[PATCH] soc: fsl: guts: us devm_kstrdup_const() for RO data

devm_kstrdup() may return NULL if internal allocation failed, but
as  machine  is from the device tree, and thus RO, devm_kstrdup_const()
can be used here, which will only copy the reference.

Signed-off-by: Nicholas Mc Guire <redacted>
Fixes: a6fc3b698130 ("soc: fsl: add GUTS driver for QorIQ platforms")
---

Problem located by experimental coccinelle script

Patch was compile tested with: multi_v7_defconfig (implies FSL_GUTS=y)

Patch is against 4.20-rc5 (localversion-next is next-20181207)

 drivers/soc/fsl/guts.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/soc/fsl/guts.c b/drivers/soc/fsl/guts.c
index 302e0c8..15071ec 100644
--- a/drivers/soc/fsl/guts.c
+++ b/drivers/soc/fsl/guts.c

@@ -157,7 +157,8 @@ static int fsl_guts_probe(struct platform_device *pdev)
 		of_property_read_string_index(root, "compatible", 0, &machine);
 	of_node_put(root);
 	if (machine)
-		soc_dev_attr.machine = devm_kstrdup(dev, machine, GFP_KERNEL);
+		soc_dev_attr.machine = devm_kstrdup_const(dev, machine,
+							  GFP_KERNEL);
 
 	svr = fsl_guts_get_svr();
 	soc_die = fsl_soc_die_match(svr, fsl_soc_die);

-- 
2.1.4

On Fri, 2018-12-07 at 09:22 +0100, Nicholas Mc Guire wrote:

devm_kstrdup() may return NULL if internal allocation failed, but
as  machine  is from the device tree, and thus RO, devm_kstrdup_const()
can be used here, which will only copy the reference.

Is it really going to only copy the reference?  That would require that
is_kernel_rodata(machine) be true, which it shouldn't be since it's not part
of the kernel image.

-Scott

On Fri, Dec 21, 2018 at 08:29:56PM -0600, Scott Wood wrote:

On Fri, 2018-12-07 at 09:22 +0100, Nicholas Mc Guire wrote:

quoted

devm_kstrdup() may return NULL if internal allocation failed, but
as  machine  is from the device tree, and thus RO, devm_kstrdup_const()
can be used here, which will only copy the reference.

Is it really going to only copy the reference?  That would require that
is_kernel_rodata(machine) be true, which it shouldn't be since it's not part
of the kernel image.

I had tried to figure out what is RO and what not but was not 
able to determine that - from the discussion it seemed that the
assumption of RO is correct though I did not ask if it would
satisfy is_kernel_rodata() so that explains the incorrect assertion.
see https://lkml.org/lkml/2018/12/6/42
So then the only option is to check the return and cleanup
on allocation failure as the orriginal patch proposed.

thanks for clearifying this !
hofrat

On Sat, Dec 22, 2018 at 2:02 AM Nicholas Mc Guire [off-list ref] wrote:

On Fri, Dec 21, 2018 at 08:29:56PM -0600, Scott Wood wrote:

quoted

On Fri, 2018-12-07 at 09:22 +0100, Nicholas Mc Guire wrote:

quoted

devm_kstrdup() may return NULL if internal allocation failed, but
as  machine  is from the device tree, and thus RO, devm_kstrdup_const()
can be used here, which will only copy the reference.

Is it really going to only copy the reference?  That would require that
is_kernel_rodata(machine) be true, which it shouldn't be since it's not part
of the kernel image.

I had tried to figure out what is RO and what not but was not
able to determine that - from the discussion it seemed that the
assumption of RO is correct though I did not ask if it would
satisfy is_kernel_rodata() so that explains the incorrect assertion.
see https://lkml.org/lkml/2018/12/6/42
So then the only option is to check the return and cleanup
on allocation failure as the orriginal patch proposed.

Thanks for the good discussion. I will drop the previous patch. But
would it also be good to just have "soc_dev_attr.machine = machine"
directly?

Regards,
Leo

On Thu, Jan 10, 2019 at 01:43:01PM -0600, Li Yang wrote:

On Sat, Dec 22, 2018 at 2:02 AM Nicholas Mc Guire [off-list ref] wrote:

quoted

On Fri, Dec 21, 2018 at 08:29:56PM -0600, Scott Wood wrote:

quoted

On Fri, 2018-12-07 at 09:22 +0100, Nicholas Mc Guire wrote:

quoted

devm_kstrdup() may return NULL if internal allocation failed, but
as  machine  is from the device tree, and thus RO, devm_kstrdup_const()
can be used here, which will only copy the reference.

Is it really going to only copy the reference?  That would require that
is_kernel_rodata(machine) be true, which it shouldn't be since it's not part
of the kernel image.

I had tried to figure out what is RO and what not but was not
able to determine that - from the discussion it seemed that the
assumption of RO is correct though I did not ask if it would
satisfy is_kernel_rodata() so that explains the incorrect assertion.
see https://lkml.org/lkml/2018/12/6/42
So then the only option is to check the return and cleanup
on allocation failure as the orriginal patch proposed.

Thanks for the good discussion. I will drop the previous patch. But
would it also be good to just have "soc_dev_attr.machine = machine"
directly?

I think that the intent is to switch to 
managed devm API so that the cleanup is handled properly
currently you would get "machine" from 
 of_property_read_string_index 
  -> of_property_read_string_helper
   -> of_find_property
which does not do any allocation - so there would actually
not be anything to cleanup here - don´t see why your solution
would not be suitable given the current API. the only advantage
of the devm_kstrdup() is that underlying APIs internal changes
would have no effect.

thx!
hofrat

-----Original Message-----
From: Nicholas Mc Guire <redacted>
Sent: Thursday, January 10, 2019 8:44 PM
To: Leo Li <redacted>
Cc: Scott Wood <oss@buserror.net>; linuxppc-dev <linuxppc-
dev@lists.ozlabs.org>; lkml [off-list ref]; moderated
list:ARM/FREESCALE IMX / MXC ARM ARCHITECTURE <linux-arm-
kernel@lists.infradead.org>; Nicholas Mc Guire [off-list ref]
Subject: Re: [PATCH] soc: fsl: guts: us devm_kstrdup_const() for RO data

On Thu, Jan 10, 2019 at 01:43:01PM -0600, Li Yang wrote:

quoted

On Sat, Dec 22, 2018 at 2:02 AM Nicholas Mc Guire [off-list ref]

wrote:

quoted

quoted

On Fri, Dec 21, 2018 at 08:29:56PM -0600, Scott Wood wrote:

quoted

On Fri, 2018-12-07 at 09:22 +0100, Nicholas Mc Guire wrote:

quoted

devm_kstrdup() may return NULL if internal allocation failed,
but as  machine  is from the device tree, and thus RO,
devm_kstrdup_const() can be used here, which will only copy the

reference.

quoted

quoted

quoted

Is it really going to only copy the reference?  That would require
that
is_kernel_rodata(machine) be true, which it shouldn't be since
it's not part of the kernel image.

I had tried to figure out what is RO and what not but was not able
to determine that - from the discussion it seemed that the
assumption of RO is correct though I did not ask if it would satisfy
is_kernel_rodata() so that explains the incorrect assertion.
see

https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fl

quoted

quoted

kml.org%2Flkml%2F2018%2F12%2F6%2F42&data=02%7C01%7Cleoyang.l
i%40

quoted

quoted

nxp.com%7Cf72d70a65d1b47f6883808d6776e9d58%7C686ea1d3bc2b4c6fa92c
d99

quoted

quoted

c5c301635%7C0%7C1%7C636827714307963102&sdata=xnaO0Y7q%2Byad
Yv8sF

quoted

quoted

VPFtkfllgnwpEIkkTIgw0K%2Fovg%3D&reserved=0
So then the only option is to check the return and cleanup on
allocation failure as the orriginal patch proposed.

Thanks for the good discussion. I will drop the previous patch. But
would it also be good to just have "soc_dev_attr.machine = machine"
directly?

I think that the intent is to switch to managed devm API so that the cleanup is
handled properly currently you would get "machine" from
of_property_read_string_index
  -> of_property_read_string_helper
   -> of_find_property
which does not do any allocation - so there would actually not be anything to
cleanup here - don´t see why your solution would not be suitable given the
current API. the only advantage of the devm_kstrdup() is that underlying
APIs internal changes would have no effect.

Thanks.  I will sent out a new version.

Regards,
Leo