x86: tsc: more resilient cpuid parsing

[kevans]

Summary

It seems unlikely, but we might as well defend against garbage in the
lower bytes -- better to return false more directly than to act on
elements outside of the brand array.

[imp]

Why 5, but regardless of why, this is correct

[markj]

In D48331#1102183, @imp wrote:

Why 5, but regardless of why, this is correct

A bit further down there's an assignment p -= 5.

[kevans]

In D48331#1102305, @markj wrote:

In D48331#1102183, @imp wrote:

Why 5, but regardless of why, this is correct

A bit further down there's an assignment p -= 5.

Yeah, I was trying to explain this with the comment. and don't seem to have quite worded it well enough.. we expect four digits + (M, G, or T) before we see 'Hz', so logically it makes sense to start the search there since we won't be able to parse a valid frequency in-bounds before that.

[markj]

sys/x86/x86/tsc.c

217

220

The mention of "potential garbage" perhaps distracts from the explanation of where 5 comes from. Maybe explain that in a separate paragraph? What kind of garbage might come up?

[imp]

In D48331#1102305, @markj wrote:

In D48331#1102183, @imp wrote:

Why 5, but regardless of why, this is correct

A bit further down there's an assignment p -= 5.

Where does that 5 come from was my real question. What spec says we know this format

[kevans]

In D48331#1102324, @imp wrote:

In D48331#1102305, @markj wrote:

In D48331#1102183, @imp wrote:

Why 5, but regardless of why, this is correct

A bit further down there's an assignment p -= 5.

Where does that 5 come from was my real question. What spec says we know this format

In the document referenced earlier: http://www.intel.com/assets/pdf/appnote/241618.pdf -- figure 3-10

[jkim]

LGTM

[kevans]

Drop the garbage reference entirely, expand on the description to note the
figure used. Also note some possible future work to rewrite the later part of
this to more closely follow the algorithm described, as they don't seem to want
to commit to D.DD[MGT]Hz or DDDD[MGT]Hz -- they specifically recommend scanning
for a prior space. This would also offer us an opportunity to validate that
all of the involved digits are actually [0-9] or '.'.

[kib]

Note that appnote is deprecated. Intel incorporated its content into SDM and always references CPUID description in SDM now.

Also when doing 'hard' references like page numbers/picture numbers, please specify the document revision.