Support interrupts binding in GICv3 and ITS

[zbb]

Summary

• Add MOVI command and routine for the LPI migration
• Allow to search for the ITS device descriptor using not only devID but also LPI number.
• Bind SPIs in the Distributor
• Don't bind its_dev to collection. Keep track of the collection IDs for each LPI.

Reviewed by:
Obtained from: Semihalf
Sponsored by: Cavium
Differential Revision:

[emaste]

sys/arm64/arm64/gic_v3_its.c

968–970 ↗

(On Diff #13123)

how is the its_dev protected outside of this region?

sys/arm64/arm64/gic_v3_var.h

136 ↗

(On Diff #13123)

Out of curiosity, what are the command #s based on?

[zbb]

sys/arm64/arm64/gic_v3_its.c

968–970 ↗

(On Diff #13123)

We lock the its_device_find_locked() to avoid mismatch in general itd_device list.
I assume that the interrupt resources that are managed through its_dev here are allocated and protected by the resource manager.
So there should be no another driver thread that is going to work on that its_dev.

However, after rethinking, I can see no point of having its_dev->col->col_id. I would probably change that in the next iteration.

sys/arm64/arm64/gic_v3_var.h

136 ↗

(On Diff #13123)

The numbers are defined by the ITS architecture. This is described in the GIC Architecture Specification PRD03-GENC-010745 24.0

[zbb]

• Don't bind its_dev to collection. Keep track of the collection IDs for each LPI.

[wma]

sys/arm64/arm64/gic_v3_its.c

921 ↗

(On Diff #13149)

Now, the execution of lpi_free_chunk is no longer guarded by the lock. I might not see all dependencies here, but please check before commiting if there is a possibility of use-after-free of lpi_col_ids if this function is reentered by mmultiple threads.

If so, it might be fixed as following:

tmp = lpic->lpi_col_ids;
lpic->lpi_col_ids = NULL;
dmb();
free(tmp);

[zbb]

sys/arm64/arm64/gic_v3_its.c

921 ↗

(On Diff #13149)

LPI chunk is per driver instance, hence per driver thread. The important thing is LPIs bitmap which is shared (and locked).