On 7/13/22 13:54, Víctor Colombo wrote:
This patch series implements the 4 instructions added in Power ISA
3.1B:
- hashchk
- hashst
- hashchkp
- hashstp
To build it, you need to apply the following patches on top of master:
<20220701133507.740619-2-lucas.coutinho@eldorado.org.br>
<20220701133507.740619-3-lucas.coutinho@eldorado.org.br>
<20220712193741.59134-2-leandro.lupori@eldorado.org.br>
<20220712193741.59134-3-leandro.lupori@eldorado.org.br>
Working branch for ease of use can be found here:
https://github.com/PPC64/qemu/tree/vccolombo-hash-to-send-v3
What do you think about the choice to implement the hash algorithm
from the ground up, following the SIMON-like algorithm presented in
Power ISA? IIUC, this algorithm is not the same as the original[1].
Other options would be to use other algorithm already implemented
in QEMU, or even make this instruction a nop for all Power versions.
Also, I was thinking about using the call to spr_register_kvm() in
init_proc_POWER10 to initialize the registers with a random value.
I'm not sure what is the behavior here, I would expect that is the job
of the OS to set the regs, but looks like KVM is not exporting them,
so they are always 0 (?). Does anyone have any insight on this?
This happens because KVM on POWER10 isn't handling these registers
appropriately. We are probably missing kernel/kvm code to do so.
Since KVM on POWER10 is on an uncertain spot at this moment I wouldn't
worry too much about it. Making the regs read/write work in TCG is good
enough for now.
Daniel