Add User Mode registers

[vinaychandra]

No description provided.

[phil-opp]

Thanks a lot for the pull request! I left a few comments, but otherwise this looks good to me. Thanks especially for adding extensive documentation for everything!

[phil-opp]

Could you clarify why you're doing - 3 here?

[vinaychandra]

The values in the registers are or'ed with 0b11 for sysret registers regardless of what the user specified. We are subtracting that back so that the user can directly match with their gdt's user mode entry rather

[phil-opp]

The problem I see is that the user might also pass a segment selector with the requested privilege bits already set to 0b11. In this case, it would be very strange to get a selector with requested privilege level 0 back, i.e. not the same that was written.

[phil-opp]

I'm not sure if automatically setting the RPL to 0b11 is the right solution. The alternative would be to require the caller to properly set these bits. What do you think is the better solution?

[vinaychandra]

The amd64 documentation says it is required but will not be tested. That is why i implicitly left it like this. This also provides direct one to one mapping to the user segment functions from gdt. Also the user can use write_raw to test it otherwise.

[phil-opp]

This also provides direct one to one mapping to the user segment functions from gdt.

The user segment functions only return a GDT descriptor, not a segment selector. You mean the add_entry function? If yes, there is currently an open issue about whether we should hardcode the requested privilege level in that function: #120

[vinaychandra]

@phil-opp, updated as per comments

[phil-opp]

Thanks for the updates! I'm still not sure whether automatically adjusting the requested privilege level is a good idea (see the inline comments). Maybe @haraldh, who opened the related #120, has an opinion on this matter?

[haraldh]

Yeah, I think we should fix #120 and then assert for the bits, if set/not set at the corresponding places. Or maybe even better and idiomatic, have different SegmentSelector types according to the privilege bits

[vinaychandra]

Another option is to remove the read altogether and just go with read raw for now. What do you think @phil-opp?

[phil-opp]

While we could merge this with only read_raw/write_raw (i.e. without read/write), I think a better solution would be the following:

• Add a simple set_rpl(&mut self, rpl: PrivilegeLevel) method to the SegmentSelector type
• Remove the - 3 operation from read.
• Remove the | 0b11 operation from write. Instead, add an assert_eq!(ss_sysret.rpl(), PrivilegeLevel::Ring3) and mention this requirement in the function documentation.

What do you think about this, @vinaychandra?

[vinaychandra]

Makes sense, I updated the PR accordingly

[phil-opp]

Thanks a lot!

[phil-opp]

Published together with #118 as version 0.9.0.