Re: [RFC PATCH] target/ppc: fix address translation bug for hash table mmus

[Richard Henderson]

On 6/8/21 7:39 AM, Bruno Piazera Larsen wrote:
> > That's odd.  We already have more arguments than the number of argument 
> > registers...  A 5x slowdown is distinctly odd.
> I did some more digging and the problem is not with ppc_radix64_check_prot, the 
> problem is ppc_radix64_xlate, which currently has 7 arguments and we're 
> increasing to 8. 7 feels like the correct number, but I couldn't find docs 
> supporting it, so I could be wrong.

According to tcg/ppc/tcg-target.c.inc, there are 8 argument registers for ppc 
hosts.  But now I see you didn't actually say on which host you observed the 
problem...  It's 6 argument registers for x86_64 host.

> That means we'd have to define radix_ctx_t (or however we call it) in 
> radix64.h, setup the struct on ppc_xlate, then pass it to ppc_radix64_xlate.

Well, if you're going to change the xlate interface, you want to do that across 
all of them.  So, not call it radix_ctx_t.

>  From looking at the code, it seems the most useful bits to put in the struct 
> are: eaddr, g_addr, h_addr, {h,g}_prot, {g,h}_page_size, mmu_idx and 
> guest_visible. They all seem reasonable to me, but I might be missing something 
> again.

I don't think h/g should be in this struct.  I think h/g should use different 
struct instances, because they are different accesses.


r~