On 2022/7/7 12:04, Xi Ruoyao wrote:Thanks for the Ruoyao's experiment.
On Thu, 2022-07-07 at 11:05 +0800, WANG Xuerui wrote:
To be frank, at this point I think you're trying to hide something.On a 3A5000LL, I did an experiment via a kernel module, which enables
(This is not your fault, blame someone else of course because they told
you the fact.) In the old-world kernel the VCSR a.k.a. FCSR16 is
certainly being saved/restored, and there's apparently no harm in doing
so. And if the contents are indeed "undefined", why are the code there
in the first place? Certainly the bits *are* meaningful, only that for
some reason you aren't revealing the semantics and pretending that they
are "undefined" and probably "do nothing externally observable" if
accessed in the first place.
LSX/LASX and tries to write and read fcsr16. I tried each bit (1, 2, 4,
8, ..., 1 << 31) one by one. The result: no matter which bit I wrote
into fcsr16, I always read out 0.
And I've objdump'ed a kernel shipped in an early Loongnix release. It
seems the only reference to fcsr16 is a "movgr2fcsr $r16, $r0"
instruction.
Hmm this is weird. I can't understand why the vcsr code was there in the first place then... I'd like to check a few Loongnix/Kylin/UOS kernels but currently I don't have the time.
If this is the case, indeed all vcsr-related code should be removed. Although I'm still not sure how to best word the commit message.