test_kernel_roundtrip_group_resilient()'s success depends on kernel HZ

[KanjiMonster]

test_kernel_roundtrip_group_resilient() sets timers and compares the set values after retrieving the objects from the kernel.

Unfortunately the timer values returned may differ depending on the configured HZ value of the running kernel. The kernel does not store the raw timer values in its database, but converts them to jiffies first, which may not convert cleanly.

I noticed this on Manjaro/arch, where the kernel uses CONFIG_HZ=300, which causes the conversion from seconds to jiffies to seconds to have a loss of precision, while e.g. Ubuntu with CONFIG_HZ=1000 is fine.

Grepping through the kernel tree, depending on the arch there are quite a few options for HZ, so I tested the values wether the conversion and back is neutral, and for most it isn't:

(seconds/input -> seconds to jiffies -> jiffies to seconds/output, with a ! if it doesn't match the initial value)

CONFIG_HZ=  24 => 15 ->   3 -> 12! 25 ->   6 -> 25 
CONFIG_HZ=  32 => 15 ->   4 -> 12! 25 ->   8 -> 25 
CONFIG_HZ=  48 => 15 ->   7 -> 14! 25 ->  12 -> 24!
CONFIG_HZ= 100 => 15 ->  15 -> 15  25 ->  25 -> 25 
CONFIG_HZ= 128 => 15 ->  19 -> 14! 25 ->  32 -> 25 
CONFIG_HZ= 250 => 15 ->  37 -> 14! 25 ->  62 -> 24!
CONFIG_HZ= 256 => 15 ->  38 -> 14! 25 ->  64 -> 25 
CONFIG_HZ= 300 => 15 ->  45 -> 14! 25 ->  75 -> 24!
CONFIG_HZ= 500 => 15 ->  75 -> 15  25 -> 125 -> 25 
CONFIG_HZ=1000 => 15 -> 150 -> 15  25 -> 250 -> 25 
CONFIG_HZ=1024 => 15 -> 153 -> 14! 25 -> 256 -> 25 
CONFIG_HZ=1200 => 15 -> 180 -> 14! 25 -> 300 -> 24!

So the only values working fine are 100, 500, and 1000.

Not sure if this counts as a kernel bug or if this is expected/allowed.

[thom311]

hi. Does #445 work for you?

[KanjiMonster]

Yep, works for me, and test now succeeds on arch with CONFIG_HZ=300. Also matches how I would have fixed that.

Not sure if we need to care for the very low hz values (<100), but since 12 is still different from the default value (AFAICT 120, which would result in 116-120), so this is fine.

[thom311]

fixed by 69231b6 .