Make mstatus[VS] dirty when write to vstart

[rez5427]

fix #569

[github-actions]

Test Results

396 tests  ±0   396 ✅ ±0   0s ⏱️ ±0s
  4 suites ±0     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit e6a8923. ± Comparison against base commit 07fa23e.

♻️ This comment has been updated with latest results.

[jrtc27]

misa[V] == 0b1 implies sys_enable_vext(), and the commit message is extremely devoid of any detail. Per CONTRIBUTING.md:

Each commit should be self-contained, provide a meaningful short summary in the subject and, if not clear from the subject alone, provide more detail in the commit description.

But that's also just a common expectation for software engineering in general, it shouldn't need to be spelled out.

[Timmmm]

I'm not sure this is correct either. I think we can get rid of dirty_v_context_if_present() and just call dirty_v_context(). Currently it is only called from ext_write_vcsr and that should only happen if V is enabled.

Emphasis on should because currently the is_CSR_defineds are missing extensionEnabled(Ext_V):

function clause is_CSR_defined (0x009) = true

I opened #624 to track that.

I believe dirty_fd_context_if_present() exists because of Zfinx... but I don't think that's correct either. When you write to fcsr it should dirty mstatus[FS] even with Zfinx but that currently doesn't happen. I opened #625 to track that.

Also this doesn't completely fix #569 - that issue also says that there are missing calls to dirty_v_context....

[rez5427]

Maybe this is good?

[Timmmm]

This looks a lot better. There are still some more places where the context needs to be dirtied, e.g. search for vcsr[vxsat] = 0b1; (I would extract it into a set_vcsr_vxsat() function or something). There's also a gazillion places where vstart is written directly. I would make a set_vstart() too.

Doesn't have to be in this PR though. In fact maybe better if it isn't...

[Timmmm]

LGTM. I'll squash and fix the commit message when merging, which I will do in a few days if nobody objects.