Security flush selftests fail when page size == 4K

Several of the selftests for the security flush mitigations fail when the page size is 4K:

test: uaccess_flush_test
tags: git_version:v6.0-rc1-434-g3dab0a9cd867
PASS (L1D misses with uaccess_flush=1: 193986539 > 190000000) [10/10 pass]
FAIL (L1D misses with uaccess_flush=0: 134015968 > 95000000) [10/10 failures]
failure: uaccess_flush_test
...
test: entry_flush_test
tags: git_version:v6.0-rc1-434-g3dab0a9cd867
PASS (L1D misses with entry_flush=1: 190602271 > 190000000) [10/10 pass]
FAIL (L1D misses with entry_flush=0: 122961742 > 95000000) [10/10 failures]
failure: entry_flush_test
...
test: rfi_flush_test
tags: git_version:v6.0-rc1-434-g3dab0a9cd867
PASS (L1D misses with rfi_flush=1: 195022410 > 190000000) [10/10 pass]
FAIL (L1D misses with rfi_flush=0: 128320653 > 95000000) [10/10 failures]
failure: rfi_flush_test

In all cases it's clear that the flushes are affecting the number of cache misses, so the mitigation is in effect. But the thresholds we have for determining pass/fail are not correct when the page size is 4K.

To fix it we probably just need to add some logic that tweaks the threshold when the page size is 4K.

Start with the named uaccess_flush_test, entry_flush_test, and rfi_flush_test selftests and inspect how their cache-miss thresholds are selected. Adjust the 4K page-size threshold logic so the tests pass while still distinguishing enabled and disabled mitigations.