Re: [PATCH v2 5/8] platform/x86/amd/pmc: Update IP information structure for newer SoCs

[Shyam Sundar S K <Shyam-sundar.S-k@amd.com>]

On 11/1/2024 17:34, Ilpo Järvinen wrote:
> On Tue, 29 Oct 2024, Shyam Sundar S K wrote:
> 
>> The latest AMD processors include additional IP blocks that must be turned
>> off before transitioning to low power. PMFW provides an interface to
>> retrieve debug information from each IP block, which is useful for
>> diagnosing issues if the system fails to enter or exit low power states,
>> or for profiling which IP block takes more time. Add support for using
>> this information within the driver.
>>
>> Co-developed-by: Sanket Goswami <Sanket.Goswami@amd.com>
>> Signed-off-by: Sanket Goswami <Sanket.Goswami@amd.com>
>> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> ---
>>  drivers/platform/x86/amd/pmc/pmc.c | 42 ++++++++++++++++++++++++++++--
>>  1 file changed, 40 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/platform/x86/amd/pmc/pmc.c b/drivers/platform/x86/amd/pmc/pmc.c
>> index f9900a03391a..0bf4065153da 100644
>> --- a/drivers/platform/x86/amd/pmc/pmc.c
>> +++ b/drivers/platform/x86/amd/pmc/pmc.c
>> @@ -95,6 +95,35 @@ struct amd_pmc_bit_map {
>>  	u32 bit_mask;
>>  };
>>  
>> +static const struct amd_pmc_bit_map soc15_ip_blk_v2[] = {
>> +	{"DISPLAY",     BIT(0)},
>> +	{"CPU",         BIT(1)},
>> +	{"GFX",         BIT(2)},
>> +	{"VDD",         BIT(3)},
>> +	{"VDD_CCX",     BIT(4)},
>> +	{"ACP",         BIT(5)},
>> +	{"VCN_0",       BIT(6)},
>> +	{"VCN_1",       BIT(7)},
>> +	{"ISP",         BIT(8)},
>> +	{"NBIO",        BIT(9)},
>> +	{"DF",          BIT(10)},
>> +	{"USB3_0",      BIT(11)},
>> +	{"USB3_1",      BIT(12)},
>> +	{"LAPIC",       BIT(13)},
>> +	{"USB3_2",      BIT(14)},
>> +	{"USB4_RT0",	BIT(15)},
>> +	{"USB4_RT1",	BIT(16)},
>> +	{"USB4_0",      BIT(17)},
>> +	{"USB4_1",      BIT(18)},
>> +	{"MPM",         BIT(19)},
>> +	{"JPEG_0",      BIT(20)},
>> +	{"JPEG_1",      BIT(21)},
>> +	{"IPU",         BIT(22)},
>> +	{"UMSCH",       BIT(23)},
>> +	{"VPE",         BIT(24)},
>> +	{}
>> +};
>> +
>>  static const struct amd_pmc_bit_map soc15_ip_blk[] = {
>>  	{"DISPLAY",	BIT(0)},
>>  	{"CPU",		BIT(1)},
>> @@ -170,7 +199,10 @@ static void amd_pmc_get_ip_info(struct amd_pmc_dev *dev)
>>  		break;
>>  	case PCI_DEVICE_ID_AMD_1AH_M20H_ROOT:
>>  	case PCI_DEVICE_ID_AMD_1AH_M60H_ROOT:
>> -		dev->num_ips = 22;
>> +		if (boot_cpu_data.x86_model == 0x70)
>> +			dev->num_ips = 25;
>> +		else
>> +			dev->num_ips = 22;
> 
> Could these use ARRAY_SIZE()? They're related to that array, aren't they?

Yes, they are. ARRAY_SIZE() does return the number of elements in the
array but there is a catch,

both soc15_ip_blk[] and soc15_ip_blk_v2[] have a last empty element,
so when ARRAY_SIZE() is used we end up getting n+1 element (i.e along
with the last empty element). So, what would you advise?

1) remove the last empty element in the array? i.e. something like this?

static const struct amd_pmc_bit_map soc15_ip_blk_v2[] = {
	{"DISPLAY",     BIT(0)},
	...
	{"VPE",         BIT(24)},
-   {} /* remove this */
};

2) or just do, `ARRAY_SIZE() - 1` so that we don't need to touch the
existing soc15_ip_blk[] and soc15_ip_blk_v2[].


I am inline with your comments on the other patches, will address them
in v3.

Thanks,
Shyam

> 
>>  		dev->smu_msg = 0x938;
>>  		break;
>>  	}
>> @@ -338,9 +370,15 @@ static int smu_fw_info_show(struct seq_file *s, void *unused)
>>  
>>  	seq_puts(s, "\n=== Active time (in us) ===\n");
>>  	for (idx = 0 ; idx < dev->num_ips ; idx++) {
>> -		if (soc15_ip_blk[idx].bit_mask & dev->active_ips)
>> +		if (dev->cpu_id == PCI_DEVICE_ID_AMD_1AH_M20H_ROOT &&
>> +		    boot_cpu_data.x86_model == 0x70) {
>> +			if (soc15_ip_blk_v2[idx].bit_mask & dev->active_ips)
>> +				seq_printf(s, "%-8s : %lld\n", soc15_ip_blk_v2[idx].name,
>> +					   table.timecondition_notmet_lastcapture[idx]);
>> +		} else if (soc15_ip_blk[idx].bit_mask & dev->active_ips) {
>>  			seq_printf(s, "%-8s : %lld\n", soc15_ip_blk[idx].name,
>>  				   table.timecondition_notmet_lastcapture[idx]);
>> +		}
> 
> Why not have amd_pmc_get_ip_info() prepare a pointer into 'dev' to the 
> relevant soc15_ip_blk_v2/soc15_ip_blk rather than trying to pick one here?
> 

Makes sense. Ack.

>>  	}
>>  
>>  	return 0;
>>
>