On Wed, Apr 29, 2020 at 11:27:10AM +0100, Kevin Brodsky wrote: > On 21/04/2020 15:25, Catalin Marinas wrote: > > diff --git a/arch/arm64/lib/mte.S b/arch/arm64/lib/mte.S > > index bd51ea7e2fcb..45be04a8c73c 100644 > > --- a/arch/arm64/lib/mte.S > > +++ b/arch/arm64/lib/mte.S > > @@ -5,6 +5,7 @@ > > #include <linux/linkage.h> > > #include <asm/assembler.h> > > +#include <asm/mte.h> > > /* > > * Compare tags of two pages > > @@ -44,3 +45,52 @@ SYM_FUNC_START(mte_memcmp_pages) > > ret > > SYM_FUNC_END(mte_memcmp_pages) > > + > > +/* > > + * Read tags from a user buffer (one tag per byte) and set the corresponding > > + * tags at the given kernel address. Used by PTRACE_POKEMTETAGS. > > + * x0 - kernel address (to) > > + * x1 - user buffer (from) > > + * x2 - number of tags/bytes (n) > > + * Returns: > > + * x0 - number of tags read/set > > + */ > > +SYM_FUNC_START(mte_copy_tags_from_user) > > + mov x3, x1 > > +1: > > +USER(2f, ldtrb w4, [x1]) > > Here we are making either of the following assumptions: > 1. The __user pointer (here `from`) actually points to user memory, not > kernel memory (and we have set_fs(USER_DS) in place). > 2. CONFIG_ARM64_UAO is enabled and the hardware implements UAO. > > 1. is currently true because these functions are only used for the new > ptrace requests, which indeed pass pointers to user memory. However, future > users of these functions may not know about this requirement. > 2. is not necessarily true because ARM64_MTE does not depend on ARM64_UAO. > > It is unlikely that future users of these functions actually need to pass > __user pointers to kernel memory, so adding a comment spelling out the first > assumption is probably fine. I found it easier to add uao_user_alternative rather than writing a comment ;). Thanks. -- Catalin
