powerpc/64s: Implement interrupt exit logic in C

Implement the bulk of interrupt return logic in C. The asm return code
must handle a few cases: restoring full GPRs, and emulating stack
store.

The stack store emulation is significantly simplfied, rather than
creating a new return frame and switching to that before performing
the store, it uses the PACA to keep a scratch register around to
perform the store.

The asm return code is moved into 64e for now. The new logic has made
allowance for 64e, but I don't have a full environment that works well
to test it, and even booting in emulated qemu is not great for stress
testing. 64e shouldn't be too far off working with this, given a bit
more testing and auditing of the logic.

This is slightly faster on a POWER9 (page fault speed increases about
1.1%), probably due to reduced mtmsrd.

mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE
handling (including the fast_interrupt_return path), to remove
trace_hardirqs_on(), and fixes the interrupt-return part of the
MSR_VSX restore bug caught by tm-unavailable selftest.

mpe: Incorporate fix from Nick:

The return-to-kernel path has to replay any soft-pending interrupts if
it is returning to a context that had interrupts soft-enabled. It has
to do this carefully and avoid plain enabling interrupts if this is an
irq context, which can cause multiple nesting of interrupts on the
stack, and other unexpected issues.

The code which avoided this case got the soft-mask state wrong, and
marked interrupts as enabled before going around again to retry. This
seems to be mostly harmless except when PREEMPT=y, this calls
preempt_schedule_irq with irqs apparently enabled and runs into a BUG
in kernel/sched/core.c

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com

1 files changed, 117 insertions, 370 deletions

diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 29949bbe857b..5d782acb86d4 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -16,6 +16,7 @@
 
 #include <linux/errno.h>
 #include <linux/err.h>
+#include <asm/cache.h>
 #include <asm/unistd.h>
 #include <asm/processor.h>
 #include <asm/page.h>
@@ -228,6 +229,7 @@ _GLOBAL(ret_from_kernel_thread)
 	li	r3,0
 	b	.Lsyscall_exit
 
+#ifdef CONFIG_PPC_BOOK3E
 /* Save non-volatile GPRs, if not already saved. */
 _GLOBAL(save_nvgprs)
 	ld	r11,_TRAP(r1)
@@ -238,6 +240,7 @@ _GLOBAL(save_nvgprs)
 	std	r0,_TRAP(r1)
 	blr
 _ASM_NOKPROBE_SYMBOL(save_nvgprs);
+#endif
 
 #ifdef CONFIG_PPC_BOOK3S_64
 
@@ -301,7 +304,7 @@ flush_count_cache:
  * state of one is saved on its kernel stack.  Then the state
  * of the other is restored from its kernel stack.  The memory
  * management hardware is updated to the second process's state.
- * Finally, we can return to the second process, via ret_from_except.
+ * Finally, we can return to the second process, via interrupt_return.
  * On entry, r3 points to the THREAD for the current task, r4
  * points to the THREAD for the new task.
  *
@@ -453,408 +456,152 @@ END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
 	addi	r1,r1,SWITCH_FRAME_SIZE
 	blr
 
-	.align	7
-_GLOBAL(ret_from_except)
-	ld	r11,_TRAP(r1)
-	andi.	r0,r11,1
-	bne	ret_from_except_lite
-	REST_NVGPRS(r1)
-
-_GLOBAL(ret_from_except_lite)
+#ifdef CONFIG_PPC_BOOK3S
 	/*
-	 * Disable interrupts so that current_thread_info()->flags
-	 * can't change between when we test it and when we return
-	 * from the interrupt.
-	 */
-#ifdef CONFIG_PPC_BOOK3E
-	wrteei	0
-#else
-	li	r10,MSR_RI
-	mtmsrd	r10,1		  /* Update machine state */
-#endif /* CONFIG_PPC_BOOK3E */
+	 * If MSR EE/RI was never enabled, IRQs not reconciled, NVGPRs not
+	 * touched, AMR not set, no exit work created, then this can be used.
+	 */
+	.balign IFETCH_ALIGN_BYTES
+	.globl fast_interrupt_return
+fast_interrupt_return:
+_ASM_NOKPROBE_SYMBOL(fast_interrupt_return)
+	ld	r4,_MSR(r1)
+	andi.	r0,r4,MSR_PR
+	bne	.Lfast_user_interrupt_return
+	andi.	r0,r4,MSR_RI
+	li	r3,0 /* 0 return value, no EMULATE_STACK_STORE */
+	bne+	.Lfast_kernel_interrupt_return
+	addi	r3,r1,STACK_FRAME_OVERHEAD
+	bl	unrecoverable_exception
+	b	. /* should not get here */
 
-	ld	r9, PACA_THREAD_INFO(r13)
-	ld	r3,_MSR(r1)
-#ifdef CONFIG_PPC_BOOK3E
-	ld	r10,PACACURRENT(r13)
-#endif /* CONFIG_PPC_BOOK3E */
-	ld	r4,TI_FLAGS(r9)
-	andi.	r3,r3,MSR_PR
-	beq	resume_kernel
-#ifdef CONFIG_PPC_BOOK3E
-	lwz	r3,(THREAD+THREAD_DBCR0)(r10)
-#endif /* CONFIG_PPC_BOOK3E */
+	.balign IFETCH_ALIGN_BYTES
+	.globl interrupt_return
+interrupt_return:
+_ASM_NOKPROBE_SYMBOL(interrupt_return)
+	REST_NVGPRS(r1)
 
-	/* Check current_thread_info()->flags */
-	andi.	r0,r4,_TIF_USER_WORK_MASK
-	bne	1f
-#ifdef CONFIG_PPC_BOOK3E
-	/*
-	 * Check to see if the dbcr0 register is set up to debug.
-	 * Use the internal debug mode bit to do this.
-	 */
-	andis.	r0,r3,DBCR0_IDM@h
-	beq	restore
-	mfmsr	r0
-	rlwinm	r0,r0,0,~MSR_DE	/* Clear MSR.DE */
-	mtmsr	r0
-	mtspr	SPRN_DBCR0,r3
-	li	r10, -1
-	mtspr	SPRN_DBSR,r10
-	b	restore
-#else
-	addi	r3,r1,STACK_FRAME_OVERHEAD
-	bl	restore_math
-	b	restore
-#endif
-1:	andi.	r0,r4,_TIF_NEED_RESCHED
-	beq	2f
-	bl	restore_interrupts
-	SCHEDULE_USER
-	b	ret_from_except_lite
-2:
-#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
-	andi.	r0,r4,_TIF_USER_WORK_MASK & ~_TIF_RESTORE_TM
-	bne	3f		/* only restore TM if nothing else to do */
+	.balign IFETCH_ALIGN_BYTES
+	.globl interrupt_return_lite
+interrupt_return_lite:
+_ASM_NOKPROBE_SYMBOL(interrupt_return_lite)
+	ld	r4,_MSR(r1)
+	andi.	r0,r4,MSR_PR
+	beq	.Lkernel_interrupt_return
 	addi	r3,r1,STACK_FRAME_OVERHEAD
-	bl	restore_tm_state
-	b	restore
-3:
-#endif
-	bl	save_nvgprs
-	/*
-	 * Use a non volatile GPR to save and restore our thread_info flags
-	 * across the call to restore_interrupts.
-	 */
-	mr	r30,r4
-	bl	restore_interrupts
-	mr	r4,r30
-	addi	r3,r1,STACK_FRAME_OVERHEAD
-	bl	do_notify_resume
-	b	ret_from_except
-
-resume_kernel:
-	/* check current_thread_info, _TIF_EMULATE_STACK_STORE */
-	andis.	r8,r4,_TIF_EMULATE_STACK_STORE@h
-	beq+	1f
+	bl	interrupt_exit_user_prepare
+	cmpdi	r3,0
+	bne-	.Lrestore_nvgprs
 
-	addi	r8,r1,INT_FRAME_SIZE	/* Get the kprobed function entry */
+.Lfast_user_interrupt_return:
+	ld	r11,_NIP(r1)
+	ld	r12,_MSR(r1)
+BEGIN_FTR_SECTION
+	ld	r10,_PPR(r1)
+	mtspr	SPRN_PPR,r10
+END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
+	mtspr	SPRN_SRR0,r11
+	mtspr	SPRN_SRR1,r12
 
-	ld	r3,GPR1(r1)
-	subi	r3,r3,INT_FRAME_SIZE	/* dst: Allocate a trampoline exception frame */
-	mr	r4,r1			/* src:  current exception frame */
-	mr	r1,r3			/* Reroute the trampoline frame to r1 */
+BEGIN_FTR_SECTION
+	stdcx.	r0,0,r1		/* to clear the reservation */
+FTR_SECTION_ELSE
+	ldarx	r0,0,r1
+ALT_FTR_SECTION_END_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
 
-	/* Copy from the original to the trampoline. */
-	li	r5,INT_FRAME_SIZE/8	/* size: INT_FRAME_SIZE */
-	li	r6,0			/* start offset: 0 */
-	mtctr	r5
-2:	ldx	r0,r6,r4
-	stdx	r0,r6,r3
-	addi	r6,r6,8
-	bdnz	2b
-
-	/* Do real store operation to complete stdu */
-	ld	r5,GPR1(r1)
-	std	r8,0(r5)
-
-	/* Clear _TIF_EMULATE_STACK_STORE flag */
-	lis	r11,_TIF_EMULATE_STACK_STORE@h
-	addi	r5,r9,TI_FLAGS
-0:	ldarx	r4,0,r5
-	andc	r4,r4,r11
-	stdcx.	r4,0,r5
-	bne-	0b
-1:
-
-#ifdef CONFIG_PREEMPTION
-	/* Check if we need to preempt */
-	andi.	r0,r4,_TIF_NEED_RESCHED
-	beq+	restore
-	/* Check that preempt_count() == 0 and interrupts are enabled */
-	lwz	r8,TI_PREEMPT(r9)
-	cmpwi	cr0,r8,0
-	bne	restore
-	ld	r0,SOFTE(r1)
-	andi.	r0,r0,IRQS_DISABLED
-	bne	restore
+	ld	r3,_CCR(r1)
+	ld	r4,_LINK(r1)
+	ld	r5,_CTR(r1)
+	ld	r6,_XER(r1)
+	li	r0,0
 
-	/*
-	 * Here we are preempting the current task. We want to make
-	 * sure we are soft-disabled first and reconcile irq state.
-	 */
-	RECONCILE_IRQ_STATE(r3,r4)
-	bl	preempt_schedule_irq
+	REST_4GPRS(7, r1)
+	REST_2GPRS(11, r1)
+	REST_GPR(13, r1)
 
-	/*
-	 * arch_local_irq_restore() from preempt_schedule_irq above may
-	 * enable hard interrupt but we really should disable interrupts
-	 * when we return from the interrupt, and so that we don't get
-	 * interrupted after loading SRR0/1.
-	 */
-#ifdef CONFIG_PPC_BOOK3E
-	wrteei	0
-#else
-	li	r10,MSR_RI
-	mtmsrd	r10,1		  /* Update machine state */
-#endif /* CONFIG_PPC_BOOK3E */
-#endif /* CONFIG_PREEMPTION */
+	mtcr	r3
+	mtlr	r4
+	mtctr	r5
+	mtspr	SPRN_XER,r6
 
-	.globl	fast_exc_return_irq
-fast_exc_return_irq:
-restore:
-	/*
-	 * This is the main kernel exit path. First we check if we
-	 * are about to re-enable interrupts
-	 */
-	ld	r5,SOFTE(r1)
-	lbz	r6,PACAIRQSOFTMASK(r13)
-	andi.	r5,r5,IRQS_DISABLED
-	bne	.Lrestore_irq_off
+	REST_4GPRS(2, r1)
+	REST_GPR(6, r1)
+	REST_GPR(0, r1)
+	REST_GPR(1, r1)
+	RFI_TO_USER
+	b	.	/* prevent speculative execution */
 
-	/* We are enabling, were we already enabled ? Yes, just return */
-	andi.	r6,r6,IRQS_DISABLED
-	beq	cr0,.Ldo_restore
+.Lrestore_nvgprs:
+	REST_NVGPRS(r1)
+	b	.Lfast_user_interrupt_return
 
-	/*
-	 * We are about to soft-enable interrupts (we are hard disabled
-	 * at this point). We check if there's anything that needs to
-	 * be replayed first.
-	 */
-	lbz	r0,PACAIRQHAPPENED(r13)
-	cmpwi	cr0,r0,0
-	bne-	.Lrestore_check_irq_replay
+	.balign IFETCH_ALIGN_BYTES
+.Lkernel_interrupt_return:
+	addi	r3,r1,STACK_FRAME_OVERHEAD
+	bl	interrupt_exit_kernel_prepare
 
-	/*
-	 * Get here when nothing happened while soft-disabled, just
-	 * soft-enable and move-on. We will hard-enable as a side
-	 * effect of rfi
-	 */
-.Lrestore_no_replay:
-	TRACE_ENABLE_INTS
-	li	r0,IRQS_ENABLED
-	stb	r0,PACAIRQSOFTMASK(r13);
+.Lfast_kernel_interrupt_return:
+	cmpdi	cr1,r3,0
+	ld	r11,_NIP(r1)
+	ld	r12,_MSR(r1)
+	mtspr	SPRN_SRR0,r11
+	mtspr	SPRN_SRR1,r12
 
-	/*
-	 * Final return path. BookE is handled in a different file
-	 */
-.Ldo_restore:
-#ifdef CONFIG_PPC_BOOK3E
-	b	exception_return_book3e
-#else
-	/*
-	 * Clear the reservation. If we know the CPU tracks the address of
-	 * the reservation then we can potentially save some cycles and use
-	 * a larx. On POWER6 and POWER7 this is significantly faster.
-	 */
 BEGIN_FTR_SECTION
 	stdcx.	r0,0,r1		/* to clear the reservation */
 FTR_SECTION_ELSE
-	ldarx	r4,0,r1
+	ldarx	r0,0,r1
 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
 
-	/*
-	 * Some code path such as load_up_fpu or altivec return directly
-	 * here. They run entirely hard disabled and do not alter the
-	 * interrupt state. They also don't use lwarx/stwcx. and thus
-	 * are known not to leave dangling reservations.
-	 */
-	.globl	fast_exception_return
-fast_exception_return:
-	ld	r3,_MSR(r1)
+	ld	r3,_LINK(r1)
 	ld	r4,_CTR(r1)
-	ld	r0,_LINK(r1)
-	mtctr	r4
-	mtlr	r0
-	ld	r4,_XER(r1)
-	mtspr	SPRN_XER,r4
-
-	kuap_check_amr r5, r6
-
-	REST_8GPRS(5, r1)
-
-	andi.	r0,r3,MSR_RI
-	beq-	.Lunrecov_restore
-
-	/*
-	 * Clear RI before restoring r13.  If we are returning to
-	 * userspace and we take an exception after restoring r13,
-	 * we end up corrupting the userspace r13 value.
-	 */
-	li	r4,0
-	mtmsrd	r4,1
-
-#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
-	/* TM debug */
-	std	r3, PACATMSCRATCH(r13) /* Stash returned-to MSR */
-#endif
-	/*
-	 * r13 is our per cpu area, only restore it if we are returning to
-	 * userspace the value stored in the stack frame may belong to
-	 * another CPU.
-	 */
-	andi.	r0,r3,MSR_PR
-	beq	1f
-BEGIN_FTR_SECTION
-	/* Restore PPR */
-	ld	r2,_PPR(r1)
-	mtspr	SPRN_PPR,r2
-END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
-	ACCOUNT_CPU_USER_EXIT(r13, r2, r4)
-	REST_GPR(13, r1)
-
-	/*
-	 * We don't need to restore AMR on the way back to userspace for KUAP.
-	 * The value of AMR only matters while we're in the kernel.
-	 */
-	mtspr	SPRN_SRR1,r3
-
-	ld	r2,_CCR(r1)
-	mtcrf	0xFF,r2
-	ld	r2,_NIP(r1)
-	mtspr	SPRN_SRR0,r2
-
-	ld	r0,GPR0(r1)
-	ld	r2,GPR2(r1)
-	ld	r3,GPR3(r1)
-	ld	r4,GPR4(r1)
-	ld	r1,GPR1(r1)
-	RFI_TO_USER
-	b	.	/* prevent speculative execution */
+	ld	r5,_XER(r1)
+	ld	r6,_CCR(r1)
+	li	r0,0
 
-1:	mtspr	SPRN_SRR1,r3
+	REST_4GPRS(7, r1)
+	REST_2GPRS(11, r1)
 
-	ld	r2,_CCR(r1)
-	mtcrf	0xFF,r2
-	ld	r2,_NIP(r1)
-	mtspr	SPRN_SRR0,r2
+	mtlr	r3
+	mtctr	r4
+	mtspr	SPRN_XER,r5
 
 	/*
 	 * Leaving a stale exception_marker on the stack can confuse
 	 * the reliable stack unwinder later on. Clear it.
 	 */
-	li	r2,0
-	std	r2,STACK_FRAME_OVERHEAD-16(r1)
+	std	r0,STACK_FRAME_OVERHEAD-16(r1)
 
-	ld	r0,GPR0(r1)
-	ld	r2,GPR2(r1)
-	ld	r3,GPR3(r1)
+	REST_4GPRS(2, r1)
 
-	kuap_restore_amr r4
-
-	ld	r4,GPR4(r1)
-	ld	r1,GPR1(r1)
+	bne-	cr1,1f /* emulate stack store */
+	mtcr	r6
+	REST_GPR(6, r1)
+	REST_GPR(0, r1)
+	REST_GPR(1, r1)
 	RFI_TO_KERNEL
 	b	.	/* prevent speculative execution */
 
-#endif /* CONFIG_PPC_BOOK3E */
-
-	/*
-	 * We are returning to a context with interrupts soft disabled.
-	 *
-	 * However, we may also about to hard enable, so we need to
-	 * make sure that in this case, we also clear PACA_IRQ_HARD_DIS
-	 * or that bit can get out of sync and bad things will happen
-	 */
-.Lrestore_irq_off:
-	ld	r3,_MSR(r1)
-	lbz	r7,PACAIRQHAPPENED(r13)
-	andi.	r0,r3,MSR_EE
-	beq	1f
-	rlwinm	r7,r7,0,~PACA_IRQ_HARD_DIS
-	stb	r7,PACAIRQHAPPENED(r13)
-1:
-#if defined(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG) && defined(CONFIG_BUG)
-	/* The interrupt should not have soft enabled. */
-	lbz	r7,PACAIRQSOFTMASK(r13)
-1:	tdeqi	r7,IRQS_ENABLED
-	EMIT_BUG_ENTRY 1b,__FILE__,__LINE__,BUGFLAG_WARNING
-#endif
-	b	.Ldo_restore
-
-	/*
-	 * Something did happen, check if a re-emit is needed
-	 * (this also clears paca->irq_happened)
-	 */
-.Lrestore_check_irq_replay:
-	/* XXX: We could implement a fast path here where we check
-	 * for irq_happened being just 0x01, in which case we can
-	 * clear it and return. That means that we would potentially
-	 * miss a decrementer having wrapped all the way around.
-	 *
-	 * Still, this might be useful for things like hash_page
-	 */
-	bl	__check_irq_replay
-	cmpwi	cr0,r3,0
-	beq	.Lrestore_no_replay
- 
-	/*
-	 * We need to re-emit an interrupt. We do so by re-using our
-	 * existing exception frame. We first change the trap value,
-	 * but we need to ensure we preserve the low nibble of it
-	 */
-	ld	r4,_TRAP(r1)
-	clrldi	r4,r4,60
-	or	r4,r4,r3
-	std	r4,_TRAP(r1)
-
-	/*
-	 * PACA_IRQ_HARD_DIS won't always be set here, so set it now
-	 * to reconcile the IRQ state. Tracing is already accounted for.
-	 */
-	lbz	r4,PACAIRQHAPPENED(r13)
-	ori	r4,r4,PACA_IRQ_HARD_DIS
-	stb	r4,PACAIRQHAPPENED(r13)
-
-	/*
-	 * Then find the right handler and call it. Interrupts are
-	 * still soft-disabled and we keep them that way.
-	*/
-	cmpwi	cr0,r3,0x500
-	bne	1f
-	addi	r3,r1,STACK_FRAME_OVERHEAD;
- 	bl	do_IRQ
-	b	ret_from_except
-1:	cmpwi	cr0,r3,0xf00
-	bne	1f
-	addi	r3,r1,STACK_FRAME_OVERHEAD;
-	bl	performance_monitor_exception
-	b	ret_from_except
-1:	cmpwi	cr0,r3,0xe60
-	bne	1f
-	addi	r3,r1,STACK_FRAME_OVERHEAD;
-	bl	handle_hmi_exception
-	b	ret_from_except
-1:	cmpwi	cr0,r3,0x900
-	bne	1f
-	addi	r3,r1,STACK_FRAME_OVERHEAD;
-	bl	timer_interrupt
-	b	ret_from_except
-#ifdef CONFIG_PPC_DOORBELL
-1:
-#ifdef CONFIG_PPC_BOOK3E
-	cmpwi	cr0,r3,0x280
-#else
-	cmpwi	cr0,r3,0xa00
-#endif /* CONFIG_PPC_BOOK3E */
-	bne	1f
-	addi	r3,r1,STACK_FRAME_OVERHEAD;
-	bl	doorbell_exception
-#endif /* CONFIG_PPC_DOORBELL */
-1:	b	ret_from_except /* What else to do here ? */
- 
-.Lunrecov_restore:
-	addi	r3,r1,STACK_FRAME_OVERHEAD
-	bl	unrecoverable_exception
-	b	.Lunrecov_restore
-
-_ASM_NOKPROBE_SYMBOL(ret_from_except);
-_ASM_NOKPROBE_SYMBOL(ret_from_except_lite);
-_ASM_NOKPROBE_SYMBOL(resume_kernel);
-_ASM_NOKPROBE_SYMBOL(fast_exc_return_irq);
-_ASM_NOKPROBE_SYMBOL(restore);
-_ASM_NOKPROBE_SYMBOL(fast_exception_return);
+1:	/*
+	 * Emulate stack store with update. New r1 value was already calculated
+	 * and updated in our interrupt regs by emulate_loadstore, but we can't
+	 * store the previous value of r1 to the stack before re-loading our
+	 * registers from it, otherwise they could be clobbered.  Use
+	 * PACA_EXGEN as temporary storage to hold the store data, as
+	 * interrupts are disabled here so it won't be clobbered.
+	 */
+	mtcr	r6
+	std	r9,PACA_EXGEN+0(r13)
+	addi	r9,r1,INT_FRAME_SIZE /* get original r1 */
+	REST_GPR(6, r1)
+	REST_GPR(0, r1)
+	REST_GPR(1, r1)
+	std	r9,0(r1) /* perform store component of stdu */
+	ld	r9,PACA_EXGEN+0(r13)
 
+	RFI_TO_KERNEL
+	b	.	/* prevent speculative execution */
+#endif /* CONFIG_PPC_BOOK3S */
 
 #ifdef CONFIG_PPC_RTAS
 /*