On Tue, Sep 7, 2021 at 7:57 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> After looking at the kcore code, it is found that kcore does not implement
> mmap, resulting in frequent context switching triggered by read.
> Therefore, we want to add mmap interface to optimize performance. Since
> vmalloc and module areas will change with allocation and release,
> consistency cannot be guaranteed, so mmap interface only maps KCORE_TEXT
> and KCORE_RAM.
Honestly, I still hate this patch.
The last time people wanted to speed up /dev/kcore accesses, it was
all for black-hat reasons and speeding up kernel attacks.
And this code just makes me nervous even aside from that, because I do
not understand what the heck it's doing.
> + if (kern_addr_valid(start)) {
> + if (m->type == KCORE_RAM)
> + pfn = __pa(start) >> PAGE_SHIFT;
> + else if (m->type == KCORE_TEXT)
> + pfn = __pa_symbol(start) >> PAGE_SHIFT;
Why is "__pa(start)" right in one situation, and "__pa_symbol(start)"
in another.
So this just makes me go "this is all confusing, dangerous, and the
use-case is dubious".
Mapping kernel memory is dangerous. The use-cases for it are dubious.
The patch isn't obvious.
All of that screams "I'll skip this".
Linus
