Skip remaining PCI devices when device 0 is absent on a bus

[jbreitbart]

Summary

• Apply standard PCI enumeration optimization: if device 0 function 0 on a bus returns vendor ID 0xFFFF, skip the remaining device slots on that bus
• Reduces PCI legacy scan from 32 buses × 32 devices = 1024 probes to ~63 probes on uhyve (which has a single device at bus 0 device 0)

Performance

Measured with RDTSC-based boot instrumentation on uhyve (3.7 GHz TSC):

• PCI legacy scan: 13,237 us → 797 us (16.6x faster)
• Total boot to application: ~17,900 us → ~5,500 us (3.3x faster)

Test plan

• Built and booted hello_world example via uhyve, device at 00:00 still discovered correctly

🤖 Generated with Claude Code

[github-actions]

Benchmark Results

Details

Benchmark	Current: 596d628	Previous: dd4adf7	Performance Ratio
startup_benchmark Build Time	99.42 s	100.04 s	0.99 ❗
startup_benchmark File Size	0.86 MB	0.86 MB	1.00 ❗
Startup Time - 1 core	0.87 s (±0.03 s)	0.97 s (±0.03 s)	0.90 ❗
Startup Time - 2 cores	0.90 s (±0.03 s)	0.97 s (±0.03 s)	0.93 ❗
Startup Time - 4 cores	0.89 s (±0.03 s)	0.96 s (±0.04 s)	0.93 ❗
multithreaded_benchmark Build Time	102.32 s	101.89 s	1.00 ❗
multithreaded_benchmark File Size	0.96 MB	0.96 MB	1.00 ❗
Multithreaded Pi Efficiency - 2 Threads	90.72 % (±9.15 %)	91.00 % (±9.06 %)	1.00
Multithreaded Pi Efficiency - 4 Threads	44.55 % (±4.15 %)	43.85 % (±3.35 %)	1.02
Multithreaded Pi Efficiency - 8 Threads	26.26 % (±2.46 %)	25.74 % (±2.15 %)	1.02
micro_benchmarks Build Time	111.89 s	113.74 s	0.98 ❗
micro_benchmarks File Size	0.96 MB	0.96 MB	1.00 ❗
Scheduling time - 1 thread	71.27 ticks (±3.18 ticks)	70.24 ticks (±3.59 ticks)	1.01
Scheduling time - 2 threads	39.86 ticks (±5.29 ticks)	39.12 ticks (±4.49 ticks)	1.02
Micro - Time for syscall (getpid)	2.97 ticks (±0.20 ticks)	2.99 ticks (±0.29 ticks)	0.99
Memcpy speed - (built_in) block size 4096	63635.92 MByte/s (±45158.67 MByte/s)	65087.76 MByte/s (±46625.38 MByte/s)	0.98
Memcpy speed - (built_in) block size 1048576	30045.91 MByte/s (±25007.78 MByte/s)	29936.81 MByte/s (±24889.06 MByte/s)	1.00
Memcpy speed - (built_in) block size 16777216	25077.57 MByte/s (±20994.58 MByte/s)	25527.51 MByte/s (±21459.57 MByte/s)	0.98
Memset speed - (built_in) block size 4096	64370.40 MByte/s (±45726.54 MByte/s)	65536.56 MByte/s (±46897.60 MByte/s)	0.98
Memset speed - (built_in) block size 1048576	30813.46 MByte/s (±25447.25 MByte/s)	30723.73 MByte/s (±25327.45 MByte/s)	1.00
Memset speed - (built_in) block size 16777216	25807.48 MByte/s (±21447.22 MByte/s)	26327.86 MByte/s (±21978.97 MByte/s)	0.98
Memcpy speed - (rust) block size 4096	57979.32 MByte/s (±42784.34 MByte/s)	57856.95 MByte/s (±42652.78 MByte/s)	1.00
Memcpy speed - (rust) block size 1048576	30065.26 MByte/s (±25142.63 MByte/s)	30116.15 MByte/s (±25100.80 MByte/s)	1.00
Memcpy speed - (rust) block size 16777216	24377.68 MByte/s (±20550.82 MByte/s)	24868.32 MByte/s (±20982.14 MByte/s)	0.98
Memset speed - (rust) block size 4096	58833.97 MByte/s (±43243.81 MByte/s)	58845.68 MByte/s (±43409.33 MByte/s)	1.00
Memset speed - (rust) block size 1048576	30866.86 MByte/s (±25573.40 MByte/s)	30915.72 MByte/s (±25554.06 MByte/s)	1.00
Memset speed - (rust) block size 16777216	25161.37 MByte/s (±21071.74 MByte/s)	25661.65 MByte/s (±21502.61 MByte/s)	0.98
alloc_benchmarks Build Time	103.92 s	106.25 s	0.98 ❗
alloc_benchmarks File Size	0.93 MB	0.93 MB	1.00 ❗
Allocations - Allocation success	100.00 %	100.00 %	1
Allocations - Deallocation success	100.00 %	100.00 %	1
Allocations - Pre-fail Allocations	100.00 %	100.00 %	1
Allocations - Average Allocation time	7698.02 Ticks (±309.70 Ticks)	13150.35 Ticks (±161.03 Ticks)	0.59 ❗
Allocations - Average Allocation time (no fail)	7698.02 Ticks (±309.70 Ticks)	13150.35 Ticks (±161.03 Ticks)	0.59 ❗
Allocations - Average Deallocation time	2230.97 Ticks (±605.34 Ticks)	1127.38 Ticks (±770.43 Ticks)	1.98 ❗
mutex_benchmark Build Time	104.88 s	104.47 s	1.00 ❗
mutex_benchmark File Size	0.96 MB	0.96 MB	1.00 ❗
Mutex Stress Test Average Time per Iteration - 1 Threads	13.10 ns (±0.67 ns)	13.02 ns (±0.58 ns)	1.01
Mutex Stress Test Average Time per Iteration - 2 Threads	15.84 ns (±0.78 ns)	15.66 ns (±0.86 ns)	1.01

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

[stlankes]

Claud is right. I checked it on https://wiki.osdev.org/PCI.

[mkroening]

The device at bus 0, device 0 is the PCI host controller.

I can't find a source for the claim that we can just skip the whole bus if device 0 does not exist.

The closest I got is from the PCIe spec:

For non-VFs, a Vendor ID value of 0xffff indicates that no function is present.

That is only about functions, though, not buses. The recursive scan code from the OSDev wiki also checks entire buses but is smart about which buses to check via handling single and multiple PCI host controllers explicitly.