MLK15034-3: ARM: cpuidle imx7d: Use a single counter for lpi flow

The current code for deciding which CPU runs the complete lpi flow is
too complicated. Since all enter/exit code now runs under the same lock
we can just use a single non-atomic counter of cpus inside lpi.

Another variable is used to make num_online_cpus() available to ASM code
but idle code can treat it as a constant.

Unlike on imx_4.9.y num_online_cpus is fetched every time idle is
entered becuase hotplug notifiers are gone.

Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>

2 files changed, 47 insertions, 91 deletions

diff --git a/arch/arm/mach-imx/cpuidle-imx7d.c b/arch/arm/mach-imx/cpuidle-imx7d.c
index d19d76ef75d7..641cdcade415 100644
--- a/arch/arm/mach-imx/cpuidle-imx7d.c
+++ b/arch/arm/mach-imx/cpuidle-imx7d.c
@@ -62,10 +62,9 @@ struct imx7_cpuidle_pm_info {
 	phys_addr_t pbase; /* The physical address of pm_info. */
 	phys_addr_t resume_addr; /* The physical resume address for asm code */
 	u32 pm_info_size;
-	int last_cpu;
 	u32 ttbr;
-	u32 cpu1_wfi;
-	u32 lpi_enter;
+	u32 num_online_cpus;
+	u32 num_lpi_cpus;
 	atomic_t val;
 	atomic_t flag0;
 	atomic_t flag1;
@@ -78,7 +77,6 @@ struct imx7_cpuidle_pm_info {
 	struct imx7_pm_base gic_dist_base;
 } __aligned(8);
 
-static atomic_t master_lpi = ATOMIC_INIT(0);
 static atomic_t master_wait = ATOMIC_INIT(0);
 
 static void (*imx7d_wfi_in_iram_fn)(void __iomem *iram_vbase);
@@ -140,6 +138,7 @@ static int imx7d_enter_low_power_idle(struct cpuidle_device *dev,
 {
 	int mode = get_bus_freq_mode();
 
+
 	if ((index == 1) || ((mode != BUS_FREQ_LOW) && index == 2)) {
 		index = 1;
 		if (atomic_inc_return(&master_wait) == num_online_cpus())
@@ -151,9 +150,11 @@ static int imx7d_enter_low_power_idle(struct cpuidle_device *dev,
 		imx_gpcv2_set_lpm_mode(WAIT_CLOCKED);
 	} else {
 		imx_pen_lock(dev->cpu);
+		cpuidle_pm_info->num_online_cpus = num_online_cpus();
+		++cpuidle_pm_info->num_lpi_cpus;
 		cpu_pm_enter();
-		if (atomic_inc_return(&master_lpi) == num_online_cpus() &&
-			cpuidle_pm_info->last_cpu == -1) {
+		if (cpuidle_pm_info->num_lpi_cpus ==
+				cpuidle_pm_info->num_online_cpus) {
 			/*
 			 * GPC will not wake on SGIs so check for them
 			 * manually here. At this point we know the other cpu
@@ -167,26 +168,22 @@ static int imx7d_enter_low_power_idle(struct cpuidle_device *dev,
 			imx_gpcv2_set_lpm_mode(WAIT_UNCLOCKED);
 			imx_gpcv2_set_cpu_power_gate_in_idle(true);
 			cpu_cluster_pm_enter();
-
-			cpuidle_pm_info->last_cpu = dev->cpu;
-
 		} else {
 			imx_set_cpu_jump(dev->cpu, ca7_cpu_resume);
 		}
 
 		cpu_suspend(0, imx7d_idle_finish);
 
-		if (atomic_dec_return(&master_lpi) == (num_online_cpus() - 1)) {
+		if (cpuidle_pm_info->num_lpi_cpus ==
+				cpuidle_pm_info->num_online_cpus) {
 			cpu_cluster_pm_exit();
 			imx_gpcv2_set_cpu_power_gate_in_idle(false);
 			imx_gpcv2_set_lpm_mode(WAIT_CLOCKED);
 		}
 
-		if (cpuidle_pm_info->last_cpu == dev->cpu)
-			cpuidle_pm_info->last_cpu = -1;
-
 skip_lpi_flow:
 		cpu_pm_exit();
+		--cpuidle_pm_info->num_lpi_cpus;
 		imx_pen_unlock(dev->cpu);
 	}
 
@@ -296,13 +293,7 @@ int __init imx7d_cpuidle_init(void)
 	cpuidle_pm_info->pbase = (phys_addr_t) wfi_iram_base_phys;
 	cpuidle_pm_info->pm_info_size = sizeof(*cpuidle_pm_info);
 	cpuidle_pm_info->resume_addr = virt_to_phys(ca7_cpu_resume);
-	if (num_online_cpus() == 1)
-		cpuidle_pm_info->cpu1_wfi = 1;
-	else
-		cpuidle_pm_info->cpu1_wfi = 0;
-	cpuidle_pm_info->lpi_enter = 0;
-	/* initialize the last cpu id to invalid here */
-	cpuidle_pm_info->last_cpu = -1;
+	cpuidle_pm_info->num_online_cpus = num_online_cpus();
 
 	cpuidle_pm_info->ddrc_base.pbase = MX7D_DDRC_BASE_ADDR;
 	cpuidle_pm_info->ddrc_base.vbase =
diff --git a/arch/arm/mach-imx/imx7d_low_power_idle.S b/arch/arm/mach-imx/imx7d_low_power_idle.S
index 72a2ce728c43..1551d2ce1acf 100644
--- a/arch/arm/mach-imx/imx7d_low_power_idle.S
+++ b/arch/arm/mach-imx/imx7d_low_power_idle.S
@@ -18,27 +18,26 @@
 #define PM_INFO_PBASE_OFFSET			0x4
 #define PM_INFO_RESUME_ADDR_OFFSET		0x8
 #define PM_INFO_PM_INFO_SIZE_OFFSET		0xc
-#define PM_INFO_PM_INFO_LAST_CPU_OFFSET		0x10
-#define PM_INFO_PM_INFO_TTBR_OFFSET		0x14
-#define PM_INFO_PM_INFO_CPU1_IN_WFI_OFFSET	0x18
-#define PM_INFO_PM_INFO_LPI_ENTER_OFFSET	0x1c
-#define PM_INFO_VAL_OFFSET			0x20
-#define PM_INFO_FLAG0_OFFSET			0x24
-#define PM_INFO_FLAG1_OFFSET			0x28
-#define PM_INFO_MX7D_DDRC_P_OFFSET		0x2c
-#define PM_INFO_MX7D_DDRC_V_OFFSET		0x30
-#define PM_INFO_MX7D_CCM_P_OFFSET		0x34
-#define PM_INFO_MX7D_CCM_V_OFFSET		0x38
-#define PM_INFO_MX7D_ANATOP_P_OFFSET		0x3c
-#define PM_INFO_MX7D_ANATOP_V_OFFSET		0x40
-#define PM_INFO_MX7D_SRC_P_OFFSET		0x44
-#define PM_INFO_MX7D_SRC_V_OFFSET		0x48
-#define PM_INFO_MX7D_IOMUXC_GPR_P_OFFSET	0x4c
-#define PM_INFO_MX7D_IOMUXC_GPR_V_OFFSET	0x50
-#define PM_INFO_MX7D_GPC_P_OFFSET		0x54
-#define PM_INFO_MX7D_GPC_V_OFFSET		0x58
-#define PM_INFO_MX7D_GIC_DIST_P_OFFSET		0x5c
-#define PM_INFO_MX7D_GIC_DIST_V_OFFSET		0x60
+#define PM_INFO_PM_INFO_TTBR_OFFSET		0x10
+#define PM_INFO_PM_INFO_NUM_ONLINE_CPUS_OFFSET	0x14
+#define PM_INFO_PM_INFO_NUM_LPI_CPUS_OFFSET	0x18
+#define PM_INFO_VAL_OFFSET			0x1c
+#define PM_INFO_FLAG0_OFFSET			0x20
+#define PM_INFO_FLAG1_OFFSET			0x24
+#define PM_INFO_MX7D_DDRC_P_OFFSET		0x28
+#define PM_INFO_MX7D_DDRC_V_OFFSET		0x2c
+#define PM_INFO_MX7D_CCM_P_OFFSET		0x30
+#define PM_INFO_MX7D_CCM_V_OFFSET		0x34
+#define PM_INFO_MX7D_ANATOP_P_OFFSET		0x38
+#define PM_INFO_MX7D_ANATOP_V_OFFSET		0x3c
+#define PM_INFO_MX7D_SRC_P_OFFSET		0x40
+#define PM_INFO_MX7D_SRC_V_OFFSET		0x44
+#define PM_INFO_MX7D_IOMUXC_GPR_P_OFFSET	0x48
+#define PM_INFO_MX7D_IOMUXC_GPR_V_OFFSET	0x4c
+#define PM_INFO_MX7D_GPC_P_OFFSET		0x50
+#define PM_INFO_MX7D_GPC_V_OFFSET		0x54
+#define PM_INFO_MX7D_GIC_DIST_P_OFFSET		0x58
+#define PM_INFO_MX7D_GIC_DIST_V_OFFSET		0x5c
 
 #define MX7D_SRC_GPR1	0x74
 #define MX7D_SRC_GPR2	0x78
@@ -612,13 +611,11 @@ ENTRY(imx7d_low_power_idle)
 
 	tlb_set_to_ocram
 
-	ldr	r6, [r0, #PM_INFO_PM_INFO_LAST_CPU_OFFSET]
-	cmp	r11, r6
-	bne	first_cpu
-last_cpu:
-	ldr	r7, [r0, #PM_INFO_PM_INFO_CPU1_IN_WFI_OFFSET]
-	cmp	r7, #0x0
-	beq	lpi_enter_done
+	/* check last to sleep */
+	ldr	r6, [r0, #PM_INFO_PM_INFO_NUM_ONLINE_CPUS_OFFSET]
+	ldr	r7, [r0, #PM_INFO_PM_INFO_NUM_LPI_CPUS_OFFSET]
+	cmp	r6, r7
+	bne	lpi_enter_done
 
 	ddrc_enter_self_refresh
 	ccm_enter_idle
@@ -679,14 +676,6 @@ rbc_loop:
 	subs	r4, r4, #0x1
 	bne	rbc_loop
 
-	/* set low power idle enter flag */
-	ldr	r7, =0x1
-	str	r7, [r0, #PM_INFO_PM_INFO_LPI_ENTER_OFFSET]
-	b	lpi_enter_done
-first_cpu:
-	/* set first cpu wfi flag */
-	ldr	r7, =0x1
-	str	r7, [r0, #PM_INFO_PM_INFO_CPU1_IN_WFI_OFFSET]
 lpi_enter_done:
 
 	imx_pen_unlock
@@ -725,17 +714,11 @@ lpi_enter_done:
 
 	imx_pen_lock
 
-	ldr	r6, [r0, #PM_INFO_PM_INFO_LAST_CPU_OFFSET]
-	cmp	r11, r6
-	beq	do_exit_wfi
-
-	/* clear first cpu wfi flag */
-	ldr	r7, =0x0
-	str	r7, [r0, #PM_INFO_PM_INFO_CPU1_IN_WFI_OFFSET]
-do_exit_wfi:
-	ldr	r7, [r0, #PM_INFO_PM_INFO_LPI_ENTER_OFFSET]
-	cmp	r7, #0
-	beq	skip_lpi_flow
+	/* check first to wake */
+	ldr	r6, [r0, #PM_INFO_PM_INFO_NUM_ONLINE_CPUS_OFFSET]
+	ldr	r7, [r0, #PM_INFO_PM_INFO_NUM_LPI_CPUS_OFFSET]
+	cmp	r6, r7
+	bne	skip_lpi_flow
 
 	ldr	r5, =0x0
 	anatop_exit_idle
@@ -747,11 +730,7 @@ do_exit_wfi:
 	ldr	r8, =0x1000
 	str	r7, [r10, r8]
 
-	/* clear lpi enter flag */
-	ldr	r7, =0x0
-	str	r7, [r0, #PM_INFO_PM_INFO_LPI_ENTER_OFFSET]
 skip_lpi_flow:
-
 	tlb_back_to_ddr
 
 #ifdef CONFIG_SMP
@@ -785,34 +764,20 @@ wakeup:
 	mcr	p15, 0, r1, c1, c0, 0
 	isb
 
-	/* r11 is cpu id */
-	mrc	p15, 0, r11, c0, c0, 5
-	and	r11, r11, #3
-	cmp	r11, #0x0
-
 	imx_pen_lock
 
-	ldr	r6, [r0, #PM_INFO_PM_INFO_LAST_CPU_OFFSET]
-	cmp	r11, r6
-	beq	do_exit_lp_idle
-
-	/* clear first wfi flag */
-	ldr	r7, =0x0
-	str	r7, [r0, #PM_INFO_PM_INFO_CPU1_IN_WFI_OFFSET]
-do_exit_lp_idle:
-	ldr	r7, [r0, #PM_INFO_PM_INFO_LPI_ENTER_OFFSET]
-	cmp	r7, #0
-	beq	wakeup_skip_lpi_flow
+	/* check first to wake */
+	ldr	r6, [r0, #PM_INFO_PM_INFO_NUM_ONLINE_CPUS_OFFSET]
+	ldr	r7, [r0, #PM_INFO_PM_INFO_NUM_LPI_CPUS_OFFSET]
+	cmp	r6, r7
+	bne	wakeup_skip_lpi_flow
 
 	ldr	r5, =0x1
 	anatop_exit_idle
 	ccm_exit_idle
 	ddrc_exit_self_refresh
-	/* clear lpi enter flag */
-	ldr	r7, =0x0
-	str	r7, [r0, #PM_INFO_PM_INFO_LPI_ENTER_OFFSET]
-wakeup_skip_lpi_flow:
 
+wakeup_skip_lpi_flow:
 	/* get physical resume address from pm_info. */
 	ldr	lr, [r0, #PM_INFO_RESUME_ADDR_OFFSET]