Skip to contentSkip to conversation

ARM: fix building NOMMU ARMv4/v5 kernels

PATCH In review Arnd Bergmann 5 participants 14 messages Mar 9, 2022 – Mar 15, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 ARM: fix building NOMMU ARMv4/v5 kernels

From: Arnd Bergmann <[email protected]>

The removal of the old-style irq entry broke obscure NOMMU configurations on machines that have an MMU:

ld.lld: error: undefined symbol: generic_handle_arch_irq
referenced by kernel/entry-armv.o:(__irq_svc) in archive arch/arm/built-in.a

A follow-up patch to convert nvic to the generic_handle_arch_irq() could have fixed this by removing the Kconfig conditional, but did it differently.

Change the Kconfig logic so ARM machines now unconditionally enable the feature.

I have also submitted a patch to remove support for the configurations that broke, but fixing the regression first is a trivial and correct change.

Fixes 54f481a2308e ("ARM: remove old-style irq entry")
Fixes 52d240871760 ("irqchip: nvic: Use GENERIC_IRQ_MULTI_HANDLER")

---
This patch should go into the arm/devel-stable branch that has the arm-irq-and-vmap-stacks-for-rmk patches from Ard

---
 arch/arm/Kconfig        | 2 +-
 drivers/irqchip/Kconfig | 1 -
 2 files changed, 1 insertion(+), 2 deletions(-)
arch/arm/Kconfig +1 -1
index e9975ddd5034..5f0b40bab4fb 100644
@@ -60,7 +60,7 @@ config ARM
 	select GENERIC_CPU_AUTOPROBE
 	select GENERIC_EARLY_IOREMAP
 	select GENERIC_IDLE_POLL_SETUP
-	select GENERIC_IRQ_MULTI_HANDLER if MMU
+	select GENERIC_IRQ_MULTI_HANDLER
 	select GENERIC_IRQ_PROBE
 	select GENERIC_IRQ_SHOW
 	select GENERIC_IRQ_SHOW_LEVEL
drivers/irqchip/Kconfig +0 -1
index 488eaa14d3a7..7038957f4a77 100644
@@ -58,7 +58,6 @@ config ARM_NVIC
 	bool
 	select IRQ_DOMAIN_HIERARCHY
 	select GENERIC_IRQ_CHIP
-	select GENERIC_IRQ_MULTI_HANDLER
 
 config ARM_VIC
 	bool
Signature · 1 line
2.29.2
#1 lore raw

PATCH 2/2 ARM: remove support for NOMMU ARMv4/v5

From: Arnd Bergmann <[email protected]>

It is possible to build MMU-less kernels for Cortex-M base microcrontrollers as well as a couple of older platforms that have not been converted to CONFIG_ARCH_MULTIPLATFORM, specifically ep93xx, footbridge, dove, sa1100 and s3c24xx.

It seems unlikely that anybody has tested those configurations in recent years, as even building them is frequently broken. A patch I submitted caused another build time regression in this configuration. I sent a patch for that, but it seems better to also remove the option entirely, leaving ARMv7-M as the only supported Arm NOMMU target for simplicity.

This addresses several build failures in randconfig builds that have accumulated over the years.

Cc Vladimir Murzin <[email protected]>
Cc Russell King <[email protected]>

---
If there are no objections, I'd apply this patch to the soc tree for 5.18 as a cleanup.

---
 arch/arm/Kconfig                              | 34 +++++++------------
 .../mach-footbridge/include/mach/hardware.h   | 20 ++++-------
 arch/arm/mach-footbridge/include/mach/io.h    | 20 -----------
 arch/arm/mach-integrator/hardware.h           |  5 ---
 4 files changed, 20 insertions(+), 59 deletions(-)
 delete mode 100644 arch/arm/mach-footbridge/include/mach/io.h
arch/arm/Kconfig +13 -21
index 5f0b40bab4fb..d184377ce2ae 100644
@@ -304,6 +304,17 @@ config MMU
 	  Select if you want MMU-based virtualised addressing space
 	  support by paged memory management. If unsure, say 'Y'.
 
+config ARM_SINGLE_ARMV7M
+	def_bool !MMU
+	select ARM_NVIC
+	select AUTO_ZRELADDR
+	select TIMER_OF
+	select COMMON_CLK
+	select CPU_V7M
+	select NO_IOPORT_MAP
+	select SPARSE_IRQ
+	select USE_OF
+
 config ARCH_MMAP_RND_BITS_MIN
 	default 8
 
@@ -318,12 +329,11 @@ config ARCH_MMAP_RND_BITS_MAX
 #
 choice
 	prompt "ARM system type"
-	default ARM_SINGLE_ARMV7M if !MMU
-	default ARCH_MULTIPLATFORM if MMU
+	depends on MMU
+	default ARCH_MULTIPLATFORM
 
 config ARCH_MULTIPLATFORM
 	bool "Allow multiple platforms to be selected"
-	depends on MMU
 	select ARCH_FLATMEM_ENABLE
 	select ARCH_SPARSEMEM_ENABLE
 	select ARCH_SELECT_MEMORY_MODEL
@@ -337,18 +347,6 @@ config ARCH_MULTIPLATFORM
 	select SPARSE_IRQ
 	select USE_OF
 
-config ARM_SINGLE_ARMV7M
-	bool "ARMv7-M based platforms (Cortex-M0/M3/M4)"
-	depends on !MMU
-	select ARM_NVIC
-	select AUTO_ZRELADDR
-	select TIMER_OF
-	select COMMON_CLK
-	select CPU_V7M
-	select NO_IOPORT_MAP
-	select SPARSE_IRQ
-	select USE_OF
-
 config ARCH_EP93XX
 	bool "EP93xx-based"
 	select ARCH_SPARSEMEM_ENABLE
@@ -367,7 +365,6 @@ config ARCH_FOOTBRIDGE
 	bool "FootBridge"
 	select CPU_SA110
 	select FOOTBRIDGE
-	select NEED_MACH_IO_H if !MMU
 	select NEED_MACH_MEMORY_H
 	help
 	  Support for systems based on the DC21285 companion chip
@@ -375,7 +372,6 @@ config ARCH_FOOTBRIDGE
 
 config ARCH_IOP32X
 	bool "IOP32x-based"
-	depends on MMU
 	select CPU_XSCALE
 	select GPIO_IOP
 	select GPIOLIB
@@ -387,7 +383,6 @@ config ARCH_IOP32X
 
 config ARCH_IXP4XX
 	bool "IXP4xx-based"
-	depends on MMU
 	select ARCH_SUPPORTS_BIG_ENDIAN
 	select ARM_PATCH_PHYS_VIRT
 	select CPU_XSCALE
@@ -418,7 +413,6 @@ config ARCH_DOVE
 
 config ARCH_PXA
 	bool "PXA2xx/PXA3xx-based"
-	depends on MMU
 	select ARCH_MTD_XIP
 	select ARM_CPU_SUSPEND if PM
 	select AUTO_ZRELADDR
@@ -437,7 +431,6 @@ config ARCH_PXA
 
 config ARCH_RPC
 	bool "RiscPC"
-	depends on MMU
 	depends on !CC_IS_CLANG && GCC_VERSION < 90100 && GCC_VERSION >= 60000
 	select ARCH_ACORN
 	select ARCH_MAY_HAVE_PC_FDC
@@ -493,7 +486,6 @@ config ARCH_S3C24XX
 
 config ARCH_OMAP1
 	bool "TI OMAP1"
-	depends on MMU
 	select ARCH_OMAP
 	select CLKSRC_MMIO
 	select GENERIC_IRQ_CHIP
arch/arm/mach-footbridge/include/mach/hardware.h +7 -13
index ecaf6e7388d9..985ad3a95671 100644
@@ -21,32 +21,26 @@
  * 0xf0000000	0x80000000	16MB	ISA memory
  */
 
-#ifdef CONFIG_MMU
-#define MMU_IO(a, b)	(a)
-#else
-#define MMU_IO(a, b)	(b)
-#endif
-
 #define XBUS_SIZE		0x00100000
-#define XBUS_BASE		MMU_IO(0xff800000, 0x40000000)
+#define XBUS_BASE		0xff800000
 
 #define ARMCSR_SIZE		0x00100000
-#define ARMCSR_BASE		MMU_IO(0xfe000000, 0x42000000)
+#define ARMCSR_BASE		0xfe000000
 
 #define WFLUSH_SIZE		0x00100000
-#define WFLUSH_BASE		MMU_IO(0xfd000000, 0x78000000)
+#define WFLUSH_BASE		0xfd000000
 
 #define PCIIACK_SIZE		0x00100000
-#define PCIIACK_BASE		MMU_IO(0xfc000000, 0x79000000)
+#define PCIIACK_BASE		0xfc000000
 
 #define PCICFG1_SIZE		0x01000000
-#define PCICFG1_BASE		MMU_IO(0xfb000000, 0x7a000000)
+#define PCICFG1_BASE		0xfb000000
 
 #define PCICFG0_SIZE		0x01000000
-#define PCICFG0_BASE		MMU_IO(0xfa000000, 0x7b000000)
+#define PCICFG0_BASE		0xfa000000
 
 #define PCIMEM_SIZE		0x01000000
-#define PCIMEM_BASE		MMU_IO(0xf0000000, 0x80000000)
+#define PCIMEM_BASE		0xf0000000
 
 #define XBUS_CS2		0x40012000
 
arch/arm/mach-footbridge/include/mach/io.h deleted +0 -20
deleted file mode 100644
index 4e18b921373f..000000000000
@@ -1,20 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0-only */
-/*
- *  arch/arm/mach-footbridge/include/mach/io.h
- *
- *  Copyright (C) 1997-1999 Russell King
- *
- *  Modifications:
- *   06-12-1997	RMK	Created.
- *   07-04-1999	RMK	Major cleanup
- */
-#ifndef __ASM_ARM_ARCH_IO_H
-#define __ASM_ARM_ARCH_IO_H
-
-/*
- * Translation of various i/o addresses to host addresses for !CONFIG_MMU
- */
-#define PCIO_BASE       0x7c000000
-#define __io(a)			((void __iomem *)(PCIO_BASE + (a)))
-
-#endif
arch/arm/mach-integrator/hardware.h +0 -5
index 4d6ade3dd4ee..81ce09e3ad45 100644
@@ -16,12 +16,7 @@
 #define IO_START		INTEGRATOR_HDR_BASE        // PA of IO
 
 /* macro to get at IO space when running virtually */
-#ifdef CONFIG_MMU
 #define IO_ADDRESS(x)	(((x) & 0x000fffff) | (((x) >> 4) & 0x0ff00000) | IO_BASE)
-#else
-#define IO_ADDRESS(x)	(x)
-#endif
-
 #define __io_address(n)		((void __iomem *)IO_ADDRESS(n))
 
 /*
Signature · 1 line
2.29.2
#2 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

Quoted 7 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

It is possible to build MMU-less kernels for Cortex-M base microcrontrollers as well as a couple of older platforms that have not been converted to CONFIG_ARCH_MULTIPLATFORM, specifically ep93xx, footbridge, dove, sa1100 and s3c24xx.

Quoted 6 lines

It seems unlikely that anybody has tested those configurations in recent years, as even building them is frequently broken. A patch I submitted caused another build time regression in this configuration. I sent a patch for that, but it seems better to also remove the option entirely, leaving ARMv7-M as the only supported Arm NOMMU target for simplicity.

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Robin.

Quoted 208 lines

This addresses several build failures in randconfig builds that have accumulated over the years.

Cc Vladimir Murzin <[email protected]>
Cc Russell King <[email protected]>

---
If there are no objections, I'd apply this patch to the soc tree for 5.18 as a cleanup.

---
  arch/arm/Kconfig                              | 34 +++++++------------
  .../mach-footbridge/include/mach/hardware.h   | 20 ++++-------
  arch/arm/mach-footbridge/include/mach/io.h    | 20 -----------
  arch/arm/mach-integrator/hardware.h           |  5 ---
  4 files changed, 20 insertions(+), 59 deletions(-)

delete mode 100644 arch/arm/mach-footbridge/include/mach/io.h

arch/arm/Kconfig +13 -21
index 5f0b40bab4fb..d184377ce2ae 100644
@@ -304,6 +304,17 @@ config MMU
  	  Select if you want MMU-based virtualised addressing space
  	  support by paged memory management. If unsure, say 'Y'.
 
+config ARM_SINGLE_ARMV7M
+	def_bool !MMU
+	select ARM_NVIC
+	select AUTO_ZRELADDR
+	select TIMER_OF
+	select COMMON_CLK
+	select CPU_V7M
+	select NO_IOPORT_MAP
+	select SPARSE_IRQ
+	select USE_OF
+
  config ARCH_MMAP_RND_BITS_MIN
  	default 8
 
@@ -318,12 +329,11 @@ config ARCH_MMAP_RND_BITS_MAX
  #
  choice
  	prompt "ARM system type"
-	default ARM_SINGLE_ARMV7M if !MMU
-	default ARCH_MULTIPLATFORM if MMU
+	depends on MMU
+	default ARCH_MULTIPLATFORM
 
  config ARCH_MULTIPLATFORM
  	bool "Allow multiple platforms to be selected"
-	depends on MMU
  	select ARCH_FLATMEM_ENABLE
  	select ARCH_SPARSEMEM_ENABLE
  	select ARCH_SELECT_MEMORY_MODEL
@@ -337,18 +347,6 @@ config ARCH_MULTIPLATFORM
  	select SPARSE_IRQ
  	select USE_OF
 
-config ARM_SINGLE_ARMV7M
-	bool "ARMv7-M based platforms (Cortex-M0/M3/M4)"
-	depends on !MMU
-	select ARM_NVIC
-	select AUTO_ZRELADDR
-	select TIMER_OF
-	select COMMON_CLK
-	select CPU_V7M
-	select NO_IOPORT_MAP
-	select SPARSE_IRQ
-	select USE_OF
-
  config ARCH_EP93XX
  	bool "EP93xx-based"
  	select ARCH_SPARSEMEM_ENABLE
@@ -367,7 +365,6 @@ config ARCH_FOOTBRIDGE
  	bool "FootBridge"
  	select CPU_SA110
  	select FOOTBRIDGE
-	select NEED_MACH_IO_H if !MMU
  	select NEED_MACH_MEMORY_H
  	help
  	  Support for systems based on the DC21285 companion chip
@@ -375,7 +372,6 @@ config ARCH_FOOTBRIDGE
 
  config ARCH_IOP32X
  	bool "IOP32x-based"
-	depends on MMU
  	select CPU_XSCALE
  	select GPIO_IOP
  	select GPIOLIB
@@ -387,7 +383,6 @@ config ARCH_IOP32X
 
  config ARCH_IXP4XX
  	bool "IXP4xx-based"
-	depends on MMU
  	select ARCH_SUPPORTS_BIG_ENDIAN
  	select ARM_PATCH_PHYS_VIRT
  	select CPU_XSCALE
@@ -418,7 +413,6 @@ config ARCH_DOVE
 
  config ARCH_PXA
  	bool "PXA2xx/PXA3xx-based"
-	depends on MMU
  	select ARCH_MTD_XIP
  	select ARM_CPU_SUSPEND if PM
  	select AUTO_ZRELADDR
@@ -437,7 +431,6 @@ config ARCH_PXA
 
  config ARCH_RPC
  	bool "RiscPC"
-	depends on MMU
  	depends on !CC_IS_CLANG && GCC_VERSION < 90100 && GCC_VERSION >= 60000
  	select ARCH_ACORN
  	select ARCH_MAY_HAVE_PC_FDC
@@ -493,7 +486,6 @@ config ARCH_S3C24XX
 
  config ARCH_OMAP1
  	bool "TI OMAP1"
-	depends on MMU
  	select ARCH_OMAP
  	select CLKSRC_MMIO
  	select GENERIC_IRQ_CHIP
arch/arm/mach-footbridge/include/mach/hardware.h +7 -13
index ecaf6e7388d9..985ad3a95671 100644
@@ -21,32 +21,26 @@
   * 0xf0000000	0x80000000	16MB	ISA memory
   */
 
-#ifdef CONFIG_MMU
-#define MMU_IO(a, b)	(a)
-#else
-#define MMU_IO(a, b)	(b)
-#endif
-
  #define XBUS_SIZE		0x00100000
-#define XBUS_BASE		MMU_IO(0xff800000, 0x40000000)
+#define XBUS_BASE		0xff800000
 
  #define ARMCSR_SIZE		0x00100000
-#define ARMCSR_BASE		MMU_IO(0xfe000000, 0x42000000)
+#define ARMCSR_BASE		0xfe000000
 
  #define WFLUSH_SIZE		0x00100000
-#define WFLUSH_BASE		MMU_IO(0xfd000000, 0x78000000)
+#define WFLUSH_BASE		0xfd000000
 
  #define PCIIACK_SIZE		0x00100000
-#define PCIIACK_BASE		MMU_IO(0xfc000000, 0x79000000)
+#define PCIIACK_BASE		0xfc000000
 
  #define PCICFG1_SIZE		0x01000000
-#define PCICFG1_BASE		MMU_IO(0xfb000000, 0x7a000000)
+#define PCICFG1_BASE		0xfb000000
 
  #define PCICFG0_SIZE		0x01000000
-#define PCICFG0_BASE		MMU_IO(0xfa000000, 0x7b000000)
+#define PCICFG0_BASE		0xfa000000
 
  #define PCIMEM_SIZE		0x01000000
-#define PCIMEM_BASE		MMU_IO(0xf0000000, 0x80000000)
+#define PCIMEM_BASE		0xf0000000
 
  #define XBUS_CS2		0x40012000
 
arch/arm/mach-footbridge/include/mach/io.h deleted +0 -20
deleted file mode 100644
index 4e18b921373f..000000000000
@@ -1,20 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0-only */
-/*
- *  arch/arm/mach-footbridge/include/mach/io.h
- *
- *  Copyright (C) 1997-1999 Russell King
- *
- *  Modifications:
- *   06-12-1997	RMK	Created.
- *   07-04-1999	RMK	Major cleanup
- */
-#ifndef __ASM_ARM_ARCH_IO_H
-#define __ASM_ARM_ARCH_IO_H
-
-/*
- * Translation of various i/o addresses to host addresses for !CONFIG_MMU
- */
-#define PCIO_BASE       0x7c000000
-#define __io(a)			((void __iomem *)(PCIO_BASE + (a)))
-
-#endif
arch/arm/mach-integrator/hardware.h +0 -5
index 4d6ade3dd4ee..81ce09e3ad45 100644
@@ -16,12 +16,7 @@
  #define IO_START		INTEGRATOR_HDR_BASE        // PA of IO
 
  /* macro to get at IO space when running virtually */
-#ifdef CONFIG_MMU
  #define IO_ADDRESS(x)	(((x) & 0x000fffff) | (((x) >> 4) & 0x0ff00000) | IO_BASE)
-#else
-#define IO_ADDRESS(x)	(x)
-#endif
-
  #define __io_address(n)		((void __iomem *)IO_ADDRESS(n))
 
  /*
#3 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

Quoted 16 lines · Robin Murphy
Quoted 7 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

It is possible to build MMU-less kernels for Cortex-M base microcrontrollers as well as a couple of older platforms that have not been converted to CONFIG_ARCH_MULTIPLATFORM, specifically ep93xx, footbridge, dove, sa1100 and s3c24xx.

Quoted 6 lines

It seems unlikely that anybody has tested those configurations in recent years, as even building them is frequently broken. A patch I submitted caused another build time regression in this configuration. I sent a patch for that, but it seems better to also remove the option entirely, leaving ARMv7-M as the only supported Arm NOMMU target for simplicity.

Quoted 3 lines

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Right, good idea. These are only selected by CONFIG_ARCH_INTEGRATOR, but that in turn doesn't build for CONFIG_MMU=n because it depends on ARCH_MULTIPLATFORM. I'll send a patch for these.

I also went back and forth between removing the #ifdefs in arch/arm/mm/proc-*.S for this patch, though I did not do that in the end. We may still want them at least for proc-v7.S because that is a small chance that someone will want to support Cortex-R.

Arnd

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#4 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

Quoted 21 lines · Arnd Bergmann
Quoted 16 lines · Robin Murphy
Quoted 7 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

It is possible to build MMU-less kernels for Cortex-M base microcrontrollers as well as a couple of older platforms that have not been converted to CONFIG_ARCH_MULTIPLATFORM, specifically ep93xx, footbridge, dove, sa1100 and s3c24xx.

Quoted 6 lines

It seems unlikely that anybody has tested those configurations in recent years, as even building them is frequently broken. A patch I submitted caused another build time regression in this configuration. I sent a patch for that, but it seems better to also remove the option entirely, leaving ARMv7-M as the only supported Arm NOMMU target for simplicity.

Quoted 3 lines

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Quoted 9 lines

Right, good idea. These are only selected by CONFIG_ARCH_INTEGRATOR, but that in turn doesn't build for CONFIG_MMU=n because it depends on ARCH_MULTIPLATFORM. I'll send a patch for these.

I also went back and forth between removing the #ifdefs in arch/arm/mm/proc-*.S for this patch, though I did not do that in the end. We may still want them at least for proc-v7.S because that is a small chance that someone will want to support Cortex-R.

I'd appreciate if you keep them since I run Cortex-R52...

Cheers
Vladimir

Quoted 2 lines

Arnd

#5 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

Quoted 26 lines · Vladimir Murzin
Quoted 25 lines · Arnd Bergmann
Quoted 16 lines · Robin Murphy
Quoted 7 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

It is possible to build MMU-less kernels for Cortex-M base microcrontrollers as well as a couple of older platforms that have not been converted to CONFIG_ARCH_MULTIPLATFORM, specifically ep93xx, footbridge, dove, sa1100 and s3c24xx.

Quoted 6 lines

It seems unlikely that anybody has tested those configurations in recent years, as even building them is frequently broken. A patch I submitted caused another build time regression in this configuration. I sent a patch for that, but it seems better to also remove the option entirely, leaving ARMv7-M as the only supported Arm NOMMU target for simplicity.

Quoted 3 lines

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Right, good idea. These are only selected by CONFIG_ARCH_INTEGRATOR, but that in turn doesn't build for CONFIG_MMU=n because it depends on ARCH_MULTIPLATFORM. I'll send a patch for these.

Quoted 7 lines
Quoted 5 lines

I also went back and forth between removing the #ifdefs in arch/arm/mm/proc-*.S for this patch, though I did not do that in the end. We may still want them at least for proc-v7.S because that is a small chance that someone will want to support Cortex-R.

I'd appreciate if you keep them since I run Cortex-R52...

Ok, no problem.

Any opinion on removing CPU_ARM{7TDMI,740T,940T,946E}?

Arnd

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Reviewed-by on v1 1/2
#12 lore raw
In reply to Arnd Bergmann
Quoted 15 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

The removal of the old-style irq entry broke obscure NOMMU configurations on machines that have an MMU:

ld.lld: error: undefined symbol: generic_handle_arch_irq
referenced by kernel/entry-armv.o:(__irq_svc) in archive arch/arm/built-in.a

A follow-up patch to convert nvic to the generic_handle_arch_irq() could have fixed this by removing the Kconfig conditional, but did it differently.

Change the Kconfig logic so ARM machines now unconditionally enable the feature.

Quoted 8 lines

I have also submitted a patch to remove support for the configurations that broke, but fixing the regression first is a trivial and correct change.

Fixes 54f481a2308e ("ARM: remove old-style irq entry")
Fixes 52d240871760 ("irqchip: nvic: Use GENERIC_IRQ_MULTI_HANDLER")

Thanks for fixing this so quickly!

Yours,
Linus Walleij

#7 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

In reply to Arnd Bergmann
Quoted 8 lines · Arnd Bergmann

Robin

Quoted 3 lines

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Right, good idea. These are only selected by CONFIG_ARCH_INTEGRATOR, but that in turn doesn't build for CONFIG_MMU=n because it depends on ARCH_MULTIPLATFORM. I'll send a patch for these.

Just delete these, I do have these CPU tiles around but they are so obscure and I never got around to even testing to boot them.

Yours,
Linus Walleij

#6 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

In reply to Arnd Bergmann
Quoted 1 line · Arnd Bergmann

Any opinion on removing CPU_ARM{7TDMI,740T,940T,946E}?

7TDMI was/is extremely prolific, it is everywhere (such as in CSR/SiRF GPS chips, SD cards, hard disks, a lot of sensors and what not).

That said I have never seen one of these systems with enough memory to boot a recent Linux kernel. I think with the failure to push the kernel footprint down (due to lack of interest) they don't really have a future as Linux targets. They all run $small_os.

Yours,
Linus Walleij

#8 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

In reply to Linus Walleij
Quoted 7 lines · Linus Walleij
Quoted 5 lines · Arnd Bergmann

Robin

Quoted 3 lines

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Quoted 6 lines
Quoted 3 lines

Right, good idea. These are only selected by CONFIG_ARCH_INTEGRATOR, but that in turn doesn't build for CONFIG_MMU=n because it depends on ARCH_MULTIPLATFORM. I'll send a patch for these.

Just delete these, I do have these CPU tiles around but they are so obscure and I never got around to even testing to boot them.

Right, of course you couldn't boot test them because it has been impossible to even select them in Kconfig for years. I've added a patch to completely remove the five v4/v5 NOMMU cores now, will send that later.

There are five more cores that are only referenced by mach-integrator that are supposed to work (922T, 1020, 1020E, 1022, 1026). Have you ever tested those, or should we consider removing them as well?

At some point, there was a proposal to add an arm10 based SoC to mainline, but that never happened and I'm fairly sure it won't come back now.

Arnd

#9 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

Quoted 20 lines · Arnd Bergmann
Quoted 7 lines · Linus Walleij
Quoted 5 lines · Arnd Bergmann

Robin

Quoted 3 lines

In that case, it would probably make sense to garbage-collect all the configs, setup code and other stuff relating to older MMU-less CPU cores like ARM1156, ARM940, etc. at the same time.

Quoted 6 lines
Quoted 3 lines

Right, good idea. These are only selected by CONFIG_ARCH_INTEGRATOR, but that in turn doesn't build for CONFIG_MMU=n because it depends on ARCH_MULTIPLATFORM. I'll send a patch for these.

Just delete these, I do have these CPU tiles around but they are so obscure and I never got around to even testing to boot them.

Right, of course you couldn't boot test them because it has been impossible to even select them in Kconfig for years. I've added a patch to completely remove the five v4/v5 NOMMU cores now, will send that later.

Quoted 7 lines

There are five more cores that are only referenced by mach-integrator that are supposed to work (922T, 1020, 1020E, 1022, 1026). Have you ever tested those, or should we consider removing them as well?

At some point, there was a proposal to add an arm10 based SoC to mainline, but that never happened and I'm fairly sure it won't come back now.

FWIW I have an Integrator/CM1026 that I'd love to play with purely for giggles, but it needs some serious archaeology to find and flash some firmware for it first.

I also have a couple of "real" ARM1026s in the form of Conexant Solos-based routers[1] which I've done a little hacking on, but don't see myself ever realistically having enough time or motivation to actually upstream anything. And at this point I don't see OpenWRT having any more interest in boards with only 15MB of RAM (yes, they really do boot with "mem=15M" to reserve 1MB for the modem...)

Cheers,
Robin.

[1] https://openwrt.org/toh/linksys/wag54g2

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#13 lore raw
In reply to Arnd Bergmann
Quoted 16 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

The removal of the old-style irq entry broke obscure NOMMU configurations on machines that have an MMU:

ld.lld: error: undefined symbol: generic_handle_arch_irq
referenced by kernel/entry-armv.o:(__irq_svc) in archive arch/arm/built-in.a

A follow-up patch to convert nvic to the generic_handle_arch_irq() could have fixed this by removing the Kconfig conditional, but did it differently.

Change the Kconfig logic so ARM machines now unconditionally enable the feature.

Quoted 8 lines

I have also submitted a patch to remove support for the configurations that broke, but fixing the regression first is a trivial and correct change.

Fixes 54f481a2308e ("ARM: remove old-style irq entry")
Fixes 52d240871760 ("irqchip: nvic: Use GENERIC_IRQ_MULTI_HANDLER")

Reviewed by: Ard Biesheuvel <[email protected]>

Quoted 36 lines

---
This patch should go into the arm/devel-stable branch that has the arm-irq-and-vmap-stacks-for-rmk patches from Ard

---
 arch/arm/Kconfig        | 2 +-
 drivers/irqchip/Kconfig | 1 -
 2 files changed, 1 insertion(+), 2 deletions(-)
arch/arm/Kconfig +1 -1
index e9975ddd5034..5f0b40bab4fb 100644
@@ -60,7 +60,7 @@ config ARM
        select GENERIC_CPU_AUTOPROBE
        select GENERIC_EARLY_IOREMAP
        select GENERIC_IDLE_POLL_SETUP
-       select GENERIC_IRQ_MULTI_HANDLER if MMU
+       select GENERIC_IRQ_MULTI_HANDLER
        select GENERIC_IRQ_PROBE
        select GENERIC_IRQ_SHOW
        select GENERIC_IRQ_SHOW_LEVEL
drivers/irqchip/Kconfig +0 -2
index 488eaa14d3a7..7038957f4a77 100644
@@ -58,7 +58,6 @@ config ARM_NVIC
        bool
        select IRQ_DOMAIN_HIERARCHY
        select GENERIC_IRQ_CHIP
-       select GENERIC_IRQ_MULTI_HANDLER
 
 config ARM_VIC
        bool
--

2.29.2

#10 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

In reply to Arnd Bergmann
Quoted 7 lines · Arnd Bergmann
Quoted 2 lines · Linus Walleij

Just delete these, I do have these CPU tiles around but they are so obscure and I never got around to even testing to boot them.

Right, of course you couldn't boot test them because it has been impossible to even select them in Kconfig for years.

Correct, I mean never got around to the whole cycle of "let's make this thing boot" and then stumbling over things like that :D

Quoted 3 lines

I've added
a patch to completely remove the five v4/v5 NOMMU cores now, will send that later.

Excellent.

Quoted 3 lines

There are five more cores that are only referenced by mach-integrator that are supposed to work (922T, 1020, 1020E, 1022, 1026). Have you ever tested those, or should we consider removing them as well?

Not the 10xx variants, but 920T and 922T I have tested and I still have the tiles, while my favourite tile that is usually fitted is ARM926EJ-S.

Yours,
Linus Walleij

#11 lore raw

Re: [PATCH 2/2] ARM: remove support for NOMMU ARMv4/v5

Quoted 18 lines · Linus Walleij
Quoted 7 lines · Arnd Bergmann
Quoted 2 lines · Linus Walleij

Just delete these, I do have these CPU tiles around but they are so obscure and I never got around to even testing to boot them.

Right, of course you couldn't boot test them because it has been impossible to even select them in Kconfig for years.

Correct, I mean never got around to the whole cycle of "let's make this thing boot" and then stumbling over things like that :D

Quoted 3 lines

I've added
a patch to completely remove the five v4/v5 NOMMU cores now, will send that later.

Excellent.

Quoted 7 lines
Quoted 3 lines

There are five more cores that are only referenced by mach-integrator that are supposed to work (922T, 1020, 1020E, 1022, 1026). Have you ever tested those, or should we consider removing them as well?

Not the 10xx variants, but 920T and 922T I have tested and I still have the tiles, while my favourite tile that is usually fitted is ARM926EJ-S.

Ok, then I would suggest we keep 922t and 1026 (which Robin mentioned) but remove the others (1020,1020e, 1022).

Arnd