Le 26/02/2020 à 07:24, Russell Currey a écrit :
There are a couple of reasons that the set_memory() functions are problematic when STRICT_KERNEL_RWX isn't enabled: - The linear mapping is a different size and apply_to_page_range() may modify a giant section, breaking everything
I don't understand.
- patch_instruction() doesn't know to work around a page being marked RO, and will subsequently crash
Is patch_instruction() involved at all ?
The latter can be replicated by building a kernel with the set_memory() patches but with STRICT_KERNEL_RWX off and running ftracetest. Reported-by: Jordan Niethe <jniethe5@xxxxxxxxx> Signed-off-by: Russell Currey <ruscur@xxxxxxxxxx> --- v4: new arch/powerpc/mm/pageattr.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/arch/powerpc/mm/pageattr.c b/arch/powerpc/mm/pageattr.c index ee6b5e3b7604..ff111930cf5e 100644 --- a/arch/powerpc/mm/pageattr.c +++ b/arch/powerpc/mm/pageattr.c @@ -96,12 +96,17 @@ static int set_page_attr(pte_t *ptep, unsigned long addr, void *data)int set_memory_attr(unsigned long addr, int numpages, pgprot_t prot)
Isn't it change_memory_attr() that is a problem for you ?
{
- unsigned long start = ALIGN_DOWN(addr, PAGE_SIZE);
- unsigned long sz = numpages * PAGE_SIZE;
+ unsigned long start, size;
+
+ if (!IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
+ return 0;
Doing this you break patch 7: mark_initmem_nx() is called regardless of CONFIG_STRICT_KERNEL_RWX__kernel_map_pages() depends on CONFIG_DEBUG_PAGEALLOC which doesn't depend on CONFIG_STRICT_KERNEL_RWX
if (!numpages)return 0;- return apply_to_page_range(&init_mm, start, sz, set_page_attr,+ start = ALIGN_DOWN(addr, PAGE_SIZE); + size = numpages * PAGE_SIZE; + + return apply_to_page_range(&init_mm, start, size, set_page_attr,
You don't need to move start and size calculation and change the above.
(void *)pgprot_val(prot)); }
Christophe
