vm: Stop reducing vm_pageout_page_count

[alc]

Summary

Reducing vm_pageout_page_count when the machine has less than 8MB of
physical memory is pointless, since we haven't run on such machines in
ages.

[alc]

Should we also const-ify vm_pageout_page_count?

[kib]

If making it const, what is the point of keeping it as a variable (4 bytes + accessing them in code) instead of pre-processor symbol? This would also remove some VLAs from the kernel.

[markj]

In D46206#1053444, @alc wrote:

Should we also const-ify vm_pageout_page_count?

Yes please, it's used to size some stack-allocated arrays. It should probably be defined using the preprocessor, if only because that's what we'd typically do today when reimplementing that code.

[alc]

Convert vm_pageout_page_count to a constant.

[alc]

Tweak comment.

[kib]

BTW, I believe that the current size (32) was tailored for the old MAXPHYS 128k value.

Do we want to utilize the full maxphys-sized writes (now default to 1M, tunable), or at least tweak the pageout size according to maxphys? If yes, then in fact making vm_pageout_page_count constant is not right, but we also need to remove VLAs.

[alc]

In D46206#1053585, @kib wrote:

BTW, I believe that the current size (32) was tailored for the old MAXPHYS 128k value.

Me too.

Do we want to utilize the full maxphys-sized writes (now default to 1M, tunable), or at least tweak the pageout size according to maxphys? If yes, then in fact making vm_pageout_page_count constant is not right, but we also need to remove VLAs.

Probably, but I'm not sure. It seems more likely to me that vm_object_page_collect_flush() would benefit, than vm_pageout_cluster(). Do we have any existing benchmarks?

For now, I propose to commit just the originally proposed change to vm_pageout_init(). Okay?

[kib]

In D46206#1053659, @alc wrote:

For now, I propose to commit just the originally proposed change to vm_pageout_init(). Okay?

Sure.