Re: [RFC PATCH hurd] Add partial /proc/cpuinfo implementation

[Luca]

Il 08/01/25 22:58, Diego Nieto Cid ha scritto:
> On Wed, Jan 08, 2025 at 10:25:41AM +0100, Luca wrote:
> > Il 08/01/25 02:30, dnietoc@gmail.com ha scritto:
> > > +
> > > +static char * features_edx[] =
> > > +  {
> > > +    "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce", "cx8", "apic",
> > > +    NULL, "sep", "mtrr", "pge", "mca", "cmov", "pat", "pse36", "psn", "clfsh",
> > > +    NULL, "ds", "acpi", "mmx", "fxsr", "sse", "sse2", "ss", "htt", "tm",
> > > +   "ia64", "pbe"
> > > +  };
> > > +
> > > +static char * features_ecx[] =
> > > +  {
> > > +    "sse3", "pclmulqdq", "dtes64", "monitor", "ds-cpl", "vmx", "smx", "est",
> > > +    "tm3", "ssse3", "cnxt-id", "sdbg", "fma", "cx16", "xtpr", "pdcm",
> > > +    NULL, "pcid", "dca", "sse4_1", "sse4_2", "x2apic", "movbe", "popcnt",
> > > +   "tsc_deadline", "aes_ni", "xsave", "osxsave", "avx", "f16c", "rdrnd", 
> > > "hypervisor"
> > > +  };
> >
> > Are these names reflecting the gcc header cpuid.h? It seems there is some
> > difference with e.g. Linux naming (rdrnd vs rdrand), I'm not sure what would
> > be the best reference.
>
> These were taken from the Wikipedia article and replaced some dots and dashes
> with underscores.
>
> I didn't know if looking at Linux code would break some licensing stuff :/
> and the output of /proc/cpuinfo didn't tell me what bit to look for in each
> case.

From a quick check it seems the names on the publicly available Intel 
manuals are the same as Linux procinfo, (just uppercase), it might be 
worth to check what other OSs use, like some BSDs, if they have such file.

About licensing, I would be a bit surprised if these flag names couldn't 
be used in GPL source code, if they come from the manuals, but I am not 
an expert on this.

> > maybe this can be already made arch-specific.
>
> Hmm, maybe. As is it only makes sense for x86 based architectures.

There was some work from Sergey towards porting to aarch64, I guess 
currently the hurd should compile there, and without a guard it would break.

> What would be the right approach at making pieces of code architechture
> based on the Hurd repo? There isn't much code in that fashion, probably
> pfinet only.

Probably guarding the code with an #ifdef is enough, there are already 
some in proc and exec covering also aarch64. Otherwise you can move the 
arch-specific part to dedicated files.


Luca