mirror of
https://github.com/openwrt/openwrt.git
synced 2024-12-20 22:23:27 +00:00
d3b47909b1
The nand driver normally while waiting for the device to become ready; this is normally fine, but xway_nand holds the ebu_lock spinlock, and this can cause lockups if other threads which use ebu_lock are interleaved. Fix this by waiting instead of polling. This mainly showed up as crashes in ath9k_pci_owl_loader (see https://github.com/openwrt/openwrt/issues/9829 ), but turning on spinlock debugging shows this happening in other places too. This doesn't seem to measurably impact boot time. Tested on bt_homehub-v5a with 5.10 and 5.15. Signed-off-by: Thomas Nixon <tom@tomn.co.uk> [Add commit description into patch] Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
39 lines
1.4 KiB
Diff
39 lines
1.4 KiB
Diff
From 416f25a948d11ef15733f2e31658d31b5cc7bef6 Mon Sep 17 00:00:00 2001
|
|
From: Thomas Nixon <tom@tomn.co.uk>
|
|
Date: Sun, 26 Mar 2023 11:08:49 +0100
|
|
Subject: [PATCH] mtd: rawnand: xway: don't yield while holding spinlock
|
|
|
|
The nand driver normally while waiting for the device to become ready;
|
|
this is normally fine, but xway_nand holds the ebu_lock spinlock, and
|
|
this can cause lockups if other threads which use ebu_lock are
|
|
interleaved. Fix this by waiting instead of polling.
|
|
|
|
This mainly showed up as crashes in ath9k_pci_owl_loader (see
|
|
https://github.com/openwrt/openwrt/issues/9829 ), but turning on
|
|
spinlock debugging shows this happening in other places too.
|
|
|
|
This doesn't seem to measurably impact boot time.
|
|
|
|
Signed-off-by: Thomas Nixon <tom@tomn.co.uk>
|
|
---
|
|
drivers/mtd/nand/raw/xway_nand.c | 8 +++++++-
|
|
1 file changed, 7 insertions(+), 1 deletion(-)
|
|
|
|
--- a/drivers/mtd/nand/raw/xway_nand.c
|
|
+++ b/drivers/mtd/nand/raw/xway_nand.c
|
|
@@ -175,7 +175,13 @@ static void xway_cmd_ctrl(struct nand_ch
|
|
|
|
static int xway_dev_ready(struct nand_chip *chip)
|
|
{
|
|
- return ltq_ebu_r32(EBU_NAND_WAIT) & NAND_WAIT_RD;
|
|
+ /*
|
|
+ * wait until ready, as otherwise the driver will yield in nand_wait or
|
|
+ * nand_wait_ready, which is a bad idea when we're holding ebu_lock
|
|
+ */
|
|
+ while ((ltq_ebu_r32(EBU_NAND_WAIT) & NAND_WAIT_RD) == 0)
|
|
+ cpu_relax();
|
|
+ return 1;
|
|
}
|
|
|
|
static unsigned char xway_read_byte(struct nand_chip *chip)
|