linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [RESEND PATCH v2 1/2] serial: samsung: Fix ERR pointer dereference on deferred probe
@ 2016-06-16  6:27 Krzysztof Kozlowski
  2016-06-16  6:27 ` [RESEND PATCH v2 2/2] serial: samsung: Fix possible out of bounds access on non-DT platform Krzysztof Kozlowski
  0 siblings, 1 reply; 2+ messages in thread
From: Krzysztof Kozlowski @ 2016-06-16  6:27 UTC (permalink / raw)
  To: Greg Kroah-Hartman, Jiri Slaby, linux-serial, linux-kernel
  Cc: Kevin Hilman, Javier Martinez Canillas, arnd,
	Krzysztof Kozlowski, Bartlomiej Zolnierkiewicz, stable

When the clk_get() of "uart" clock returns EPROBE_DEFER, the next re-probe
finishes with success but uses invalid (ERR_PTR) values.  This leads to
dereferencing of ERR_PTR stored under ourport->clk:

	12c30000.serial: Controller clock not found
	(...)
	12c30000.serial: ttySAC3 at MMIO 0x12c30000 (irq = 61, base_baud = 0) is a S3C6400/10
	Unable to handle kernel paging request at virtual address fffffdfb

	(clk_prepare) from [<c039f7d0>] (s3c24xx_serial_pm+0x20/0x128)
	(s3c24xx_serial_pm) from [<c0395414>] (uart_change_pm+0x38/0x40)
	(uart_change_pm) from [<c039689c>] (uart_add_one_port+0x31c/0x44c)
	(uart_add_one_port) from [<c03a035c>] (s3c24xx_serial_probe+0x2a8/0x418)
	(s3c24xx_serial_probe) from [<c03ee110>] (platform_drv_probe+0x50/0xb0)
	(platform_drv_probe) from [<c03ecb44>] (driver_probe_device+0x1f4/0x2b0)
	(driver_probe_device) from [<c03eb0c0>] (bus_for_each_drv+0x44/0x8c)
	(bus_for_each_drv) from [<c03ec8c8>] (__device_attach+0x9c/0x100)
	(__device_attach) from [<c03ebf54>] (bus_probe_device+0x84/0x8c)
	(bus_probe_device) from [<c03ec388>] (deferred_probe_work_func+0x60/0x8c)
	(deferred_probe_work_func) from [<c012fee4>] (process_one_work+0x120/0x328)
	(process_one_work) from [<c0130150>] (worker_thread+0x2c/0x4ac)
	(worker_thread) from [<c0135320>] (kthread+0xd8/0xf4)
	(kthread) from [<c0107978>] (ret_from_fork+0x14/0x3c)

The first unsuccessful clk_get() causes s3c24xx_serial_init_port() to
exit with failure but the s3c24xx_uart_port is left half-configured
(e.g. port->mapbase is set, clk contains ERR_PTR).  On next re-probe,
the function s3c24xx_serial_init_port() will exit early with success
because of configured port->mapbase and driver will use old values,
including the ERR_PTR as clock.

Fix this by cleaning the port->mapbase on error path so each re-probe
will initialize all of the port settings.

Fixes: 60e93575476f ("serial: samsung: enable clock before clearing pending interrupts during init")
Cc: <stable@vger.kernel.org>
Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Reviewed-by: Javier Martinez Canillas <javier@osg.samsung.com>
Tested-by: Javier Martinez Canillas <javier@osg.samsung.com>
Tested-by: Kevin Hilman <khilman@baylibre.com>
---

Changes since v1:
1. Apply suggestion from Bartlomiej (offline) that early check for
   port->mapbase in s3c24xx_serial_init_port() should fail if it is
   already set.  The port->mapbase is now always cleared when probe
   fails, so re-probing with 'port->mapbase != 0' is unexpected.
---
 drivers/tty/serial/samsung.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

diff --git a/drivers/tty/serial/samsung.c b/drivers/tty/serial/samsung.c
index 99bb23161dd6..f0bd2ec0db59 100644
--- a/drivers/tty/serial/samsung.c
+++ b/drivers/tty/serial/samsung.c
@@ -1684,7 +1684,7 @@ static int s3c24xx_serial_init_port(struct s3c24xx_uart_port *ourport,
 		return -ENODEV;
 
 	if (port->mapbase != 0)
-		return 0;
+		return -EINVAL;
 
 	/* setup info for port */
 	port->dev	= &platdev->dev;
@@ -1738,22 +1738,25 @@ static int s3c24xx_serial_init_port(struct s3c24xx_uart_port *ourport,
 		ourport->dma = devm_kzalloc(port->dev,
 					    sizeof(*ourport->dma),
 					    GFP_KERNEL);
-		if (!ourport->dma)
-			return -ENOMEM;
+		if (!ourport->dma) {
+			ret = -ENOMEM;
+			goto err;
+		}
 	}
 
 	ourport->clk	= clk_get(&platdev->dev, "uart");
 	if (IS_ERR(ourport->clk)) {
 		pr_err("%s: Controller clock not found\n",
 				dev_name(&platdev->dev));
-		return PTR_ERR(ourport->clk);
+		ret = PTR_ERR(ourport->clk);
+		goto err;
 	}
 
 	ret = clk_prepare_enable(ourport->clk);
 	if (ret) {
 		pr_err("uart: clock failed to prepare+enable: %d\n", ret);
 		clk_put(ourport->clk);
-		return ret;
+		goto err;
 	}
 
 	/* Keep all interrupts masked and cleared */
@@ -1769,7 +1772,12 @@ static int s3c24xx_serial_init_port(struct s3c24xx_uart_port *ourport,
 
 	/* reset the fifos (and setup the uart) */
 	s3c24xx_serial_resetport(port, cfg);
+
 	return 0;
+
+err:
+	port->mapbase = 0;
+	return ret;
 }
 
 /* Device driver serial port probe */
-- 
1.9.1

^ permalink raw reply	[flat|nested] 2+ messages in thread

* [RESEND PATCH v2 2/2] serial: samsung: Fix possible out of bounds access on non-DT platform
  2016-06-16  6:27 [RESEND PATCH v2 1/2] serial: samsung: Fix ERR pointer dereference on deferred probe Krzysztof Kozlowski
@ 2016-06-16  6:27 ` Krzysztof Kozlowski
  0 siblings, 0 replies; 2+ messages in thread
From: Krzysztof Kozlowski @ 2016-06-16  6:27 UTC (permalink / raw)
  To: Greg Kroah-Hartman, Jiri Slaby, linux-serial, linux-kernel
  Cc: Kevin Hilman, Javier Martinez Canillas, arnd,
	Krzysztof Kozlowski, Bartlomiej Zolnierkiewicz

On non-DeviceTree platforms, the index of serial device is a static
variable incremented on each probe.  It is incremented even if deferred
probe happens when getting the clock in s3c24xx_serial_init_port().

This index is used for referencing elements of statically allocated
s3c24xx_serial_ports array.  In case of re-probe, the index will point
outside of this array leading to memory corruption.

Increment the index only on successful probe.

Reported-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Fixes: b497549a035c ("[ARM] S3C24XX: Split serial driver into core and per-cpu drivers")
Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>

---

Not cc-ing stable because:
1. It is just a possibility, not really reproduced (I don't have non-DT
   platform).
2. I am not sure whether deferred probe may happen on non-DT platform.

Changes since v1:
1. New patch.
---
 drivers/tty/serial/samsung.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/serial/samsung.c b/drivers/tty/serial/samsung.c
index f0bd2ec0db59..4d2924f61e0b 100644
--- a/drivers/tty/serial/samsung.c
+++ b/drivers/tty/serial/samsung.c
@@ -1844,8 +1844,6 @@ static int s3c24xx_serial_probe(struct platform_device *pdev)
 	ourport->min_dma_size = max_t(int, ourport->port.fifosize,
 				    dma_get_cache_alignment());
 
-	probe_index++;
-
 	dbg("%s: initialising port %p...\n", __func__, ourport);
 
 	ret = s3c24xx_serial_init_port(ourport, pdev);
@@ -1875,6 +1873,8 @@ static int s3c24xx_serial_probe(struct platform_device *pdev)
 	if (ret < 0)
 		dev_err(&pdev->dev, "failed to add cpufreq notifier\n");
 
+	probe_index++;
+
 	return 0;
 }
 
-- 
1.9.1

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2016-06-16  6:27 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-06-16  6:27 [RESEND PATCH v2 1/2] serial: samsung: Fix ERR pointer dereference on deferred probe Krzysztof Kozlowski
2016-06-16  6:27 ` [RESEND PATCH v2 2/2] serial: samsung: Fix possible out of bounds access on non-DT platform Krzysztof Kozlowski

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).