iflib: Add sysctl to request extra MSIX vectors on driver load

[erj]

Summary

Intended to be used with upcoming feature to add sub-interfaces, since
those new interfaces will be dynamically created and will need to have
spare MSI-X interrupts already allocated for them on driver load.

While here, fix some alignment of sysctl declarations as well.

Signed-off-by: Eric Joyner <[e-mail]>

[marius]

sys/net/iflib.c

199

Currently, there's a mix and match of different sizes for actually boolean SYSCTLs. 'ifc_sysctl_use_extra_msix_vectors' also sounds like it's intent is to be an on/off knob rather than a count of vectors. Is there a particular reason for it to be of type uint16_t?

[erj]

sys/net/iflib.c

199

It is actually intended to be a count -- I think the name might not make it clear (especially since it comes after use_logical_cores which is supposed to act like a bool); I wasn't sure on what to use, so I went with this and thought about revisiting it later. Do you have a suggestion?

The variable doesn't get used in iflib proper because ice(4) handles its own MSI-X interrupt allocations (hence the public accessor function), but it's located here just in case other drivers would like to use it or whatever enhanced functionality ice(4) has gets pulled into iflib.

[marius]

sys/net/iflib.c

199

Well, I'd just drop the 'use_' parts then and make the SYSCTL description something like 'attempted reservation count of extra MSI-X vectors for runtime interface creation'. The latter might be worded more precisley depending on what 'try' means in this context, i. e. is failure to reserve these vectors fatal or not, will their number be reduced to what can acutally be handled etc.

[erj]

Update variable names and sysctl help string

[erj]

Rebase on latest main

[gallatin]

sys/net/iflib.c

6819

Why is this a tunable? It would seem to be calculated at driver load, when the vectors are allocated.

Speaking of that, maybe I'm being dense, but I don't see anything in this change that actually allocates extra vectors. What am I missing?

[gallatin]

sys/net/iflib.c

6819

Why is this a tunable? It would seem to be calculated at driver load, when the vectors are allocated.

Speaking of that, maybe I'm being dense, but I don't see anything in this change that actually allocates extra vectors. What am I missing?

Oh, this is some kind of diff stack. I can't seem to figure how how to see related changes in the web interface :(

[erj]

@gallatin Do you see the Stack tab?

Weirdly, the later diffs in the stack have the label in red with the number of open diffs in parentheses, like this: