powerpc/64s: Avoid cpabort in context switch when possible

The ISA v3.0B copy-paste facility only requires cpabort when switching
to a process that has foreign real addresses mapped (direct access to
accelerators), to clear a potential copy buffer filled by a previous
thread. There is no accelerator driver implemented yet, so cpabort can
be removed. It can be be re-added when a driver is implemented.

POWER9 DD1 requires the copy buffer to always be cleared on context
switch, but if accelerators are not in use, then an unpaired copy from
a dummy region is sufficient to clear data out of the copy buffer.

This increases context switch performance by about 5% on POWER9.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

arch/powerpc/include/asm/ppc-opcode.h

@@ -189,8 +189,7 @@
 /* sorted alphabetically */
 #define PPC_INST_BHRBE			0x7c00025c
 #define PPC_INST_CLRBHRB		0x7c00035c
-#define PPC_INST_COPY			0x7c00060c
-#define PPC_INST_COPY_FIRST		0x7c20060c
+#define PPC_INST_COPY			0x7c20060c
 #define PPC_INST_CP_ABORT		0x7c00068c
 #define PPC_INST_DCBA			0x7c0005ec
 #define PPC_INST_DCBA_MASK		0xfc0007fe
@@ -223,8 +222,7 @@
 #define PPC_INST_MSGSNDP		0x7c00011c
 #define PPC_INST_MTTMR			0x7c0003dc
 #define PPC_INST_NOP			0x60000000
-#define PPC_INST_PASTE			0x7c00070c
-#define PPC_INST_PASTE_LAST		0x7c20070d
+#define PPC_INST_PASTE			0x7c20070d
 #define PPC_INST_POPCNTB		0x7c0000f4
 #define PPC_INST_POPCNTB_MASK		0xfc0007fe
 #define PPC_INST_POPCNTD		0x7c0003f4
@@ -392,6 +390,8 @@
 
 /* Deal with instructions that older assemblers aren't aware of */
 #define	PPC_CP_ABORT		stringify_in_c(.long PPC_INST_CP_ABORT)
+#define	PPC_COPY(a, b)		stringify_in_c(.long PPC_INST_COPY | \
+					___PPC_RA(a) | ___PPC_RB(b))
 #define	PPC_DCBAL(a, b)		stringify_in_c(.long PPC_INST_DCBAL | \
 					__PPC_RA(a) | __PPC_RB(b))
 #define	PPC_DCBZL(a, b)		stringify_in_c(.long PPC_INST_DCBZL | \

arch/powerpc/kernel/entry_64.S

@@ -536,15 +536,6 @@ _GLOBAL(_switch)
 	 * which contains larx/stcx, which will clear any reservation
 	 * of the task being switched.
 	 */
-
-BEGIN_FTR_SECTION
-/*
- * A cp_abort (copy paste abort) here ensures that when context switching, a
- * copy from one process can't leak into the paste of another.
- */
-	PPC_CP_ABORT
-END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
-
 #ifdef CONFIG_PPC_BOOK3S
 /* Cancel all explict user streams as they will have no use after context
  * switch and will stop the HW from creating streams itself

arch/powerpc/kernel/process.c

@@ -1137,6 +1137,11 @@ static inline void restore_sprs(struct thread_struct *old_thread,
 #endif
 }
 
+#ifdef CONFIG_PPC_BOOK3S_64
+#define CP_SIZE 128
+static const u8 dummy_copy_buffer[CP_SIZE] __attribute__((aligned(CP_SIZE)));
+#endif
+
 struct task_struct *__switch_to(struct task_struct *prev,
 	struct task_struct *new)
 {
@@ -1226,8 +1231,28 @@ struct task_struct *__switch_to(struct task_struct *prev,
 		batch->active = 1;
 	}
 
-	if (current_thread_info()->task->thread.regs)
+	if (current_thread_info()->task->thread.regs) {
 		restore_math(current_thread_info()->task->thread.regs);
+
+		/*
+		 * The copy-paste buffer can only store into foreign real
+		 * addresses, so unprivileged processes can not see the
+		 * data or use it in any way unless they have foreign real
+		 * mappings. We don't have a VAS driver that allocates those
+		 * yet, so no cpabort is required.
+		 */
+		if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
+			/*
+			 * DD1 allows paste into normal system memory, so we
+			 * do an unpaired copy here to clear the buffer and
+			 * prevent a covert channel being set up.
+			 *
+			 * cpabort is not used because it is quite expensive.
+			 */
+			asm volatile(PPC_COPY(%0, %1)
+					: : "r"(dummy_copy_buffer), "r"(0));
+		}
+	}
 #endif /* CONFIG_PPC_STD_MMU_64 */
 
 	return last;