On 2025-01-24 at 08:14:41 -0800, Dave Hansen wrote:
>On 11/27/24 09:35, Maciej Wieczor-Retman wrote:
>> -/* Check 5-level page table feature in CPUID.(EAX=07H, ECX=00H):ECX.[bit 16] */
>> static inline int cpu_has_la57(void)
>> {
>> - unsigned int cpuinfo[4];
>> -
>> - __cpuid_count(0x7, 0, cpuinfo[0], cpuinfo[1], cpuinfo[2], cpuinfo[3]);
>> -
>> - return (cpuinfo[2] & (1 << 16));
>> + return !system("grep -wq la57 /proc/cpuinfo");
>> }
>
>I would rather we find another way.
>
>First, we've documented the behavior a bit in here:
>
> https://docs.kernel.org/arch/x86/cpuinfo.html
>
>The important part is:
>
> "The absence of a flag in /proc/cpuinfo by itself means almost
> nothing to an end user."
>
>Even worse, let's say there's a CPU bug and we say define a bug bit:
>
> bugs : spectre_v1 spectre_v2 ... la57_is_broken
>
>How is that grep going to work out? ;)
>
>Could you poke around and see if there is any existing ABI that we can
>use to query LA57 support? Maybe one of the things KVM exports, or some
>TASK_SIZE_MAX comparisons?
Sure, I'll try to find some other way.
My previous tactic was to munmap() a high address and see if it works. Does that
sound okay in case there isn't anything else would indicate la57 to userspace
reliably?
>
--
Kind regards
Maciej Wieczór-Retman
