mm: tmpfs: huge=always Kconfig help understates severe space amplification (follow up)

[Artem S. Tashkinov]

Hi Dmitry,

One additional observation concerning diagnostic tools:

In this situation, ordinary du reports 45 GiB for a tree containing only 
1.4 GiB of apparent data.

Although this is technically correct allocation accounting, it makes 
identifying the cause quite difficult. Both df and du show an almost 
full filesystem, while neither explains that huge-page allocation 
overhead is responsible for most of the consumption.

Only comparing du against du --apparent-size exposed the scale of the 
discrepancy.

This reinforces the need for explicit Kconfig documentation: users 
investigating unexpectedly exhausted tmpfs filesystems may reasonably 
conclude that they have accumulated excessive data, rather than 
recognising that the selected allocation policy is consuming vastly more 
space than the contents require.

I am not suggesting that GNU du is itself defective. Rather, this is a 
diagnostic trap worth documenting alongside the configuration option.

Regards,
Artem