nvme: Convert adjacent bools to a bitfield

[imp]

Summary

Neither timeout (meaning this request is allowed to timeout), nor
payload_valid (meaning this request has a memdesc that's good) are set
in a way that race anything (these are set during initializaiton and then
only tested).

Sponsored by: Netflix

[mav]

I suppose it may not save anything on 64-bit platforms due to alignment.

[jhb]

I kind of doubt this makes a difference due to padding (I suspect you have padding here both before and after if you check ptype /o in GDB). Probably there's a hole after retries that you could move these two down into that might remove some padding making the structure smaller.

[imp]

In D41312#941078, @jhb wrote:

I kind of doubt this makes a difference due to padding (I suspect you have padding here both before and after if you check ptype /o in GDB). Probably there's a hole after retries that you could move these two down into that might remove some padding making the structure smaller.

This has more to do with wanting more booleans... but I suppose that there's a total of 8 before we overflow so maybe I should just abandon this...

[chuck]

If you are looking to add more flag values, I'd move this to after retries and change the type to uint32_t or uint16_t if you think more might be needed in the future.

[imp]

In D41312#941277, @chuck wrote:

If you are looking to add more flag values, I'd move this to after retries and change the type to uint32_t or uint16_t if you think more might be needed in the future.

I've dropped this, and likely will move the bools to a 64-bit slot and make it be an array of 8 - #used-booleans to be explicit about padding.