On Tue, Sep 22, 2020 at 07:57:49AM +0200, Marc Kleine-Budde wrote:
> On 9/21/20 9:33 PM, Mark Brown wrote:

> >> +	if (priv->map_buf_nocrc_rx) {
> >> +		devm_kfree(&priv->spi->dev, priv->map_buf_nocrc_rx);
> >> +		priv->map_buf_nocrc_rx = NULL;
> >> +	}

> > Why explicitly free managed allocations like this?

> The driver covers two chips, the mcp2517fd and the mcp2518fd. Due to erratas we
> need different quirks for these variants. Further you can configure if regmap
> should use transfers with or without CRC checks (separately for register access,
> RAM access in RX and TX path).

> It's possible to distinguish between chip variants by reading/writing a
> register, so we first init the regmap with CRC mode, autodetect the chip
> variant, and then re-init the regmap with the quirks of that detected variant.

> The mcp25xxfd regmap init is written so that resources that are allocated first,
> but not needed due to different quirks are then freed (both regmap and the
> buffers for rx/tx).

This feels like a non-idiomatic way of doing this - usually you'd
enumerate then allocate the extra maps (using regmap_reinit_cache() to
replace the regmap used to do the enumeration).