Re: [PATCH v3 2/4] mm: Make generic arch_is_kernel_initmem_freed() do what it says

Le 04/11/2021 à 22:44, Andrew Morton a écrit :

On Fri, 01 Oct 2021 17:14:41 +1000 Daniel Axtens [off-list ref] wrote:

quoted

quoted

  #ifdef __KERNEL__
+/*
+ * Check if an address is part of freed initmem. After initmem is freed,
+ * memory can be allocated from it, and such allocations would then have
+ * addresses within the range [_stext, _end].
+ */
+#ifndef arch_is_kernel_initmem_freed
+static int arch_is_kernel_initmem_freed(unsigned long addr)
+{
+	if (system_state < SYSTEM_FREEING_INITMEM)
+		return 0;
+
+	return init_section_contains((void *)addr, 1);

Is init_section_contains sufficient here?

include/asm-generic/sections.h says:
  * [__init_begin, __init_end]: contains .init.* sections, but .init.text.*
  *                   may be out of this range on some architectures.
  * [_sinittext, _einittext]: contains .init.text.* sections

init_section_contains only checks __init_*:
static inline bool init_section_contains(void *virt, size_t size)
{
	return memory_contains(__init_begin, __init_end, virt, size);
}

Do we need to check against _sinittext and _einittext?

Your proposed generic code will work for powerpc and s390 because those
archs only test against __init_* anyway. I don't know if any platform
actually does place .init.text outside of __init_begin=>__init_end, but
the comment seems to suggest that they could.

Christophe?

Sorry for answering late.

I've been thorugh free_initmem() in each architecture. The only sections 
involved in the freeing actions are [__init_begin, __init_end], so I 
think checking against __init_being, __init_end is enough.

If some architecture has init text outside of this section, then it is 
not freed hence not necessary to check.

Christophe