| Message ID | 545df946044fc1fc05a4217cdf0054be7a79e49e.1619161112.git.christophe.jaillet@wanadoo.fr (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
| Series | clk: mvebu: ap-cpu-clk: Fix a memory leak in error handling paths | expand |
Quoting Christophe JAILLET (2021-04-23 00:02:26) > If we exit the for_each_of_cpu_node loop early, the reference on the > current node must be decremented, otherwise there is a leak. > > Fixes: f756e362d938 ("clk: mvebu: add CPU clock driver for Armada 7K/8K") > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > --- > Also, I wonder if the drivers in drivers/clk/mvebu are used by anyone. > In order to compile-test the changes, I also had to change the 'bool' in Kconfig > by 'bool "blah"'. Without this change, it was not possible to set > CONFIG_ARMADA_AP_CPU_CLK required by Makefile. > > I don't know if I did something wrong, if it is an issue only on my environment > or if something got broken at some time in the build chain but it looks > spurious. > > If I'm right and that these drivers never compile and no-one noticed it, > maybe removing them is better than fixing some unlikely issues and style. > If these drivers should stay, Kconfig may need some love from someone. Nobody has said anything on this patch. So I'm not really sure what's going on. Probably we never take the error path, or the whole system fails to boot?
On Sat, Aug 28, 2021 at 10:27:17PM -0700, Stephen Boyd wrote: > Quoting Christophe JAILLET (2021-04-23 00:02:26) > > If we exit the for_each_of_cpu_node loop early, the reference on the > > current node must be decremented, otherwise there is a leak. > > > > Fixes: f756e362d938 ("clk: mvebu: add CPU clock driver for Armada 7K/8K") > > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > > --- > > Also, I wonder if the drivers in drivers/clk/mvebu are used by anyone. > > In order to compile-test the changes, I also had to change the 'bool' in Kconfig > > by 'bool "blah"'. Without this change, it was not possible to set > > CONFIG_ARMADA_AP_CPU_CLK required by Makefile. > > > > I don't know if I did something wrong, if it is an issue only on my environment > > or if something got broken at some time in the build chain but it looks > > spurious. > > > > If I'm right and that these drivers never compile and no-one noticed it, > > maybe removing them is better than fixing some unlikely issues and style. > > If these drivers should stay, Kconfig may need some love from someone. > > Nobody has said anything on this patch. So I'm not really sure what's > going on. Probably we never take the error path, or the whole system > fails to boot? You probably take this one. 106 /* If cpu2 or cpu3 is enabled */ 107 if (cpu & APN806_CLUSTER_NUM_MASK) { 108 nclusters = 2; 109 + of_node_put(dn); 110 break; 111 } But, yeah, probably on one carse of "dn" can't be freed in real life. Still probably worth fixing though just for correctness. Otherwise it makes static analysis less effective if people deliberately leave bugs in the code. regards, dan carpenter
Quoting Dan Carpenter (2021-08-30 04:59:31) > On Sat, Aug 28, 2021 at 10:27:17PM -0700, Stephen Boyd wrote: > > Quoting Christophe JAILLET (2021-04-23 00:02:26) > > > If we exit the for_each_of_cpu_node loop early, the reference on the > > > current node must be decremented, otherwise there is a leak. > > > > > > Fixes: f756e362d938 ("clk: mvebu: add CPU clock driver for Armada 7K/8K") > > > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > > > --- > > > Also, I wonder if the drivers in drivers/clk/mvebu are used by anyone. > > > In order to compile-test the changes, I also had to change the 'bool' in Kconfig > > > by 'bool "blah"'. Without this change, it was not possible to set > > > CONFIG_ARMADA_AP_CPU_CLK required by Makefile. > > > > > > I don't know if I did something wrong, if it is an issue only on my environment > > > or if something got broken at some time in the build chain but it looks > > > spurious. > > > > > > If I'm right and that these drivers never compile and no-one noticed it, > > > maybe removing them is better than fixing some unlikely issues and style. > > > If these drivers should stay, Kconfig may need some love from someone. > > > > Nobody has said anything on this patch. So I'm not really sure what's > > going on. Probably we never take the error path, or the whole system > > fails to boot? > > You probably take this one. > > 106 /* If cpu2 or cpu3 is enabled */ > 107 if (cpu & APN806_CLUSTER_NUM_MASK) { > 108 nclusters = 2; > 109 + of_node_put(dn); > 110 break; > 111 } > > But, yeah, probably on one carse of "dn" can't be freed in real life. > > Still probably worth fixing though just for correctness. Otherwise it > makes static analysis less effective if people deliberately leave bugs > in the code. Thanks. Can we turn that into a Reviewed-by?
On Fri, Apr 23, 2021 at 09:02:26AM +0200, Christophe JAILLET wrote: > If we exit the for_each_of_cpu_node loop early, the reference on the > current node must be decremented, otherwise there is a leak. > > Fixes: f756e362d938 ("clk: mvebu: add CPU clock driver for Armada 7K/8K") > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > --- Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com> regards, dan carpenter
Quoting Christophe JAILLET (2021-04-23 00:02:26) > If we exit the for_each_of_cpu_node loop early, the reference on the > current node must be decremented, otherwise there is a leak. > > Fixes: f756e362d938 ("clk: mvebu: add CPU clock driver for Armada 7K/8K") > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > --- Applied to clk-next
diff --git a/drivers/clk/mvebu/ap-cpu-clk.c b/drivers/clk/mvebu/ap-cpu-clk.c index 08ba59ec3fb1..71bdd7c3ff03 100644 --- a/drivers/clk/mvebu/ap-cpu-clk.c +++ b/drivers/clk/mvebu/ap-cpu-clk.c @@ -256,12 +256,15 @@ static int ap_cpu_clock_probe(struct platform_device *pdev) int cpu, err; err = of_property_read_u32(dn, "reg", &cpu); - if (WARN_ON(err)) + if (WARN_ON(err)) { + of_node_put(dn); return err; + } /* If cpu2 or cpu3 is enabled */ if (cpu & APN806_CLUSTER_NUM_MASK) { nclusters = 2; + of_node_put(dn); break; } } @@ -288,8 +291,10 @@ static int ap_cpu_clock_probe(struct platform_device *pdev) int cpu, err; err = of_property_read_u32(dn, "reg", &cpu); - if (WARN_ON(err)) + if (WARN_ON(err)) { + of_node_put(dn); return err; + } cluster_index = cpu & APN806_CLUSTER_NUM_MASK; cluster_index >>= APN806_CLUSTER_NUM_OFFSET; @@ -301,6 +306,7 @@ static int ap_cpu_clock_probe(struct platform_device *pdev) parent = of_clk_get(np, cluster_index); if (IS_ERR(parent)) { dev_err(dev, "Could not get the clock parent\n"); + of_node_put(dn); return -EINVAL; } parent_name = __clk_get_name(parent); @@ -319,8 +325,10 @@ static int ap_cpu_clock_probe(struct platform_device *pdev) init.parent_names = &parent_name; ret = devm_clk_hw_register(dev, &ap_cpu_clk[cluster_index].hw); - if (ret) + if (ret) { + of_node_put(dn); return ret; + } ap_cpu_data->hws[cluster_index] = &ap_cpu_clk[cluster_index].hw; }
If we exit the for_each_of_cpu_node loop early, the reference on the current node must be decremented, otherwise there is a leak. Fixes: f756e362d938 ("clk: mvebu: add CPU clock driver for Armada 7K/8K") Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> --- Also, I wonder if the drivers in drivers/clk/mvebu are used by anyone. In order to compile-test the changes, I also had to change the 'bool' in Kconfig by 'bool "blah"'. Without this change, it was not possible to set CONFIG_ARMADA_AP_CPU_CLK required by Makefile. I don't know if I did something wrong, if it is an issue only on my environment or if something got broken at some time in the build chain but it looks spurious. If I'm right and that these drivers never compile and no-one noticed it, maybe removing them is better than fixing some unlikely issues and style. If these drivers should stay, Kconfig may need some love from someone. --- drivers/clk/mvebu/ap-cpu-clk.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-)