target/riscv: Select hart in update_dcsr()

[timsifive]

Otherwise we may end up modifying DCSR of a different hart.

Fixes a bug introduced in f089815.

Change-Id: I39bde21a1444623ed150f2b3d504b9318b9d6191

[JanMatCodasip]

I have checked this only visually. Several notes:

1
I think this is a wider issue, which I'd like to address in the future: Virtually all functions accept the target as the argument but it is not very clear

• a) which of them will select the corresponding hart
• b) which of them simply assume that the caller (or some grand-caller) has selected the target already, and it can be directly used

For the functions in the b) category, my preference would be to add an assertion that checks the number of the last selected hart. I assume that would have caught this bug as well.

2
The commit message says ... in update_dcsr() but the change is actually made in register_read_direct().

3
register_write_direct() does not contain the call to dm013_select_target but register_read_direct() now has it. This asymmetry does not look right. It seems to me that adding the select_target call to update_dcsr() would make more sense.

[timsifive]

1. Agreed, but I don't think it's that big a problem. Just adding the occasional call to dm013_select_target() fixes it, and that seems better than an assert().
2. I'll move the change back to update_dcsr() then.

[JanMatCodasip]

Re-checked visually and looks all right. Thanks.