Re: [PATCH v2 4/4] platform/x86: pmc_atom: improve critclk_systems matching for Siemens PCs

[Henning Schild]

Am Mon, 15 Mar 2021 10:57:10 +0100
schrieb Henning Schild <henning.schild@xxxxxxxxxxx>:

> Siemens industrial PCs unfortunately can not always be properly
> identified the way we used to. An earlier commit introduced code that
> allows proper identification without looking at DMI strings that could
> differ based on product branding.
> Switch over to that proper way and revert commits that used to collect
> the machines based on unstable strings.
> 
> Fixes: 648e921888ad ("clk: x86: Stop marking clocks as
> CLK_IS_CRITICAL") Fixes: e8796c6c69d1 ("platform/x86: pmc_atom: Add
> Siemens CONNECT ...") Fixes: f110d252ae79 ("platform/x86: pmc_atom:
> Add Siemens SIMATIC ...") Fixes: ad0d315b4d4e ("platform/x86:
> pmc_atom: Add Siemens SIMATIC ...") Tested-by: Michael Haener
> <michael.haener@xxxxxxxxxxx> Signed-off-by: Henning Schild
> <henning.schild@xxxxxxxxxxx> ---
>  drivers/platform/x86/pmc_atom.c | 47
> +++++++++++++++++++-------------- 1 file changed, 27 insertions(+),
> 20 deletions(-)
> 
> diff --git a/drivers/platform/x86/pmc_atom.c
> b/drivers/platform/x86/pmc_atom.c index ca684ed760d1..38542d547f29
> 100644 --- a/drivers/platform/x86/pmc_atom.c
> +++ b/drivers/platform/x86/pmc_atom.c
> @@ -13,6 +13,7 @@
>  #include <linux/io.h>
>  #include <linux/platform_data/x86/clk-pmc-atom.h>
>  #include <linux/platform_data/x86/pmc_atom.h>
> +#include <linux/platform_data/x86/simatic-ipc.h>
>  #include <linux/platform_device.h>
>  #include <linux/pci.h>
>  #include <linux/seq_file.h>
> @@ -362,6 +363,23 @@ static void pmc_dbgfs_register(struct pmc_dev
> *pmc) }
>  #endif /* CONFIG_DEBUG_FS */
>  
> +static bool pmc_clk_is_critical = true;
> +
> +static int siemens_clk_is_critical(const struct dmi_system_id *d)
> +{
> +	u32 st_id;
> +
> +	if (dmi_walk(simatic_ipc_find_dmi_entry_helper, &st_id))
> +		goto out;
> +
> +	if (st_id == SIMATIC_IPC_IPC227E || st_id ==
> SIMATIC_IPC_IPC277E)
> +		return 1;
> +
> +out:
> +	pmc_clk_is_critical = false;
> +	return 1;
> +}
> +
>  /*
>   * Some systems need one or more of their pmc_plt_clks to be
>   * marked as critical.
> @@ -424,24 +442,10 @@ static const struct dmi_system_id
> critclk_systems[] = { },
>  	},
>  	{
> -		.ident = "SIMATIC IPC227E",
> -		.matches = {
> -			DMI_MATCH(DMI_SYS_VENDOR, "SIEMENS AG"),
> -			DMI_MATCH(DMI_PRODUCT_VERSION, "6ES7647-8B"),
> -		},
> -	},
> -	{
> -		.ident = "SIMATIC IPC277E",
> -		.matches = {
> -			DMI_MATCH(DMI_SYS_VENDOR, "SIEMENS AG"),
> -			DMI_MATCH(DMI_PRODUCT_VERSION, "6AV7882-0"),
> -		},
> -	},
> -	{
> -		.ident = "CONNECT X300",
> +		.callback = siemens_clk_is_critical,
> +		.ident = "SIEMENS AG",
>  		.matches = {
>  			DMI_MATCH(DMI_SYS_VENDOR, "SIEMENS AG"),
> -			DMI_MATCH(DMI_PRODUCT_VERSION,
> "A5E45074588"), },
>  	},
>  
> @@ -453,7 +457,7 @@ static int pmc_setup_clks(struct pci_dev *pdev,
> void __iomem *pmc_regmap, {
>  	struct platform_device *clkdev;
>  	struct pmc_clk_data *clk_data;
> -	const struct dmi_system_id *d =
> dmi_first_match(critclk_systems);
> +	const struct dmi_system_id *d;
>  
>  	clk_data = kzalloc(sizeof(*clk_data), GFP_KERNEL);
>  	if (!clk_data)
> @@ -461,9 +465,12 @@ static int pmc_setup_clks(struct pci_dev *pdev,
> void __iomem *pmc_regmap, 
>  	clk_data->base = pmc_regmap; /* offset is added by client */
>  	clk_data->clks = pmc_data->clks;
> -	if (d) {
> -		clk_data->critical = true;
> -		pr_info("%s critclks quirk enabled\n", d->ident);
> +	if (dmi_check_system(critclk_systems)) {

Had to switch to check_system to get the callback to work.

> +		clk_data->critical = pmc_clk_is_critical;
> +		if (clk_data->critical) {
> +			d = dmi_first_match(critclk_systems);
> +			pr_info("%s critclks quirk enabled\n",
> d->ident);

Now need a double match here just to print the ident. Not too happy
with that but proposing it like this to keep the ident printing.

I guess it could be improved by not printing the ident or having a
global variable and global callback to remember the ident to print
later. I would propose to not print the ident if the double-match does
not find traction.

Henning

> +		}
>  	}
>  
>  	clkdev = platform_device_register_data(&pdev->dev,
> "clk-pmc-atom",