Linux-Renesas-SoC Archive on lore.kernel.org
 help / color / Atom feed
From: Geert Uytterhoeven <geert+renesas@glider.be>
To: Michael Turquette <mturquette@baylibre.com>,
	Stephen Boyd <sboyd@kernel.org>
Cc: Yao Lihua <Lihua.Yao@desay-svautomotive.com>,
	Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>,
	Wolfram Sang <wsa+renesas@sang-engineering.com>,
	linux-clk@vger.kernel.org, linux-renesas-soc@vger.kernel.org,
	linux-kernel@vger.kernel.org,
	Geert Uytterhoeven <geert+renesas@glider.be>
Subject: [PATCH] clk: renesas: cpg-mssr: Fix reset control race condition
Date: Thu, 11 Jul 2019 15:03:59 +0200
Message-ID: <20190711130359.1060-1-geert+renesas@glider.be> (raw)

The module reset code in the Renesas CPG/MSSR driver uses
read-modify-write (RMW) operations to write to a Software Reset Register
(SRCRn), and simple writes to write to a Software Reset Clearing
Register (SRSTCLRn), as was mandated by the R-Car Gen2 and Gen3 Hardware
User's Manuals.

However, this may cause a race condition when two devices are reset in
parallel: if the reset for device A completes in the middle of the RMW
operation for device B, device A may be reset again, causing subtle
failures (e.g. i2c timeouts):

	thread A			thread B
	--------			--------

	val = SRCRn
	val |= bit A
	SRCRn = val

	delay

					val = SRCRn (bit A is set)

	SRSTCLRn = bit A
	(bit A in SRCRn is cleared)

					val |= bit B
					SRCRn = val (bit A and B are set)

This can be reproduced on e.g. Salvator-XS using:

    $ while true; do i2cdump -f -y 4 0x6A b > /dev/null; done &
    $ while true; do i2cdump -f -y 2 0x10 b > /dev/null; done &

    i2c-rcar e6510000.i2c: error -110 : 40000002
    i2c-rcar e66d8000.i2c: error -110 : 40000002

According to the R-Car Gen3 Hardware Manual Errata for Rev.
0.80 of Feb 28, 2018, reflected in Rev. 1.00 of the R-Car Gen3 Hardware
User's Manual, writes to SRCRn do not require read-modify-write cycles.

Note that the R-Car Gen2 Hardware User's Manual has not been updated
yet, and still says a read-modify-write sequence is required.  According
to the hardware team, the reset hardware block is the same on both R-Car
Gen2 and Gen3, though.

Hence fix the issue by replacing the read-modify-write operations on
SRCRn by simple writes.

Reported-by: Yao Lihua <Lihua.Yao@desay-svautomotive.com>
Fixes: 6197aa65c4905532 ("clk: renesas: cpg-mssr: Add support for reset control")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
So far I haven't been able to reproduce the issue on R-Car Gen2 (after
forcing i2c reset on Gen2, too).  Perhaps my Koelsch doesn't have enough
CPU cores.  What about Lager?

Hi Mike, Stephen,

As this is a bugfix, can you please take this directly, if accepted?

Thanks!
---
 drivers/clk/renesas/renesas-cpg-mssr.c | 16 ++--------------
 1 file changed, 2 insertions(+), 14 deletions(-)

diff --git a/drivers/clk/renesas/renesas-cpg-mssr.c b/drivers/clk/renesas/renesas-cpg-mssr.c
index 52bbb9ce3807db31..d4075b13067429cd 100644
--- a/drivers/clk/renesas/renesas-cpg-mssr.c
+++ b/drivers/clk/renesas/renesas-cpg-mssr.c
@@ -572,17 +572,11 @@ static int cpg_mssr_reset(struct reset_controller_dev *rcdev,
 	unsigned int reg = id / 32;
 	unsigned int bit = id % 32;
 	u32 bitmask = BIT(bit);
-	unsigned long flags;
-	u32 value;
 
 	dev_dbg(priv->dev, "reset %u%02u\n", reg, bit);
 
 	/* Reset module */
-	spin_lock_irqsave(&priv->rmw_lock, flags);
-	value = readl(priv->base + SRCR(reg));
-	value |= bitmask;
-	writel(value, priv->base + SRCR(reg));
-	spin_unlock_irqrestore(&priv->rmw_lock, flags);
+	writel(bitmask, priv->base + SRCR(reg));
 
 	/* Wait for at least one cycle of the RCLK clock (@ ca. 32 kHz) */
 	udelay(35);
@@ -599,16 +593,10 @@ static int cpg_mssr_assert(struct reset_controller_dev *rcdev, unsigned long id)
 	unsigned int reg = id / 32;
 	unsigned int bit = id % 32;
 	u32 bitmask = BIT(bit);
-	unsigned long flags;
-	u32 value;
 
 	dev_dbg(priv->dev, "assert %u%02u\n", reg, bit);
 
-	spin_lock_irqsave(&priv->rmw_lock, flags);
-	value = readl(priv->base + SRCR(reg));
-	value |= bitmask;
-	writel(value, priv->base + SRCR(reg));
-	spin_unlock_irqrestore(&priv->rmw_lock, flags);
+	writel(bitmask, priv->base + SRCR(reg));
 	return 0;
 }
 
-- 
2.17.1


             reply index

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-07-11 13:03 Geert Uytterhoeven [this message]
2019-07-12  9:19 ` Yoshihiro Shimoda

Reply instructions:

You may reply publically 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=20190711130359.1060-1-geert+renesas@glider.be \
    --to=geert+renesas@glider.be \
    --cc=Lihua.Yao@desay-svautomotive.com \
    --cc=linux-clk@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-renesas-soc@vger.kernel.org \
    --cc=mturquette@baylibre.com \
    --cc=sboyd@kernel.org \
    --cc=wsa+renesas@sang-engineering.com \
    --cc=yoshihiro.shimoda.uh@renesas.com \
    /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

Linux-Renesas-SoC Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-renesas-soc/0 linux-renesas-soc/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-renesas-soc linux-renesas-soc/ https://lore.kernel.org/linux-renesas-soc \
		linux-renesas-soc@vger.kernel.org linux-renesas-soc@archiver.kernel.org
	public-inbox-index linux-renesas-soc


Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-renesas-soc


AGPL code for this site: git clone https://public-inbox.org/ public-inbox