src/mte_tag: tighten the first stage page table conditions for MTAG

[deepak0414]

Quite often loader or other piece of code might read code pages. If a code page is set as MTAG=1, then those loads (to code page) will look for tags. This change clarifies in specification that loads/stores to pages marked as executable always generate unchecked memory accesses.

Setting MTAG for XWR=111 and XWR=010 was disallowed. Instead of raising page fault, raise access-fault exception to signify fatality.

[deepak0414]

It came to my attention that there was a bug on normal (non-MPRV, non-HLV/HSV)
load/store side as well. A load to code page shouldn't lead to checked accesses.
@fmayer , I assume that should be fine. Let me know if it isn't.

@radimkrcmar, if you could spend few minutes on this as well. That would be appreciated.

[radimkrcmar]

@radimkrcmar, if you could spend few minutes on this as well. That would be appreciated.

I am a bit out of water here as I don't understand why XWR is reserved, as XR seems practically the same, so any decision about how it should interact with tagging seems arbitrary to me.

[deepak0414]

@radimkrcmar, if you could spend few minutes on this as well. That would be appreciated.

I am a bit out of water here as I don't understand why XWR is reserved, as XR seems practically the same, so any decision about how it should interact with tagging seems arbitrary to me.

MTAG = 1 with XWR because it creates a situation where code can be self modifying or code is operating on data which is on same page. As per the rules tag checks should elided. However any such situation wanting tag checks or elision doesn't make sense. So totally avoiding that encoding to be possible and keeping it reserved.

[deepak0414]

@radimkrcmar, if you could spend few minutes on this as well. That would be appreciated.

I am a bit out of water here as I don't understand why XWR is reserved, as XR seems practically the same, so any decision about how it should interact with tagging seems arbitrary to me.

MTAG = 1 with XWR because it creates a situation where code can be self modifying or code is operating on data which is on same page. As per the rules tag checks should elided. However any such situation wanting tag checks or elision doesn't make sense. So totally avoiding that encoding to be possible and keeping it reserved.

Aah
I think I got your point. In case of XWR, it's probably better to ignore MTAG like in case of XR with MTAG.
Basically all loads to a page with X=1 will generate unchecked mem accesses. And thus avoid special casing. Kind of make sense. Let me think on that for a bit.

[deepak0414]

@radimkrcmar, if you could spend few minutes on this as well. That would be appreciated.

I am a bit out of water here as I don't understand why XWR is reserved, as XR seems practically the same, so any decision about how it should interact with tagging seems arbitrary to me.

MTAG = 1 with XWR because it creates a situation where code can be self modifying or code is operating on data which is on same page. As per the rules tag checks should elided. However any such situation wanting tag checks or elision doesn't make sense. So totally avoiding that encoding to be possible and keeping it reserved.

Aah I think I got your point. In case of XWR, it's probably better to ignore MTAG like in case of XR with MTAG. Basically all loads to a page with X=1 will generate unchecked mem accesses. And thus avoid special casing. Kind of make sense. Let me think on that for a bit.

Did an update on PR.

[deepak0414]

@fmayer says that its easier to say simply that checked accesses come executable page with mtag=0 towards a data page with mtag = 1.
Revise PR and try to simplify rules.

[deepak0414]

Can anyone take a look, approve or drop a comment?