refactor(interrupts): construct handler mappings during device init

[cagatay-y]

MSI-X will require support for more than one interrupt vector per driver and more information from the driver side to decide how to handle the interrupts correctly. Revise how the mappings are constructed to make the construction more flexible and prepare for that use case.

[github-actions]

Benchmark Results

Details

Benchmark	Current: b2838e5	Previous: 3fddec8	Performance Ratio
startup_benchmark Build Time	77.40 s	77.54 s	1.00 ❗
startup_benchmark File Size	0.77 MB	0.77 MB	1.00 ❗
Startup Time - 1 core	0.75 s (±0.02 s)	0.74 s (±0.02 s)	1.00
Startup Time - 2 cores	0.74 s (±0.02 s)	0.74 s (±0.02 s)	1.00
Startup Time - 4 cores	0.74 s (±0.02 s)	0.74 s (±0.02 s)	1.00
multithreaded_benchmark Build Time	79.03 s	78.02 s	1.01 ❗
multithreaded_benchmark File Size	0.87 MB	0.83 MB	1.04 ❗
Multithreaded Pi Efficiency - 2 Threads	87.98 % (±7.06 %)	88.15 % (±10.12 %)	1.00
Multithreaded Pi Efficiency - 4 Threads	43.84 % (±2.80 %)	44.00 % (±2.70 %)	1.00
Multithreaded Pi Efficiency - 8 Threads	25.52 % (±1.27 %)	25.61 % (±1.15 %)	1.00
micro_benchmarks Build Time	78.56 s	78.90 s	1.00 ❗
micro_benchmarks File Size	0.87 MB	0.83 MB	1.04 ❗
Scheduling time - 1 thread	65.78 ticks (±2.50 ticks)	63.55 ticks (±2.19 ticks)	1.04
Scheduling time - 2 threads	35.03 ticks (±3.48 ticks)	37.36 ticks (±5.81 ticks)	0.94
Micro - Time for syscall (getpid)	3.34 ticks (±0.54 ticks)	3.42 ticks (±0.66 ticks)	0.98
Memcpy speed - (built_in) block size 4096	82491.13 MByte/s (±57337.46 MByte/s)	85755.51 MByte/s (±59154.88 MByte/s)	0.96
Memcpy speed - (built_in) block size 1048576	30864.70 MByte/s (±24833.00 MByte/s)	30798.00 MByte/s (±24808.58 MByte/s)	1.00
Memcpy speed - (built_in) block size 16777216	27680.72 MByte/s (±22873.02 MByte/s)	27735.11 MByte/s (±22922.42 MByte/s)	1.00
Memset speed - (built_in) block size 4096	82666.97 MByte/s (±57435.51 MByte/s)	85561.11 MByte/s (±59022.96 MByte/s)	0.97
Memset speed - (built_in) block size 1048576	31599.45 MByte/s (±25274.35 MByte/s)	31540.44 MByte/s (±25246.30 MByte/s)	1.00
Memset speed - (built_in) block size 16777216	28478.29 MByte/s (±23379.32 MByte/s)	28522.65 MByte/s (±23408.89 MByte/s)	1.00
Memcpy speed - (rust) block size 4096	76981.89 MByte/s (±53610.29 MByte/s)	76278.05 MByte/s (±53195.57 MByte/s)	1.01
Memcpy speed - (rust) block size 1048576	30689.11 MByte/s (±24813.32 MByte/s)	30571.08 MByte/s (±24694.20 MByte/s)	1.00
Memcpy speed - (rust) block size 16777216	28015.11 MByte/s (±23151.44 MByte/s)	27092.68 MByte/s (±22306.35 MByte/s)	1.03
Memset speed - (rust) block size 4096	77428.31 MByte/s (±53927.51 MByte/s)	77081.89 MByte/s (±53714.43 MByte/s)	1.00
Memset speed - (rust) block size 1048576	31450.78 MByte/s (±25261.92 MByte/s)	31329.56 MByte/s (±25140.07 MByte/s)	1.00
Memset speed - (rust) block size 16777216	28814.21 MByte/s (±23653.36 MByte/s)	27893.05 MByte/s (±22822.05 MByte/s)	1.03
alloc_benchmarks Build Time	73.04 s	74.00 s	0.99 ❗
alloc_benchmarks File Size	0.85 MB	0.86 MB	1.00 ❗
Allocations - Allocation success	91.32 %	91.32 %	1
Allocations - Deallocation success	100.00 %	100.00 %	1
Allocations - Pre-fail Allocations	61.46 %	61.46 %	1
Allocations - Average Allocation time	2736.03 Ticks (±405.51 Ticks)	2517.26 Ticks (±526.46 Ticks)	1.09
Allocations - Average Allocation time (no fail)	3736.99 Ticks (±349.82 Ticks)	3544.10 Ticks (±455.05 Ticks)	1.05
Allocations - Average Deallocation time	859.27 Ticks (±92.57 Ticks)	674.85 Ticks (±85.89 Ticks)	1.27 ❗
mutex_benchmark Build Time	75.04 s	74.68 s	1.00 ❗
mutex_benchmark File Size	0.87 MB	0.84 MB	1.04 ❗
Mutex Stress Test Average Time per Iteration - 1 Threads	11.94 ns (±0.37 ns)	12.00 ns (±0.20 ns)	0.99
Mutex Stress Test Average Time per Iteration - 2 Threads	38.48 ns (±3.00 ns)	39.30 ns (±3.77 ns)	0.98

This comment was automatically generated by workflow using github-action-benchmark.

[mkroening]

Could you expand a bit on the overall design and how it is necessitated by MSI-X?

Why have fn init_dev(&mut self, handlers: &mut Map, irq: Option<Irq>) instead of fn interrupts(&self) -> impl Iterator<Item = Irq>, for example?

[cagatay-y]

Why have fn init_dev(&mut self, handlers: &mut Map, irq: Option<Irq>) instead of fn interrupts(&self) -> impl Iterator<Item = Irq>, for example?

MSI-X can have different interrupt vectors for device configuration interrupts and different queues. An implementation that uses an iterator based approach would require the caller to have a lot of knowledge about the functioning of the device, for example which number is for which kind of queue, which queues we are interested in the interrupts of, etc. An iterator that returns vector-handler pairs could overcome this issue but giving the drivers access to the handler mapping allows the drivers to determine which spots are free and thus avoid vector collisions, reducing the latency for handling.

[mkroening]

Thanks! :)