Re: [PATCH -V3 3/4] acpi, hmat: calculate abstract distance with HMAT

From: Alistair Popple
Date: Tue Sep 19 2023 - 01:31:30 EST



Thanks for making changes here, looks better to me at least.

Huang Ying <ying.huang@xxxxxxxxx> writes:

> static __init void hmat_free_structures(void)
> {
> struct memory_target *target, *tnext;
> @@ -801,6 +857,7 @@ static __init int hmat_init(void)
> struct acpi_table_header *tbl;
> enum acpi_hmat_type i;
> acpi_status status;
> + int usage;
>
> if (srat_disabled() || hmat_disable)
> return 0;
> @@ -841,7 +898,10 @@ static __init int hmat_init(void)
> hmat_register_targets();
>
> /* Keep the table and structures if the notifier may use them */
> - if (!hotplug_memory_notifier(hmat_callback, HMAT_CALLBACK_PRI))
> + usage = !hotplug_memory_notifier(hmat_callback, HMAT_CALLBACK_PRI);
> + if (!hmat_set_default_dram_perf())
> + usage += !register_mt_adistance_algorithm(&hmat_adist_nb);
> + if (usage)
> return 0;

Can we simplify the error handling here? As I understand it
hotplug_memory_notifier() and register_mt_adistance_algorithm() aren't
expected to ever fail because hmat_init() should only be called once and
the notifier register shouldn't fail. So wouldn't the below be
effectively the same thing but clearer?

if (hotplug_memory_notifier(hmat_callback, HMAT_CALLBACK_PRI))
goto out_put;

if (!hmat_set_default_dram_perf())
register_mt_adistance_algorithm(&hmat_adist_nb);

return 0;

> out_put:
> hmat_free_structures();

Also as an aside while looking at this patch I noticed a minor bug:

status = acpi_get_table(ACPI_SIG_HMAT, 0, &tbl);
if (ACPI_FAILURE(status))
goto out_put;

This will call acpi_put_table(tbl) even though we failed to get the
table.