在 2024/3/28 6:05, Borislav Petkov 写道:
On Sun, Feb 18, 2024 at 06:08:14PM +0800, Tong Tiangen wrote:
1. The copy_mc_to_kernel() is used in the coredump, KSM, and COW
scenarios, in these scenarios, the src mem stores the user data and the
kernel use kernel address to access the src mem(using kmap()).
2. the src mem of copy_mc_to_user() is currently only used by the DAX:
You mean just because it currently is used somewhere which probably is
ok - no clue what DAX does - and even if the source address is still
*kernel* memory and even at the danger that someone else might use it in
the future and think the handling on a potential #MC is ok, you're still
arguing that this is the right thing to do perhaps because it fits your
use case?!
Sorry Tiangen, not gonna happen.
I left the office last week and felt sorry for the lateness of the
reply.
You are right. Our current processing is based on "experience" rather
than interface constraints.
is_copy_from_user() determines whether a user is a "copy user" based on
fault_in_kernel_space(). Therefore, it returns false for
copy_mc_to_kernel()/copy_mc_to_user(). As a result, MCE_IN_KERNEL_COPYIN
cannot be set in error_context().
Comprehensive consideration of all factors, it is better to manually
call memory_failure_queue() to handle this problem case by case.
Finally, do we consider accepting only the patch 1/3 ?
Thanks,
Tong.
