_uint32(&sc->sc_rnd_source, irq);
- if (irq & SUNXI_CAN_INT_TX_FLAG) {
- sunxi_can_tx_intr(sc);
- }
if ((irq & (SUNXI_CAN_INT_RX_FLAG | SUNXI_CAN_INT_DATA_OR)) ==
SUNXI_CAN_INT_RX_FLAG) {
while (sts & SUNXI_CAN_STA_RX_RDY) {
sunxi_can_rx_intr(sc);
sts = sunxi_can_read(sc, SUNXI_CAN_STA_REG);
}
+ /*
+ * Don't write SUNXI_CAN_INT_RX_FLAG to the interrupt
+ * register, this may clear the RX pending flag
+ * while there is indeed a packet pending.
+ * Reading packets should have cleared the RX interrupt,
+ * so just restart the loop and re-read the interrupt
+ * register. In the common case irq will now be 0.
+ */
+ continue;
+ }
+ if (irq & SUNXI_CAN_INT_TX_FLAG) {
+ sunxi_can_tx_intr(sc);
}
if (irq & SUNXI_CAN_INT_ALLERRS) {
sunxi_can_err_intr(sc, irq, sts);
}
sunxi_can_write(sc, SUNXI_CAN_INT_REG, irq);
- rnd_add_uint32(&sc->sc_rnd_source, irq);
-
}
mutex_exit(&sc->sc_intr_lock);
Home |
Main Index |
Thread Index |
Old Index