Skip to contentSkip to conversation

ARMv7: PSCI: add PSCI v1.0 support

PATCH Superseded macro.wave.z at gmail.com 5 participants 34 messages Jun 14, 2016 – Jul 21, 2016

Subsystems (4)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.
- Move the patch of factoring out psci_cpu_on_common to earlier place 2/8, so that the following patches can only update the psci_cpu_on_common instead of touching each platform's psci_cpu_on, this reduces patches size obviously and make it easier for review.

v4 changes:
- since there is already PSCI v0.2 function IDs definition merged in 5a07abb, I give up my previous patches 1/11 and 2/11, and move previous 7/11 "ARMv7:
PSCI: add PSCI v1.0 functions skeleton" as current first one 1/9
- accept Andre's comment to add the missed "arm,psci-0,2" into compatible string

v3 changes:
- patch 3/11, re-init the stack pointer to address like start of page instead of page end, because for ARM push operation, the stack pointer is encreased before storing data.
- patch 10/11, delete the previous un-implemented cpu_suspend function for ls1 platform, because there is default blank functions for all those are not implemented in specific platform.

v2 changes:
- re-organize psci_cpu_on_common, this code should be called by each platform's psci_cpu_on, should not be a function calling each psci_cpu_on, all related functions are updated due to this change
- update some registers usage, e.g. if r10 is used without push/pop, u-boot cannot launch rt-kernel
- update some comments to be clearer, re-organize all patches for easier review
- add patch to check already_on or on_pending for LS102XA

This patch set contains two parts:
ARMv7 PSCI common framework: fix some issues and add v1.0 support NXP (was Freescale) LS102XA: codes enhancement and add v1.0 implementation And this patch set was initially created by Dongsheng Wang.

Hongbo Zhang (8):
ARMv7: PSCI: add PSCI v1.0 functions skeleton
ARMv7: PSCI: factor out reusable psci_cpu_on_common
ARMv7: PSCI: update the place of saving target PC
ARMv7: PSCI: add codes to save context ID for CPU_ON
ARMv7: PSCI: ls102xa: check target CPU ID before further operations
ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON
ARMv7: PSCI: ls102xa: add more PSCI v1.0 functions implemention
ARMv7: PSCI: ls102xa: move secure text section into OCRAM

 arch/arm/cpu/armv7/ls102xa/psci.S          | 169 +++++++++++++++++++++++++++--
 arch/arm/cpu/armv7/mx7/psci.S              |   5 +-
 arch/arm/cpu/armv7/nonsec_virt.S           |   7 ++
 arch/arm/cpu/armv7/psci.S                  |  99 ++++++++++++++++-
 arch/arm/cpu/armv7/sunxi/psci_sun6i.S      |   5 +-
 arch/arm/cpu/armv7/sunxi/psci_sun7i.S      |   5 +-
 arch/arm/cpu/armv7/virt-dt.c               |  10 +-
 arch/arm/include/asm/arch-ls102xa/config.h |   3 +-
 arch/arm/include/asm/psci.h                |  31 ++++++
 arch/arm/mach-tegra/psci.S                 |   5 +-
 board/freescale/ls1021aqds/Makefile        |   1 +
 board/freescale/ls1021aqds/psci.S          |  36 ++++++
 board/freescale/ls1021atwr/Makefile        |   1 +
 board/freescale/ls1021atwr/psci.S          |  28 +++++
 include/configs/ls1021aqds.h               |   3 +
 include/configs/ls1021atwr.h               |   3 +
 16 files changed, 383 insertions(+), 28 deletions(-)
 create mode 100644 board/freescale/ls1021aqds/psci.S
 create mode 100644 board/freescale/ls1021atwr/psci.S
Signature · 1 line
2.1.4
#1 lore raw

PATCH v5 1/8 ARMv7: PSCI: add PSCI v1.0 functions skeleton

From: Hongbo Zhang <[email protected]>

This patch adds all the PSCI v1.0 functions in to the common framework, with all the functions returning "not implemented" by default, as a common framework all the dummy functions are added here, it is up to every platform developer to decide which version of PSCI and which functions to implement.

---
 arch/arm/cpu/armv7/psci.S    | 78 ++++++++++++++++++++++++++++++++++++++++++++
 arch/arm/cpu/armv7/virt-dt.c | 10 +++++-
 arch/arm/include/asm/psci.h  | 16 +++++++++
 3 files changed, 103 insertions(+), 1 deletion(-)
arch/arm/cpu/armv7/psci.S +78 -0
index 87c0c0b..8e25300 100644
@@ -46,20 +46,62 @@ ENTRY(default_psci_vector)
 ENDPROC(default_psci_vector)
 .weak default_psci_vector
 
+ENTRY(psci_version)
 ENTRY(psci_cpu_suspend)
 ENTRY(psci_cpu_off)
 ENTRY(psci_cpu_on)
+ENTRY(psci_affinity_info)
 ENTRY(psci_migrate)
+ENTRY(psci_migrate_info_type)
+ENTRY(psci_migrate_info_up_cpu)
+ENTRY(psci_system_off)
+ENTRY(psci_system_reset)
+ENTRY(psci_features)
+ENTRY(psci_cpu_freeze)
+ENTRY(psci_cpu_default_suspend)
+ENTRY(psci_node_hw_state)
+ENTRY(psci_system_suspend)
+ENTRY(psci_set_suspend_mode)
+ENTRY(psi_stat_residency)
+ENTRY(psci_stat_count)
 	mov	r0, #ARM_PSCI_RET_NI	@ Return -1 (Not Implemented)
 	mov	pc, lr
+ENDPROC(psci_stat_count)
+ENDPROC(psi_stat_residency)
+ENDPROC(psci_set_suspend_mode)
+ENDPROC(psci_system_suspend)
+ENDPROC(psci_node_hw_state)
+ENDPROC(psci_cpu_default_suspend)
+ENDPROC(psci_cpu_freeze)
+ENDPROC(psci_features)
+ENDPROC(psci_system_reset)
+ENDPROC(psci_system_off)
+ENDPROC(psci_migrate_info_up_cpu)
+ENDPROC(psci_migrate_info_type)
 ENDPROC(psci_migrate)
+ENDPROC(psci_affinity_info)
 ENDPROC(psci_cpu_on)
 ENDPROC(psci_cpu_off)
 ENDPROC(psci_cpu_suspend)
+ENDPROC(psci_version)
+.weak psci_version
 .weak psci_cpu_suspend
 .weak psci_cpu_off
 .weak psci_cpu_on
+.weak psci_affinity_info
 .weak psci_migrate
+.weak psci_migrate_info_type
+.weak psci_migrate_info_up_cpu
+.weak psci_system_off
+.weak psci_system_reset
+.weak psci_features
+.weak psci_cpu_freeze
+.weak psci_cpu_default_suspend
+.weak psci_node_hw_state
+.weak psci_system_suspend
+.weak psci_set_suspend_mode
+.weak psi_stat_residency
+.weak psci_stat_count
 
 _psci_table:
 	.word	ARM_PSCI_FN_CPU_SUSPEND
@@ -70,6 +112,42 @@ _psci_table:
 	.word	psci_cpu_on
 	.word	ARM_PSCI_FN_MIGRATE
 	.word	psci_migrate
+	.word	ARM_PSCI_0_2_FN_PSCI_VERSION
+	.word	psci_version
+	.word	ARM_PSCI_0_2_FN_CPU_SUSPEND
+	.word	psci_cpu_suspend
+	.word	ARM_PSCI_0_2_FN_CPU_OFF
+	.word	psci_cpu_off
+	.word	ARM_PSCI_0_2_FN_CPU_ON
+	.word	psci_cpu_on
+	.word	ARM_PSCI_0_2_FN_AFFINITY_INFO
+	.word	psci_affinity_info
+	.word	ARM_PSCI_0_2_FN_MIGRATE
+	.word	psci_migrate
+	.word	ARM_PSCI_0_2_FN_MIGRATE_INFO_TYPE
+	.word	psci_migrate_info_type
+	.word	ARM_PSCI_0_2_FN_MIGRATE_INFO_UP_CPU
+	.word	psci_migrate_info_up_cpu
+	.word	ARM_PSCI_0_2_FN_SYSTEM_OFF
+	.word	psci_system_off
+	.word	ARM_PSCI_0_2_FN_SYSTEM_RESET
+	.word	psci_system_reset
+	.word	ARM_PSCI_1_0_FN_PSCI_FEATURES
+	.word	psci_features
+	.word	ARM_PSCI_1_0_FN_CPU_FREEZE
+	.word	psci_cpu_freeze
+	.word	ARM_PSCI_1_0_FN_CPU_DEFAULT_SUSPEND
+	.word	psci_cpu_default_suspend
+	.word	ARM_PSCI_1_0_FN_NODE_HW_STATE
+	.word	psci_node_hw_state
+	.word	ARM_PSCI_1_0_FN_SYSTEM_SUSPEND
+	.word	psci_system_suspend
+	.word	ARM_PSCI_1_0_FN_SET_SUSPEND_MODE
+	.word	psci_set_suspend_mode
+	.word	ARM_PSCI_1_0_FN_STAT_RESIDENCY
+	.word	psi_stat_residency
+	.word	ARM_PSCI_1_0_FN_STAT_COUNT
+	.word	psci_stat_count
 	.word	0
 	.word	0
 
arch/arm/cpu/armv7/virt-dt.c +9 -1
index 32c368f..5e31891 100644
@@ -67,7 +67,15 @@ static int fdt_psci(void *fdt)
 			return nodeoff;
 	}
 
-	tmp = fdt_setprop_string(fdt, nodeoff, "compatible", "arm,psci");
+#ifdef CONFIG_ARMV7_PSCI_1_0
+	tmp = fdt_setprop_string(fdt, nodeoff, "compatible", "arm,psci-1.0");
+	if (tmp)
+		return tmp;
+	tmp = fdt_appendprop_string(fdt, nodeoff, "compatible", "arm,psci-0.2");
+	if (tmp)
+		return tmp;
+#endif
+	tmp = fdt_appendprop_string(fdt, nodeoff, "compatible", "arm,psci");
 	if (tmp)
 		return tmp;
 	tmp = fdt_setprop_string(fdt, nodeoff, "method", "smc");
arch/arm/include/asm/psci.h +16 -0
index 3704f07..2367ec0 100644
@@ -31,6 +31,12 @@
 #define ARM_PSCI_RET_NI			(-1)
 #define ARM_PSCI_RET_INVAL		(-2)
 #define ARM_PSCI_RET_DENIED		(-3)
+#define ARM_PSCI_RET_ALREADY_ON		(-4)
+#define ARM_PSCI_RET_ON_PENDING		(-5)
+#define ARM_PSCI_RET_INTERNAL_FAILURE	(-6)
+#define ARM_PSCI_RET_NOT_PRESENT	(-7)
+#define ARM_PSCI_RET_DISABLED		(-8)
+#define ARM_PSCI_RET_INVALID_ADDRESS	(-9)
 
 /* PSCI 0.2 interface */
 #define ARM_PSCI_0_2_FN_BASE			0x84000000
@@ -47,6 +53,16 @@
 #define ARM_PSCI_0_2_FN_SYSTEM_OFF		ARM_PSCI_0_2_FN(8)
 #define ARM_PSCI_0_2_FN_SYSTEM_RESET		ARM_PSCI_0_2_FN(9)
 
+/* PSCI 1.0 interface */
+#define ARM_PSCI_1_0_FN_PSCI_FEATURES		ARM_PSCI_0_2_FN(10)
+#define ARM_PSCI_1_0_FN_CPU_FREEZE		ARM_PSCI_0_2_FN(11)
+#define ARM_PSCI_1_0_FN_CPU_DEFAULT_SUSPEND	ARM_PSCI_0_2_FN(12)
+#define ARM_PSCI_1_0_FN_NODE_HW_STATE		ARM_PSCI_0_2_FN(13)
+#define ARM_PSCI_1_0_FN_SYSTEM_SUSPEND		ARM_PSCI_0_2_FN(14)
+#define ARM_PSCI_1_0_FN_SET_SUSPEND_MODE	ARM_PSCI_0_2_FN(15)
+#define ARM_PSCI_1_0_FN_STAT_RESIDENCY		ARM_PSCI_0_2_FN(16)
+#define ARM_PSCI_1_0_FN_STAT_COUNT		ARM_PSCI_0_2_FN(17)
+
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
Signature · 1 line
2.1.4
#2 lore raw

PATCH v5 2/8 ARMv7: PSCI: factor out reusable psci_cpu_on_common

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

There are codes for saving target PC in each platform psci_cpu_on routines, these can be factored out as psci_cpu_on_common.
Another purpose is that the codes of saving target PC will be changed a bit and a new context ID will be saved in the following patches, then we can update the psci_cpu_on_common mainly instead of touching every platform's psci_cpu_on functions more times, this makes it easier for coding and being reviewed.

---
 arch/arm/cpu/armv7/ls102xa/psci.S     |  5 +----
 arch/arm/cpu/armv7/mx7/psci.S         |  5 +----
 arch/arm/cpu/armv7/psci.S             | 12 ++++++++++++
 arch/arm/cpu/armv7/sunxi/psci_sun6i.S |  5 +----
 arch/arm/cpu/armv7/sunxi/psci_sun7i.S |  5 +----
 arch/arm/mach-tegra/psci.S            |  5 +----
 6 files changed, 17 insertions(+), 20 deletions(-)
arch/arm/cpu/armv7/ls102xa/psci.S +1 -4
index cf5cd48..548c507 100644
@@ -35,10 +35,7 @@ psci_cpu_on:
 	@ r1 = 0xf01
 	and	r1, r1, #0xff
 
-	mov	r0, r1
-	bl	psci_get_cpu_stack_top
-	str	r2, [r0]
-	dsb
+	bl	psci_cpu_on_common
 
 	@ Get DCFG base address
 	movw	r4, #(CONFIG_SYS_FSL_GUTS_ADDR & 0xffff)
arch/arm/cpu/armv7/mx7/psci.S +1 -4
index 34c6ab3..74fdc4d 100644
@@ -29,10 +29,7 @@ psci_arch_init:
 psci_cpu_on:
 	push	{lr}
 
-	mov	r0, r1
-	bl	psci_get_cpu_stack_top
-	str	r2, [r0]
-	dsb
+	bl	psci_cpu_on_common
 
 	ldr	r2, =psci_cpu_entry
 	bl	imx_cpu_on
arch/arm/cpu/armv7/psci.S +12 -0
index 8e25300..3b92f1d 100644
@@ -253,6 +253,18 @@ ENTRY(psci_enable_smp)
 ENDPROC(psci_enable_smp)
 .weak psci_enable_smp
 
+/* expects target CPU in r1, target PC in r2 */
+ENTRY(psci_cpu_on_common)
+	push	{lr}
+
+	mov	r0, r1
+	bl	psci_get_cpu_stack_top		@ get stack top of target CPU
+	str	r2, [r0]			@ store target PC at stack top
+	dsb
+
+	pop	{pc}
+ENDPROC(psci_cpu_on_common)
+
 ENTRY(psci_cpu_off_common)
 	push	{lr}
 
arch/arm/cpu/armv7/sunxi/psci_sun6i.S +1 -4
index 90b5bfd..016e491 100644
@@ -134,10 +134,7 @@ out:	mcr	p15, 0, r7, c1, c1, 0
 psci_cpu_on:
 	push	{lr}
 
-	mov	r0, r1
-	bl	psci_get_cpu_stack_top	@ get stack top of target CPU
-	str	r2, [r0]		@ store target PC at stack top
-	dsb
+	bl	psci_cpu_on_common
 
 	movw	r0, #(SUN6I_CPUCFG_BASE & 0xffff)
 	movt	r0, #(SUN6I_CPUCFG_BASE >> 16)
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +1 -4
index e15d587..0ebb30e 100644
@@ -123,10 +123,7 @@ out:	mcr	p15, 0, r7, c1, c1, 0
 psci_cpu_on:
 	push	{lr}
 
-	mov	r0, r1
-	bl	psci_get_cpu_stack_top	@ get stack top of target CPU
-	str	r2, [r0]		@ store target PC at stack top
-	dsb
+	bl	psci_cpu_on_common
 
 	movw	r0, #(SUN7I_CPUCFG_BASE & 0xffff)
 	movt	r0, #(SUN7I_CPUCFG_BASE >> 16)
arch/arm/mach-tegra/psci.S +1 -4
index b836da1..8a0147c 100644
@@ -90,10 +90,7 @@ ENDPROC(psci_cpu_off)
 ENTRY(psci_cpu_on)
 	push	{lr}
 
-	mov	r0, r1
-	bl	psci_get_cpu_stack_top	@ get stack top of target CPU
-	str	r2, [r0]		@ store target PC at stack top
-	dsb
+	bl	psci_cpu_on_common
 
 	ldr	r6, =TEGRA_RESET_EXCEPTION_VECTOR
 	ldr	r5, =psci_cpu_entry
Signature · 1 line
2.1.4
#3 lore raw

PATCH v5 2/8 ARMv7: PSCI: factor out reusable psci_cpu_on_common

Hi,

Quoted 62 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

There are codes for saving target PC in each platform psci_cpu_on routines, these can be factored out as psci_cpu_on_common.
Another purpose is that the codes of saving target PC will be changed a bit and a new context ID will be saved in the following patches, then we can update the psci_cpu_on_common mainly instead of touching every platform's psci_cpu_on functions more times, this makes it easier for coding and being reviewed.

---
 arch/arm/cpu/armv7/ls102xa/psci.S     |  5 +----
 arch/arm/cpu/armv7/mx7/psci.S         |  5 +----
 arch/arm/cpu/armv7/psci.S             | 12 ++++++++++++
 arch/arm/cpu/armv7/sunxi/psci_sun6i.S |  5 +----
 arch/arm/cpu/armv7/sunxi/psci_sun7i.S |  5 +----
 arch/arm/mach-tegra/psci.S            |  5 +----
 6 files changed, 17 insertions(+), 20 deletions(-)
arch/arm/cpu/armv7/ls102xa/psci.S +1 -4
index cf5cd48..548c507 100644
@@ -35,10 +35,7 @@ psci_cpu_on:
        @ r1 = 0xf01
        and     r1, r1, #0xff
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top
-       str     r2, [r0]
-       dsb
+       bl      psci_cpu_on_common
 
        @ Get DCFG base address
        movw    r4, #(CONFIG_SYS_FSL_GUTS_ADDR & 0xffff)
arch/arm/cpu/armv7/mx7/psci.S +1 -4
index 34c6ab3..74fdc4d 100644
@@ -29,10 +29,7 @@ psci_arch_init:
 psci_cpu_on:
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top
-       str     r2, [r0]
-       dsb
+       bl      psci_cpu_on_common
 
        ldr     r2, =psci_cpu_entry
        bl      imx_cpu_on
arch/arm/cpu/armv7/psci.S +2 -0
index 8e25300..3b92f1d 100644
@@ -253,6 +253,18 @@ ENTRY(psci_enable_smp)
 ENDPROC(psci_enable_smp)
 .weak psci_enable_smp
 
+/* expects target CPU in r1, target PC in r2 */
+ENTRY(psci_cpu_on_common)

arch/arm/cpu/armv7/psci.S ENTRY(psci_enable_smp) · line 267

Quoted 10 lines
arch/arm/cpu/armv7/psci.S +10 -0
+       push    {lr}
+
+       mov     r0, r1
+       bl      psci_get_cpu_stack_top          @ get stack top of target CPU
+       str     r2, [r0]                        @ store target PC at stack top
+       dsb
+
+       pop     {pc}
+ENDPROC(psci_cpu_on_common)
+

Since you are just saving the target PC, why not just name the function like that?

See https://github.com/wens/u-boot-sunxi/commit/5a42630107e93bea9aa4da27cc16bdb604993dca

Quoted 24 lines
arch/arm/cpu/armv7/psci.S +0 -0
 ENTRY(psci_cpu_off_common)
        push    {lr}
 
arch/arm/cpu/armv7/sunxi/psci_sun6i.S +1 -4
index 90b5bfd..016e491 100644
@@ -134,10 +134,7 @@ out:       mcr     p15, 0, r7, c1, c1, 0
 psci_cpu_on:
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top  @ get stack top of target CPU
-       str     r2, [r0]                @ store target PC at stack top
-       dsb
+       bl      psci_cpu_on_common
 
        movw    r0, #(SUN6I_CPUCFG_BASE & 0xffff)
        movt    r0, #(SUN6I_CPUCFG_BASE >> 16)
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +0 -0
index e15d587..0ebb30e 100644

arch/arm/cpu/armv7/sunxi/psci_sun7i.S out: mcr p15, 0, r7, c1, c1, 0 · line 126

Quoted 8 lines
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +1 -4
@@ -123,10 +123,7 @@ out:       mcr     p15, 0, r7, c1, c1, 0
 psci_cpu_on:
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top  @ get stack top of target CPU
-       str     r2, [r0]                @ store target PC at stack top
-       dsb
+       bl      psci_cpu_on_common

The sunxi parts no longer apply. The partial rewrite in C has landed in master.

Regards
ChenYu

Quoted 22 lines
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +0 -0
        movw    r0, #(SUN7I_CPUCFG_BASE & 0xffff)
        movt    r0, #(SUN7I_CPUCFG_BASE >> 16)
arch/arm/mach-tegra/psci.S +1 -5
index b836da1..8a0147c 100644
@@ -90,10 +90,7 @@ ENDPROC(psci_cpu_off)
 ENTRY(psci_cpu_on)
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top  @ get stack top of target CPU
-       str     r2, [r0]                @ store target PC at stack top
-       dsb
+       bl      psci_cpu_on_common
 
        ldr     r6, =TEGRA_RESET_EXCEPTION_VECTOR
        ldr     r5, =psci_cpu_entry
--

2.1.4

#4 lore raw

PATCH v5 2/8 ARMv7: PSCI: factor out reusable psci_cpu_on_common

Quoted 63 lines · Chen-Yu Tsai

Hi,

Quoted 60 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

There are codes for saving target PC in each platform psci_cpu_on routines, these can be factored out as psci_cpu_on_common.
Another purpose is that the codes of saving target PC will be changed a bit and a new context ID will be saved in the following patches, then we can update the psci_cpu_on_common mainly instead of touching every platform's psci_cpu_on functions more times, this makes it easier for coding and being reviewed.

---
 arch/arm/cpu/armv7/ls102xa/psci.S     |  5 +----
 arch/arm/cpu/armv7/mx7/psci.S         |  5 +----
 arch/arm/cpu/armv7/psci.S             | 12 ++++++++++++
 arch/arm/cpu/armv7/sunxi/psci_sun6i.S |  5 +----
 arch/arm/cpu/armv7/sunxi/psci_sun7i.S |  5 +----
 arch/arm/mach-tegra/psci.S            |  5 +----
 6 files changed, 17 insertions(+), 20 deletions(-)
arch/arm/cpu/armv7/ls102xa/psci.S +1 -4
index cf5cd48..548c507 100644
@@ -35,10 +35,7 @@ psci_cpu_on:
        @ r1 = 0xf01
        and     r1, r1, #0xff
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top
-       str     r2, [r0]
-       dsb
+       bl      psci_cpu_on_common
 
        @ Get DCFG base address
        movw    r4, #(CONFIG_SYS_FSL_GUTS_ADDR & 0xffff)
arch/arm/cpu/armv7/mx7/psci.S +1 -4
index 34c6ab3..74fdc4d 100644
@@ -29,10 +29,7 @@ psci_arch_init:
 psci_cpu_on:
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top
-       str     r2, [r0]
-       dsb
+       bl      psci_cpu_on_common
 
        ldr     r2, =psci_cpu_entry
        bl      imx_cpu_on
arch/arm/cpu/armv7/psci.S +0 -0
index 8e25300..3b92f1d 100644
@@ -253,6 +253,18 @@ ENTRY(psci_enable_smp)
 ENDPROC(psci_enable_smp)
 .weak psci_enable_smp
 
Quoted 16 lines

arch/arm/cpu/armv7/psci.S ENTRY(psci_enable_smp) · line 267

Quoted 12 lines
arch/arm/cpu/armv7/psci.S +12 -0
+/* expects target CPU in r1, target PC in r2 */
+ENTRY(psci_cpu_on_common)
+       push    {lr}
+
+       mov     r0, r1
+       bl      psci_get_cpu_stack_top          @ get stack top of target CPU
+       str     r2, [r0]                        @ store target PC at stack top
+       dsb
+
+       pop     {pc}
+ENDPROC(psci_cpu_on_common)
+

Since you are just saving the target PC, why not just name the function like that?

No, not only saving target PC, you will see later that the newly added context ID is saved too.
In my previous patches, I save target PC here, and then in later patch save target ID too, and at last the common function is factored out, in this way every platform's code are touched too many times.

Copy one paragraph of my cover letter here:
"Move the patch of factoring out psci_cpu_on_common to earlier place 2/8, so that the following patches can only update the psci_cpu_on_common instead of touching each platform's psci_cpu_on, this reduces patches size obviously and make it easier for review."

Quoted 29 lines

See https://github.com/wens/u-boot-sunxi/commit/5a42630107e93bea9aa4da27cc16bdb604993dca

Quoted 27 lines
arch/arm/cpu/armv7/psci.S +0 -0
 ENTRY(psci_cpu_off_common)
        push    {lr}
 
arch/arm/cpu/armv7/sunxi/psci_sun6i.S +1 -4
index 90b5bfd..016e491 100644
@@ -134,10 +134,7 @@ out:       mcr     p15, 0, r7, c1, c1, 0
 psci_cpu_on:
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top  @ get stack top of target CPU
-       str     r2, [r0]                @ store target PC at stack top
-       dsb
+       bl      psci_cpu_on_common
 
        movw    r0, #(SUN6I_CPUCFG_BASE & 0xffff)
        movt    r0, #(SUN6I_CPUCFG_BASE >> 16)
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +0 -0
index e15d587..0ebb30e 100644
@@ -123,10 +123,7 @@ out:       mcr     p15, 0, r7, c1, c1, 0
 psci_cpu_on:
        push    {lr}
 
Quoted 9 lines

arch/arm/cpu/armv7/sunxi/psci_sun7i.S out: mcr p15, 0, r7, c1, c1, 0 · line 126

Quoted 5 lines
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +1 -4
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top  @ get stack top of target CPU
-       str     r2, [r0]                @ store target PC at stack top
-       dsb
+       bl      psci_cpu_on_common

The sunxi parts no longer apply. The partial rewrite in C has landed in master.

OK, will check sunxi's codes again.

Quoted 25 lines

Regards
ChenYu

Quoted 22 lines
arch/arm/cpu/armv7/sunxi/psci_sun7i.S +0 -0
        movw    r0, #(SUN7I_CPUCFG_BASE & 0xffff)
        movt    r0, #(SUN7I_CPUCFG_BASE >> 16)
arch/arm/mach-tegra/psci.S +1 -5
index b836da1..8a0147c 100644
@@ -90,10 +90,7 @@ ENDPROC(psci_cpu_off)
 ENTRY(psci_cpu_on)
        push    {lr}
 
-       mov     r0, r1
-       bl      psci_get_cpu_stack_top  @ get stack top of target CPU
-       str     r2, [r0]                @ store target PC at stack top
-       dsb
+       bl      psci_cpu_on_common
 
        ldr     r6, =TEGRA_RESET_EXCEPTION_VECTOR
        ldr     r5, =psci_cpu_entry
--

2.1.4

#5 lore raw

PATCH v5 3/8 ARMv7: PSCI: update the place of saving target PC

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

The legacy code saves target PC at stack top, this patch changes it to stack bottom, because we will save more contents for PSCI v1.0, by this way we don't need to adjust the stack pointer when more contents are saved.

---
 arch/arm/cpu/armv7/psci.S   | 9 +++++----
 arch/arm/include/asm/psci.h | 4 ++++
 2 files changed, 9 insertions(+), 4 deletions(-)
arch/arm/cpu/armv7/psci.S +5 -4
index 3b92f1d..5b235df 100644
@@ -259,7 +259,8 @@ ENTRY(psci_cpu_on_common)
 
 	mov	r0, r1
 	bl	psci_get_cpu_stack_top		@ get stack top of target CPU
-	str	r2, [r0]			@ store target PC at stack top
+	sub	r5, r0, #PSCI_TARGET_PC_OFFSET
+	str	r2, [r5]			@ store target PC
 	dsb
 
 	pop	{pc}
@@ -286,14 +287,13 @@ ENDPROC(psci_cpu_off_common)
 
 @ expects CPU ID in r0 and returns stack top in r0
 ENTRY(psci_get_cpu_stack_top)
-	mov	r5, #0x400			@ 1kB of stack per CPU
+	mov	r5, #PSCI_PERCPU_STACK_SIZE	@ 1kB of stack per CPU
 	mul	r0, r0, r5
 
 	ldr	r5, =psci_text_end		@ end of monitor text
 	add	r5, r5, #0x2000			@ Skip two pages
 	lsr	r5, r5, #12			@ Align to start of page
 	lsl	r5, r5, #12
-	sub	r5, r5, #4			@ reserve 1 word for target PC
 	sub	r0, r5, r0			@ here's our stack!
 
 	bx	lr
@@ -306,7 +306,8 @@ ENTRY(psci_cpu_entry)
 
 	bl	psci_get_cpu_id			@ CPU ID => r0
 	bl	psci_get_cpu_stack_top		@ stack top => r0
-	ldr	r0, [r0]			@ target PC at stack top
+	sub	r0, r0, #PSCI_TARGET_PC_OFFSET
+	ldr	r0, [r0]			@ get target PC
 	b	_do_nonsec_entry
 ENDPROC(psci_cpu_entry)
 
arch/arm/include/asm/psci.h +4 -0
index 2367ec0..cb08544 100644
@@ -63,6 +63,10 @@
 #define ARM_PSCI_1_0_FN_STAT_RESIDENCY		ARM_PSCI_0_2_FN(16)
 #define ARM_PSCI_1_0_FN_STAT_COUNT		ARM_PSCI_0_2_FN(17)
 
+/* size of percpu stack, 1kB */
+#define PSCI_PERCPU_STACK_SIZE		0x400
+#define PSCI_TARGET_PC_OFFSET		(PSCI_PERCPU_STACK_SIZE - 4)
+
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
Signature · 1 line
2.1.4
#6 lore raw

PATCH v5 3/8 ARMv7: PSCI: update the place of saving target PC

Quoted 18 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

The legacy code saves target PC at stack top, this patch changes it to stack bottom, because we will save more contents for PSCI v1.0, by this way we don't need to adjust the stack pointer when more contents are saved.

---
 arch/arm/cpu/armv7/psci.S   | 9 +++++----
 arch/arm/include/asm/psci.h | 4 ++++
 2 files changed, 9 insertions(+), 4 deletions(-)
arch/arm/cpu/armv7/psci.S +0 -0
index 3b92f1d..5b235df 100644

arch/arm/cpu/armv7/psci.S ENTRY(psci_cpu_on_common) · line 263

Quoted 6 lines
arch/arm/cpu/armv7/psci.S +2 -1
@@ -259,7 +259,8 @@ ENTRY(psci_cpu_on_common)
 
        mov     r0, r1
        bl      psci_get_cpu_stack_top          @ get stack top of target CPU
-       str     r2, [r0]                        @ store target PC at stack top
+       sub     r5, r0, #PSCI_TARGET_PC_OFFSET
+       str     r2, [r5]                        @ store target PC

IMO having a separate function for this would be better. It would be clearer, and easier to reuse or replace.

Also you should save and restore r5 across this function.

Quoted 7 lines
arch/arm/cpu/armv7/psci.S +0 -0
        dsb
 
        pop     {pc}
@@ -286,14 +287,13 @@ ENDPROC(psci_cpu_off_common)
 
 @ expects CPU ID in r0 and returns stack top in r0
 ENTRY(psci_get_cpu_stack_top)

arch/arm/cpu/armv7/psci.S ENDPROC(psci_cpu_off_common) · line 297

Quoted 10 lines
arch/arm/cpu/armv7/psci.S +1 -2
-       mov     r5, #0x400                      @ 1kB of stack per CPU
+       mov     r5, #PSCI_PERCPU_STACK_SIZE     @ 1kB of stack per CPU
        mul     r0, r0, r5
 
        ldr     r5, =psci_text_end              @ end of monitor text
        add     r5, r5, #0x2000                 @ Skip two pages
        lsr     r5, r5, #12                     @ Align to start of page
        lsl     r5, r5, #12
-       sub     r5, r5, #4                      @ reserve 1 word for target PC
        sub     r0, r5, r0                      @ here's our stack!

This does not apply. r5 was changed to r3.

Quoted 17 lines
arch/arm/cpu/armv7/psci.S +2 -1
        bx      lr
@@ -306,7 +306,8 @@ ENTRY(psci_cpu_entry)
 
        bl      psci_get_cpu_id                 @ CPU ID => r0
        bl      psci_get_cpu_stack_top          @ stack top => r0
-       ldr     r0, [r0]                        @ target PC at stack top
+       sub     r0, r0, #PSCI_TARGET_PC_OFFSET
+       ldr     r0, [r0]                        @ get target PC
        b       _do_nonsec_entry
 ENDPROC(psci_cpu_entry)
 
arch/arm/include/asm/psci.h +0 -0
index 2367ec0..cb08544 100644

arch/arm/include/asm/psci.h line 68

Quoted 6 lines
arch/arm/include/asm/psci.h +3 -0
@@ -63,6 +63,10 @@
 #define ARM_PSCI_1_0_FN_STAT_RESIDENCY         ARM_PSCI_0_2_FN(16)
 #define ARM_PSCI_1_0_FN_STAT_COUNT             ARM_PSCI_0_2_FN(17)
 
+/* size of percpu stack, 1kB */
+#define PSCI_PERCPU_STACK_SIZE         0x400
+#define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)

I think you want PSCI_PERCPU_STACK_SIZE?

A stack starts at 0x400 and goes down to 0x0. You want to store the target PC at 0x0, not 0x4.

Regards
ChenYu

Quoted 7 lines
arch/arm/include/asm/psci.h +1 -1
+
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
--

2.1.4

#7 lore raw

PATCH v5 3/8 ARMv7: PSCI: update the place of saving target PC

Quoted 20 lines · Chen-Yu Tsai
Quoted 19 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

The legacy code saves target PC at stack top, this patch changes it to stack bottom, because we will save more contents for PSCI v1.0, by this way we don't need to adjust the stack pointer when more contents are saved.

---
 arch/arm/cpu/armv7/psci.S   | 9 +++++----
 arch/arm/include/asm/psci.h | 4 ++++
 2 files changed, 9 insertions(+), 4 deletions(-)
arch/arm/cpu/armv7/psci.S +0 -0
index 3b92f1d..5b235df 100644
@@ -259,7 +259,8 @@ ENTRY(psci_cpu_on_common)
 
Quoted 9 lines

arch/arm/cpu/armv7/psci.S ENTRY(psci_cpu_on_common) · line 263

Quoted 5 lines
arch/arm/cpu/armv7/psci.S +2 -1
        mov     r0, r1
        bl      psci_get_cpu_stack_top          @ get stack top of target CPU
-       str     r2, [r0]                        @ store target PC at stack top
+       sub     r5, r0, #PSCI_TARGET_PC_OFFSET
+       str     r2, [r5]                        @ store target PC

IMO having a separate function for this would be better. It would be clearer, and easier to reuse or replace.

Renaming psci_cpu_on_common to psci_save_target_pc doesn't apply because saving context id will be added.
Do you mean the psci_cpu_on_common calls two functions psci_save_target_pc and psci_save_context_id?

Quoted 1 line

Also you should save and restore r5 across this function.

I did pay attention to this, and found that in the psci codes, general purpose registers are used carefully without push/pop, and there is example of using r5 without push/pop, so I just follow this pattern. Sure saving register context makes it safer.

Quoted 12 lines
Quoted 12 lines
arch/arm/cpu/armv7/psci.S +1 -1
        dsb
 
        pop     {pc}
@@ -286,14 +287,13 @@ ENDPROC(psci_cpu_off_common)
 
 @ expects CPU ID in r0 and returns stack top in r0
 ENTRY(psci_get_cpu_stack_top)
-       mov     r5, #0x400                      @ 1kB of stack per CPU
+       mov     r5, #PSCI_PERCPU_STACK_SIZE     @ 1kB of stack per CPU
        mul     r0, r0, r5
 
Quoted 9 lines

arch/arm/cpu/armv7/psci.S ENDPROC(psci_cpu_off_common) · line 297

Quoted 6 lines
arch/arm/cpu/armv7/psci.S +0 -1
        ldr     r5, =psci_text_end              @ end of monitor text
        add     r5, r5, #0x2000                 @ Skip two pages
        lsr     r5, r5, #12                     @ Align to start of page
        lsl     r5, r5, #12
-       sub     r5, r5, #4                      @ reserve 1 word for target PC
        sub     r0, r5, r0                      @ here's our stack!

This does not apply. r5 was changed to r3.

I noticed you sent such a patch before, but when I sent out this patch set, it wasn't pulled in.
Will re-check all the updates.

Quoted 20 lines
Quoted 20 lines
arch/arm/cpu/armv7/psci.S +2 -1
        bx      lr
@@ -306,7 +306,8 @@ ENTRY(psci_cpu_entry)
 
        bl      psci_get_cpu_id                 @ CPU ID => r0
        bl      psci_get_cpu_stack_top          @ stack top => r0
-       ldr     r0, [r0]                        @ target PC at stack top
+       sub     r0, r0, #PSCI_TARGET_PC_OFFSET
+       ldr     r0, [r0]                        @ get target PC
        b       _do_nonsec_entry
 ENDPROC(psci_cpu_entry)
 
arch/arm/include/asm/psci.h +0 -0
index 2367ec0..cb08544 100644
@@ -63,6 +63,10 @@
 #define ARM_PSCI_1_0_FN_STAT_RESIDENCY         ARM_PSCI_0_2_FN(16)
 #define ARM_PSCI_1_0_FN_STAT_COUNT             ARM_PSCI_0_2_FN(17)
 
Quoted 6 lines

arch/arm/include/asm/psci.h line 68

Quoted 3 lines
arch/arm/include/asm/psci.h +3 -0
+/* size of percpu stack, 1kB */
+#define PSCI_PERCPU_STACK_SIZE         0x400
+#define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)

I think you want PSCI_PERCPU_STACK_SIZE?

Yes.

Quoted 2 lines

A stack starts at 0x400 and goes down to 0x0. You want to store the target PC at 0x0, not 0x4.

But there are more details here.
Suppose the stack bottom is 0x0 and stack size is 0x400, the top of stack should be 0x3fc (32bits world, covering 4 bytes 0x3fc ~ 0x3ff), 0x400 is out of the stack range.
If the psci_get_stack_top returns 0x3fc, then (PSCI_PERCPU_STACK_SIZE
- 4) is right value.
But the psci_get_stack_top returns 0x400 actually, I've tried to change the code to return value like 0x3fc but found out that it isn't needed, because when push operation executed , sp is subtracted by 4 first and then data is saved into where pointed by the decreased sp. So I canceled changed to stack top but forgot to update this macro definiation.

Quoted 11 lines

Regards
ChenYu

Quoted 7 lines
arch/arm/include/asm/psci.h +1 -1
+
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
--

2.1.4

#8 lore raw

PATCH v5 3/8 ARMv7: PSCI: update the place of saving target PC

Quoted 30 lines · Hongbo Zhang
Quoted 20 lines · Chen-Yu Tsai
Quoted 19 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

The legacy code saves target PC at stack top, this patch changes it to stack bottom, because we will save more contents for PSCI v1.0, by this way we don't need to adjust the stack pointer when more contents are saved.

---
 arch/arm/cpu/armv7/psci.S   | 9 +++++----
 arch/arm/include/asm/psci.h | 4 ++++
 2 files changed, 9 insertions(+), 4 deletions(-)
arch/arm/cpu/armv7/psci.S +0 -0
index 3b92f1d..5b235df 100644
@@ -259,7 +259,8 @@ ENTRY(psci_cpu_on_common)
 
Quoted 9 lines

arch/arm/cpu/armv7/psci.S ENTRY(psci_cpu_on_common) · line 263

Quoted 5 lines
arch/arm/cpu/armv7/psci.S +2 -1
        mov     r0, r1
        bl      psci_get_cpu_stack_top          @ get stack top of target CPU
-       str     r2, [r0]                        @ store target PC at stack top
+       sub     r5, r0, #PSCI_TARGET_PC_OFFSET
+       str     r2, [r5]                        @ store target PC

IMO having a separate function for this would be better. It would be clearer, and easier to reuse or replace.

Quoted 4 lines

Renaming psci_cpu_on_common to psci_save_target_pc doesn't apply because saving context id will be added.
Do you mean the psci_cpu_on_common calls two functions psci_save_target_pc and psci_save_context_id?

That's the idea. Having common save/get helper functions for target PC, context ID, power state, etc. which others can use in their platform specific code should reduce duplicate code and make it easier to maintain. The platform code shouldn't care how or where this stuff is stored, as long as it can be reliably used. With your patches they are stored at the bottom of the stack. With mine they are moved to a secure data section.

Quoted 6 lines
Quoted 2 lines

Also you should save and restore r5 across this function.

I did pay attention to this, and found that in the psci codes, general purpose registers are used carefully without push/pop, and there is example of using r5 without push/pop, so I just follow this pattern.

Originally the PSCI implementation was contained in one file, written by one person. It was easy to check if registers were used. Now it's split up into a number of files across different platforms, with sunxi even using C. It's time to start paying attention to these details.

Quoted 1 line

Sure saving register context makes it safer.

Thanks. With sunxi C code, the compiler might use these registers, and we might get weird issues after calling such a common function.

Quoted 22 lines
Quoted 13 lines
Quoted 13 lines
arch/arm/cpu/armv7/psci.S +1 -1
        dsb
 
        pop     {pc}
@@ -286,14 +287,13 @@ ENDPROC(psci_cpu_off_common)
 
 @ expects CPU ID in r0 and returns stack top in r0
 ENTRY(psci_get_cpu_stack_top)
-       mov     r5, #0x400                      @ 1kB of stack per CPU
+       mov     r5, #PSCI_PERCPU_STACK_SIZE     @ 1kB of stack per CPU
        mul     r0, r0, r5
 
Quoted 9 lines

arch/arm/cpu/armv7/psci.S ENDPROC(psci_cpu_off_common) · line 297

Quoted 6 lines
arch/arm/cpu/armv7/psci.S +0 -1
        ldr     r5, =psci_text_end              @ end of monitor text
        add     r5, r5, #0x2000                 @ Skip two pages
        lsr     r5, r5, #12                     @ Align to start of page
        lsl     r5, r5, #12
-       sub     r5, r5, #4                      @ reserve 1 word for target PC
        sub     r0, r5, r0                      @ here's our stack!

This does not apply. r5 was changed to r3.

Quoted 3 lines

I noticed you sent such a patch before, but when I sent out this patch set, it wasn't pulled in.
Will re-check all the updates.

Thanks. The series was pulled in later than expected.

Quoted 26 lines
Quoted 20 lines
Quoted 20 lines
arch/arm/cpu/armv7/psci.S +2 -1
        bx      lr
@@ -306,7 +306,8 @@ ENTRY(psci_cpu_entry)
 
        bl      psci_get_cpu_id                 @ CPU ID => r0
        bl      psci_get_cpu_stack_top          @ stack top => r0
-       ldr     r0, [r0]                        @ target PC at stack top
+       sub     r0, r0, #PSCI_TARGET_PC_OFFSET
+       ldr     r0, [r0]                        @ get target PC
        b       _do_nonsec_entry
 ENDPROC(psci_cpu_entry)
 
arch/arm/include/asm/psci.h +0 -0
index 2367ec0..cb08544 100644
@@ -63,6 +63,10 @@
 #define ARM_PSCI_1_0_FN_STAT_RESIDENCY         ARM_PSCI_0_2_FN(16)
 #define ARM_PSCI_1_0_FN_STAT_COUNT             ARM_PSCI_0_2_FN(17)
 
Quoted 6 lines

arch/arm/include/asm/psci.h line 68

Quoted 3 lines
arch/arm/include/asm/psci.h +3 -0
+/* size of percpu stack, 1kB */
+#define PSCI_PERCPU_STACK_SIZE         0x400
+#define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)

I think you want PSCI_PERCPU_STACK_SIZE?

Quoted 8 lines

Yes.

Quoted 2 lines

A stack starts at 0x400 and goes down to 0x0. You want to store the target PC at 0x0, not 0x4.

But there are more details here.
Suppose the stack bottom is 0x0 and stack size is 0x400, the top of stack should be 0x3fc (32bits world, covering 4 bytes 0x3fc ~ 0x3ff), 0x400 is out of the stack range.

Let's define "stack top" as the address of SP when the stack is empty.

Quoted 7 lines

If the psci_get_stack_top returns 0x3fc, then (PSCI_PERCPU_STACK_SIZE
- 4) is right value.
But the psci_get_stack_top returns 0x400 actually, I've tried to change the code to return value like 0x3fc but found out that it isn't needed, because when push operation executed , sp is subtracted by 4 first and then data is saved into where pointed by the decreased sp. So I canceled changed to stack top but forgot to update this macro definiation.

Yes. You are right. This is what I alluded to.

Regards
ChenYu

Quoted 12 lines
Quoted 12 lines

Regards
ChenYu

Quoted 7 lines
arch/arm/include/asm/psci.h +1 -1
+
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
--

2.1.4

#9 lore raw

PATCH v5 4/8 ARMv7: PSCI: add codes to save context ID for CPU_ON

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

According to latest PSCI specification, the context ID is needed by CPU_ON. This patch saves context ID to the second lowest address of the stack (next to where target PC is saved), and restores it to r0 when needed while target CPU booting up.

---
 arch/arm/cpu/armv7/nonsec_virt.S | 7 +++++++
 arch/arm/cpu/armv7/psci.S        | 4 +++-
 arch/arm/include/asm/psci.h      | 1 +
 3 files changed, 11 insertions(+), 1 deletion(-)
arch/arm/cpu/armv7/nonsec_virt.S +7 -0
index b7563ed..6566643 100644
@@ -11,6 +11,7 @@
 #include <asm/gic.h>
 #include <asm/armv7.h>
 #include <asm/proc-armv/ptrace.h>
+#include <asm/psci.h>
 
 .arch_extension sec
 .arch_extension virt
@@ -89,6 +90,12 @@ _secure_monitor:
 	movne	r4, #0
 	mcrrne	p15, 4, r4, r4, c14		@ Reset CNTVOFF to zero
 1:
+#ifdef CONFIG_ARMV7_PSCI
+	bl	psci_get_cpu_id
+	bl	psci_get_cpu_stack_top
+	sub	r0, r0, #PSCI_CONTEXT_ID_OFFSET
+	ldr	r0, [r0]			@ get Context ID in r0
+#endif
 	mov	lr, ip
 	mov	ip, #(F_BIT | I_BIT | A_BIT)	@ Set A, I and F
 	tst	lr, #1				@ Check for Thumb PC
arch/arm/cpu/armv7/psci.S +3 -1
index 5b235df..3ba9e51 100644
@@ -253,7 +253,7 @@ ENTRY(psci_enable_smp)
 ENDPROC(psci_enable_smp)
 .weak psci_enable_smp
 
-/* expects target CPU in r1, target PC in r2 */
+/* expects target CPU in r1, target PC in r2, target conetxt ID in r3 */
 ENTRY(psci_cpu_on_common)
 	push	{lr}
 
@@ -261,6 +261,8 @@ ENTRY(psci_cpu_on_common)
 	bl	psci_get_cpu_stack_top		@ get stack top of target CPU
 	sub	r5, r0, #PSCI_TARGET_PC_OFFSET
 	str	r2, [r5]			@ store target PC
+	sub	r5, r0, #PSCI_CONTEXT_ID_OFFSET
+	str	r3, [r5]			@ store target context ID
 	dsb
 
 	pop	{pc}
arch/arm/include/asm/psci.h +1 -0
index cb08544..bedcd30 100644
@@ -66,6 +66,7 @@
 /* size of percpu stack, 1kB */
 #define PSCI_PERCPU_STACK_SIZE		0x400
 #define PSCI_TARGET_PC_OFFSET		(PSCI_PERCPU_STACK_SIZE - 4)
+#define PSCI_CONTEXT_ID_OFFSET		(PSCI_PERCPU_STACK_SIZE - 8)
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
Signature · 1 line
2.1.4
#10 lore raw

PATCH v5 4/8 ARMv7: PSCI: add codes to save context ID for CPU_ON

Hi,

Quoted 6 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

According to latest PSCI specification, the context ID is needed by CPU_ON. This patch saves context ID to the second lowest address of the stack (next to where target PC is saved), and restores it to r0 when needed while target CPU booting up.

Interesting. This doesn't seem to be used by Linux yet. This led me using r3 as a scratch register. See below.

Quoted 22 lines
---
 arch/arm/cpu/armv7/nonsec_virt.S | 7 +++++++
 arch/arm/cpu/armv7/psci.S        | 4 +++-
 arch/arm/include/asm/psci.h      | 1 +
 3 files changed, 11 insertions(+), 1 deletion(-)
arch/arm/cpu/armv7/nonsec_virt.S +1 -0
index b7563ed..6566643 100644
@@ -11,6 +11,7 @@
 #include <asm/gic.h>
 #include <asm/armv7.h>
 #include <asm/proc-armv/ptrace.h>
+#include <asm/psci.h>
 
 .arch_extension sec
 .arch_extension virt

arch/arm/cpu/armv7/nonsec_virt.S _secure_monitor: · line 97

Quoted 8 lines
arch/arm/cpu/armv7/nonsec_virt.S +5 -0
@@ -89,6 +90,12 @@ _secure_monitor:
        movne   r4, #0
        mcrrne  p15, 4, r4, r4, c14             @ Reset CNTVOFF to zero
 1:
+#ifdef CONFIG_ARMV7_PSCI
+       bl      psci_get_cpu_id
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CONTEXT_ID_OFFSET
+       ldr     r0, [r0]                        @ get Context ID in r0

You should do this in psci_cpu_entry, for a couple of reasons:

- That is also where the target PC is loaded.

- It is PSCI specific.

- All first time SMC calls would go through _secure_monitor. This includes when U-boot jumps into the kernel through _do_nonsec_entry, at which point you don't have a proper value stored, while U-boot passes 0 here. See:

http://git.denx.de/?p=u-boot.git;a=blob;f=arch/arm/lib/bootm.c;h=0838d89907b9a2eb81f4ebb31d8c045e031c5e11;hb=HEAD#l324

Quoted 18 lines
arch/arm/cpu/armv7/nonsec_virt.S +1 -0
+#endif
        mov     lr, ip
        mov     ip, #(F_BIT | I_BIT | A_BIT)    @ Set A, I and F
        tst     lr, #1                          @ Check for Thumb PC
arch/arm/cpu/armv7/psci.S +1 -1
index 5b235df..3ba9e51 100644
@@ -253,7 +253,7 @@ ENTRY(psci_enable_smp)
 ENDPROC(psci_enable_smp)
 .weak psci_enable_smp
 
-/* expects target CPU in r1, target PC in r2 */
+/* expects target CPU in r1, target PC in r2, target conetxt ID in r3 */
 ENTRY(psci_cpu_on_common)
        push    {lr}
 

arch/arm/cpu/armv7/psci.S ENTRY(psci_cpu_on_common) · line 261

Quoted 1 line
arch/arm/cpu/armv7/psci.S +0 -0
@@ -261,6 +261,8 @@ ENTRY(psci_cpu_on_common)
        bl      psci_get_cpu_stack_top          @ get stack top of target CPU

r3 get overwritten in psci_get_cpu_stack_top in

http://git.denx.de/?p=u-boot.git;a=commitdiff;h=dae08d228122e4ad296077106520a4db3ca17872

Some of the functions now follow the ARM calling conventions. You should save r3 and possibly other registers across any function calls.

Regards
ChenYu

Quoted 22 lines
arch/arm/cpu/armv7/psci.S +2 -0
        sub     r5, r0, #PSCI_TARGET_PC_OFFSET
        str     r2, [r5]                        @ store target PC
+       sub     r5, r0, #PSCI_CONTEXT_ID_OFFSET
+       str     r3, [r5]                        @ store target context ID
        dsb
 
        pop     {pc}
arch/arm/include/asm/psci.h +1 -1
index cb08544..bedcd30 100644
@@ -66,6 +66,7 @@
 /* size of percpu stack, 1kB */
 #define PSCI_PERCPU_STACK_SIZE         0x400
 #define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)
+#define PSCI_CONTEXT_ID_OFFSET         (PSCI_PERCPU_STACK_SIZE - 8)
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
--

2.1.4

#11 lore raw

PATCH v5 4/8 ARMv7: PSCI: add codes to save context ID for CPU_ON

Quoted 5 lines · Chen-Yu Tsai

Hi,

Quoted 2 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

Quoted 8 lines
Quoted 4 lines

According to latest PSCI specification, the context ID is needed by CPU_ON. This patch saves context ID to the second lowest address of the stack (next to where target PC is saved), and restores it to r0 when needed while target CPU booting up.

Interesting. This doesn't seem to be used by Linux yet. This led me using r3 as a scratch register. See below.

Context ID is introduced by PSCI v1.0, since we are adding v1.0 support, I think we should add this although Linux kernel doesn't use this till now, it is spec.
And what's more, r0, r1, r2 are already used by current codes, and in the spec paragraph 5.2.1:
"For PSCI functions that use only 32-bit parameters, the arguments are passed in R0 to R3 (AArch32) or W0 to W3 (AArch64), with return values in R0 or W0."
All of these make the r3 the only best register for context ID I think.

Quoted 19 lines
Quoted 19 lines
---
 arch/arm/cpu/armv7/nonsec_virt.S | 7 +++++++
 arch/arm/cpu/armv7/psci.S        | 4 +++-
 arch/arm/include/asm/psci.h      | 1 +
 3 files changed, 11 insertions(+), 1 deletion(-)
arch/arm/cpu/armv7/nonsec_virt.S +1 -0
index b7563ed..6566643 100644
@@ -11,6 +11,7 @@
 #include <asm/gic.h>
 #include <asm/armv7.h>
 #include <asm/proc-armv/ptrace.h>
+#include <asm/psci.h>
 
Quoted 14 lines

arch/arm/cpu/armv7/nonsec_virt.S _secure_monitor: · line 97

Quoted 11 lines
arch/arm/cpu/armv7/nonsec_virt.S +5 -0
 .arch_extension sec
 .arch_extension virt
@@ -89,6 +90,12 @@ _secure_monitor:
        movne   r4, #0
        mcrrne  p15, 4, r4, r4, c14             @ Reset CNTVOFF to zero
 1:
+#ifdef CONFIG_ARMV7_PSCI
+       bl      psci_get_cpu_id
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CONTEXT_ID_OFFSET
+       ldr     r0, [r0]                        @ get Context ID in r0

You should do this in psci_cpu_entry, for a couple of reasons:

Yes, thanks.

Quoted 33 lines

- That is also where the target PC is loaded.

- It is PSCI specific.

- All first time SMC calls would go through _secure_monitor. This includes when U-boot jumps into the kernel through _do_nonsec_entry, at which point you don't have a proper value stored, while U-boot passes 0 here. See:

http://git.denx.de/?p=u-boot.git;a=blob;f=arch/arm/lib/bootm.c;h=0838d89907b9a2eb81f4ebb31d8c045e031c5e11;hb=HEAD#l324

Quoted 18 lines
arch/arm/cpu/armv7/nonsec_virt.S +1 -0
+#endif
        mov     lr, ip
        mov     ip, #(F_BIT | I_BIT | A_BIT)    @ Set A, I and F
        tst     lr, #1                          @ Check for Thumb PC
arch/arm/cpu/armv7/psci.S +1 -1
index 5b235df..3ba9e51 100644
@@ -253,7 +253,7 @@ ENTRY(psci_enable_smp)
 ENDPROC(psci_enable_smp)
 .weak psci_enable_smp
 
-/* expects target CPU in r1, target PC in r2 */
+/* expects target CPU in r1, target PC in r2, target conetxt ID in r3 */
 ENTRY(psci_cpu_on_common)
        push    {lr}
 

arch/arm/cpu/armv7/psci.S ENTRY(psci_cpu_on_common) · line 261

Quoted 1 line
arch/arm/cpu/armv7/psci.S +0 -0
@@ -261,6 +261,8 @@ ENTRY(psci_cpu_on_common)
        bl      psci_get_cpu_stack_top          @ get stack top of target CPU

r3 get overwritten in psci_get_cpu_stack_top in

Quoted 6 lines

http://git.denx.de/?p=u-boot.git;a=commitdiff;h=dae08d228122e4ad296077106520a4db3ca17872

Some of the functions now follow the ARM calling conventions. You should save r3 and possibly other registers across any function calls.

OK, will consider register saving.

Quoted 25 lines

Regards
ChenYu

Quoted 22 lines
arch/arm/cpu/armv7/psci.S +2 -0
        sub     r5, r0, #PSCI_TARGET_PC_OFFSET
        str     r2, [r5]                        @ store target PC
+       sub     r5, r0, #PSCI_CONTEXT_ID_OFFSET
+       str     r3, [r5]                        @ store target context ID
        dsb
 
        pop     {pc}
arch/arm/include/asm/psci.h +1 -1
index cb08544..bedcd30 100644
@@ -66,6 +66,7 @@
 /* size of percpu stack, 1kB */
 #define PSCI_PERCPU_STACK_SIZE         0x400
 #define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)
+#define PSCI_CONTEXT_ID_OFFSET         (PSCI_PERCPU_STACK_SIZE - 8)
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
--

2.1.4

#12 lore raw

PATCH v5 5/8 ARMv7: PSCI: ls102xa: check target CPU ID before further operations

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

The input parameter CPU ID needs to be validated before furher oprations such as CPU_ON, this patch introduces the function to do this.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 34 +++++++++++++++++++++++++++++++++-
 1 file changed, 33 insertions(+), 1 deletion(-)
arch/arm/cpu/armv7/ls102xa/psci.S +33 -1
index 548c507..a4482e4 100644
@@ -25,6 +25,34 @@
 #define	ONE_MS		(GENERIC_TIMER_CLK / 1000)
 #define	RESET_WAIT	(30 * ONE_MS)
 
+.globl	psci_check_target_cpu_id
+psci_check_target_cpu_id:
+	@ Get the real CPU number
+	and	r0, r1, #0xff
+
+	@ Verify bit[31:24], bits must be zero.
+	tst	r1, #0xff000000
+	bne	out_psci_invalid_target_cpu_id
+
+	@ Verify Affinity level 2: Cluster, only one cluster in LS1021xa SoC.
+	tst	r1, #0xff0000
+	bne	out_psci_invalid_target_cpu_id
+
+	@ Verify Affinity level 1: Processors, should be in 0xf00 format.
+	lsr	r1, r1, #8
+	teq	r1, #0xf
+	bne	out_psci_invalid_target_cpu_id
+
+	@ Verify Affinity level 0: CPU, only 0, 1 are valid values.
+	cmp	r0, #2
+	bge	out_psci_invalid_target_cpu_id
+
+	bx	lr
+
+out_psci_invalid_target_cpu_id:
+	mov	r0, #ARM_PSCI_RET_INVAL
+	bx	lr
+
 	@ r1 = target CPU
 	@ r2 = target PC
 .globl	psci_cpu_on
@@ -33,7 +61,10 @@ psci_cpu_on:
 
 	@ Clear and Get the correct CPU number
 	@ r1 = 0xf01
-	and	r1, r1, #0xff
+	bl	psci_check_target_cpu_id
+	cmp	r0, #ARM_PSCI_RET_INVAL
+	beq	out_psci_cpu_on
+	mov	r1, r0
 
 	bl	psci_cpu_on_common
 
@@ -98,6 +129,7 @@ holdoff_release:
 	@ Return
 	mov	r0, #ARM_PSCI_RET_SUCCESS
 
+out_psci_cpu_on:
 	pop	{lr}
 	bx	lr
 
Signature · 1 line
2.1.4
#13 lore raw

PATCH v5 5/8 ARMv7: PSCI: ls102xa: check target CPU ID before further operations

Quoted 16 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

The input parameter CPU ID needs to be validated before furher oprations such as CPU_ON, this patch introduces the function to do this.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 34 +++++++++++++++++++++++++++++++++-
 1 file changed, 33 insertions(+), 1 deletion(-)
arch/arm/cpu/armv7/ls102xa/psci.S +0 -0
index 548c507..a4482e4 100644

arch/arm/cpu/armv7/ls102xa/psci.S line 27

Quoted 3 lines
arch/arm/cpu/armv7/ls102xa/psci.S +0 -0
@@ -25,6 +25,34 @@
 #define        ONE_MS          (GENERIC_TIMER_CLK / 1000)
 #define        RESET_WAIT      (30 * ONE_MS)
 

A note describing the arguments, the return value, and any other affected registers would be nice. For example r1 is clobbered.

arch/arm/cpu/armv7/ls102xa/psci.S line 29

Quoted 2 lines
arch/arm/cpu/armv7/ls102xa/psci.S +2 -0
+.globl psci_check_target_cpu_id
+psci_check_target_cpu_id:

You probably don't need .globl.

Also you can use ENTRY() or LENTRY() here. These and ENDPROC() below are defined in linux/linkage.h

Quoted 16 lines
arch/arm/cpu/armv7/ls102xa/psci.S +16 -0
+       @ Get the real CPU number
+       and     r0, r1, #0xff
+
+       @ Verify bit[31:24], bits must be zero.
+       tst     r1, #0xff000000
+       bne     out_psci_invalid_target_cpu_id
+
+       @ Verify Affinity level 2: Cluster, only one cluster in LS1021xa SoC.
+       tst     r1, #0xff0000
+       bne     out_psci_invalid_target_cpu_id
+
+       @ Verify Affinity level 1: Processors, should be in 0xf00 format.
+       lsr     r1, r1, #8
+       teq     r1, #0xf
+       bne     out_psci_invalid_target_cpu_id
+

arch/arm/cpu/armv7/ls102xa/psci.S line 55

Quoted 10 lines
arch/arm/cpu/armv7/ls102xa/psci.S +10 -0
+       @ Verify Affinity level 0: CPU, only 0, 1 are valid values.
+       cmp     r0, #2
+       bge     out_psci_invalid_target_cpu_id
+
+       bx      lr
+
+out_psci_invalid_target_cpu_id:
+       mov     r0, #ARM_PSCI_RET_INVAL
+       bx      lr
+

And you could have an ENDPROC() here.

About the whole function, you could use an extra scratch register, store ARM_PSCI_RET_INVAL in r0 at the beginning, and directly return at errors. Kind of like:

if (bit[31:24] != 0)

        return ARM_PSCI_RET_INVAL;
if (cluster != 0)
        return ARM_PSCI_RET_INVAL;
if (processor != 0xf)
        return ARM_PSCI_RET_INVAL;
if (cpu >= 2)
        return ARM_PSCI_RET_INVAL;

return cpu;

It's just a different style though. Feel free to keep the current structure.

The code itself looks good.

Regards
ChenYu

Quoted 26 lines
arch/arm/cpu/armv7/ls102xa/psci.S +5 -2
        @ r1 = target CPU
        @ r2 = target PC
 .globl psci_cpu_on
@@ -33,7 +61,10 @@ psci_cpu_on:
 
        @ Clear and Get the correct CPU number
        @ r1 = 0xf01
-       and     r1, r1, #0xff
+       bl      psci_check_target_cpu_id
+       cmp     r0, #ARM_PSCI_RET_INVAL
+       beq     out_psci_cpu_on
+       mov     r1, r0
 
        bl      psci_cpu_on_common
 
@@ -98,6 +129,7 @@ holdoff_release:
        @ Return
        mov     r0, #ARM_PSCI_RET_SUCCESS
 
+out_psci_cpu_on:
        pop     {lr}
        bx      lr
 
--

2.1.4

#14 lore raw

PATCH v5 5/8 ARMv7: PSCI: ls102xa: check target CPU ID before further operations

Quoted 12 lines · Chen-Yu Tsai
Quoted 11 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

The input parameter CPU ID needs to be validated before furher oprations such as CPU_ON, this patch introduces the function to do this.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 34 +++++++++++++++++++++++++++++++++-
 1 file changed, 33 insertions(+), 1 deletion(-)
Quoted 11 lines

arch/arm/cpu/armv7/ls102xa/psci.S line 27

Quoted 8 lines
arch/arm/cpu/armv7/ls102xa/psci.S +0 -0
index 548c507..a4482e4 100644
@@ -25,6 +25,34 @@
 #define        ONE_MS          (GENERIC_TIMER_CLK / 1000)
 #define        RESET_WAIT      (30 * ONE_MS)
 

A note describing the arguments, the return value, and any other affected registers would be nice. For example r1 is clobbered.

Good suggestion, thanks.

Quoted 6 lines

arch/arm/cpu/armv7/ls102xa/psci.S line 29

Quoted 3 lines
arch/arm/cpu/armv7/ls102xa/psci.S +2 -0
+.globl psci_check_target_cpu_id
+psci_check_target_cpu_id:

You probably don't need .globl.

Oh, this is due to a carelessness.

Quoted 3 lines

Also you can use ENTRY() or LENTRY() here. These and ENDPROC() below are defined in linux/linkage.h

Yes, sure.

Quoted 41 lines
Quoted 16 lines
arch/arm/cpu/armv7/ls102xa/psci.S +16 -0
+       @ Get the real CPU number
+       and     r0, r1, #0xff
+
+       @ Verify bit[31:24], bits must be zero.
+       tst     r1, #0xff000000
+       bne     out_psci_invalid_target_cpu_id
+
+       @ Verify Affinity level 2: Cluster, only one cluster in LS1021xa SoC.
+       tst     r1, #0xff0000
+       bne     out_psci_invalid_target_cpu_id
+
+       @ Verify Affinity level 1: Processors, should be in 0xf00 format.
+       lsr     r1, r1, #8
+       teq     r1, #0xf
+       bne     out_psci_invalid_target_cpu_id
+

arch/arm/cpu/armv7/ls102xa/psci.S line 55

Quoted 10 lines
arch/arm/cpu/armv7/ls102xa/psci.S +10 -0
+       @ Verify Affinity level 0: CPU, only 0, 1 are valid values.
+       cmp     r0, #2
+       bge     out_psci_invalid_target_cpu_id
+
+       bx      lr
+
+out_psci_invalid_target_cpu_id:
+       mov     r0, #ARM_PSCI_RET_INVAL
+       bx      lr
+

And you could have an ENDPROC() here.

About the whole function, you could use an extra scratch register, store ARM_PSCI_RET_INVAL in r0 at the beginning, and directly return at errors. Kind of like:

if (bit[31:24] != 0)

        return ARM_PSCI_RET_INVAL;
if (cluster != 0)
        return ARM_PSCI_RET_INVAL;
if (processor != 0xf)
        return ARM_PSCI_RET_INVAL;
if (cpu >= 2)
        return ARM_PSCI_RET_INVAL;
Quoted 4 lines

return cpu;

It's just a different style though. Feel free to keep the current structure.

Yes, thanks.

Quoted 32 lines

The code itself looks good.

Regards
ChenYu

Quoted 26 lines
arch/arm/cpu/armv7/ls102xa/psci.S +5 -2
        @ r1 = target CPU
        @ r2 = target PC
 .globl psci_cpu_on
@@ -33,7 +61,10 @@ psci_cpu_on:
 
        @ Clear and Get the correct CPU number
        @ r1 = 0xf01
-       and     r1, r1, #0xff
+       bl      psci_check_target_cpu_id
+       cmp     r0, #ARM_PSCI_RET_INVAL
+       beq     out_psci_cpu_on
+       mov     r1, r0
 
        bl      psci_cpu_on_common
 
@@ -98,6 +129,7 @@ holdoff_release:
        @ Return
        mov     r0, #ARM_PSCI_RET_SUCCESS
 
+out_psci_cpu_on:
        pop     {lr}
        bx      lr
 
--

2.1.4

#15 lore raw

PATCH v5 6/8 ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

For the robustness of codes, while powering on a CPU, it is better to check if the target CPU is already on or in the process of power on, if yes the power on routine shouldn't be executed further and should return with the corresponding status immediately.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 29 +++++++++++++++++++++++++++++
 arch/arm/include/asm/psci.h       |  5 +++++
 2 files changed, 34 insertions(+)
arch/arm/cpu/armv7/ls102xa/psci.S +29 -0
index a4482e4..0188ade 100644
@@ -66,6 +66,22 @@ psci_cpu_on:
 	beq	out_psci_cpu_on
 	mov	r1, r0
 
+	bl	psci_get_cpu_stack_top
+	sub	r0, r0, #PSCI_CPU_STATUS_OFFSET
+	ldr	r5, [r0]
+
+	cmp	r5, #PSCI_CPU_STATUS_ON
+	moveq	r0, #ARM_PSCI_RET_ALREADY_ON
+	beq	out_psci_cpu_on
+
+	cmp	r5, #PSCI_CPU_STATUS_ON_PENDING
+	moveq	r0, #ARM_PSCI_RET_ON_PENDING
+	beq	out_psci_cpu_on
+
+	mov	r5, #PSCI_CPU_STATUS_ON_PENDING
+	str	r5, [r0]
+	dsb
+
 	bl	psci_cpu_on_common
 
 	@ Get DCFG base address
@@ -123,6 +139,12 @@ holdoff_release:
 	rev	r6, r6
 	str	r6, [r4, #DCFG_CCSR_SCRATCHRW1]
 
+	mov	r0, r1
+	bl	psci_get_cpu_stack_top
+	sub	r0, r0, #PSCI_CPU_STATUS_OFFSET
+	mov	r5, #PSCI_CPU_STATUS_ON
+	str	r5, [r0]
+
 	isb
 	dsb
 
@@ -137,6 +159,13 @@ out_psci_cpu_on:
 psci_cpu_off:
 	bl	psci_cpu_off_common
 
+	bl	psci_get_cpu_id
+	bl	psci_get_cpu_stack_top
+	sub	r0, r0, #PSCI_CPU_STATUS_OFFSET
+	mov	r5, #PSCI_CPU_STATUS_OFF
+	str	r5, [r0]
+	dsb
+
 1:	wfi
 	b	1b
 
arch/arm/include/asm/psci.h +5 -0
index bedcd30..89a1ba5 100644
@@ -67,6 +67,11 @@
 #define PSCI_PERCPU_STACK_SIZE		0x400
 #define PSCI_TARGET_PC_OFFSET		(PSCI_PERCPU_STACK_SIZE - 4)
 #define PSCI_CONTEXT_ID_OFFSET		(PSCI_PERCPU_STACK_SIZE - 8)
+#define PSCI_CPU_STATUS_OFFSET		(PSCI_PERCPU_STACK_SIZE - 12)
+
+#define PSCI_CPU_STATUS_OFF		0
+#define PSCI_CPU_STATUS_ON		1
+#define PSCI_CPU_STATUS_ON_PENDING	2
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
Signature · 1 line
2.1.4
#16 lore raw

PATCH v5 6/8 ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON

Quoted 54 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

For the robustness of codes, while powering on a CPU, it is better to check if the target CPU is already on or in the process of power on, if yes the power on routine shouldn't be executed further and should return with the corresponding status immediately.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 29 +++++++++++++++++++++++++++++
 arch/arm/include/asm/psci.h       |  5 +++++
 2 files changed, 34 insertions(+)
arch/arm/cpu/armv7/ls102xa/psci.S +22 -0
index a4482e4..0188ade 100644
@@ -66,6 +66,22 @@ psci_cpu_on:
        beq     out_psci_cpu_on
        mov     r1, r0
 
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       ldr     r5, [r0]
+
+       cmp     r5, #PSCI_CPU_STATUS_ON
+       moveq   r0, #ARM_PSCI_RET_ALREADY_ON
+       beq     out_psci_cpu_on
+
+       cmp     r5, #PSCI_CPU_STATUS_ON_PENDING
+       moveq   r0, #ARM_PSCI_RET_ON_PENDING
+       beq     out_psci_cpu_on
+
+       mov     r5, #PSCI_CPU_STATUS_ON_PENDING
+       str     r5, [r0]
+       dsb
+
        bl      psci_cpu_on_common
 
        @ Get DCFG base address
@@ -123,6 +139,12 @@ holdoff_release:
        rev     r6, r6
        str     r6, [r4, #DCFG_CCSR_SCRATCHRW1]
 
+       mov     r0, r1
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       mov     r5, #PSCI_CPU_STATUS_ON
+       str     r5, [r0]
+
        isb
        dsb
 

arch/arm/cpu/armv7/ls102xa/psci.S out_psci_cpu_on: · line 167

Quoted 9 lines
arch/arm/cpu/armv7/ls102xa/psci.S +6 -0
@@ -137,6 +159,13 @@ out_psci_cpu_on:
 psci_cpu_off:
        bl      psci_cpu_off_common
 
+       bl      psci_get_cpu_id
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       mov     r5, #PSCI_CPU_STATUS_OFF
+       str     r5, [r0]
+       dsb

psci_cpu_off_common flushes and disables caches, and turns off SMP. So the code you're adding might not work as expected? ARM folks might know more.

The rest looks good, though I expect them to be pulled out into common helpers and rewritten in C. :)

Regards
ChenYu

Quoted 23 lines
arch/arm/cpu/armv7/ls102xa/psci.S +1 -0
+
 1:     wfi
        b       1b
 
arch/arm/include/asm/psci.h +5 -1
index bedcd30..89a1ba5 100644
@@ -67,6 +67,11 @@
 #define PSCI_PERCPU_STACK_SIZE         0x400
 #define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)
 #define PSCI_CONTEXT_ID_OFFSET         (PSCI_PERCPU_STACK_SIZE - 8)
+#define PSCI_CPU_STATUS_OFFSET         (PSCI_PERCPU_STACK_SIZE - 12)
+
+#define PSCI_CPU_STATUS_OFF            0
+#define PSCI_CPU_STATUS_ON             1
+#define PSCI_CPU_STATUS_ON_PENDING     2
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
--

2.1.4

#17 lore raw

PATCH v5 6/8 ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON

Quoted 58 lines · Chen-Yu Tsai
Quoted 57 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

For the robustness of codes, while powering on a CPU, it is better to check if the target CPU is already on or in the process of power on, if yes the power on routine shouldn't be executed further and should return with the corresponding status immediately.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 29 +++++++++++++++++++++++++++++
 arch/arm/include/asm/psci.h       |  5 +++++
 2 files changed, 34 insertions(+)
arch/arm/cpu/armv7/ls102xa/psci.S +22 -0
index a4482e4..0188ade 100644
@@ -66,6 +66,22 @@ psci_cpu_on:
        beq     out_psci_cpu_on
        mov     r1, r0
 
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       ldr     r5, [r0]
+
+       cmp     r5, #PSCI_CPU_STATUS_ON
+       moveq   r0, #ARM_PSCI_RET_ALREADY_ON
+       beq     out_psci_cpu_on
+
+       cmp     r5, #PSCI_CPU_STATUS_ON_PENDING
+       moveq   r0, #ARM_PSCI_RET_ON_PENDING
+       beq     out_psci_cpu_on
+
+       mov     r5, #PSCI_CPU_STATUS_ON_PENDING
+       str     r5, [r0]
+       dsb
+
        bl      psci_cpu_on_common
 
        @ Get DCFG base address
@@ -123,6 +139,12 @@ holdoff_release:
        rev     r6, r6
        str     r6, [r4, #DCFG_CCSR_SCRATCHRW1]
 
+       mov     r0, r1
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       mov     r5, #PSCI_CPU_STATUS_ON
+       str     r5, [r0]
+
        isb
        dsb
 
@@ -137,6 +159,13 @@ out_psci_cpu_on:
 psci_cpu_off:
        bl      psci_cpu_off_common
 
Quoted 11 lines

arch/arm/cpu/armv7/ls102xa/psci.S out_psci_cpu_on: · line 167

Quoted 6 lines
arch/arm/cpu/armv7/ls102xa/psci.S +6 -0
+       bl      psci_get_cpu_id
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       mov     r5, #PSCI_CPU_STATUS_OFF
+       str     r5, [r0]
+       dsb

psci_cpu_off_common flushes and disables caches, and turns off SMP. So the code you're adding might not work as expected? ARM folks might know more.

Moving this code ahead of psci_cpu_off_common seems better? I just though such a code should be placed as later as it can.

Quoted 29 lines

The rest looks good, though I expect them to be pulled out into common helpers and rewritten in C. :)

Regards
ChenYu

Quoted 23 lines
arch/arm/cpu/armv7/ls102xa/psci.S +1 -0
+
 1:     wfi
        b       1b
 
arch/arm/include/asm/psci.h +5 -1
index bedcd30..89a1ba5 100644
@@ -67,6 +67,11 @@
 #define PSCI_PERCPU_STACK_SIZE         0x400
 #define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)
 #define PSCI_CONTEXT_ID_OFFSET         (PSCI_PERCPU_STACK_SIZE - 8)
+#define PSCI_CPU_STATUS_OFFSET         (PSCI_PERCPU_STACK_SIZE - 12)
+
+#define PSCI_CPU_STATUS_OFF            0
+#define PSCI_CPU_STATUS_ON             1
+#define PSCI_CPU_STATUS_ON_PENDING     2
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
--

2.1.4

#18 lore raw

PATCH v5 6/8 ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON

Quoted 70 lines · Hongbo Zhang
Quoted 58 lines · Chen-Yu Tsai
Quoted 57 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

For the robustness of codes, while powering on a CPU, it is better to check if the target CPU is already on or in the process of power on, if yes the power on routine shouldn't be executed further and should return with the corresponding status immediately.

---
 arch/arm/cpu/armv7/ls102xa/psci.S | 29 +++++++++++++++++++++++++++++
 arch/arm/include/asm/psci.h       |  5 +++++
 2 files changed, 34 insertions(+)
arch/arm/cpu/armv7/ls102xa/psci.S +22 -0
index a4482e4..0188ade 100644
@@ -66,6 +66,22 @@ psci_cpu_on:
        beq     out_psci_cpu_on
        mov     r1, r0
 
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       ldr     r5, [r0]
+
+       cmp     r5, #PSCI_CPU_STATUS_ON
+       moveq   r0, #ARM_PSCI_RET_ALREADY_ON
+       beq     out_psci_cpu_on
+
+       cmp     r5, #PSCI_CPU_STATUS_ON_PENDING
+       moveq   r0, #ARM_PSCI_RET_ON_PENDING
+       beq     out_psci_cpu_on
+
+       mov     r5, #PSCI_CPU_STATUS_ON_PENDING
+       str     r5, [r0]
+       dsb
+
        bl      psci_cpu_on_common
 
        @ Get DCFG base address
@@ -123,6 +139,12 @@ holdoff_release:
        rev     r6, r6
        str     r6, [r4, #DCFG_CCSR_SCRATCHRW1]
 
+       mov     r0, r1
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       mov     r5, #PSCI_CPU_STATUS_ON
+       str     r5, [r0]
+
        isb
        dsb
 
@@ -137,6 +159,13 @@ out_psci_cpu_on:
 psci_cpu_off:
        bl      psci_cpu_off_common
 
Quoted 11 lines

arch/arm/cpu/armv7/ls102xa/psci.S out_psci_cpu_on: · line 167

Quoted 6 lines
arch/arm/cpu/armv7/ls102xa/psci.S +6 -0
+       bl      psci_get_cpu_id
+       bl      psci_get_cpu_stack_top
+       sub     r0, r0, #PSCI_CPU_STATUS_OFFSET
+       mov     r5, #PSCI_CPU_STATUS_OFF
+       str     r5, [r0]
+       dsb

psci_cpu_off_common flushes and disables caches, and turns off SMP. So the code you're adding might not work as expected? ARM folks might know more.

Quoted 2 lines

Moving this code ahead of psci_cpu_off_common seems better? I just though such a code should be placed as later as it can.

I agree it should be as late as possible. I'm just not sure about the cache and AMP/SMP settings. Best to have someone more familiar with this comment on it.

ChenYu

Quoted 29 lines
Quoted 29 lines

The rest looks good, though I expect them to be pulled out into common helpers and rewritten in C. :)

Regards
ChenYu

Quoted 23 lines
arch/arm/cpu/armv7/ls102xa/psci.S +1 -0
+
 1:     wfi
        b       1b
 
arch/arm/include/asm/psci.h +5 -1
index bedcd30..89a1ba5 100644
@@ -67,6 +67,11 @@
 #define PSCI_PERCPU_STACK_SIZE         0x400
 #define PSCI_TARGET_PC_OFFSET          (PSCI_PERCPU_STACK_SIZE - 4)
 #define PSCI_CONTEXT_ID_OFFSET         (PSCI_PERCPU_STACK_SIZE - 8)
+#define PSCI_CPU_STATUS_OFFSET         (PSCI_PERCPU_STACK_SIZE - 12)
+
+#define PSCI_CPU_STATUS_OFF            0
+#define PSCI_CPU_STATUS_ON             1
+#define PSCI_CPU_STATUS_ON_PENDING     2
 
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
--

2.1.4

#19 lore raw

PATCH v5 7/8 ARMv7: PSCI: ls102xa: add more PSCI v1.0 functions implemention

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

This patch implements PSCI functions for ls102xa SoC following PSCI v1.0, they are as the list:
psci_version,
psci_features,
psci_cpu_suspend,
psci_affinity_info,
psci_system_reset,
psci_system_off.

Tested on LS1021aQDS, LS1021aTWR.

---
 arch/arm/cpu/armv7/ls102xa/psci.S          | 105 +++++++++++++++++++++++++++--
 arch/arm/include/asm/arch-ls102xa/config.h |   1 +
 arch/arm/include/asm/psci.h                |   5 ++
 board/freescale/ls1021aqds/Makefile        |   1 +
 board/freescale/ls1021aqds/psci.S          |  36 ++++++++++
 board/freescale/ls1021atwr/Makefile        |   1 +
 board/freescale/ls1021atwr/psci.S          |  28 ++++++++
 include/configs/ls1021aqds.h               |   3 +
 include/configs/ls1021atwr.h               |   1 +
 9 files changed, 177 insertions(+), 4 deletions(-)
 create mode 100644 board/freescale/ls1021aqds/psci.S
 create mode 100644 board/freescale/ls1021atwr/psci.S
arch/arm/cpu/armv7/ls102xa/psci.S +101 -4
index 0188ade..084f914 100644
@@ -12,19 +12,72 @@
 #include <asm/arch-armv7/generictimer.h>
 #include <asm/psci.h>
 
+#define RCPM_TWAITSR		0x04C
+
 #define SCFG_CORE0_SFT_RST      0x130
 #define SCFG_CORESRENCR         0x204
 
-#define DCFG_CCSR_BRR           0x0E4
-#define DCFG_CCSR_SCRATCHRW1    0x200
+#define DCFG_CCSR_RSTCR			0x0B0
+#define DCFG_CCSR_RSTCR_RESET_REQ	0x2
+#define DCFG_CCSR_BRR			0x0E4
+#define DCFG_CCSR_SCRATCHRW1		0x200
+
+#define PSCI_FN_PSCI_VERSION_FEATURE_MASK	0x0
+#define PSCI_FN_CPU_SUSPEND_FEATURE_MASK	0x0
+#define PSCI_FN_CPU_OFF_FEATURE_MASK		0x0
+#define PSCI_FN_CPU_ON_FEATURE_MASK		0x0
+#define PSCI_FN_AFFINITY_INFO_FEATURE_MASK	0x0
+#define PSCI_FN_SYSTEM_OFF_FEATURE_MASK		0x0
+#define PSCI_FN_SYSTEM_RESET_FEATURE_MASK	0x0
 
 	.pushsection ._secure.text, "ax"
 
 	.arch_extension sec
 
+	.align	5
+
 #define	ONE_MS		(GENERIC_TIMER_CLK / 1000)
 #define	RESET_WAIT	(30 * ONE_MS)
 
+.globl	psci_version
+psci_version:
+	movw	r0, #0
+	movt	r0, #1
+
+	bx	lr
+
+_ls102x_psci_supported_table:
+	.word	ARM_PSCI_0_2_FN_PSCI_VERSION
+	.word	PSCI_FN_PSCI_VERSION_FEATURE_MASK
+	.word	ARM_PSCI_0_2_FN_CPU_SUSPEND
+	.word	PSCI_FN_CPU_SUSPEND_FEATURE_MASK
+	.word	ARM_PSCI_0_2_FN_CPU_OFF
+	.word	PSCI_FN_CPU_OFF_FEATURE_MASK
+	.word	ARM_PSCI_0_2_FN_CPU_ON
+	.word	PSCI_FN_CPU_ON_FEATURE_MASK
+	.word	ARM_PSCI_0_2_FN_AFFINITY_INFO
+	.word	PSCI_FN_AFFINITY_INFO_FEATURE_MASK
+	.word	ARM_PSCI_0_2_FN_SYSTEM_OFF
+	.word	PSCI_FN_SYSTEM_OFF_FEATURE_MASK
+	.word	ARM_PSCI_0_2_FN_SYSTEM_RESET
+	.word	PSCI_FN_SYSTEM_RESET_FEATURE_MASK
+	.word	0
+	.word	ARM_PSCI_RET_NI
+
+.globl	psci_features
+psci_features:
+	adr	r2, _ls102x_psci_supported_table
+1:	ldr	r3, [r2]
+	cmp	r3, #0
+	beq	out_psci_features
+	cmp	r1, r3
+	addne	r2, r2, #8
+	bne	1b
+
+out_psci_features:
+	ldr	r0, [r2, #4]
+	bx	lr
+
 .globl	psci_check_target_cpu_id
 psci_check_target_cpu_id:
 	@ Get the real CPU number
@@ -169,6 +222,52 @@ psci_cpu_off:
 1:	wfi
 	b	1b
 
+.globl	psci_affinity_info
+psci_affinity_info:
+	push	{lr}
+
+	mov	r0, #ARM_PSCI_RET_INVAL
+
+	@ Verify Affinity level
+	cmp	r2, #0
+	bne	out_affinity_info
+
+	bl	psci_check_target_cpu_id
+	cmp	r0, #ARM_PSCI_RET_INVAL
+	beq	out_affinity_info
+	mov	r1, r0
+
+	@ Get RCPM base address
+	movw	r4, #(CONFIG_SYS_FSL_RCPM_ADDR & 0xffff)
+	movt	r4, #(CONFIG_SYS_FSL_RCPM_ADDR >> 16)
+
+	mov	r0, #PSCI_AFFINITY_LEVEL_ON
+
+	@ Detect target CPU state
+	ldr	r2, [r4, #RCPM_TWAITSR]
+	rev	r2, r2
+	lsr	r2, r2, r1
+	ands	r2, r2, #1
+	beq	out_affinity_info
+
+	mov	r0, #PSCI_AFFINITY_LEVEL_OFF
+
+out_affinity_info:
+	pop	{pc}
+
+.globl	psci_system_reset
+psci_system_reset:
+	@ Get DCFG base address
+	movw	r1, #(CONFIG_SYS_FSL_GUTS_ADDR & 0xffff)
+	movt	r1, #(CONFIG_SYS_FSL_GUTS_ADDR >> 16)
+
+	mov	r2, #DCFG_CCSR_RSTCR_RESET_REQ
+	rev	r2, r2
+	str	r2, [r1, #DCFG_CCSR_RSTCR]
+
+1:	wfi
+	b	1b
+
 .globl	psci_arch_init
 psci_arch_init:
 	mov	r6, lr
@@ -179,6 +278,4 @@ psci_arch_init:
 
 	bx	r6
 
-	.globl psci_text_end
-psci_text_end:
 	.popsection
arch/arm/include/asm/arch-ls102xa/config.h +1 -0
index 04abec4..7a0e4bf 100644
@@ -32,6 +32,7 @@
 #define CONFIG_SYS_FSL_SERDES_ADDR		(CONFIG_SYS_IMMR + 0x00ea0000)
 #define CONFIG_SYS_FSL_GUTS_ADDR		(CONFIG_SYS_IMMR + 0x00ee0000)
 #define CONFIG_SYS_FSL_LS1_CLK_ADDR		(CONFIG_SYS_IMMR + 0x00ee1000)
+#define CONFIG_SYS_FSL_RCPM_ADDR		(CONFIG_SYS_IMMR + 0x00ee2000)
 #define CONFIG_SYS_NS16550_COM1			(CONFIG_SYS_IMMR + 0x011c0500)
 #define CONFIG_SYS_NS16550_COM2			(CONFIG_SYS_IMMR + 0x011d0500)
 #define CONFIG_SYS_DCU_ADDR			(CONFIG_SYS_IMMR + 0x01ce0000)
arch/arm/include/asm/psci.h +5 -0
index 89a1ba5..1c6d819 100644
@@ -73,6 +73,11 @@
 #define PSCI_CPU_STATUS_ON		1
 #define PSCI_CPU_STATUS_ON_PENDING	2
 
+/* PSCI affinity level state returned by AFFINITY_INFO */
+#define PSCI_AFFINITY_LEVEL_ON		0
+#define PSCI_AFFINITY_LEVEL_OFF		1
+#define PSCI_AFFINITY_LEVEL_ON_PENDING	2
+
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
board/freescale/ls1021aqds/Makefile +1 -0
index ab02344..f0390c1 100644
@@ -8,3 +8,4 @@ obj-y += ls1021aqds.o
 obj-y += ddr.o
 obj-y += eth.o
 obj-$(CONFIG_FSL_DCU_FB) += dcu.o
+obj-$(CONFIG_ARMV7_PSCI) += psci.o
board/freescale/ls1021aqds/psci.S new file +36 -0
new file mode 100644
index 0000000..6b75595
@@ -0,0 +1,36 @@
+/*
+ * Copyright 2016 NXP Semiconductor.
+ * Author: Wang Dongsheng <[email protected]>
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+#include <config.h>
+#include <linux/linkage.h>
+
+#include <asm/armv7.h>
+#include <asm/psci.h>
+
+	.pushsection ._secure.text, "ax"
+
+	.arch_extension sec
+
+	.align	5
+
+.globl	psci_system_off
+psci_system_off:
+	@ Get QIXIS base address
+	movw	r1, #(QIXIS_BASE & 0xffff)
+	movt	r1, #(QIXIS_BASE >> 16)
+
+	ldrb	r2, [r1, #QIXIS_PWR_CTL]
+	orr	r2, r2, #QIXIS_PWR_CTL_POWEROFF
+	strb	r2, [r1, #QIXIS_PWR_CTL]
+
+1:	wfi
+	b	1b
+
+.globl	psci_text_end
+psci_text_end:
+	nop
+	.popsection
board/freescale/ls1021atwr/Makefile +1 -0
index 01296c0..5238b15 100644
@@ -6,3 +6,4 @@
 
 obj-y += ls1021atwr.o
 obj-$(CONFIG_FSL_DCU_FB) += dcu.o
+obj-$(CONFIG_ARMV7_PSCI) += psci.o
board/freescale/ls1021atwr/psci.S new file +28 -0
new file mode 100644
index 0000000..adf5b85
@@ -0,0 +1,28 @@
+/*
+ * Copyright 2016 NXP Semiconductor.
+ * Author: Wang Dongsheng <[email protected]>
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+#include <config.h>
+#include <linux/linkage.h>
+
+#include <asm/armv7.h>
+#include <asm/psci.h>
+
+	.pushsection ._secure.text, "ax"
+
+	.arch_extension sec
+
+	.align	5
+
+.globl	psci_system_off
+psci_system_off:
+1:	wfi
+	b	1b
+
+.globl	psci_text_end
+psci_text_end:
+	nop
+	.popsection
include/configs/ls1021aqds.h +3 -0
index db684d2..2359fb0 100644
@@ -10,6 +10,7 @@
 #define CONFIG_LS102XA
 
 #define CONFIG_ARMV7_PSCI
+#define CONFIG_ARMV7_PSCI_1_0
 
 #define CONFIG_SYS_FSL_CLK
 
@@ -279,6 +280,8 @@ unsigned long get_board_ddr_clk(void);
 #define QIXIS_LBMAP_SHIFT		0
 #define QIXIS_LBMAP_DFLTBANK		0x00
 #define QIXIS_LBMAP_ALTBANK		0x04
+#define QIXIS_PWR_CTL			0x21
+#define QIXIS_PWR_CTL_POWEROFF		0x80
 #define QIXIS_RST_CTL_RESET		0x44
 #define QIXIS_RCFG_CTL_RECONFIG_IDLE	0x20
 #define QIXIS_RCFG_CTL_RECONFIG_START	0x21
include/configs/ls1021atwr.h +1 -0
index 0fb28ef..9d0c4fe 100644
@@ -10,6 +10,7 @@
 #define CONFIG_LS102XA
 
 #define CONFIG_ARMV7_PSCI
+#define CONFIG_ARMV7_PSCI_1_0
 
 #define CONFIG_SYS_FSL_CLK
 
Signature · 1 line
2.1.4
#20 lore raw

PATCH v5 7/8 ARMv7: PSCI: ls102xa: add more PSCI v1.0 functions implemention

Quoted 128 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

This patch implements PSCI functions for ls102xa SoC following PSCI v1.0, they are as the list:
psci_version,
psci_features,
psci_cpu_suspend,
psci_affinity_info,
psci_system_reset,
psci_system_off.

Tested on LS1021aQDS, LS1021aTWR.

---
 arch/arm/cpu/armv7/ls102xa/psci.S          | 105 +++++++++++++++++++++++++++--
 arch/arm/include/asm/arch-ls102xa/config.h |   1 +
 arch/arm/include/asm/psci.h                |   5 ++
 board/freescale/ls1021aqds/Makefile        |   1 +
 board/freescale/ls1021aqds/psci.S          |  36 ++++++++++
 board/freescale/ls1021atwr/Makefile        |   1 +
 board/freescale/ls1021atwr/psci.S          |  28 ++++++++
 include/configs/ls1021aqds.h               |   3 +
 include/configs/ls1021atwr.h               |   1 +
 9 files changed, 177 insertions(+), 4 deletions(-)
 create mode 100644 board/freescale/ls1021aqds/psci.S
 create mode 100644 board/freescale/ls1021atwr/psci.S
arch/arm/cpu/armv7/ls102xa/psci.S +71 -2
index 0188ade..084f914 100644
@@ -12,19 +12,72 @@
 #include <asm/arch-armv7/generictimer.h>
 #include <asm/psci.h>
 
+#define RCPM_TWAITSR           0x04C
+
 #define SCFG_CORE0_SFT_RST      0x130
 #define SCFG_CORESRENCR         0x204
 
-#define DCFG_CCSR_BRR           0x0E4
-#define DCFG_CCSR_SCRATCHRW1    0x200
+#define DCFG_CCSR_RSTCR                        0x0B0
+#define DCFG_CCSR_RSTCR_RESET_REQ      0x2
+#define DCFG_CCSR_BRR                  0x0E4
+#define DCFG_CCSR_SCRATCHRW1           0x200
+
+#define PSCI_FN_PSCI_VERSION_FEATURE_MASK      0x0
+#define PSCI_FN_CPU_SUSPEND_FEATURE_MASK       0x0
+#define PSCI_FN_CPU_OFF_FEATURE_MASK           0x0
+#define PSCI_FN_CPU_ON_FEATURE_MASK            0x0
+#define PSCI_FN_AFFINITY_INFO_FEATURE_MASK     0x0
+#define PSCI_FN_SYSTEM_OFF_FEATURE_MASK                0x0
+#define PSCI_FN_SYSTEM_RESET_FEATURE_MASK      0x0
 
        .pushsection ._secure.text, "ax"
 
        .arch_extension sec
 
+       .align  5
+
 #define        ONE_MS          (GENERIC_TIMER_CLK / 1000)
 #define        RESET_WAIT      (30 * ONE_MS)
 
+.globl psci_version
+psci_version:
+       movw    r0, #0
+       movt    r0, #1
+
+       bx      lr
+
+_ls102x_psci_supported_table:
+       .word   ARM_PSCI_0_2_FN_PSCI_VERSION
+       .word   PSCI_FN_PSCI_VERSION_FEATURE_MASK
+       .word   ARM_PSCI_0_2_FN_CPU_SUSPEND
+       .word   PSCI_FN_CPU_SUSPEND_FEATURE_MASK
+       .word   ARM_PSCI_0_2_FN_CPU_OFF
+       .word   PSCI_FN_CPU_OFF_FEATURE_MASK
+       .word   ARM_PSCI_0_2_FN_CPU_ON
+       .word   PSCI_FN_CPU_ON_FEATURE_MASK
+       .word   ARM_PSCI_0_2_FN_AFFINITY_INFO
+       .word   PSCI_FN_AFFINITY_INFO_FEATURE_MASK
+       .word   ARM_PSCI_0_2_FN_SYSTEM_OFF
+       .word   PSCI_FN_SYSTEM_OFF_FEATURE_MASK
+       .word   ARM_PSCI_0_2_FN_SYSTEM_RESET
+       .word   PSCI_FN_SYSTEM_RESET_FEATURE_MASK
+       .word   0
+       .word   ARM_PSCI_RET_NI
+
+.globl psci_features
+psci_features:
+       adr     r2, _ls102x_psci_supported_table
+1:     ldr     r3, [r2]
+       cmp     r3, #0
+       beq     out_psci_features
+       cmp     r1, r3
+       addne   r2, r2, #8
+       bne     1b
+
+out_psci_features:
+       ldr     r0, [r2, #4]
+       bx      lr
+
 .globl psci_check_target_cpu_id
 psci_check_target_cpu_id:
        @ Get the real CPU number
@@ -169,6 +222,52 @@ psci_cpu_off:
 1:     wfi
        b       1b
 
+.globl psci_affinity_info
+psci_affinity_info:
+       push    {lr}
+
+       mov     r0, #ARM_PSCI_RET_INVAL
+
+       @ Verify Affinity level
+       cmp     r2, #0
+       bne     out_affinity_info
+
+       bl      psci_check_target_cpu_id
+       cmp     r0, #ARM_PSCI_RET_INVAL
+       beq     out_affinity_info
+       mov     r1, r0
+
+       @ Get RCPM base address

arch/arm/cpu/armv7/ls102xa/psci.S psci_cpu_off: · line 250

Quoted 10 lines
arch/arm/cpu/armv7/ls102xa/psci.S +10 -0
+       movw    r4, #(CONFIG_SYS_FSL_RCPM_ADDR & 0xffff)
+       movt    r4, #(CONFIG_SYS_FSL_RCPM_ADDR >> 16)
+
+       mov     r0, #PSCI_AFFINITY_LEVEL_ON
+
+       @ Detect target CPU state
+       ldr     r2, [r4, #RCPM_TWAITSR]
+       rev     r2, r2
+       lsr     r2, r2, r1
+       ands    r2, r2, #1

tst should suffice.

Quoted 47 lines
arch/arm/cpu/armv7/ls102xa/psci.S +20 -2
+       beq     out_affinity_info
+
+       mov     r0, #PSCI_AFFINITY_LEVEL_OFF
+
+out_affinity_info:
+       pop     {pc}
+
+.globl psci_system_reset
+psci_system_reset:
+       @ Get DCFG base address
+       movw    r1, #(CONFIG_SYS_FSL_GUTS_ADDR & 0xffff)
+       movt    r1, #(CONFIG_SYS_FSL_GUTS_ADDR >> 16)
+
+       mov     r2, #DCFG_CCSR_RSTCR_RESET_REQ
+       rev     r2, r2
+       str     r2, [r1, #DCFG_CCSR_RSTCR]
+
+1:     wfi
+       b       1b
+
 .globl psci_arch_init
 psci_arch_init:
        mov     r6, lr
@@ -179,6 +278,4 @@ psci_arch_init:
 
        bx      r6
 
-       .globl psci_text_end
-psci_text_end:
        .popsection
arch/arm/include/asm/arch-ls102xa/config.h +1 -0
index 04abec4..7a0e4bf 100644
@@ -32,6 +32,7 @@
 #define CONFIG_SYS_FSL_SERDES_ADDR             (CONFIG_SYS_IMMR + 0x00ea0000)
 #define CONFIG_SYS_FSL_GUTS_ADDR               (CONFIG_SYS_IMMR + 0x00ee0000)
 #define CONFIG_SYS_FSL_LS1_CLK_ADDR            (CONFIG_SYS_IMMR + 0x00ee1000)
+#define CONFIG_SYS_FSL_RCPM_ADDR               (CONFIG_SYS_IMMR + 0x00ee2000)
 #define CONFIG_SYS_NS16550_COM1                        (CONFIG_SYS_IMMR + 0x011c0500)
 #define CONFIG_SYS_NS16550_COM2                        (CONFIG_SYS_IMMR + 0x011d0500)
 #define CONFIG_SYS_DCU_ADDR                    (CONFIG_SYS_IMMR + 0x01ce0000)
arch/arm/include/asm/psci.h +0 -0
index 89a1ba5..1c6d819 100644

arch/arm/include/asm/psci.h line 80

Quoted 8 lines
arch/arm/include/asm/psci.h +5 -0
@@ -73,6 +73,11 @@
 #define PSCI_CPU_STATUS_ON             1
 #define PSCI_CPU_STATUS_ON_PENDING     2
 
+/* PSCI affinity level state returned by AFFINITY_INFO */
+#define PSCI_AFFINITY_LEVEL_ON         0
+#define PSCI_AFFINITY_LEVEL_OFF                1
+#define PSCI_AFFINITY_LEVEL_ON_PENDING 2
+

This should be part of the common PSCI 1.0 skeleton patch.

Quoted 82 lines
arch/arm/include/asm/psci.h +0 -0
 #ifndef __ASSEMBLY__
 int psci_update_dt(void *fdt);
 void psci_board_init(void);
board/freescale/ls1021aqds/Makefile +1 -0
index ab02344..f0390c1 100644
@@ -8,3 +8,4 @@ obj-y += ls1021aqds.o
 obj-y += ddr.o
 obj-y += eth.o
 obj-$(CONFIG_FSL_DCU_FB) += dcu.o
+obj-$(CONFIG_ARMV7_PSCI) += psci.o
board/freescale/ls1021aqds/psci.S new file +36 -0
new file mode 100644
index 0000000..6b75595
@@ -0,0 +1,36 @@
+/*
+ * Copyright 2016 NXP Semiconductor.
+ * Author: Wang Dongsheng <[email protected]>
+ *
+ * SPDX-License-Identifier:    GPL-2.0+
+ */
+
+#include <config.h>
+#include <linux/linkage.h>
+
+#include <asm/armv7.h>
+#include <asm/psci.h>
+
+       .pushsection ._secure.text, "ax"
+
+       .arch_extension sec
+
+       .align  5
+
+.globl psci_system_off
+psci_system_off:
+       @ Get QIXIS base address
+       movw    r1, #(QIXIS_BASE & 0xffff)
+       movt    r1, #(QIXIS_BASE >> 16)
+
+       ldrb    r2, [r1, #QIXIS_PWR_CTL]
+       orr     r2, r2, #QIXIS_PWR_CTL_POWEROFF
+       strb    r2, [r1, #QIXIS_PWR_CTL]
+
+1:     wfi
+       b       1b
+
+.globl psci_text_end
+psci_text_end:
+       nop
+       .popsection
board/freescale/ls1021atwr/Makefile +1 -0
index 01296c0..5238b15 100644
@@ -6,3 +6,4 @@
 
 obj-y += ls1021atwr.o
 obj-$(CONFIG_FSL_DCU_FB) += dcu.o
+obj-$(CONFIG_ARMV7_PSCI) += psci.o
board/freescale/ls1021atwr/psci.S new file +13 -0
new file mode 100644
index 0000000..adf5b85
@@ -0,0 +1,28 @@
+/*
+ * Copyright 2016 NXP Semiconductor.
+ * Author: Wang Dongsheng <[email protected]>
+ *
+ * SPDX-License-Identifier:    GPL-2.0+
+ */
+
+#include <config.h>
+#include <linux/linkage.h>
+
+#include <asm/armv7.h>
+#include <asm/psci.h>
+

board/freescale/ls1021atwr/psci.S line 23

Quoted 10 lines
board/freescale/ls1021atwr/psci.S new file +10 -0
+       .pushsection ._secure.text, "ax"
+
+       .arch_extension sec
+
+       .align  5
+
+.globl psci_system_off
+psci_system_off:
+1:     wfi
+       b       1b

This does nothing more than halt the machine.

The PSCI spec says "On a SYSTEM_OFF call the implementation completely removes power from highest power level."

I can understand you might have issues supporting this feature for external PMICs though. Allwinner/sunxi has the same problem.

Regards
ChenYu

Quoted 41 lines
board/freescale/ls1021atwr/psci.S new file +5 -0
+
+.globl psci_text_end
+psci_text_end:
+       nop
+       .popsection
include/configs/ls1021aqds.h +3 -0
index db684d2..2359fb0 100644
@@ -10,6 +10,7 @@
 #define CONFIG_LS102XA
 
 #define CONFIG_ARMV7_PSCI
+#define CONFIG_ARMV7_PSCI_1_0
 
 #define CONFIG_SYS_FSL_CLK
 
@@ -279,6 +280,8 @@ unsigned long get_board_ddr_clk(void);
 #define QIXIS_LBMAP_SHIFT              0
 #define QIXIS_LBMAP_DFLTBANK           0x00
 #define QIXIS_LBMAP_ALTBANK            0x04
+#define QIXIS_PWR_CTL                  0x21
+#define QIXIS_PWR_CTL_POWEROFF         0x80
 #define QIXIS_RST_CTL_RESET            0x44
 #define QIXIS_RCFG_CTL_RECONFIG_IDLE   0x20
 #define QIXIS_RCFG_CTL_RECONFIG_START  0x21
include/configs/ls1021atwr.h +1 -1
index 0fb28ef..9d0c4fe 100644
@@ -10,6 +10,7 @@
 #define CONFIG_LS102XA
 
 #define CONFIG_ARMV7_PSCI
+#define CONFIG_ARMV7_PSCI_1_0
 
 #define CONFIG_SYS_FSL_CLK
 
--

2.1.4

#21 lore raw

PATCH v5 8/8 ARMv7: PSCI: ls102xa: move secure text section into OCRAM

In reply to macro.wave.z at gmail.com

From: Hongbo Zhang <[email protected]>

LS1021 offers two secure OCRAM blocks for trustzone. This patch moves all the secure text sections into the OCRAM.

---
 arch/arm/include/asm/arch-ls102xa/config.h | 2 +-
 include/configs/ls1021atwr.h               | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/arm/include/asm/arch-ls102xa/config.h +1 -1
index 7a0e4bf..4729044 100644
@@ -10,7 +10,7 @@
 #define CONFIG_SYS_CACHELINE_SIZE		64
 
 #define OCRAM_BASE_ADDR				0x10000000
-#define OCRAM_SIZE				0x00020000
+#define OCRAM_SIZE				0x00010000
 #define OCRAM_BASE_S_ADDR			0x10010000
 #define OCRAM_S_SIZE				0x00010000
 
include/configs/ls1021atwr.h +2 -0
index 9d0c4fe..e6fbd77 100644
@@ -12,6 +12,8 @@
 #define CONFIG_ARMV7_PSCI
 #define CONFIG_ARMV7_PSCI_1_0
 
+#define CONFIG_ARMV7_SECURE_BASE	OCRAM_BASE_S_ADDR
+
 #define CONFIG_SYS_FSL_CLK
 
 #define CONFIG_DISPLAY_CPUINFO
Signature · 1 line
2.1.4
#22 lore raw

PATCH v5 8/8 ARMv7: PSCI: ls102xa: move secure text section into OCRAM

Quoted 39 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

LS1021 offers two secure OCRAM blocks for trustzone. This patch moves all the secure text sections into the OCRAM.

---
 arch/arm/include/asm/arch-ls102xa/config.h | 2 +-
 include/configs/ls1021atwr.h               | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/arm/include/asm/arch-ls102xa/config.h +1 -1
index 7a0e4bf..4729044 100644
@@ -10,7 +10,7 @@
 #define CONFIG_SYS_CACHELINE_SIZE              64
 
 #define OCRAM_BASE_ADDR                                0x10000000
-#define OCRAM_SIZE                             0x00020000
+#define OCRAM_SIZE                             0x00010000
 #define OCRAM_BASE_S_ADDR                      0x10010000
 #define OCRAM_S_SIZE                           0x00010000
 
include/configs/ls1021atwr.h +2 -1
index 9d0c4fe..e6fbd77 100644
@@ -12,6 +12,8 @@
 #define CONFIG_ARMV7_PSCI
 #define CONFIG_ARMV7_PSCI_1_0
 
+#define CONFIG_ARMV7_SECURE_BASE       OCRAM_BASE_S_ADDR
+
 #define CONFIG_SYS_FSL_CLK
 
 #define CONFIG_DISPLAY_CPUINFO
--
Quoted 2 lines

2.1.4

The patch itself looks good. Though I wonder if there's some register you need to configure to split access to the 2 SRAM blocks? If they are both secure-only, then everything's fine.

#23 lore raw

PATCH v5 8/8 ARMv7: PSCI: ls102xa: move secure text section into OCRAM

Quoted 38 lines · Chen-Yu Tsai
Quoted 37 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

LS1021 offers two secure OCRAM blocks for trustzone. This patch moves all the secure text sections into the OCRAM.

---
 arch/arm/include/asm/arch-ls102xa/config.h | 2 +-
 include/configs/ls1021atwr.h               | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/arm/include/asm/arch-ls102xa/config.h +1 -1
index 7a0e4bf..4729044 100644
@@ -10,7 +10,7 @@
 #define CONFIG_SYS_CACHELINE_SIZE              64
 
 #define OCRAM_BASE_ADDR                                0x10000000
-#define OCRAM_SIZE                             0x00020000
+#define OCRAM_SIZE                             0x00010000
 #define OCRAM_BASE_S_ADDR                      0x10010000
 #define OCRAM_S_SIZE                           0x00010000
 
include/configs/ls1021atwr.h +2 -0
index 9d0c4fe..e6fbd77 100644
@@ -12,6 +12,8 @@
 #define CONFIG_ARMV7_PSCI
 #define CONFIG_ARMV7_PSCI_1_0
 
+#define CONFIG_ARMV7_SECURE_BASE       OCRAM_BASE_S_ADDR
+
 #define CONFIG_SYS_FSL_CLK
 
Quoted 9 lines
Quoted 4 lines
include/configs/ls1021atwr.h +0 -1
 #define CONFIG_DISPLAY_CPUINFO
--

2.1.4

The patch itself looks good. Though I wonder if there's some register you need to configure to split access to the 2 SRAM blocks? If they are both secure-only, then everything's fine.

No other registers needed to be configured, this works fine.

Quoted 1 line

Thanks for the tag.

#24 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

In reply to macro.wave.z at gmail.com

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Regards
ChenYu

Quoted 68 lines

- Move the patch of factoring out psci_cpu_on_common to earlier place 2/8, so that the following patches can only update the psci_cpu_on_common instead of touching each platform's psci_cpu_on, this reduces patches size obviously and make it easier for review.

v4 changes:
- since there is already PSCI v0.2 function IDs definition merged in 5a07abb, I give up my previous patches 1/11 and 2/11, and move previous 7/11 "ARMv7:
PSCI: add PSCI v1.0 functions skeleton" as current first one 1/9
- accept Andre's comment to add the missed "arm,psci-0,2" into compatible string

v3 changes:
- patch 3/11, re-init the stack pointer to address like start of page instead of page end, because for ARM push operation, the stack pointer is encreased before storing data.
- patch 10/11, delete the previous un-implemented cpu_suspend function for ls1 platform, because there is default blank functions for all those are not implemented in specific platform.

v2 changes:
- re-organize psci_cpu_on_common, this code should be called by each platform's psci_cpu_on, should not be a function calling each psci_cpu_on, all related functions are updated due to this change
- update some registers usage, e.g. if r10 is used without push/pop, u-boot cannot launch rt-kernel
- update some comments to be clearer, re-organize all patches for easier review
- add patch to check already_on or on_pending for LS102XA

This patch set contains two parts:
ARMv7 PSCI common framework: fix some issues and add v1.0 support NXP (was Freescale) LS102XA: codes enhancement and add v1.0 implementation And this patch set was initially created by Dongsheng Wang.

Hongbo Zhang (8):
ARMv7: PSCI: add PSCI v1.0 functions skeleton
ARMv7: PSCI: factor out reusable psci_cpu_on_common
ARMv7: PSCI: update the place of saving target PC
ARMv7: PSCI: add codes to save context ID for CPU_ON
ARMv7: PSCI: ls102xa: check target CPU ID before further operations
ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON
ARMv7: PSCI: ls102xa: add more PSCI v1.0 functions implemention
ARMv7: PSCI: ls102xa: move secure text section into OCRAM

 arch/arm/cpu/armv7/ls102xa/psci.S          | 169 +++++++++++++++++++++++++++--
 arch/arm/cpu/armv7/mx7/psci.S              |   5 +-
 arch/arm/cpu/armv7/nonsec_virt.S           |   7 ++
 arch/arm/cpu/armv7/psci.S                  |  99 ++++++++++++++++-
 arch/arm/cpu/armv7/sunxi/psci_sun6i.S      |   5 +-
 arch/arm/cpu/armv7/sunxi/psci_sun7i.S      |   5 +-
 arch/arm/cpu/armv7/virt-dt.c               |  10 +-
 arch/arm/include/asm/arch-ls102xa/config.h |   3 +-
 arch/arm/include/asm/psci.h                |  31 ++++++
 arch/arm/mach-tegra/psci.S                 |   5 +-
 board/freescale/ls1021aqds/Makefile        |   1 +
 board/freescale/ls1021aqds/psci.S          |  36 ++++++
 board/freescale/ls1021atwr/Makefile        |   1 +
 board/freescale/ls1021atwr/psci.S          |  28 +++++
 include/configs/ls1021aqds.h               |   3 +
 include/configs/ls1021atwr.h               |   3 +
 16 files changed, 383 insertions(+), 28 deletions(-)
 create mode 100644 board/freescale/ls1021aqds/psci.S
 create mode 100644 board/freescale/ls1021atwr/psci.S

--
2.1.4

#25 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

Quoted 13 lines · On Wed, 15 Jun 2016 15:16:08 +0800 Chen-Yu Tsai <[email protected]> wrote:

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

And have silent corruption from the secure side? I hope not.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

I haven't looked at these patches, but I definitely welcome that kind of development. Please also consider providing some form of per-cpu infrastructure (which would in turn solve the issue that Hongbo Zhang wants to sidestep).

Thanks,

	M.
Signature · 1 line
Jazz is not dead. It just smells funny.
#26 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

In reply to Chen-Yu Tsai
Quoted 14 lines · Chen-Yu Tsai

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Yes, secure data section is welcome.
I am implementing our suspend-to-ram function in PSCI in C language, there are some global variables which should be placed into such secure data section.

Quoted 71 lines

Regards
ChenYu

Quoted 68 lines

- Move the patch of factoring out psci_cpu_on_common to earlier place 2/8, so that the following patches can only update the psci_cpu_on_common instead of touching each platform's psci_cpu_on, this reduces patches size obviously and make it easier for review.

v4 changes:
- since there is already PSCI v0.2 function IDs definition merged in 5a07abb, I give up my previous patches 1/11 and 2/11, and move previous 7/11 "ARMv7:
PSCI: add PSCI v1.0 functions skeleton" as current first one 1/9
- accept Andre's comment to add the missed "arm,psci-0,2" into compatible string

v3 changes:
- patch 3/11, re-init the stack pointer to address like start of page instead of page end, because for ARM push operation, the stack pointer is encreased before storing data.
- patch 10/11, delete the previous un-implemented cpu_suspend function for ls1 platform, because there is default blank functions for all those are not implemented in specific platform.

v2 changes:
- re-organize psci_cpu_on_common, this code should be called by each platform's psci_cpu_on, should not be a function calling each psci_cpu_on, all related functions are updated due to this change
- update some registers usage, e.g. if r10 is used without push/pop, u-boot cannot launch rt-kernel
- update some comments to be clearer, re-organize all patches for easier review
- add patch to check already_on or on_pending for LS102XA

This patch set contains two parts:
ARMv7 PSCI common framework: fix some issues and add v1.0 support NXP (was Freescale) LS102XA: codes enhancement and add v1.0 implementation And this patch set was initially created by Dongsheng Wang.

Hongbo Zhang (8):
ARMv7: PSCI: add PSCI v1.0 functions skeleton
ARMv7: PSCI: factor out reusable psci_cpu_on_common
ARMv7: PSCI: update the place of saving target PC
ARMv7: PSCI: add codes to save context ID for CPU_ON
ARMv7: PSCI: ls102xa: check target CPU ID before further operations
ARMv7: PSCI: ls102xa: check ALREADY_ON or ON_PENDING for CPU_ON
ARMv7: PSCI: ls102xa: add more PSCI v1.0 functions implemention
ARMv7: PSCI: ls102xa: move secure text section into OCRAM

 arch/arm/cpu/armv7/ls102xa/psci.S          | 169 +++++++++++++++++++++++++++--
 arch/arm/cpu/armv7/mx7/psci.S              |   5 +-
 arch/arm/cpu/armv7/nonsec_virt.S           |   7 ++
 arch/arm/cpu/armv7/psci.S                  |  99 ++++++++++++++++-
 arch/arm/cpu/armv7/sunxi/psci_sun6i.S      |   5 +-
 arch/arm/cpu/armv7/sunxi/psci_sun7i.S      |   5 +-
 arch/arm/cpu/armv7/virt-dt.c               |  10 +-
 arch/arm/include/asm/arch-ls102xa/config.h |   3 +-
 arch/arm/include/asm/psci.h                |  31 ++++++
 arch/arm/mach-tegra/psci.S                 |   5 +-
 board/freescale/ls1021aqds/Makefile        |   1 +
 board/freescale/ls1021aqds/psci.S          |  36 ++++++
 board/freescale/ls1021atwr/Makefile        |   1 +
 board/freescale/ls1021atwr/psci.S          |  28 +++++
 include/configs/ls1021aqds.h               |   3 +
 include/configs/ls1021atwr.h               |   3 +
 16 files changed, 383 insertions(+), 28 deletions(-)
 create mode 100644 board/freescale/ls1021aqds/psci.S
 create mode 100644 board/freescale/ls1021atwr/psci.S

--
2.1.4

#27 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

In reply to Chen-Yu Tsai
Quoted 14 lines · Chen-Yu Tsai

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

#28 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

Quoted 26 lines · york sun
Quoted 14 lines · Chen-Yu Tsai

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Quoted 8 lines

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

Thank Chenyu for valuable review comments, thank you York. Will send another version soon.

#29 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

I said I would send a new iteration of this series, but there are code dependencies/conflicts with Chenyu's work, so I'd like to wait for a few more days to see his v2 sent out or even been merged. During this stage, any review comments are still appreciated, I am always looking at this.
Thanks.

Quoted 37 lines · Hongbo Zhang
Quoted 26 lines · york sun
Quoted 14 lines · Chen-Yu Tsai

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Quoted 8 lines

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

Thank Chenyu for valuable review comments, thank you York. Will send another version soon.

#30 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

Hi,

Quoted 5 lines · Hongbo Zhang

I said I would send a new iteration of this series, but there are code dependencies/conflicts with Chenyu's work, so I'd like to wait for a few more days to see his v2 sent out or even been merged. During this stage, any review comments are still appreciated, I am always looking at this.

In that case you can probably best base your work on the u-boot-sunxi next branch, that currently still has v1 of Chen-Yu's patches, but I do not think things will change that much:

http://git.denx.de/?p=u-boot/u-boot-sunxi.git;a=shortlog;h=refs/heads/next

Regards,

Hans

Quoted 40 lines

Thanks.

Quoted 37 lines · Hongbo Zhang
Quoted 26 lines · york sun
Quoted 14 lines · Chen-Yu Tsai

Hi,

Quoted 10 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end when there is more CPUs in system. Because I just want to keep this series as simple as it could be: adding basic PSCI v1.0 support and adding more PSCI v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective reviews, even I think there is potential bug of PSCI stack, let's fix it in sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Quoted 8 lines

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

Thank Chenyu for valuable review comments, thank you York. Will send another version soon.

#31 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

Quoted 11 lines · Hans de Goede

Hi,

Quoted 6 lines · Hongbo Zhang

I said I would send a new iteration of this series, but there are code dependencies/conflicts with Chenyu's work, so I'd like to wait for a few more days to see his v2 sent out or even been merged. During this stage, any review comments are still appreciated, I am always looking at this.

Quoted 6 lines

In that case you can probably best base your work on the u-boot-sunxi next branch, that currently still has v1 of Chen-Yu's patches, but I do not think things will change that much:

http://git.denx.de/?p=u-boot/u-boot-sunxi.git;a=shortlog;h=refs/heads/next

OK, I see

Quoted 57 lines

Regards,

Hans

Quoted 51 lines

Thanks.

Quoted 47 lines · Hongbo Zhang
Quoted 35 lines · york sun
Quoted 22 lines · Chen-Yu Tsai

Hi,

Quoted 16 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end
when there is more CPUs in system. Because I just want to keep this series as
simple as it could be: adding basic PSCI v1.0 support and adding more PSCI
v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective
reviews, even I think there is potential bug of PSCI stack, let's fix it in
sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Quoted 8 lines

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

Thank Chenyu for valuable review comments, thank you York. Will send another version soon.

#32 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

Quoted 12 lines · Hongbo Zhang
Quoted 11 lines · Hans de Goede

Hi,

Quoted 6 lines · Hongbo Zhang

I said I would send a new iteration of this series, but there are code dependencies/conflicts with Chenyu's work, so I'd like to wait for a few more days to see his v2 sent out or even been merged. During this stage, any review comments are still appreciated, I am always looking at this.

Quoted 66 lines
Quoted 6 lines

In that case you can probably best base your work on the u-boot-sunxi next branch, that currently still has v1 of Chen-Yu's patches, but I do not think things will change that much:

http://git.denx.de/?p=u-boot/u-boot-sunxi.git;a=shortlog;h=refs/heads/next

OK, I see

Quoted 57 lines

Regards,

Hans

Quoted 51 lines

Thanks.

Quoted 47 lines · Hongbo Zhang
Quoted 35 lines · york sun
Quoted 22 lines · Chen-Yu Tsai

Hi,

Quoted 16 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end
when there is more CPUs in system. Because I just want to keep this series as
simple as it could be: adding basic PSCI v1.0 support and adding more PSCI
v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective
reviews, even I think there is potential bug of PSCI stack, let's fix it in
sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Quoted 8 lines

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

Thank Chenyu for valuable review comments, thank you York. Will send another version soon.

Hongbo,

Where are we on this thread? Do you have an update?

York

#33 lore raw

v5 cover ARMv7: PSCI: add PSCI v1.0 support

Hi York,
I saw sunxi's patches were merged, so I am preparing my patches, will send it out soon.

Quoted 85 lines · york sun
Quoted 12 lines · Hongbo Zhang
Quoted 11 lines · Hans de Goede

Hi,

Quoted 6 lines · Hongbo Zhang

I said I would send a new iteration of this series, but there are code dependencies/conflicts with Chenyu's work, so I'd like to wait for a few more days to see his v2 sent out or even been merged. During this stage, any review comments are still appreciated, I am always looking at this.

Quoted 66 lines
Quoted 6 lines

In that case you can probably best base your work on the u-boot-sunxi next branch, that currently still has v1 of Chen-Yu's patches, but I do not think things will change that much:

http://git.denx.de/?p=u-boot/u-boot-sunxi.git;a=shortlog;h=refs/heads/next

OK, I see

Quoted 57 lines

Regards,

Hans

Quoted 51 lines

Thanks.

Quoted 47 lines · Hongbo Zhang
Quoted 35 lines · york sun
Quoted 22 lines · Chen-Yu Tsai

Hi,

Quoted 16 lines · On Tue, Jun 14, 2016 at 3:01 PM, <[email protected]> wrote:

From: Hongbo Zhang <[email protected]>

v5 changes:
- Give up fixing the potential bug of PSCI stack overlap with secure text end
when there is more CPUs in system. Because I just want to keep this series as
simple as it could be: adding basic PSCI v1.0 support and adding more PSCI
v1.0 implements of our platform.
While too compplicated patches in assembly language won't call for effective
reviews, even I think there is potential bug of PSCI stack, let's fix it in
sepetated patch later.

Quoted 9 lines

I've done some patches fixing the stack allocation issue:

https://github.com/wens/u-boot-sunxi/commits/c-psci-part2

These patches are ready, but I want to add a secure data section still. The data section will be for variables such as target PC, core/cluster power status, context ID, etc.. IMHO this is better than putting stuff at the top or bottom of the stack, and can also be referenced directly.

Quoted 8 lines

Chenyu,

I see you have been working on PSCI for ARMv7. Can you review Hongbo's patch set and give your comment or ack?

York

Thank Chenyu for valuable review comments, thank you York. Will send another version soon.

Hongbo,

Where are we on this thread? Do you have an update?

York