Rework managing hotplug commands with command completions.

[jhb]

Summary

Rework managing hotplug commands with command completions.

Previously the command completion interrupt would post any pending
command immediately, but before pcib_pcie_hotplug_update() had been
run to inspect the current status. Now, the command completion
interrupt merely clears the flag and stops the timer assuming that
the caller is always going to call pcib_pcie_hotplug_update() to
generate the next hotplug command if one is needed.

Test Plan

• only tested ExpressCard to ensure no regressions, but that doesn't actually use command completion interrupts. :(

[jhb]

sys/dev/pci/pci_pci.c

893–902

Since pcib_pcie_hotplug_update() always updates the full mask/value now, I think I can probably get rid of the pending things entirely and just write whatever values are needed for the current state after a command completion interrupt posts.

1022

This was the only place that wasn't already setting the mask on every update. Here though we want to be sure to not change the EI's state (so write a zero to the bit) if it's in the correct setting already.

[jhb]

• Drop the pending link fields entirely.
• Write the new value, not the old.

[jhb]

Updated commit message for removal of pending fields.

[jhb]

Refresh