Re: [RESEND] mtd: spi-nor: Fix Cadence QSPI RCU Schedule Stall

From: Tudor.Ambarus
Date: Wed Aug 21 2019 - 04:22:27 EST




On 08/16/2019 01:55 AM, thor.thayer@xxxxxxxxxxxxxxx wrote:
> External E-Mail
>
>
> From: Thor Thayer <thor.thayer@xxxxxxxxxxxxxxx>
>
> The current Cadence QSPI driver sometimes caused a
> "rcu_sched self-detected stall" while writing large files.
>
> Stall Report:
> '# mtd_debug write /dev/mtd1 0 48816464 blob.img
> [ 1815.454227] rcu: INFO: rcu_sched self-detected stall on CPU
> [ 1815.459789] rcu: 0-....: (2099 ticks this GP) idle=8c6/1/0x40000002
> softirq=6492/6492 fqs=935
> [ 1815.468442] rcu: (t=2100 jiffies g=8749 q=247)
> <snip> (abbreviated backtrace)
> [ 1815.772086] [<c05a3ea0>] (cqspi_exec_flash_cmd) (cqspi_read_reg)
> [ 1815.786203] [<c05a5488>] (cqspi_read_reg) from (read_sr)
> [ 1815.803790] [<c05a0330>] (read_sr) from
> (spi_nor_wait_till_ready_with_timeout)
> [ 1815.816610] [<c05a182c>] (spi_nor_wait_till_ready_with_timeout) from
> (spi_nor_write+0x104/0x1d0)
> [ 1815.836791] [<c05a1a44>] (spi_nor_write) from (part_write+0x50/0x58)
> <snip>
> [ 1815.997961] cadence-qspi ff809000.spi: Flash command execution timed out.
> [ 1816.004733] error -110 reading SR
> file_to_flash: write, size 0x2e8e150, n 0x2e8e150
> write(): Connection timed out
>
> This was caused by a tight loop in cqspi_wait_for_bit(). Fix by using
> readl_relaxed_poll_timeout() which sleeps 10us while polling a register.
>
> Fit onto 80 character line by truncating the bool clear parameter
>
> Fixes: 140623410536 ("mtd: spi-nor: Add driver for Cadence Quad SPI Flash Controller")
> Signed-off-by: Thor Thayer <thor.thayer@xxxxxxxxxxxxxxx>
> ---
> drivers/mtd/spi-nor/cadence-quadspi.c | 19 +++++--------------
> 1 file changed, 5 insertions(+), 14 deletions(-)
>


Since the bug was not introduced in the previous release and we are quite late
for mtd/fixes,

Applied to https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git,
spi-nor/next branch.

Thanks,
ta