Skip to contentSkip to conversation

dm: avoid dev->req_seq overflow

PATCH New Robert Baldyga 2 participants 4 messages Sep 18, 2014 – Sep 19, 2014

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dm: avoid dev->req_seq overflow

Since dev->req_seq value is initialized from "reg" property of fdt node, there is posibility, that address value contained in fdt is greater than INT_MAX, and then value in dev->req_seq is negative which led to probe() fail.

This patch fix this problem by ensuring that req_seq is positive, unless it's one of errno codes.

---
 drivers/core/device.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/core/device.c +2 -0
index 166b073..35ffce0 100644
@@ -107,6 +107,8 @@ int device_bind(struct udevice *parent, struct driver *drv, const char *name,
 	 * when the device is probed.
 	 */
 	dev->req_seq = fdtdec_get_int(gd->fdt_blob, of_offset, "reg", -1);
+	if (!IS_ERR_VALUE(dev->req_seq))
+		dev->req_seq &= INT_MAX;
 	dev->seq = -1;
 	if (uc->uc_drv->name && of_offset != -1) {
 		fdtdec_get_alias_seq(gd->fdt_blob, uc->uc_drv->name, of_offset,
Signature · 1 line
1.9.1
#1 lore raw

PATCH 1/1 dm: avoid dev->req_seq overflow

Hi Robert,

Quoted 8 lines · Robert Baldyga

Since dev->req_seq value is initialized from "reg" property of fdt node, there is posibility, that address value contained in fdt is greater than INT_MAX, and then value in dev->req_seq is negative which led to probe() fail.

This patch fix this problem by ensuring that req_seq is positive, unless it's one of errno codes.

Wouldn't this be a bug in the device tree file? What does it mean to have a -ve value?

Quoted 19 lines
---
 drivers/core/device.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/core/device.c +2 -0
index 166b073..35ffce0 100644
@@ -107,6 +107,8 @@ int device_bind(struct udevice *parent, struct driver
 *drv, const char *name,
         * when the device is probed.
         */
        dev->req_seq = fdtdec_get_int(gd->fdt_blob, of_offset, "reg", -1);
+       if (!IS_ERR_VALUE(dev->req_seq))
+               dev->req_seq &= INT_MAX;
        dev->seq = -1;
        if (uc->uc_drv->name && of_offset != -1) {
                fdtdec_get_alias_seq(gd->fdt_blob, uc->uc_drv->name,
Quoted 5 lines

of_offset,
--
1.9.1

Regards,
Simon

#2 lore raw

PATCH 1/1 dm: avoid dev->req_seq overflow

Quoted 10 lines · Simon Glass

Hi Robert,

On 18 September 2014 09:13, Robert Baldyga <[email protected] <mailto:[email protected]>> wrote:

Since dev->req_seq value is initialized from "reg" property of fdt node, there is posibility, that address value contained in fdt is greater than INT_MAX, and then value in dev->req_seq is negative which led to probe() fail.

Quoted 7 lines

This patch fix this problem by ensuring that req_seq is positive, unless it's one of errno codes.

Wouldn't this be a bug in the device tree file? What does it mean to have a -ve value?

Device tree seems to be ok. We have:

	pinctrl0: pinctrl at e0200000 {
		compatible = "samsung,s5pc110-pinctrl";
		reg = <0xe0200000 0x1000>;
	};

So when we take address from "reg" as dev->req_seq, then value 0xe0200000 after casting to int gives -534773760. Function uclass_resolve_seq() returns it as proper seq number, because it's unique. But then in file drivers/core/device.c, in function device_probe() we have:

	seq = uclass_resolve_seq(dev);
	if (seq < 0) {
		ret = seq;
		goto fail;
	}

And it will obviously fail.

Using "reg" value as req_seq doesn't work when this value is greater than INT_MAX.

Quoted 17 lines
    Signed-off-by: Robert Baldyga <[email protected]
    <mailto:[email protected]>>
    ---
     drivers/core/device.c | 2 ++
     1 file changed, 2 insertions(+)

    diff --git a/drivers/core/device.c b/drivers/core/device.c
    index 166b073..35ffce0 100644
    --- a/drivers/core/device.c
    +++ b/drivers/core/device.c
    @@ -107,6 +107,8 @@ int device_bind(struct udevice *parent, struct
    driver *drv, const char *name,
             * when the device is probed.
             */
            dev->req_seq = fdtdec_get_int(gd->fdt_blob, of_offset,
    "reg", -1);
Quoted 9 lines
    +       if (!IS_ERR_VALUE(dev->req_seq))
    +               dev->req_seq &= INT_MAX;
            dev->seq = -1;
            if (uc->uc_drv->name && of_offset != -1) {
                    fdtdec_get_alias_seq(gd->fdt_blob, uc->uc_drv->name,
    of_offset,
    --
    1.9.1

Thanks,
Robert Baldyga

#3 lore raw

PATCH 1/1 dm: avoid dev->req_seq overflow

Hi Robert,

Quoted 39 lines · Robert Baldyga
Quoted 10 lines · Simon Glass

Hi Robert,

On 18 September 2014 09:13, Robert Baldyga <[email protected] <mailto:[email protected]>> wrote:

Since dev->req_seq value is initialized from "reg" property of fdt node, there is posibility, that address value contained in fdt is greater than INT_MAX, and then value in dev->req_seq is negative which led to probe() fail.

Quoted 7 lines

This patch fix this problem by ensuring that req_seq is positive, unless it's one of errno codes.

Wouldn't this be a bug in the device tree file? What does it mean to have a -ve value?

Device tree seems to be ok. We have:

        pinctrl0: pinctrl at e0200000 {
                compatible = "samsung,s5pc110-pinctrl";
                reg = <0xe0200000 0x1000>;
        };

So when we take address from "reg" as dev->req_seq, then value 0xe0200000 after casting to int gives -534773760. Function uclass_resolve_seq() returns it as proper seq number, because it's unique. But then in file drivers/core/device.c, in function device_probe() we have:

        seq = uclass_resolve_seq(dev);
        if (seq < 0) {
                ret = seq;
                goto fail;
        }
Quoted 4 lines

And it will obviously fail.

Using "reg" value as req_seq doesn't work when this value is greater than INT_MAX.

OK I see. Thanks for the clear explanation. We don't really want req_seq for things that are not buses, but there is no real concept of that in DM at this stage. Let's see how things pan out. For now, this patch is a good solution.

Regards,
Simon