Skip to contentSkip to conversation

Detect UDIV/SDIV support from ISAR0

PATCH In review Stephen Boyd 4 participants 16 messages Mar 18, 2013 – May 7, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v2 cover Detect UDIV/SDIV support from ISAR0

While attempting to upstream a patch to add the IDIV hwcap for Krait processors, Will suggested we move the code to read the ISAR0 register. This patchset does that and also works around the early Krait CPU designs that don't follow the latest ARM ARM for the ISAR0 register.

Stepan Moskovchenko (1):
ARM: Work around faulty ISAR0 register in some Krait CPUs

Stephen Boyd (2):
ARM: Clear IDIVT hwcap if CONFIG_ARM_THUMB=n
ARM: Detect support for SDIV/UDIV from ISAR0 register

 arch/arm/kernel/setup.c | 22 +++++++++++++++++++++-
 arch/arm/mm/proc-v7.S   | 19 +++++++++++++++++--
 2 files changed, 38 insertions(+), 3 deletions(-)
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation
#1 lore raw

PATCH v2 1/3 ARM: Clear IDIVT hwcap if CONFIG_ARM_THUMB=n

Don't advertise support for the SDIV/UDIV thumb instructions if the kernel is not compiled with support for thumb userspace. This is in line with how we remove the THUMB hwcap in these configurations.

---
 arch/arm/kernel/setup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/setup.c +1 -1
index 3f6cbb2..e2c8bbf 100644
@@ -484,7 +484,7 @@ static void __init setup_processor(void)
 		 list->elf_name, ENDIANNESS);
 	elf_hwcap = list->elf_hwcap;
 #ifndef CONFIG_ARM_THUMB
-	elf_hwcap &= ~HWCAP_THUMB;
+	elf_hwcap &= ~(HWCAP_THUMB | HWCAP_IDIVT);
 #endif
 
 	feat_v6_fixup();
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation
#2 lore raw

PATCH v2 2/3 ARM: Detect support for SDIV/UDIV from ISAR0 register

In reply to Stephen Boyd

The ISAR0 register indicates support for the SDIV and UDIV instructions in both the Thumb and ARM instruction set. Read the register to detect the supported instructions and update the elf_hwcap mask as appropriate. This is better than adding more and more cpuid checks in proc-v7.S for each new cpu variant that supports these instructions.

---
 arch/arm/kernel/setup.c | 20 ++++++++++++++++++++
 arch/arm/mm/proc-v7.S   |  4 ++--
 2 files changed, 22 insertions(+), 2 deletions(-)
arch/arm/kernel/setup.c +20 -0
index e2c8bbf..f3ac13f 100644
@@ -353,6 +353,23 @@ void __init early_print(const char *str, ...)
 	printk("%s", buf);
 }
 
+static void __init cpuid_init_hwcaps(void)
+{
+	unsigned int divide_instrs;
+
+	if (cpu_architecture() < CPU_ARCH_ARMv7)
+		return;
+
+	divide_instrs = (read_cpuid_ext(CPUID_EXT_ISAR0) & 0x0f000000) >> 24;
+
+	switch (divide_instrs) {
+	case 2:
+		elf_hwcap |= HWCAP_IDIVA;
+	case 1:
+		elf_hwcap |= HWCAP_IDIVT;
+	}
+}
+
 static void __init feat_v6_fixup(void)
 {
 	int id = read_cpuid_id();
@@ -483,6 +500,9 @@ static void __init setup_processor(void)
 	snprintf(elf_platform, ELF_PLATFORM_SIZE, "%s%c",
 		 list->elf_name, ENDIANNESS);
 	elf_hwcap = list->elf_hwcap;
+
+	cpuid_init_hwcaps();
+
 #ifndef CONFIG_ARM_THUMB
 	elf_hwcap &= ~(HWCAP_THUMB | HWCAP_IDIVT);
 #endif
arch/arm/mm/proc-v7.S +2 -2
index 3a3c015..bcd3d48 100644
@@ -420,7 +420,7 @@ __v7_pj4b_proc_info:
 __v7_ca7mp_proc_info:
 	.long	0x410fc070
 	.long	0xff0ffff0
-	__v7_proc __v7_ca7mp_setup, hwcaps = HWCAP_IDIV
+	__v7_proc __v7_ca7mp_setup
 	.size	__v7_ca7mp_proc_info, . - __v7_ca7mp_proc_info
 
 	/*
@@ -430,7 +430,7 @@ __v7_ca7mp_proc_info:
 __v7_ca15mp_proc_info:
 	.long	0x410fc0f0
 	.long	0xff0ffff0
-	__v7_proc __v7_ca15mp_setup, hwcaps = HWCAP_IDIV
+	__v7_proc __v7_ca15mp_setup
 	.size	__v7_ca15mp_proc_info, . - __v7_ca15mp_proc_info
 
 	/*
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation
#3 lore raw

Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0 register

Hello Stephen,

Quoted 6 lines · Stephen Boyd

The ISAR0 register indicates support for the SDIV and UDIV instructions in both the Thumb and ARM instruction set. Read the register to detect the supported instructions and update the elf_hwcap mask as appropriate. This is better than adding more and more cpuid checks in proc-v7.S for each new cpu variant that supports these instructions.

you pointed out yesterday that this could work on v7-m, too. As I based my patches on 3.9-rc1 this patch (8164f7af88) wasn't included. When updating this results in a warning, because I have

6ebd4d0 (ARM: stub out read_cpuid and read_cpuid_ext for CPU_CP15=n)

from rmk's devel-stable branch.

Quoted 14 lines
---
 arch/arm/kernel/setup.c | 20 ++++++++++++++++++++
 arch/arm/mm/proc-v7.S   |  4 ++--
 2 files changed, 22 insertions(+), 2 deletions(-)
arch/arm/kernel/setup.c +0 -0
index e2c8bbf..f3ac13f 100644
@@ -353,6 +353,23 @@ void __init early_print(const char *str, ...)
 	printk("%s", buf);

arch/arm/kernel/setup.c void __init early_print(const char *str, ...) · line 363

Quoted 10 lines
arch/arm/kernel/setup.c +8 -0
 }
 
+static void __init cpuid_init_hwcaps(void)
+{
+	unsigned int divide_instrs;
+
+	if (cpu_architecture() < CPU_ARCH_ARMv7)
+		return;
+
+	divide_instrs = (read_cpuid_ext(CPUID_EXT_ISAR0) & 0x0f000000) >> 24;

The problem is that read_cpuid_ext is called which doesn't map to something useful for v7-m. So maybe add a check:

	if (!IS_ENABLED(CONFIG_CPU_CP15))
		return;

?

Best regards
Uwe

Signature · 2 lines
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
#4 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

Hello,

Quoted 42 lines · Uwe Kleine-K?nig
Quoted 6 lines · Stephen Boyd

The ISAR0 register indicates support for the SDIV and UDIV instructions in both the Thumb and ARM instruction set. Read the register to detect the supported instructions and update the elf_hwcap mask as appropriate. This is better than adding more and more cpuid checks in proc-v7.S for each new cpu variant that supports these instructions.

you pointed out yesterday that this could work on v7-m, too. As I based my patches on 3.9-rc1 this patch (8164f7af88) wasn't included. When updating this results in a warning, because I have

6ebd4d0 (ARM: stub out read_cpuid and read_cpuid_ext for CPU_CP15=n)

from rmk's devel-stable branch.

Quoted 14 lines
---
 arch/arm/kernel/setup.c | 20 ++++++++++++++++++++
 arch/arm/mm/proc-v7.S   |  4 ++--
 2 files changed, 22 insertions(+), 2 deletions(-)
arch/arm/kernel/setup.c +0 -0
index e2c8bbf..f3ac13f 100644
@@ -353,6 +353,23 @@ void __init early_print(const char *str, ...)
 	printk("%s", buf);

arch/arm/kernel/setup.c void __init early_print(const char *str, ...) · line 363

Quoted 10 lines
arch/arm/kernel/setup.c +8 -0
 }
 
+static void __init cpuid_init_hwcaps(void)
+{
+	unsigned int divide_instrs;
+
+	if (cpu_architecture() < CPU_ARCH_ARMv7)
+		return;
+
+	divide_instrs = (read_cpuid_ext(CPUID_EXT_ISAR0) & 0x0f000000) >> 24;

The problem is that read_cpuid_ext is called which doesn't map to something useful for v7-m. So maybe add a check:

Quoted 4 lines
	if (!IS_ENABLED(CONFIG_CPU_CP15))
		return;

?

The patch below fixes the issue for me on V7-M. The only drawback is that the list of registers isn't shared. Maybe someone has a nice idea?

BTW, there doesn't seem to be an equivalent register set for CPUID_{CACHETYPE,TCM,TLBTYPE,MPIDR} on v7-M.

Best regards
Uwe

arch/arm/include/asm/cputype.h +34 -4
index 4eb94a3..ec635ff 100644
@@ -10,6 +10,22 @@
 #define CPUID_TLBTYPE	3
 #define CPUID_MPIDR	5
 
+#ifdef CONFIG_CPU_V7M
+#define CPUID_EXT_PFR0	0x40
+#define CPUID_EXT_PFR1	0x44
+#define CPUID_EXT_DFR0	0x48
+#define CPUID_EXT_AFR0	0x4c
+#define CPUID_EXT_MMFR0	0x50
+#define CPUID_EXT_MMFR1	0x54
+#define CPUID_EXT_MMFR2	0x58
+#define CPUID_EXT_MMFR3	0x5c
+#define CPUID_EXT_ISAR0	0x60
+#define CPUID_EXT_ISAR1	0x64
+#define CPUID_EXT_ISAR2	0x68
+#define CPUID_EXT_ISAR3	0x6c
+#define CPUID_EXT_ISAR4	0x70
+#define CPUID_EXT_ISAR5	0x74
+#else
 #define CPUID_EXT_PFR0	"c1, 0"
 #define CPUID_EXT_PFR1	"c1, 1"
 #define CPUID_EXT_DFR0	"c1, 2"
@@ -24,6 +40,7 @@
 #define CPUID_EXT_ISAR3	"c2, 3"
 #define CPUID_EXT_ISAR4	"c2, 4"
 #define CPUID_EXT_ISAR5	"c2, 5"
+#endif
 
 #define MPIDR_SMP_BITMASK (0x3 << 30)
 #define MPIDR_SMP_VALUE (0x2 << 30)
@@ -79,7 +96,23 @@ extern unsigned int processor_id;
 		__val;							\
 	})
 
-#else /* ifdef CONFIG_CPU_CP15 */
+#elif defined(CONFIG_CPU_V7M)
+
+#include <asm/io.h>
+#include <asm/v7m.h>
+
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})
+
+static inline unsigned int __attribute_const__ read_cpuid_ext(unsigned offset)
+{
+	return readl(BASEADDR_V7M_SCB + offset);
+}
+
+#else /* ifdef CONFIG_CPU_CP15 / elif defined (CONFIG_CPU_V7M) */
 
 /*
  * read_cpuid and read_cpuid_ext should only ever be called on machines that
@@ -108,9 +141,6 @@ static inline unsigned int __attribute_const__ read_cpuid_id(void)
 
 #elif defined(CONFIG_CPU_V7M)
 
-#include <asm/io.h>
-#include <asm/v7m.h>
-
 static inline unsigned int __attribute_const__ read_cpuid_id(void)
 {
 	return readl(BASEADDR_V7M_SCB + V7M_SCB_CPUID);
arch/arm/mm/proc-v7m.S +1 -1
index 000499c..0c93588 100644
@@ -144,7 +144,7 @@ __v7m_proc_info:
 	b	__v7m_setup		@ proc_info_list.__cpu_flush
 	.long	cpu_arch_name
 	.long	cpu_elf_name
-	.long	HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT|HWCAP_IDIVT
+	.long	HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT
 	.long	cpu_v7m_name
 	.long	v7m_processor_functions	@ proc_info_list.proc
 	.long	0			@ proc_info_list.tlb
Signature · 2 lines
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
#5 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

Quoted 45 lines · Uwe Kleine-K?nig

Hello,

Quoted 42 lines · Uwe Kleine-K?nig
Quoted 6 lines · Stephen Boyd

The ISAR0 register indicates support for the SDIV and UDIV instructions in both the Thumb and ARM instruction set. Read the register to detect the supported instructions and update the elf_hwcap mask as appropriate. This is better than adding more and more cpuid checks in proc-v7.S for each new cpu variant that supports these instructions.

you pointed out yesterday that this could work on v7-m, too. As I based my patches on 3.9-rc1 this patch (8164f7af88) wasn't included. When updating this results in a warning, because I have

6ebd4d0 (ARM: stub out read_cpuid and read_cpuid_ext for CPU_CP15=n)

from rmk's devel-stable branch.

Quoted 14 lines
---
 arch/arm/kernel/setup.c | 20 ++++++++++++++++++++
 arch/arm/mm/proc-v7.S   |  4 ++--
 2 files changed, 22 insertions(+), 2 deletions(-)
arch/arm/kernel/setup.c +0 -0
index e2c8bbf..f3ac13f 100644
@@ -353,6 +353,23 @@ void __init early_print(const char *str, ...)
 	printk("%s", buf);

arch/arm/kernel/setup.c void __init early_print(const char *str, ...) · line 363

Quoted 10 lines
arch/arm/kernel/setup.c +8 -0
 }
 
+static void __init cpuid_init_hwcaps(void)
+{
+	unsigned int divide_instrs;
+
+	if (cpu_architecture() < CPU_ARCH_ARMv7)
+		return;
+
+	divide_instrs = (read_cpuid_ext(CPUID_EXT_ISAR0) & 0x0f000000) >> 24;

The problem is that read_cpuid_ext is called which doesn't map to something useful for v7-m. So maybe add a check:

Quoted 6 lines
Quoted 4 lines
	if (!IS_ENABLED(CONFIG_CPU_CP15))
		return;

?

The patch below fixes the issue for me on V7-M. The only drawback is that the list of registers isn't shared. Maybe someone has a nice idea?

The patch looks ok to me.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
#6 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

In reply to Uwe Kleine-König
Quoted 1 line · Uwe Kleine-K?nig

Hello,

Hi Uwe,

Quoted 6 lines · Uwe Kleine-K?nig

The patch below fixes the issue for me on V7-M. The only drawback is that the list of registers isn't shared. Maybe someone has a nice idea?

BTW, there doesn't seem to be an equivalent register set for CPUID_{CACHETYPE,TCM,TLBTYPE,MPIDR} on v7-M.

That make sense: we don't have caches, MMUs or SMP.

Quoted 43 lines

Best regards
Uwe

arch/arm/include/asm/cputype.h +17 -1
index 4eb94a3..ec635ff 100644
@@ -10,6 +10,22 @@
 #define CPUID_TLBTYPE	3
 #define CPUID_MPIDR	5
 
+#ifdef CONFIG_CPU_V7M
+#define CPUID_EXT_PFR0	0x40
+#define CPUID_EXT_PFR1	0x44
+#define CPUID_EXT_DFR0	0x48
+#define CPUID_EXT_AFR0	0x4c
+#define CPUID_EXT_MMFR0	0x50
+#define CPUID_EXT_MMFR1	0x54
+#define CPUID_EXT_MMFR2	0x58
+#define CPUID_EXT_MMFR3	0x5c
+#define CPUID_EXT_ISAR0	0x60
+#define CPUID_EXT_ISAR1	0x64
+#define CPUID_EXT_ISAR2	0x68
+#define CPUID_EXT_ISAR3	0x6c
+#define CPUID_EXT_ISAR4	0x70
+#define CPUID_EXT_ISAR5	0x74
+#else
 #define CPUID_EXT_PFR0	"c1, 0"
 #define CPUID_EXT_PFR1	"c1, 1"
 #define CPUID_EXT_DFR0	"c1, 2"
@@ -24,6 +40,7 @@
 #define CPUID_EXT_ISAR3	"c2, 3"
 #define CPUID_EXT_ISAR4	"c2, 4"
 #define CPUID_EXT_ISAR5	"c2, 5"
+#endif
 
 #define MPIDR_SMP_BITMASK (0x3 << 30)
 #define MPIDR_SMP_VALUE (0x2 << 30)
@@ -79,7 +96,23 @@ extern unsigned int processor_id;
 		__val;							\
 	})
 
-#else /* ifdef CONFIG_CPU_CP15 */

arch/arm/include/asm/cputype.h extern unsigned int processor_id; · line 108

Quoted 10 lines
arch/arm/include/asm/cputype.h +10 -0
+#elif defined(CONFIG_CPU_V7M)
+
+#include <asm/io.h>
+#include <asm/v7m.h>
+
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})

We do have an MIDR in PMSAv7 (not to be confused with MPIDR), so why not read that here?

Will

#7 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

Hello Will,

Quoted 9 lines · Will Deacon
Quoted 6 lines · Uwe Kleine-K?nig

On Thu, Apr 18, 2013 at 11:10:19AM +0200, Uwe Kleine-K?nig wrote:
The patch below fixes the issue for me on V7-M. The only drawback is that the list of registers isn't shared. Maybe someone has a nice idea?

BTW, there doesn't seem to be an equivalent register set for CPUID_{CACHETYPE,TCM,TLBTYPE,MPIDR} on v7-M.

That make sense: we don't have caches, MMUs or SMP.

IMHO the sensible thing is to implement the registers then and let them
tell: I don't have these things.

Quoted 39 lines
Quoted 39 lines
arch/arm/include/asm/cputype.h +17 -0
index 4eb94a3..ec635ff 100644
@@ -10,6 +10,22 @@
 #define CPUID_TLBTYPE	3
 #define CPUID_MPIDR	5
 
+#ifdef CONFIG_CPU_V7M
+#define CPUID_EXT_PFR0	0x40
+#define CPUID_EXT_PFR1	0x44
+#define CPUID_EXT_DFR0	0x48
+#define CPUID_EXT_AFR0	0x4c
+#define CPUID_EXT_MMFR0	0x50
+#define CPUID_EXT_MMFR1	0x54
+#define CPUID_EXT_MMFR2	0x58
+#define CPUID_EXT_MMFR3	0x5c
+#define CPUID_EXT_ISAR0	0x60
+#define CPUID_EXT_ISAR1	0x64
+#define CPUID_EXT_ISAR2	0x68
+#define CPUID_EXT_ISAR3	0x6c
+#define CPUID_EXT_ISAR4	0x70
+#define CPUID_EXT_ISAR5	0x74
+#else
 #define CPUID_EXT_PFR0	"c1, 0"
 #define CPUID_EXT_PFR1	"c1, 1"
 #define CPUID_EXT_DFR0	"c1, 2"
@@ -24,6 +40,7 @@
 #define CPUID_EXT_ISAR3	"c2, 3"
 #define CPUID_EXT_ISAR4	"c2, 4"
 #define CPUID_EXT_ISAR5	"c2, 5"
+#endif
 
 #define MPIDR_SMP_BITMASK (0x3 << 30)
 #define MPIDR_SMP_VALUE (0x2 << 30)
@@ -79,7 +96,23 @@ extern unsigned int processor_id;
 		__val;							\
 	})
 
Quoted 14 lines

arch/arm/include/asm/cputype.h extern unsigned int processor_id; · line 108

Quoted 11 lines
arch/arm/include/asm/cputype.h +10 -1
-#else /* ifdef CONFIG_CPU_CP15 */
+#elif defined(CONFIG_CPU_V7M)
+
+#include <asm/io.h>
+#include <asm/v7m.h>
+
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})

We do have an MIDR in PMSAv7 (not to be confused with MPIDR), so why not read that here?

I don't know it and it's unrelated to this patch. But I will note it and check it. In the meantime I'm taking patches of course :-)

Uwe

Signature · 2 lines
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
#8 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

Quoted 1 line · Uwe Kleine-K?nig

Hello Will,

Hello,

Quoted 15 lines · Will Deacon
Quoted 6 lines · Uwe Kleine-K?nig
Quoted 5 lines
arch/arm/include/asm/cputype.h +5 -0
+#elif defined(CONFIG_CPU_V7M)
+
+#include <asm/io.h>
+#include <asm/v7m.h>
+
Quoted 8 lines

arch/arm/include/asm/cputype.h extern unsigned int processor_id; · line 108

Quoted 5 lines
arch/arm/include/asm/cputype.h +5 -0
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})

We do have an MIDR in PMSAv7 (not to be confused with MPIDR), so why not read that here?

Quoted 2 lines

I don't know it and it's unrelated to this patch. But I will note it and check it. In the meantime I'm taking patches of course :-)

I don't think it is unrelated to this patch: I'm suggesting that you add an MIDR read to read_cpuid, instead of the unconditional WARN_ON_ONCE which you're adding here.

Then you can remove your v7m-specific read_cpuid_id implementation, because read_cpuid(CPUID_ID) will do the right thing.

...or am I missing something?

Will

#9 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

In reply to Will Deacon

Hi Will,

Quoted 11 lines · Will Deacon

arch/arm/include/asm/cputype.h extern unsigned int processor_id; · line 108

Quoted 6 lines · Uwe Kleine-K?nig
arch/arm/include/asm/cputype.h +6 -0
+
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})

We do have an MIDR in PMSAv7 (not to be confused with MPIDR), so why not read that here?

We don't need to add/remove anything for PMSAv7 as that's already implemented...

For V7M we have something called CPUID Base Register, not MIDR - the layout looks the same though. That's the only one of the CPUID_X that we have, but it is never requested directly - only via read_cpuid_id (this is to support a hardcoded cpuid from kconfig) so what Uwe's got does work, though I agree that it'd be much nicer to implement the underlying function and have the accessor general for both AR and M.

Jonny

#10 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

Quoted 17 lines · Jonathan Austin

Hi Will,

Quoted 11 lines · Will Deacon

arch/arm/include/asm/cputype.h extern unsigned int processor_id; · line 108

Quoted 6 lines · Uwe Kleine-K?nig
arch/arm/include/asm/cputype.h +6 -0
+
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})

We do have an MIDR in PMSAv7 (not to be confused with MPIDR), so why not read that here?

We don't need to add/remove anything for PMSAv7 as that's already implemented...

Quoted 7 lines

For V7M we have something called CPUID Base Register, not MIDR - the layout looks the same though. That's the only one of the CPUID_X that we have, but it is never requested directly - only via read_cpuid_id (this is to support a hardcoded cpuid from kconfig) so what Uwe's got does work, though I agree that it'd be much nicer to implement the underlying function and have the accessor general for both AR and M.

The current state (that is next + the patch from this thread) is:

#ifdef CONFIG_CPU_CP15
define read_cpuid and read_cpuid_ext using mrc p15 read_cpuid_id = read_cpuid(CPUID_ID)
#elif defined (CONFIG_CPU_V7M)
read_cpuid returns 0 (with a warning) read_cpuid_ext reads from SCB
read_cpuid_id reads from V7M_SCB_CPUID
#else
read_cpuid and read_cpuid_ext return 0 (with a warning) read_cpuid_id returns CONFIG_PROCESSOR_ID (a bit simplified)
#endif

Now, what is the "underlying function" that should be used to implement which accessor? I assume you want to have

read_cpuid_id = read_cpuid(CPUID_ID)

on v7-M, too? So what is the suggestion for read_cpuid?:

	if (reg == CPUID_ID)
		return readl(V7M_SCB_CPUID);
	else
		return 0;

That doesn't make much sense in my eyes. If I got you wrong, please give more details.

Best regards
Uwe

Signature · 2 lines
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
#11 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

Quoted 25 lines · Uwe Kleine-K?nig
Quoted 7 lines · Jonathan Austin

For V7M we have something called CPUID Base Register, not MIDR - the layout looks the same though. That's the only one of the CPUID_X that we have, but it is never requested directly - only via read_cpuid_id (this is to support a hardcoded cpuid from kconfig) so what Uwe's got does work, though I agree that it'd be much nicer to implement the underlying function and have the accessor general for both AR and M.

The current state (that is next + the patch from this thread) is:

#ifdef CONFIG_CPU_CP15
define read_cpuid and read_cpuid_ext using mrc p15 read_cpuid_id = read_cpuid(CPUID_ID)
#elif defined (CONFIG_CPU_V7M)
read_cpuid returns 0 (with a warning) read_cpuid_ext reads from SCB
read_cpuid_id reads from V7M_SCB_CPUID
#else
read_cpuid and read_cpuid_ext return 0 (with a warning) read_cpuid_id returns CONFIG_PROCESSOR_ID (a bit simplified)
#endif

Now, what is the "underlying function" that should be used to implement which accessor? I assume you want to have

Quoted 8 lines

read_cpuid_id = read_cpuid(CPUID_ID)

on v7-M, too? So what is the suggestion for read_cpuid?:

	if (reg == CPUID_ID)
		return readl(V7M_SCB_CPUID);
	else
		return 0;

Yes, either switch on the reg or define CPUID_ID to be V7M_SCB_CPUID for v7m, then just do the readl in read_cpuid. I'm not sure the warnings add much (we can do read_cpuid(random integer) with cp15 and it will blindly construct the mrc).

Quoted 2 lines

That doesn't make much sense in my eyes. If I got you wrong, please give more details.

I'm just suggesting that we try and keep the functions similar between v7m and cp15 where there isn't a good reason not to.

Up to you.

Will

#12 lore raw

RFC PATCH 1/1 implement read_cpuid_ext for v7-M (Was: Re: [PATCHv2 2/3] ARM: Detect support for SDIV/UDIV from ISAR0) register

In reply to Uwe Kleine-König
Quoted 45 lines · Uwe Kleine-K?nig

Hello,

Quoted 42 lines · Uwe Kleine-K?nig
Quoted 6 lines · Stephen Boyd

The ISAR0 register indicates support for the SDIV and UDIV instructions in both the Thumb and ARM instruction set. Read the register to detect the supported instructions and update the elf_hwcap mask as appropriate. This is better than adding more and more cpuid checks in proc-v7.S for each new cpu variant that supports these instructions.

you pointed out yesterday that this could work on v7-m, too. As I based my patches on 3.9-rc1 this patch (8164f7af88) wasn't included. When updating this results in a warning, because I have

6ebd4d0 (ARM: stub out read_cpuid and read_cpuid_ext for CPU_CP15=n)

from rmk's devel-stable branch.

Quoted 14 lines
---
  arch/arm/kernel/setup.c | 20 ++++++++++++++++++++
  arch/arm/mm/proc-v7.S   |  4 ++--
  2 files changed, 22 insertions(+), 2 deletions(-)
arch/arm/kernel/setup.c +0 -0
index e2c8bbf..f3ac13f 100644
@@ -353,6 +353,23 @@ void __init early_print(const char *str, ...)
  	printk("%s", buf);

arch/arm/kernel/setup.c void __init early_print(const char *str, ...) · line 363

Quoted 10 lines
arch/arm/kernel/setup.c +8 -0
  }
 
+static void __init cpuid_init_hwcaps(void)
+{
+	unsigned int divide_instrs;
+
+	if (cpu_architecture() < CPU_ARCH_ARMv7)
+		return;
+
+	divide_instrs = (read_cpuid_ext(CPUID_EXT_ISAR0) & 0x0f000000) >> 24;

The problem is that read_cpuid_ext is called which doesn't map to something useful for v7-m. So maybe add a check:

Quoted 6 lines
Quoted 4 lines
	if (!IS_ENABLED(CONFIG_CPU_CP15))
		return;

?

The patch below fixes the issue for me on V7-M. The only drawback is that the list of registers isn't shared. Maybe someone has a nice idea?

I looked at whether there was anything nice that we could do around this but it most things end up more ugly than a separate long list of defines... It also makes things hard to understand so I prefer having these extra defines.

On the other hand, I think that this is about the time to think about another file for V7M CPUID stuff...

This file is beginning to look a bit unruly with all the different cases (V7M, !CPU_CP15, as well as the functions like read_cpuid_part_number that have the same definition in all cases). Thoughts?

Quoted 12 lines

BTW, there doesn't seem to be an equivalent register set for CPUID_{CACHETYPE,TCM,TLBTYPE,MPIDR} on v7-M.

Best regards
Uwe

arch/arm/include/asm/cputype.h +0 -0
index 4eb94a3..ec635ff 100644

arch/arm/include/asm/cputype.h line 13

Quoted 4 lines
arch/arm/include/asm/cputype.h +1 -0
@@ -10,6 +10,22 @@
  #define CPUID_TLBTYPE	3
  #define CPUID_MPIDR	5
 
+#ifdef CONFIG_CPU_V7M

Why not stick the

#include <asm/io.h>
#include <asm/v7m.h>

here?

Quoted 75 lines
arch/arm/include/asm/cputype.h +33 -4
+#define CPUID_EXT_PFR0	0x40
+#define CPUID_EXT_PFR1	0x44
+#define CPUID_EXT_DFR0	0x48
+#define CPUID_EXT_AFR0	0x4c
+#define CPUID_EXT_MMFR0	0x50
+#define CPUID_EXT_MMFR1	0x54
+#define CPUID_EXT_MMFR2	0x58
+#define CPUID_EXT_MMFR3	0x5c
+#define CPUID_EXT_ISAR0	0x60
+#define CPUID_EXT_ISAR1	0x64
+#define CPUID_EXT_ISAR2	0x68
+#define CPUID_EXT_ISAR3	0x6c
+#define CPUID_EXT_ISAR4	0x70
+#define CPUID_EXT_ISAR5	0x74
+#else
  #define CPUID_EXT_PFR0	"c1, 0"
  #define CPUID_EXT_PFR1	"c1, 1"
  #define CPUID_EXT_DFR0	"c1, 2"
@@ -24,6 +40,7 @@
  #define CPUID_EXT_ISAR3	"c2, 3"
  #define CPUID_EXT_ISAR4	"c2, 4"
  #define CPUID_EXT_ISAR5	"c2, 5"
+#endif
 
  #define MPIDR_SMP_BITMASK (0x3 << 30)
  #define MPIDR_SMP_VALUE (0x2 << 30)
@@ -79,7 +96,23 @@ extern unsigned int processor_id;
  		__val;							\
  	})
 
-#else /* ifdef CONFIG_CPU_CP15 */
+#elif defined(CONFIG_CPU_V7M)
+
+#include <asm/io.h>
+#include <asm/v7m.h>
+
+#define read_cpuid(reg)							\
+	({								\
+		WARN_ON_ONCE(1);					\
+		0;							\
+	})
+
+static inline unsigned int __attribute_const__ read_cpuid_ext(unsigned offset)
+{
+	return readl(BASEADDR_V7M_SCB + offset);
+}
+
+#else /* ifdef CONFIG_CPU_CP15 / elif defined (CONFIG_CPU_V7M) */
 
  /*
   * read_cpuid and read_cpuid_ext should only ever be called on machines that
@@ -108,9 +141,6 @@ static inline unsigned int __attribute_const__ read_cpuid_id(void)
 
  #elif defined(CONFIG_CPU_V7M)
 
-#include <asm/io.h>
-#include <asm/v7m.h>
-
  static inline unsigned int __attribute_const__ read_cpuid_id(void)
  {
  	return readl(BASEADDR_V7M_SCB + V7M_SCB_CPUID);
arch/arm/mm/proc-v7m.S +1 -1
index 000499c..0c93588 100644
@@ -144,7 +144,7 @@ __v7m_proc_info:
  	b	__v7m_setup		@ proc_info_list.__cpu_flush
  	.long	cpu_arch_name
  	.long	cpu_elf_name
-	.long	HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT|HWCAP_IDIVT
+	.long	HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT
  	.long	cpu_v7m_name
  	.long	v7m_processor_functions	@ proc_info_list.proc
  	.long	0			@ proc_info_list.tlb
#13 lore raw

PATCH v2 3/3 ARM: Work around faulty ISAR0 register in some Krait CPUs

In reply to Stephen Boyd

From: Stepan Moskovchenko <[email protected]>

Some early versions of the Krait CPU design incorrectly indicate that they only support the UDIV and SDIV instructions in Thumb mode when they actually support them in ARM and Thumb mode. It seems that these CPUs follow the DDI0406B ARM ARM which has two possible values for the divide instructions field, instead of the DDI0406C document which has three possible values.

Work around this problem by checking the MIDR against Krait CPUs with this faulty ISAR0 register and force the hwcaps to indicate support in both modes.

[sboyd: Rewrote commit text to reflect real reasoning now that
we autodetect udiv/sdiv]

---
 arch/arm/mm/proc-v7.S | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
arch/arm/mm/proc-v7.S +15 -0
index bcd3d48..f584d3f 100644
@@ -434,6 +434,21 @@ __v7_ca15mp_proc_info:
 	.size	__v7_ca15mp_proc_info, . - __v7_ca15mp_proc_info
 
 	/*
+	 * Qualcomm Inc. Krait processors.
+	 */
+	.type	__krait_proc_info, #object
+__krait_proc_info:
+	.long	0x510f0400		@ Required ID value
+	.long	0xff0ffc00		@ Mask for ID
+	/*
+	 * Some Krait processors don't indicate support for SDIV and UDIV
+	 * instructions in the ARM instruction set, even though they actually
+	 * do support them.
+	 */
+	__v7_proc __v7_setup, hwcaps = HWCAP_IDIV
+	.size	__krait_proc_info, . - __krait_proc_info
+
+	/*
 	 * Match any ARMv7 processor core.
 	 */
 	.type	__v7_proc_info, #object
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation
Acked-by on v2 3/3
#14 lore raw

Re: [PATCHv2 3/3] ARM: Work around faulty ISAR0 register in some Krait CPUs

Quoted 13 lines · Stephen Boyd

From: Stepan Moskovchenko <[email protected]>

Some early versions of the Krait CPU design incorrectly indicate that they only support the UDIV and SDIV instructions in Thumb mode when they actually support them in ARM and Thumb mode. It seems that these CPUs follow the DDI0406B ARM ARM which has two possible values for the divide instructions field, instead of the DDI0406C document which has three possible values.

Work around this problem by checking the MIDR against Krait CPUs with this faulty ISAR0 register and force the hwcaps to indicate support in both modes.

Quoted 8 lines

[sboyd: Rewrote commit text to reflect real reasoning now that
we autodetect udiv/sdiv]

---
 arch/arm/mm/proc-v7.S | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

Will

#15 lore raw

Re: [PATCHv2 3/3] ARM: Work around faulty ISAR0 register in some Krait CPUs

Quoted 22 lines · Will Deacon
Quoted 13 lines · Stephen Boyd

From: Stepan Moskovchenko <[email protected]>

Some early versions of the Krait CPU design incorrectly indicate that they only support the UDIV and SDIV instructions in Thumb mode when they actually support them in ARM and Thumb mode. It seems that these CPUs follow the DDI0406B ARM ARM which has two possible values for the divide instructions field, instead of the DDI0406C document which has three possible values.

Work around this problem by checking the MIDR against Krait CPUs with this faulty ISAR0 register and force the hwcaps to indicate support in both modes.

Quoted 8 lines

[sboyd: Rewrote commit text to reflect real reasoning now that
we autodetect udiv/sdiv]

---
 arch/arm/mm/proc-v7.S | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
Quoted 2 lines

Thanks. Put all three in the patch tracker.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation