powerpc/32s: Do kuep_lock() and kuep_unlock() in assembly

When interrupt and syscall entries where converted to C, KUEP locking
and unlocking was also converted. It improved performance by unrolling
the loop, and allowed easily implementing boot time deactivation of
KUEP.

However, null_syscall selftest shows that KUEP is still heavy
(361 cycles with KUEP, 212 cycles without).

A way to improve more is to group 'mtsr's together, instead of
repeating 'addi' + 'mtsr' several times.

In order to do that, more registers need to be available. In C, GCC
will always be able to provide the requested number of registers, but
at the cost of saving some data on the stack, which is counter
performant here.

So let's do it in assembly, when we have full control of which
register can be used. It also has the advantage of locking earlier
and unlocking later and it helps GCC generating less tricky code.
The only drawback is to make boot time deactivation less straight
forward and require 'hand' instruction patching.

Group 'mtsr's by 4.

With this change, null_syscall selftest reports 336 cycles. Without
the change it was 361 cycles, that's a 7% reduction.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/115cb279e9b9948dfd93a065e047081c59e3a2a6.1634627931.git.christophe.leroy@csgroup.eu

1 files changed, 76 insertions, 1 deletions

diff --git a/arch/powerpc/include/asm/book3s/32/mmu-hash.h b/arch/powerpc/include/asm/book3s/32/mmu-hash.h
index f5be185cbdf8..e2f7ccc13edb 100644
--- a/arch/powerpc/include/asm/book3s/32/mmu-hash.h
+++ b/arch/powerpc/include/asm/book3s/32/mmu-hash.h
@@ -64,7 +64,82 @@ struct ppc_bat {
 #define SR_KP	0x20000000	/* User key */
 #define SR_KS	0x40000000	/* Supervisor key */
 
-#ifndef __ASSEMBLY__
+#ifdef __ASSEMBLY__
+
+#include <asm/asm-offsets.h>
+
+.macro uus_addi sr reg1 reg2 imm
+	.if NUM_USER_SEGMENTS > \sr
+	addi	\reg1,\reg2,\imm
+	.endif
+.endm
+
+.macro uus_mtsr sr reg1
+	.if NUM_USER_SEGMENTS > \sr
+	mtsr	\sr, \reg1
+	.endif
+.endm
+
+/*
+ * This isync() shouldn't be necessary as the kernel is not excepted to run
+ * any instruction in userspace soon after the update of segments and 'rfi'
+ * instruction is used to return to userspace, but hash based cores
+ * (at least G3) seem to exhibit a random behaviour when the 'isync' is not
+ * there. 603 cores don't have this behaviour so don't do the 'isync' as it
+ * saves several CPU cycles.
+ */
+.macro uus_isync
+#ifdef CONFIG_PPC_BOOK3S_604
+BEGIN_MMU_FTR_SECTION
+	isync
+END_MMU_FTR_SECTION_IFSET(MMU_FTR_HPTE_TABLE)
+#endif
+.endm
+
+.macro update_user_segments_by_4 tmp1 tmp2 tmp3 tmp4
+	uus_addi	1, \tmp2, \tmp1, 0x111
+	uus_addi	2, \tmp3, \tmp1, 0x222
+	uus_addi	3, \tmp4, \tmp1, 0x333
+
+	uus_mtsr	0, \tmp1
+	uus_mtsr	1, \tmp2
+	uus_mtsr	2, \tmp3
+	uus_mtsr	3, \tmp4
+
+	uus_addi	4, \tmp1, \tmp1, 0x444
+	uus_addi	5, \tmp2, \tmp2, 0x444
+	uus_addi	6, \tmp3, \tmp3, 0x444
+	uus_addi	7, \tmp4, \tmp4, 0x444
+
+	uus_mtsr	4, \tmp1
+	uus_mtsr	5, \tmp2
+	uus_mtsr	6, \tmp3
+	uus_mtsr	7, \tmp4
+
+	uus_addi	8, \tmp1, \tmp1, 0x444
+	uus_addi	9, \tmp2, \tmp2, 0x444
+	uus_addi	10, \tmp3, \tmp3, 0x444
+	uus_addi	11, \tmp4, \tmp4, 0x444
+
+	uus_mtsr	8, \tmp1
+	uus_mtsr	9, \tmp2
+	uus_mtsr	10, \tmp3
+	uus_mtsr	11, \tmp4
+
+	uus_addi	12, \tmp1, \tmp1, 0x444
+	uus_addi	13, \tmp2, \tmp2, 0x444
+	uus_addi	14, \tmp3, \tmp3, 0x444
+	uus_addi	15, \tmp4, \tmp4, 0x444
+
+	uus_mtsr	12, \tmp1
+	uus_mtsr	13, \tmp2
+	uus_mtsr	14, \tmp3
+	uus_mtsr	15, \tmp4
+
+	uus_isync
+.endm
+
+#else
 
 /*
  * This macro defines the mapping from contexts to VSIDs (virtual