| Message ID | 20200515124808.213538-1-linus.walleij@linaro.org (mailing list archive) |
|---|---|
| State | Mainlined |
| Commit | e1de94380af588bdf6ad6f0cc1f75004c35bc096 |
| Headers | show |
| Series | [v2] ARM: mm: make act_mm() respect THREAD_SIZE | expand |
On Fri, May 15, 2020 at 02:48:08PM +0200, Linus Walleij wrote: > Recent work with KASan exposed the folling hard-coded bitmask > in arch/arm/mm/proc-macros.S: > > bic \rd, sp, #8128 > bic \rd, \rd, #63 > > This forms the bitmask 0x1FFF that is coinciding with > (PAGE_SIZE << THREAD_SIZE_ORDER) - 1, this code was assuming > that THREAD_SIZE is always 8K (8192). > > As KASan was increasing THREAD_SIZE_ORDER to 2, I ran into > this bug. > > Fix it by this little oneline suggested by Ard: > > bic \rd, sp, #(THREAD_SIZE - 1) & ~63 > > Where THREAD_SIZE is defined using THREAD_SIZE_ORDER. > > We have to also include <linux/const.h> since the THREAD_SIZE > expands to use the _AC() macro. > > Cc: Ard Biesheuvel <ardb@kernel.org> > Cc: Florian Fainelli <f.fainelli@gmail.com> > Suggested-by: Ard Biesheuvel <ardb@kernel.org> > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> > --- > ChangeLog v1->v2: > - Change from using THREAD_SIZE_ORDER with a hardcoded > page size constant to just using THREAD_SIZE - 1 > for the mask. > --- > arch/arm/mm/proc-macros.S | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/arch/arm/mm/proc-macros.S b/arch/arm/mm/proc-macros.S > index 5461d589a1e2..60ac7c5999a9 100644 > --- a/arch/arm/mm/proc-macros.S > +++ b/arch/arm/mm/proc-macros.S > @@ -5,6 +5,7 @@ > * VMA_VM_FLAGS > * VM_EXEC > */ > +#include <linux/const.h> > #include <asm/asm-offsets.h> > #include <asm/thread_info.h> > > @@ -30,7 +31,7 @@ > * act_mm - get current->active_mm > */ > .macro act_mm, rd > - bic \rd, sp, #8128 > + bic \rd, sp, #(THREAD_SIZE - 1) & ~63 > bic \rd, \rd, #63 We have a get_thread_info macro in asm/assembler that performs the same task. Maybe this should be converted to use that, and maybe the macro should be updated to use bic, since this seems to be acceptable for Thumb and is one instruction shorter.
diff --git a/arch/arm/mm/proc-macros.S b/arch/arm/mm/proc-macros.S index 5461d589a1e2..60ac7c5999a9 100644 --- a/arch/arm/mm/proc-macros.S +++ b/arch/arm/mm/proc-macros.S @@ -5,6 +5,7 @@ * VMA_VM_FLAGS * VM_EXEC */ +#include <linux/const.h> #include <asm/asm-offsets.h> #include <asm/thread_info.h> @@ -30,7 +31,7 @@ * act_mm - get current->active_mm */ .macro act_mm, rd - bic \rd, sp, #8128 + bic \rd, sp, #(THREAD_SIZE - 1) & ~63 bic \rd, \rd, #63 ldr \rd, [\rd, #TI_TASK] .if (TSK_ACTIVE_MM > IMM12_MASK)
Recent work with KASan exposed the folling hard-coded bitmask in arch/arm/mm/proc-macros.S: bic \rd, sp, #8128 bic \rd, \rd, #63 This forms the bitmask 0x1FFF that is coinciding with (PAGE_SIZE << THREAD_SIZE_ORDER) - 1, this code was assuming that THREAD_SIZE is always 8K (8192). As KASan was increasing THREAD_SIZE_ORDER to 2, I ran into this bug. Fix it by this little oneline suggested by Ard: bic \rd, sp, #(THREAD_SIZE - 1) & ~63 Where THREAD_SIZE is defined using THREAD_SIZE_ORDER. We have to also include <linux/const.h> since the THREAD_SIZE expands to use the _AC() macro. Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Florian Fainelli <f.fainelli@gmail.com> Suggested-by: Ard Biesheuvel <ardb@kernel.org> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> --- ChangeLog v1->v2: - Change from using THREAD_SIZE_ORDER with a hardcoded page size constant to just using THREAD_SIZE - 1 for the mask. --- arch/arm/mm/proc-macros.S | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)