Re: [PATCH v3 0/8] Generic IRQ entry/exit support for powerpc

[Shrikanth Hegde <sshegde@linux.ibm.com>]

On 12/29/25 10:24 AM, Mukesh Kumar Chaurasiya wrote:
> Adding support for the generic irq entry/exit handling for PowerPC. The
> goal is to bring PowerPC in line with other architectures that already
> use the common irq entry infrastructure, reducing duplicated code and
> making it easier to share future changes in entry/exit paths.
> 
> This is slightly tested of ppc64le and ppc32.
> 
> The performance benchmarks are below:
> 
> perf bench syscall usec/op
> 
> | Syscall | Base       | New        | change % |
> | ------- | ---------- | ---------- | -------- |
> | basic   | 0.173212   | 0.133294   | -23.05   |
> | execve  | 363.176190 | 357.148150 | -1.66    |
> | fork    | 853.657880 | 840.268800 | -1.57    |
> | getpgid | 0.174832   | 0.135372   | -22.57   |
> 

Could you please add a description for whether its an improvement
or regression.

Looking at it, it seems time taken for operation to complete. So it is
an improvement. Seeing negative numbers one might think its a regression.

Also, are these average numbers?
Could you get the numbers with preempt=lazy and with context tracking on?

> perf bench syscall ops/sec
> 
> | Syscall | Base    | New     | change % |
> | ------- | ------- | ------- | -------- |
> | basic   | 6006021 | 7502236 | +24.91   |
> | execve  | 2753    | 2800    | +1.71    |
> | fork    | 1171    | 1190    | +1.62    |
> | getpgid | 5942117 | 7387040 | +24.32   |
> 
> IPI latency benchmark
> 
> | Metric         | Base (ns)     | Test (ns)     | change % |
> | -------------- | ------------- | ------------- | -------- |
> | Dry-run        | 206652.45     | 209317.37     | +1.29    |
> | Self-IPI       | 3567895.23    | 3590444.77    | +0.63    |
> | Normal IPI     | 148570416.17  | 148564173.40  | -0.00    |
> | Broadcast IPI  | 4033489673.38 | 4007319512.62 | -0.65    |
> | Broadcast lock | 4011023005.48 | 4010267885.93 | -0.02    |
> 
>