On Tue, May 23, 2017 at 1:06 PM, Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> 4.11-stable review patch. If anyone has any objections, please let me know.
No objections, but I do want to note that I verified with a
"allyesconfig" build on i386 that this code actually never triggers at
least in current -git. I might have screwed something up, but I added
a "fsin" instruction to that inline asm, and then verified that
"objdump" didn't find any fsin instructions in any object files after
the build.
There are people who do 64-bit reads from user space, but they all
seem to happily use "get_user()" that gets it right, not the
"__get_user()" helper that did not.
So it's still worth putting in stable (just in case of future or
external users), but at least it doesn't seem to be a potential
security issue (which it might have been - with the right unlucky code
generation you could make the 64-bit load basically be a pointer
following thing).
Linus