Re: [PATCH] of: restore old handling of cells_name=NULL in of_*_phandle_with_args()

[Peter Rosin]

On 2019-09-18 08:38, Uwe Kleine-KÃnig wrote:
> From: Uwe Kleine-KÃnig <uwe@xxxxxxxxxxxxxxxxx>
> 
> Before commit e42ee61017f5 ("of: Let of_for_each_phandle fallback to
> non-negative cell_count") the iterator functions calling
> of_for_each_phandle assumed a cell count of 0 if cells_name was NULL.
> This corner case was missed when implementing the fallback logic in
> e42ee61017f5 and resulted in an endless loop.
> 
> Restore the old behaviour of of_count_phandle_with_args() and
> of_parse_phandle_with_args() and add a check to
> of_phandle_iterator_init() to prevent a similar failure as a safety
> precaution. of_parse_phandle_with_args_map() doesn't need a similar fix
> as cells_name isn't NULL there.
> 
> Affected drivers are:
>  - drivers/base/power/domain.c
>  - drivers/base/power/domain.c
>  - drivers/clk/ti/clk-dra7-atl.c
>  - drivers/hwmon/ibmpowernv.c
>  - drivers/i2c/muxes/i2c-demux-pinctrl.c
>  - drivers/iommu/mtk_iommu.c
>  - drivers/net/ethernet/freescale/fman/mac.c
>  - drivers/opp/of.c
>  - drivers/perf/arm_dsu_pmu.c
>  - drivers/regulator/of_regulator.c
>  - drivers/remoteproc/imx_rproc.c
>  - drivers/soc/rockchip/pm_domains.c
>  - sound/soc/fsl/imx-audmix.c
>  - sound/soc/fsl/imx-audmix.c
>  - sound/soc/meson/axg-card.c
>  - sound/soc/samsung/tm2_wm5110.c
>  - sound/soc/samsung/tm2_wm5110.c
> 
> Thanks to Geert Uytterhoeven for reporting the issue, Peter Rosin for
> helping pinpoint the actual problem and the testers for confirming this
> fix.
> 
> Fixes: e42ee61017f5 ("of: Let of_for_each_phandle fallback to non-negative cell_count")
> Tested-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Tested-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> Signed-off-by: Uwe Kleine-KÃnig <u.kleine-koenig@xxxxxxxxxxxxxx>
> ---
> Hello,
> 
> compared to the untested patch I sent yesterday I also fixed
> of_parse_phandle_with_args which has three users that pass
> cells_name=NULL. (i.e. drivers/clk/ti/clk-dra7-atl.c,
> sound/soc/fsl/imx-audmix.c, sound/soc/samsung/tm2_wm5110.c) I didn't
> look closely, but maybe these could be converted to use of_parse_phandle
> as there are no arguments to be processed with no cells_name?!
> 
> Best regards
> Uwe
> 
>  drivers/of/base.c | 30 ++++++++++++++++++++++++++++--
>  1 file changed, 28 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/of/base.c b/drivers/of/base.c
> index 2f25d2dfecfa..25ee07c0a3cd 100644
> --- a/drivers/of/base.c
> +++ b/drivers/of/base.c
> @@ -1286,6 +1286,13 @@ int of_phandle_iterator_init(struct of_phandle_iterator *it,
>  
>  	memset(it, 0, sizeof(*it));
>  
> +	/*
> +	 * one of cell_count or cells_name must be provided to determine the
> +	 * argument length.
> +	 */
> +	if (cell_count < 0 && !cells_name)
> +		return -EINVAL;
> +
>  	list = of_get_property(np, list_name, &size);
>  	if (!list)
>  		return -ENOENT;
> @@ -1512,10 +1519,17 @@ int of_parse_phandle_with_args(const struct device_node *np, const char *list_na
>  				const char *cells_name, int index,
>  				struct of_phandle_args *out_args)
>  {
> +	int cell_count = -1;
> +
>  	if (index < 0)
>  		return -EINVAL;
> -	return __of_parse_phandle_with_args(np, list_name, cells_name, -1,
> -					    index, out_args);
> +
> +	/* If cells_name if NULL we assume a cell count of 0 */
> +	if (!cells_name)
> +		cell_count = 0;
> +
> +	return __of_parse_phandle_with_args(np, list_name, cells_name,
> +					    cell_count, index, out_args);
>  }
>  EXPORT_SYMBOL(of_parse_phandle_with_args);
>  
> @@ -1765,6 +1779,18 @@ int of_count_phandle_with_args(const struct device_node *np, const char *list_na
>  	struct of_phandle_iterator it;
>  	int rc, cur_index = 0;
>  
> +	/* If cells_name is NULL we assume a cell count of 0 */
> +	if (cells_name == NULL) {

A couple of nits.

I don't know if there are other considerations, but in the previous two
hunks you use !cells_name instead of comparing explicitly with NULL.
Personally, I find the shorter form more readable, and in the name of
consistency bla bla...

Also, the comment explaining this NULL-check didn't really make sense
to me until I realized that knowing the cell count to be zero makes
counting trivial. Something along those lines should perhaps be in the
comment?

But as I said, these are nits. Feel free to ignore.

Cheers,
Peter

> +		const __be32 *list;
> +		int size;
> +
> +		list = of_get_property(np, list_name, &size);
> +		if (!list)
> +			return -ENOENT;
> +
> +		return size / sizeof(*list);
> +	}
> +
>  	rc = of_phandle_iterator_init(&it, np, list_name, cells_name, -1);
>  	if (rc)
>  		return rc;
>