On 26-03-18, 16:52, Suman Anna wrote:
> Commit 05829d9431df ("cpufreq: ti-cpufreq: kfree opp_data when
> failure") has fixed a memory leak in the failure path, however
> kmemleak still keeps reporting a leak even on successful probes.
> This is a false-positive and is mostly a result of the opp_data
I don't agree to this reasoning for this particular patch. The code is just fine
and kmemleak is something that requires a fix.
> variable not being stored anywhere in the probe function. The
> patch also returned a positive value on the get_cpu_device()
> failure instead of a negative value.
Maybe that could have been fixed in a separate patch, cc'ing stable kernels as
well.
> unreferenced object 0xecae4d80 (size 64):
> comm "swapper/0", pid 1, jiffies 4294937673 (age 154.420s)
> hex dump (first 32 bytes):
> 10 40 d9 ee 74 b7 db ee 00 24 ac ec 20 a3 ea c0 .@..t....$.. ...
> 00 26 ac ec 00 00 00 00 00 00 00 00 00 00 00 00 .&..............
> backtrace:
> [<ec080d62>] platform_drv_probe+0x50/0xac
> [<cbde8566>] driver_probe_device+0x24c/0x330
> [<a5818eb4>] bus_for_each_drv+0x54/0xb8
> [<2c6f7021>] __device_attach+0xcc/0x13c
> [<a04478a2>] bus_probe_device+0x88/0x90
> [<b322c963>] device_add+0x38c/0x5b4
> [<6f1af99b>] platform_device_add+0x100/0x220
> [<cef42bca>] platform_device_register_full+0xf0/0x104
> [<4d492439>] ti_cpufreq_init+0x44/0x6c
> [<81222e89>] do_one_initcall+0x48/0x190
> [<3bebf42a>] kernel_init_freeable+0x1f4/0x2b8
> [<230ad7df>] kernel_init+0x8/0x110
> [<43a165c3>] ret_from_fork+0x14/0x20
> [< (null)>] (null)
> [<87288797>] 0xffffffff
>
> Fix both issues by replacing the previous logic by using the devres
> managed API for allocating the opp_data variable, and simplifying
> the get_cpu_device() failure return path.
>
> Fixes: 05829d9431df ("cpufreq: ti-cpufreq: kfree opp_data when failure")
> Cc: Zumeng Chen <zumeng.c...@gmail.com>
> Signed-off-by: Suman Anna <s-a...@ti.com>
> ---
> drivers/cpufreq/ti-cpufreq.c | 7 ++-----
> 1 file changed, 2 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/cpufreq/ti-cpufreq.c b/drivers/cpufreq/ti-cpufreq.c
> index a099b7bf74cd..7d353a21935b 100644
> --- a/drivers/cpufreq/ti-cpufreq.c
> +++ b/drivers/cpufreq/ti-cpufreq.c
> @@ -217,7 +217,7 @@ static int ti_cpufreq_probe(struct platform_device *pdev)
> if (!match)
> return -ENODEV;
>
> - opp_data = kzalloc(sizeof(*opp_data), GFP_KERNEL);
> + opp_data = devm_kzalloc(&pdev->dev, sizeof(*opp_data), GFP_KERNEL);
> if (!opp_data)
> return -ENOMEM;
>
> @@ -226,8 +226,7 @@ static int ti_cpufreq_probe(struct platform_device *pdev)
> opp_data->cpu_dev = get_cpu_device(0);
> if (!opp_data->cpu_dev) {
> pr_err("%s: Failed to get device for CPU0\n", __func__);
> - ret = ENODEV;
> - goto free_opp_data;
> + return -ENODEV;
> }
>
> opp_data->opp_node = dev_pm_opp_of_get_opp_desc_node(opp_data->cpu_dev);
> @@ -285,8 +284,6 @@ static int ti_cpufreq_probe(struct platform_device *pdev)
>
> fail_put_node:
> of_node_put(opp_data->opp_node);
> -free_opp_data:
> - kfree(opp_data);
>
> return ret;
> }
I am fine with the diff though, as that makes sense. So maybe do this ?
- send separate patch for ENODEV thing
- and another patch to move to devres with a different reason than fixing false
positive.
--
viresh
