Re: [PATCH V2 1/2] remoteproc: imx_rproc: release carveout under imx_rproc after rproc_attach() fails

[Mathieu Poirier <mathieu.poirier@linaro.org>]

Hi Xiaolei,

On Sat, Apr 26, 2025 at 02:53:47PM +0800, Xiaolei Wang wrote:
> When rproc->state = RPROC_DETACHED and rproc_attach() is used
> to attach to the remote processor, if rproc_handle_resources()
> returns a failure, the resources allocated by rproc_prepare_device()
> should be released, otherwise the following memory leak will occur.
> 
> Therefore, add imx_rproc_unprepare() to imx_rproc to release the
> memory allocated in imx_rproc_prepare().
> 
> unreferenced object 0xffff0000861c5d00 (size 128):
> comm "kworker/u12:3", pid 59, jiffies 4294893509 (age 149.220s)
> hex dump (first 32 bytes):
> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
> 00 00 02 88 00 00 00 00 00 00 10 00 00 00 00 00 ............
> backtrace:
>  [<00000000f949fe18>] slab_post_alloc_hook+0x98/0x37c
>  [<00000000adbfb3e7>] __kmem_cache_alloc_node+0x138/0x2e0
>  [<00000000521c0345>] kmalloc_trace+0x40/0x158
>  [<000000004e330a49>] rproc_mem_entry_init+0x60/0xf8
>  [<000000002815755e>] imx_rproc_prepare+0xe0/0x180
>  [<0000000003f61b4e>] rproc_boot+0x2ec/0x528
>  [<00000000e7e994ac>] rproc_add+0x124/0x17c
>  [<0000000048594076>] imx_rproc_probe+0x4ec/0x5d4
>  [<00000000efc298a1>] platform_probe+0x68/0xd8
>  [<00000000110be6fe>] really_probe+0x110/0x27c
>  [<00000000e245c0ae>] __driver_probe_device+0x78/0x12c
>  [<00000000f61f6f5e>] driver_probe_device+0x3c/0x118
>  [<00000000a7874938>] __device_attach_driver+0xb8/0xf8
>  [<0000000065319e69>] bus_for_each_drv+0x84/0xe4
>  [<00000000db3eb243>] __device_attach+0xfc/0x18c
>  [<0000000072e4e1a4>] device_initial_probe+0x14/0x20
> 
> Signed-off-by: Xiaolei Wang <xiaolei.wang@windriver.com>
> ---
>  drivers/remoteproc/imx_rproc.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 
> diff --git a/drivers/remoteproc/imx_rproc.c b/drivers/remoteproc/imx_rproc.c
> index 74299af1d7f1..c489bd15ee91 100644
> --- a/drivers/remoteproc/imx_rproc.c
> +++ b/drivers/remoteproc/imx_rproc.c
> @@ -595,6 +595,19 @@ static int imx_rproc_prepare(struct rproc *rproc)
>  	return  0;
>  }
>  
> +static int imx_rproc_unprepare(struct rproc *rproc)
> +{
> +	struct rproc_mem_entry *entry, *tmp;
> +
> +	rproc_coredump_cleanup(rproc);
> +	/* clean up carveout allocations */
> +	list_for_each_entry_safe(entry, tmp, &rproc->carveouts, node) {
> +		list_del(&entry->node);
> +		kfree(entry);
> +	}
> +	return  0;
> +}
> +

I have reflected long and hard on this.  The problem with the above approach is
that we do almost the same thing in imx_rproc_prepare() and
rproc_resource_cleanup().  Function rproc_resource_cleanup() is able to deal
with empty lists so I think it is better to fix the "goto" statements in
rproc_attach().  More specifically, replace the "unprepare_device" goto
statement with "clean_up_resources" and get rid of the "unprepare_device" label.  

Please see if that works on your side.  I am good with 2/2 of this series.

Thanks,
Mathieu

>  static int imx_rproc_parse_fw(struct rproc *rproc, const struct firmware *fw)
>  {
>  	int ret;
> @@ -675,6 +688,7 @@ imx_rproc_elf_find_loaded_rsc_table(struct rproc *rproc, const struct firmware *
>  
>  static const struct rproc_ops imx_rproc_ops = {
>  	.prepare	= imx_rproc_prepare,
> +	.unprepare	= imx_rproc_unprepare,
>  	.attach		= imx_rproc_attach,
>  	.detach		= imx_rproc_detach,
>  	.start		= imx_rproc_start,
> -- 
> 2.25.1
>