Skip to contentSkip to conversation

MMU disabling code and kexec fixes

PATCH Superseded Will Deacon 5 participants 29 messages Aug 23, 2011 – Aug 25, 2011 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

v4 cover MMU disabling code and kexec fixes

Hello,

This is version 4 of the patches originally posted here:

v1: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052157.html
v2: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052559.html
v3: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/053252.html

There are many changes since v3. Notably:
- Support for Thumb-2
- Based on 3.1-rc3 (now that some of the previous patches have been merged)
- The reserved stack page now sits below swapper and is reserved by memblock during memory init. This has the advantage of being at a known physical address for a platform, which will be helpful for non-hotplug SMP booting.

Currently, SMP kexec relies on platform_cpu_kill (a CPU hotplug callback) doing the right thing. I'm working on a generic SMP soft reboot implementation but this isn't quite working yet (and my git repo is predictably offline at the moment so I can't provide a link to the patches).

All feedback welcome.

Thanks,

Will

Will Deacon (8):
ARM: proc-v7: disable SCTLR.TE when disabling MMU
ARM: kexec: ensure new kernel is entered in ARM state
ARM: lib: add switch_stack function for safely changing stack
ARM: idmap: add header file for identity mapping functions
ARM: reset: allow kernelspace mappings to be flat mapped during reset
ARM: reset: add reset functionality for jumping to a physical address
ARM: kexec: use arm_machine_reset for branching to the reboot buffer
ARM: stop: execute platform callback from cpu_stop code

 arch/arm/Kconfig                  |    2 +-
 arch/arm/include/asm/idmap.h      |   15 +++++++
 arch/arm/include/asm/pgtable.h    |    3 -
 arch/arm/include/asm/system.h     |    1 +
 arch/arm/kernel/machine_kexec.c   |   14 +------
 arch/arm/kernel/process.c         |   76 ++++++++++++++++++++++++++++++++++--
 arch/arm/kernel/relocate_kernel.S |    2 +-
 arch/arm/kernel/smp.c             |    5 ++
 arch/arm/lib/Makefile             |    3 +-
 arch/arm/lib/switch_stack.S       |   44 +++++++++++++++++++++
 arch/arm/mm/idmap.c               |   43 ++++++++++++++++-----
 arch/arm/mm/mmu.c                 |   13 +++++-
 arch/arm/mm/proc-v7.S             |    1 +
 13 files changed, 187 insertions(+), 35 deletions(-)
 create mode 100644 arch/arm/include/asm/idmap.h
 create mode 100644 arch/arm/lib/switch_stack.S
#1 lore raw

PATCH v4 1/8 ARM: proc-v7: disable SCTLR.TE when disabling MMU

cpu_v7_reset disables the MMU and then branches to the provided address. On Thumb-2 kernels, we should take care to clear the Thumb Exception enable bit in the System Control Register, otherwise this may wreak havok in the code to which we are branching (for example, an ARM kernel image via kexec).

---
 arch/arm/mm/proc-v7.S |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/arm/mm/proc-v7.S +1 -0
index a30e785..96b872c 100644
@@ -66,6 +66,7 @@ ENDPROC(cpu_v7_proc_fin)
 ENTRY(cpu_v7_reset)
 	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
 	bic	r1, r1, #0x1			@ ...............m
+ THUMB(	bic	r1, r1, #1 << 30 )		@ Thumb exceptions
 	mcr	p15, 0, r1, c1, c0, 0		@ disable MMU
 	isb
 	mov	pc, r0
Signature · 1 line
1.7.0.4
#2 lore raw

PATCH v4 1/8 ARM: proc-v7: disable SCTLR.TE when disabling MMU

Quoted 16 lines · Will Deacon

cpu_v7_reset disables the MMU and then branches to the provided address. On Thumb-2 kernels, we should take care to clear the Thumb Exception enable bit in the System Control Register, otherwise this may wreak havok in the code to which we are branching (for example, an ARM kernel image via kexec).

---
 arch/arm/mm/proc-v7.S |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/arm/mm/proc-v7.S +0 -0
index a30e785..96b872c 100644

arch/arm/mm/proc-v7.S ENDPROC(cpu_v7_proc_fin) · line 69

Quoted 4 lines
arch/arm/mm/proc-v7.S +1 -0
@@ -66,6 +66,7 @@ ENDPROC(cpu_v7_proc_fin)
 ENTRY(cpu_v7_reset)
 	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
 	bic	r1, r1, #0x1			@ ...............m
+ THUMB(	bic	r1, r1, #1 << 30 )		@ Thumb exceptions

Can we give the architectural name for the bit here?

If we must have magic numbers, we can at least minimise the scope for confusion.

What that,

Cheers
---Dave

#3 lore raw

PATCH v4 1/8 ARM: proc-v7: disable SCTLR.TE when disabling MMU

Quoted 12 lines · Dave Martin
Quoted 11 lines · Will Deacon

cpu_v7_reset disables the MMU and then branches to the provided address. On Thumb-2 kernels, we should take care to clear the Thumb Exception enable bit in the System Control Register, otherwise this may wreak havok in the code to which we are branching (for example, an ARM kernel image via kexec).

---
 arch/arm/mm/proc-v7.S |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
Quoted 11 lines

arch/arm/mm/proc-v7.S ENDPROC(cpu_v7_proc_fin) · line 69

Quoted 9 lines
arch/arm/mm/proc-v7.S +1 -0
index a30e785..96b872c 100644
@@ -66,6 +66,7 @@ ENDPROC(cpu_v7_proc_fin)
 ENTRY(cpu_v7_reset)
 	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
 	bic	r1, r1, #0x1			@ ...............m
+ THUMB(	bic	r1, r1, #1 << 30 )		@ Thumb exceptions

Can we give the architectural name for the bit here?

Ok. It's SCTLR.TE so representing it as we do for SCTLR.M looks a bit odd unless you use two lines. Perhaps if I just add @ SCTLR.TE (Thumb exceptions) ?

Quoted 2 lines

What that,

Ta,

Will

#4 lore raw

PATCH v4 1/8 ARM: proc-v7: disable SCTLR.TE when disabling MMU

Quoted 25 lines · Will Deacon
Quoted 12 lines · Dave Martin
Quoted 11 lines · Will Deacon

cpu_v7_reset disables the MMU and then branches to the provided address. On Thumb-2 kernels, we should take care to clear the Thumb Exception enable bit in the System Control Register, otherwise this may wreak havok in the code to which we are branching (for example, an ARM kernel image via kexec).

---
 arch/arm/mm/proc-v7.S |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
Quoted 11 lines

arch/arm/mm/proc-v7.S ENDPROC(cpu_v7_proc_fin) · line 69

Quoted 9 lines
arch/arm/mm/proc-v7.S +1 -0
index a30e785..96b872c 100644
@@ -66,6 +66,7 @@ ENDPROC(cpu_v7_proc_fin)
 ENTRY(cpu_v7_reset)
 	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
 	bic	r1, r1, #0x1			@ ...............m
+ THUMB(	bic	r1, r1, #1 << 30 )		@ Thumb exceptions

Can we give the architectural name for the bit here?

Quoted 3 lines

Ok. It's SCTLR.TE so representing it as we do for SCTLR.M looks a bit odd unless you use two lines. Perhaps if I just add @ SCTLR.TE (Thumb exceptions) ?

That seems fine -- exactly how it looks is not so important as the information itself IMHO.

It's a pretty minor nit in any case.

Cheers
---Dave

#5 lore raw

PATCH v4 2/8 ARM: kexec: ensure new kernel is entered in ARM state

In reply to Will Deacon

Commit 540b5738 ("ARM: 6999/1: head, zImage: Always Enter the kernel in ARM state") mandates that the kernel should be entered in ARM state.

If a Thumb-2 kernel kexecs a new kernel image, we need to ensure that we change state when branching to the new code. This patch replaces a mov pc, lr with a bx lr so that we transition to ARM state if need be.

---
 arch/arm/kernel/relocate_kernel.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/kernel/relocate_kernel.S +1 -1
index 9cf4cbf..0908633 100644
@@ -57,7 +57,7 @@ relocate_new_kernel:
 	mov r0,#0
 	ldr r1,kexec_mach_type
 	ldr r2,kexec_boot_atags
-	mov pc,lr
+	bx lr
 
 	.align
 
Signature · 1 line
1.7.0.4
#6 lore raw

PATCH v4 2/8 ARM: kexec: ensure new kernel is entered in ARM state

Quoted 6 lines · Will Deacon

Commit 540b5738 ("ARM: 6999/1: head, zImage: Always Enter the kernel in ARM state") mandates that the kernel should be entered in ARM state.

If a Thumb-2 kernel kexecs a new kernel image, we need to ensure that we change state when branching to the new code. This patch replaces a mov pc, lr with a bx lr so that we transition to ARM state if need be.

This will break kexec support on ARMv4 targets. No idea if any ARMv4 target uses kexec though.

Quoted 27 lines
---
 arch/arm/kernel/relocate_kernel.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/kernel/relocate_kernel.S +1 -1
index 9cf4cbf..0908633 100644
@@ -57,7 +57,7 @@ relocate_new_kernel:
 	mov r0,#0
 	ldr r1,kexec_mach_type
 	ldr r2,kexec_boot_atags
-	mov pc,lr
+	bx lr
 
 	.align
 
Signature · 7 lines
1.7.0.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#7 lore raw

PATCH v4 2/8 ARM: kexec: ensure new kernel is entered in ARM state

Quoted 11 lines · Nicolas Pitre
Quoted 6 lines · Will Deacon

Commit 540b5738 ("ARM: 6999/1: head, zImage: Always Enter the kernel in ARM state") mandates that the kernel should be entered in ARM state.

If a Thumb-2 kernel kexecs a new kernel image, we need to ensure that we change state when branching to the new code. This patch replaces a mov pc, lr with a bx lr so that we transition to ARM state if need be.

This will break kexec support on ARMv4 targets. No idea if any ARMv4 target uses kexec though.

Huh, good point -- I was too hasty.

We should use ARM() THUMB() here.

Cheers
---Dave

Quoted 33 lines
Quoted 28 lines
---
 arch/arm/kernel/relocate_kernel.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/kernel/relocate_kernel.S +1 -1
index 9cf4cbf..0908633 100644
@@ -57,7 +57,7 @@ relocate_new_kernel:
 	mov r0,#0
 	ldr r1,kexec_mach_type
 	ldr r2,kexec_boot_atags
-	mov pc,lr
+	bx lr
 
 	.align
 
Signature · 7 lines
1.7.0.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#8 lore raw

PATCH v4 2/8 ARM: kexec: ensure new kernel is entered in ARM state

In reply to Nicolas Pitre
Quoted 11 lines · Nicolas Pitre
Quoted 6 lines · Will Deacon

Commit 540b5738 ("ARM: 6999/1: head, zImage: Always Enter the kernel in ARM state") mandates that the kernel should be entered in ARM state.

If a Thumb-2 kernel kexecs a new kernel image, we need to ensure that we change state when branching to the new code. This patch replaces a mov pc, lr with a bx lr so that we transition to ARM state if need be.

This will break kexec support on ARMv4 targets. No idea if any ARMv4 target uses kexec though.

It's easy enough to sort out though, so I may as well fix the code in case anybody wants to try kexec on ARMv4!

Thanks,

Will

#9 lore raw

PATCH v4 2/8 ARM: kexec: ensure new kernel is entered in ARM state

In reply to Will Deacon
Quoted 17 lines · Will Deacon

Commit 540b5738 ("ARM: 6999/1: head, zImage: Always Enter the kernel in ARM state") mandates that the kernel should be entered in ARM state.

If a Thumb-2 kernel kexecs a new kernel image, we need to ensure that we change state when branching to the new code. This patch replaces a mov pc, lr with a bx lr so that we transition to ARM state if need be.

---
 arch/arm/kernel/relocate_kernel.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/kernel/relocate_kernel.S +0 -0
index 9cf4cbf..0908633 100644

arch/arm/kernel/relocate_kernel.S relocate_new_kernel: · line 63

Quoted 8 lines
arch/arm/kernel/relocate_kernel.S +1 -1
@@ -57,7 +57,7 @@ relocate_new_kernel:
 	mov r0,#0
 	ldr r1,kexec_mach_type
 	ldr r2,kexec_boot_atags
-	mov pc,lr
+	bx lr
 
 	.align
 
Quoted 8 lines
Signature · 7 lines
1.7.0.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#10 lore raw

PATCH v4 3/8 ARM: lib: add switch_stack function for safely changing stack

In reply to Will Deacon

When disabling the MMU, it is necessary to take out a 1:1 identity map of the reset code so that it can safely be executed with and without the MMU active. To avoid the situation where the physical address of the reset code aliases with the virtual address of the active stack (which cannot be included in the 1:1 mapping), it is desirable to change to a new stack at a location which is less likely to alias.

This code adds a new lib function, switch_stack:

void switch_stack(void (*fn)(void *), void *arg, void *sp);

which changes the stack to point at the sp parameter, before invoking fn(arg) with the new stack selected.

---
 arch/arm/lib/Makefile       |    3 +-
 arch/arm/lib/switch_stack.S |   44 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+), 1 deletions(-)
 create mode 100644 arch/arm/lib/switch_stack.S
arch/arm/lib/Makefile +2 -1
index cf73a7f..cbb1bc1 100644
@@ -13,7 +13,8 @@ lib-y		:= backtrace.o changebit.o csumipv6.o csumpartial.o   \
 		   testchangebit.o testclearbit.o testsetbit.o        \
 		   ashldi3.o ashrdi3.o lshrdi3.o muldi3.o             \
 		   ucmpdi2.o lib1funcs.o div64.o                      \
-		   io-readsb.o io-writesb.o io-readsl.o io-writesl.o
+		   io-readsb.o io-writesb.o io-readsl.o io-writesl.o  \
+		   switch_stack.o
 
 mmu-y	:= clear_user.o copy_page.o getuser.o putuser.o
 
arch/arm/lib/switch_stack.S new file +44 -0

44 lines not on this page · raw message

Signature · 1 line
1.7.0.4
#11 lore raw

PATCH v4 3/8 ARM: lib: add switch_stack function for safely changing stack

Quoted 9 lines · Will Deacon

When disabling the MMU, it is necessary to take out a 1:1 identity map of the reset code so that it can safely be executed with and without the MMU active. To avoid the situation where the physical address of the reset code aliases with the virtual address of the active stack (which cannot be included in the 1:1 mapping), it is desirable to change to a new stack at a location which is less likely to alias.

This code adds a new lib function, switch_stack:

Quoted 7 lines

void switch_stack(void (*fn)(void *), void *arg, void *sp);

which changes the stack to point at the sp parameter, before invoking fn(arg) with the new stack selected.

For the implementation:

However I think switch_stack() is too vague a name for what this actually does. Maybe something like call_with_stack(fn, arg, sp) is more precise.

Quoted 79 lines
---
 arch/arm/lib/Makefile       |    3 +-
 arch/arm/lib/switch_stack.S |   44 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+), 1 deletions(-)
 create mode 100644 arch/arm/lib/switch_stack.S
arch/arm/lib/Makefile +2 -1
index cf73a7f..cbb1bc1 100644
@@ -13,7 +13,8 @@ lib-y		:= backtrace.o changebit.o csumipv6.o csumpartial.o   \
 		   testchangebit.o testclearbit.o testsetbit.o        \
 		   ashldi3.o ashrdi3.o lshrdi3.o muldi3.o             \
 		   ucmpdi2.o lib1funcs.o div64.o                      \
-		   io-readsb.o io-writesb.o io-readsl.o io-writesl.o
+		   io-readsb.o io-writesb.o io-readsl.o io-writesl.o  \
+		   switch_stack.o
 
 mmu-y	:= clear_user.o copy_page.o getuser.o putuser.o
 
arch/arm/lib/switch_stack.S new file +44 -0
new file mode 100644
index 0000000..552090d
@@ -0,0 +1,44 @@
+/*
+ * arch/arm/lib/switch_stack.S
+ *
+ * Copyright (C) 2011 ARM Ltd.
+ * Written by Will Deacon <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+
+/*
+ * void switch_stack(void (*fn)(void *), void *arg, void *sp)
+ *
+ * Change the stack to that pointed at by sp, then invoke fn(arg) with
+ * the new stack.
+ */
+ENTRY(switch_stack)
+	str	sp, [r2, #-4]!
+	str	lr, [r2, #-4]!
+
+	mov	sp, r2
+	mov	r2, r0
+	mov	r0, r1
+
+	adr	lr, BSYM(1f)
+	mov	pc, r2
+
+1:	ldr	lr, [sp]
+	ldr	sp, [sp, #4]
+	mov	pc, lr
+ENDPROC(switch_stack)
Signature · 7 lines
1.7.0.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#12 lore raw

PATCH v4 3/8 ARM: lib: add switch_stack function for safely changing stack

Quoted 19 lines · Nicolas Pitre
Quoted 9 lines · Will Deacon

When disabling the MMU, it is necessary to take out a 1:1 identity map of the reset code so that it can safely be executed with and without the MMU active. To avoid the situation where the physical address of the reset code aliases with the virtual address of the active stack (which cannot be included in the 1:1 mapping), it is desirable to change to a new stack at a location which is less likely to alias.

This code adds a new lib function, switch_stack:

Quoted 7 lines

void switch_stack(void (*fn)(void *), void *arg, void *sp);

which changes the stack to point at the sp parameter, before invoking fn(arg) with the new stack selected.

Quoted 7 lines

For the implementation:

However I think switch_stack() is too vague a name for what this actually does. Maybe something like call_with_stack(fn, arg, sp) is more precise.

call_with_stack() soundslike a good name to me.

---Dave

#13 lore raw

PATCH v4 3/8 ARM: lib: add switch_stack function for safely changing stack

In reply to Nicolas Pitre
Quoted 19 lines · Nicolas Pitre
Quoted 9 lines · Will Deacon

When disabling the MMU, it is necessary to take out a 1:1 identity map of the reset code so that it can safely be executed with and without the MMU active. To avoid the situation where the physical address of the reset code aliases with the virtual address of the active stack (which cannot be included in the 1:1 mapping), it is desirable to change to a new stack at a location which is less likely to alias.

This code adds a new lib function, switch_stack:

Quoted 7 lines

void switch_stack(void (*fn)(void *), void *arg, void *sp);

which changes the stack to point at the sp parameter, before invoking fn(arg) with the new stack selected.

Quoted 7 lines

For the implementation:

However I think switch_stack() is too vague a name for what this actually does. Maybe something like call_with_stack(fn, arg, sp) is more precise.

That's a much better name, thanks.

Will

#14 lore raw

PATCH v4 4/8 ARM: idmap: add header file for identity mapping functions

In reply to Will Deacon

The identity mappings functions are useful outside of SMP booting, so expose them through their own header file.

---
 arch/arm/include/asm/idmap.h   |   11 +++++++++++
 arch/arm/include/asm/pgtable.h |    3 ---
 arch/arm/kernel/process.c      |    3 +--
 arch/arm/kernel/smp.c          |    1 +
 arch/arm/mm/idmap.c            |    5 +++++
 5 files changed, 18 insertions(+), 5 deletions(-)
 create mode 100644 arch/arm/include/asm/idmap.h
arch/arm/include/asm/idmap.h new file +11 -0
new file mode 100644
index 0000000..fe7ea86
@@ -0,0 +1,11 @@
+#ifndef _ARM_IDMAP_H
+#define _ARM_IDMAP_H
+
+#include <asm/page.h>
+
+void identity_mapping_add(pgd_t *pgd, unsigned long addr, unsigned long end);
+void identity_mapping_del(pgd_t *pgd, unsigned long addr, unsigned long end);
+
+void setup_mm_for_reboot(char mode);
+
+#endif	/* _ARM_IDMAP_H */
arch/arm/include/asm/pgtable.h +0 -3
index 5750704..9d559a8 100644
@@ -474,9 +474,6 @@ static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 
 #define pgtable_cache_init() do { } while (0)
 
-void identity_mapping_add(pgd_t *, unsigned long, unsigned long);
-void identity_mapping_del(pgd_t *, unsigned long, unsigned long);
-
 #endif /* !__ASSEMBLY__ */
 
 #endif /* CONFIG_MMU */
arch/arm/kernel/process.c +1 -2
index 1a347f4..89fe6ee 100644
@@ -33,6 +33,7 @@
 #include <linux/cpuidle.h>
 
 #include <asm/cacheflush.h>
+#include <asm/idmap.h>
 #include <asm/leds.h>
 #include <asm/processor.h>
 #include <asm/system.h>
@@ -57,8 +58,6 @@ static const char *isa_modes[] = {
   "ARM" , "Thumb" , "Jazelle", "ThumbEE"
 };
 
-extern void setup_mm_for_reboot(char mode);
-
 static volatile int hlt_counter;
 
 #include <mach/system.h>
arch/arm/kernel/smp.c +1 -0
index d88ff02..96a7956 100644
@@ -39,6 +39,7 @@
 #include <asm/tlbflush.h>
 #include <asm/ptrace.h>
 #include <asm/localtimer.h>
+#include <asm/idmap.h>
 
 /*
  * as from 2.5, kernels no longer have an init_tasks structure
arch/arm/mm/idmap.c +5 -0
index 2be9139..e99dc38 100644
@@ -1,6 +1,7 @@
 #include <linux/kernel.h>
 
 #include <asm/cputype.h>
+#include <asm/idmap.h>
 #include <asm/pgalloc.h>
 #include <asm/pgtable.h>
 
@@ -71,6 +72,10 @@ void identity_mapping_del(pgd_t *pgd, unsigned long addr, unsigned long end)
 		idmap_del_pud(pgd, addr, next);
 	} while (pgd++, addr = next, addr != end);
 }
+#else
+void identity_mapping_del(pgd_t *pgd, unsigned long addr, unsigned long end)
+{
+}
 #endif
 
 /*
Signature · 1 line
1.7.0.4
#15 lore raw

PATCH v4 5/8 ARM: reset: allow kernelspace mappings to be flat mapped during reset

In reply to Will Deacon

Currently, switch_mm_for_reboot only takes out a 1:1 mapping from 0x0 to TASK_SIZE during reboot. For situations where we actually want to turn off the MMU (e.g. kexec, hibernate, CPU hotplug) we want to map as much memory as possible using the identity mapping so that we increase the chance of mapping our reset code.

This patch extends setup_mm_for_reboot to take a set of page tables as an optional parameter. If this is NULL then the behaviour is as before, otherwise the new tables are used to remap all of memory apart from a small window around the kernel image (TASK_SIZE - _end). The page immediately below swapper_pg_dir is reserved during boot and can be used as a temporary stack whilst the flat mapping is in place.

---
 arch/arm/include/asm/idmap.h    |    6 +++++-
 arch/arm/kernel/machine_kexec.c |    4 ++--
 arch/arm/kernel/process.c       |    2 +-
 arch/arm/mm/idmap.c             |   38 ++++++++++++++++++++++++++++----------
 arch/arm/mm/mmu.c               |   13 +++++++++++--
 5 files changed, 47 insertions(+), 16 deletions(-)
arch/arm/include/asm/idmap.h +5 -1
index fe7ea86..c2c0576 100644
@@ -2,10 +2,14 @@
 #define _ARM_IDMAP_H
 
 #include <asm/page.h>
+#include <asm/sections.h>
 
 void identity_mapping_add(pgd_t *pgd, unsigned long addr, unsigned long end);
 void identity_mapping_del(pgd_t *pgd, unsigned long addr, unsigned long end);
 
-void setup_mm_for_reboot(char mode);
+/* Page reserved below swapper. */
+#define RESERVE_STACK_PAGE	(unsigned long)swapper_pg_dir
+
+void setup_mm_for_reboot(char mode, pgd_t *pgd);
 
 #endif	/* _ARM_IDMAP_H */
arch/arm/kernel/machine_kexec.c +2 -2
index e59bbd4..dfc1784 100644
@@ -16,7 +16,7 @@
 extern const unsigned char relocate_new_kernel[];
 extern const unsigned int relocate_new_kernel_size;
 
-extern void setup_mm_for_reboot(char mode);
+extern void setup_mm_for_reboot(char mode, pgd_t *pgd);
 
 extern unsigned long kexec_start_address;
 extern unsigned long kexec_indirection_page;
@@ -113,7 +113,7 @@ void machine_kexec(struct kimage *image)
 		kexec_reinit();
 	local_irq_disable();
 	local_fiq_disable();
-	setup_mm_for_reboot(0); /* mode is not used, so just pass 0*/
+	setup_mm_for_reboot(0, NULL); /* mode is not used, so just pass 0*/
 	flush_cache_all();
 	outer_flush_all();
 	outer_disable();
arch/arm/kernel/process.c +1 -1
index 89fe6ee..84228ec 100644
@@ -102,7 +102,7 @@ void arm_machine_restart(char mode, const char *cmd)
 	 * we may need it to insert some 1:1 mappings so that
 	 * soft boot works.
 	 */
-	setup_mm_for_reboot(mode);
+	setup_mm_for_reboot(mode, NULL);
 
 	/* Clean and invalidate caches */
 	flush_cache_all();
arch/arm/mm/idmap.c +28 -10
index e99dc38..3d83ce4 100644
@@ -79,17 +79,35 @@ void identity_mapping_del(pgd_t *pgd, unsigned long addr, unsigned long end)
 #endif
 
 /*
- * In order to soft-boot, we need to insert a 1:1 mapping in place of
- * the user-mode pages.  This will then ensure that we have predictable
- * results when turning the mmu off
+ * In order to soft-boot, we need to insert a 1:1 mapping of memory.
+ * This will then ensure that we have predictable results when turning
+ * the mmu off.
  */
-void setup_mm_for_reboot(char mode)
+void setup_mm_for_reboot(char mode, pgd_t *pgd)
 {
-	/*
-	 * We need to access to user-mode page tables here. For kernel threads
-	 * we don't have any user-mode mappings so we use the context that we
-	 * "borrowed".
-	 */
-	identity_mapping_add(current->active_mm->pgd, 0, TASK_SIZE);
+	unsigned long kernel_end;
+
+	/* If we don't have a pgd, hijack the current task. */
+	if (pgd == NULL) {
+		pgd = current->active_mm->pgd;
+		identity_mapping_add(pgd, 0, TASK_SIZE);
+	} else {
+		identity_mapping_add(pgd, 0, TASK_SIZE);
+		/*
+		 * Extend the flat mapping into kernelspace.
+		 * We leave room for the kernel image and the reserved
+		 * page below swapper.
+		 */
+		kernel_end = ALIGN((unsigned long)_end, PMD_SIZE);
+		identity_mapping_add(pgd, kernel_end, 0);
+	}
+
+	/* Clean and invalidate L1. */
+	flush_cache_all();
+
+	/* Switch exclusively to kernel mappings. */
+	cpu_switch_mm(pgd, &init_mm);
+
+	/* Flush the TLB. */
 	local_flush_tlb_all();
 }
arch/arm/mm/mmu.c +11 -2
index 594d677..e664cbd 100644
@@ -894,18 +894,27 @@ static inline void prepare_page_table(void)
  */
 void __init arm_mm_memblock_reserve(void)
 {
+	phys_addr_t swapper_pa = __pa(swapper_pg_dir);
+
 	/*
 	 * Reserve the page tables.  These are already in use,
 	 * and can only be in node 0.
 	 */
-	memblock_reserve(__pa(swapper_pg_dir), PTRS_PER_PGD * sizeof(pgd_t));
+	memblock_reserve(swapper_pa, PTRS_PER_PGD * sizeof(pgd_t));
 
 #ifdef CONFIG_SA1111
 	/*
 	 * Because of the SA1111 DMA bug, we want to preserve our
 	 * precious DMA-able memory...
 	 */
-	memblock_reserve(PHYS_OFFSET, __pa(swapper_pg_dir) - PHYS_OFFSET);
+	memblock_reserve(PHYS_OFFSET, swapper_pa - PHYS_OFFSET);
+#else
+	/*
+	 * Reserve the page immediately below swapper to use as a
+	 * temporary stack and a holding area for secondary CPUs when we
+	 * are kexec'd.
+	 */
+	memblock_reserve(swapper_pa - PAGE_SIZE, PAGE_SIZE);
 #endif
 }
 
Signature · 1 line
1.7.0.4
#16 lore raw

PATCH v4 6/8 ARM: reset: add reset functionality for jumping to a physical address

In reply to Will Deacon

Tools such as kexec and CPU hotplug require a way to reset the processor and branch to some code in physical space. This requires various bits of jiggery pokery with the caches and MMU which, when it goes wrong, tends to lock up the system.

This patch implements a new function, arm_machine_reset, for consolidating this code in one place where it can be used by multiple subsystems.

---
 arch/arm/include/asm/system.h |    1 +
 arch/arm/kernel/process.c     |   71 +++++++++++++++++++++++++++++++++++++++-
 2 files changed, 70 insertions(+), 2 deletions(-)
arch/arm/include/asm/system.h +1 -0
index 832888d..cd2a3cd 100644
@@ -108,6 +108,7 @@ extern int cpu_architecture(void);
 extern void cpu_init(void);
 
 void arm_machine_restart(char mode, const char *cmd);
+void arm_machine_reset(unsigned long reset_code_phys);
 extern void (*arm_pm_restart)(char str, const char *cmd);
 
 #define UDBG_UNDEFINED	(1 << 0)
arch/arm/kernel/process.c +69 -2

95 lines not on this page · raw message

Signature · 1 line
1.7.0.4
#17 lore raw

PATCH v4 7/8 ARM: kexec: use arm_machine_reset for branching to the reboot buffer

In reply to Will Deacon

Now that there is a common way to reset the machine, let's use it instead of reinventing the wheel in the kexec backend.

---
 arch/arm/kernel/machine_kexec.c |   14 ++------------
 1 files changed, 2 insertions(+), 12 deletions(-)
arch/arm/kernel/machine_kexec.c +2 -12
index dfc1784..f84567c 100644
@@ -16,8 +16,6 @@
 extern const unsigned char relocate_new_kernel[];
 extern const unsigned int relocate_new_kernel_size;
 
-extern void setup_mm_for_reboot(char mode, pgd_t *pgd);
-
 extern unsigned long kexec_start_address;
 extern unsigned long kexec_indirection_page;
 extern unsigned long kexec_mach_type;
@@ -111,14 +109,6 @@ void machine_kexec(struct kimage *image)
 
 	if (kexec_reinit)
 		kexec_reinit();
-	local_irq_disable();
-	local_fiq_disable();
-	setup_mm_for_reboot(0, NULL); /* mode is not used, so just pass 0*/
-	flush_cache_all();
-	outer_flush_all();
-	outer_disable();
-	cpu_proc_fin();
-	outer_inv_all();
-	flush_cache_all();
-	cpu_reset(reboot_code_buffer_phys);
+
+	arm_machine_reset(reboot_code_buffer_phys);
 }
Signature · 1 line
1.7.0.4
#18 lore raw

PATCH v4 8/8 ARM: stop: execute platform callback from cpu_stop code

In reply to Will Deacon

Sending IPI_CPU_STOP to a CPU causes it to execute a busy cpu_relax loop forever. This makes it impossible to kexec successfully on an SMP system since the secondary CPUs do not reset.

This patch adds a callback to platform_cpu_kill, defined when CONFIG_HOTPLUG_CPU=y, from the ipi_cpu_stop handling code. This function currently just returns 1 on all platforms that define it but allows them to do something more sophisticated in the future.

---
 arch/arm/Kconfig      |    2 +-
 arch/arm/kernel/smp.c |    4 ++++
 2 files changed, 5 insertions(+), 1 deletions(-)
arch/arm/Kconfig +1 -1
index 5ebc5d9..d2e9d42 100644
@@ -1849,7 +1849,7 @@ config XIP_PHYS_ADDR
 
 config KEXEC
 	bool "Kexec system call (EXPERIMENTAL)"
-	depends on EXPERIMENTAL
+	depends on EXPERIMENTAL && (!SMP || HOTPLUG_CPU)
 	help
 	  kexec is a system call that implements the ability to shutdown your
 	  current kernel, and to start another kernel.  It is like a reboot
arch/arm/kernel/smp.c +4 -0
index 96a7956..3f6e8d1 100644
@@ -559,6 +559,10 @@ static void ipi_cpu_stop(unsigned int cpu)
 	local_fiq_disable();
 	local_irq_disable();
 
+#ifdef CONFIG_HOTPLUG_CPU
+	platform_cpu_kill(cpu);
+#endif
+
 	while (1)
 		cpu_relax();
 }
Signature · 1 line
1.7.0.4
#19 lore raw

v4 cover MMU disabling code and kexec fixes

In reply to Will Deacon
Quoted 8 lines · Will Deacon

Hello,

This is version 4 of the patches originally posted here:

v1: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052157.html
v2: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052559.html
v3: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/053252.html

Quoted 8 lines

There are many changes since v3. Notably:
- Support for Thumb-2
- Based on 3.1-rc3 (now that some of the previous patches have been merged)
- The reserved stack page now sits below swapper and is reserved by memblock during memory init. This has the advantage of being at a known physical address for a platform, which will be helpful for non-hotplug SMP booting.

Is this really necessary? Relying on "known physical address" is often going to lead us in situations where that physical address get encoded outside the kernel proper and then we're stuck with it forever. This would be more "reliable" if the interface would be passed that address and if the page was dynamically allocated/freed.

Nicolas

#20 lore raw

v4 cover MMU disabling code and kexec fixes

Hi Nicolas,

Thanks for looking at this lot.

Quoted 19 lines · Nicolas Pitre
Quoted 8 lines · Will Deacon

Hello,

This is version 4 of the patches originally posted here:

v1: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052157.html
v2: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052559.html
v3: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/053252.html

Quoted 8 lines

There are many changes since v3. Notably:
- Support for Thumb-2
- Based on 3.1-rc3 (now that some of the previous patches have been merged)
- The reserved stack page now sits below swapper and is reserved by memblock during memory init. This has the advantage of being at a known physical address for a platform, which will be helpful for non-hotplug SMP booting.

Quoted 5 lines

Is this really necessary? Relying on "known physical address" is often going to lead us in situations where that physical address get encoded outside the kernel proper and then we're stuck with it forever. This would be more "reliable" if the interface would be passed that address and if the page was dynamically allocated/freed.

We don't want to allocate this dynamically because we need to avoid the situation where the virtual address of this page aliases with the physical address of the reset code. By keeping the page close to the start of physical memory, we avoid this scenario.

On top of that, I'd like to use the bottom of this page as a holding pen for secondary CPUs when doing an SMP kexec. In this case, the new kernel needs to take care not to allocate the page for general use. I suppose we could fix this with a new ATAG / FDT property but that seems a bit overkill.

I was also thinking that it might be useful if bootloaders want to throw all of the CPUs at the kernel during boot, rather than implementing their own SMP pen. In this case, they could make use of this reserved page (I was planning to version the pen implementation).

The latter idea might simply be crazy, but I think the other points are valid :)

Will

#21 lore raw

v4 cover MMU disabling code and kexec fixes

Quoted 30 lines · Will Deacon

Hi Nicolas,

Thanks for looking at this lot.

Quoted 19 lines · Nicolas Pitre
Quoted 8 lines · Will Deacon

Hello,

This is version 4 of the patches originally posted here:

v1: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052157.html
v2: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052559.html
v3: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/053252.html

Quoted 8 lines

There are many changes since v3. Notably:
- Support for Thumb-2
- Based on 3.1-rc3 (now that some of the previous patches have been merged)
- The reserved stack page now sits below swapper and is reserved by memblock during memory init. This has the advantage of being at a known physical address for a platform, which will be helpful for non-hotplug SMP booting.

Quoted 5 lines

Is this really necessary? Relying on "known physical address" is often going to lead us in situations where that physical address get encoded outside the kernel proper and then we're stuck with it forever. This would be more "reliable" if the interface would be passed that address and if the page was dynamically allocated/freed.

Quoted 4 lines

We don't want to allocate this dynamically because we need to avoid the situation where the virtual address of this page aliases with the physical address of the reset code. By keeping the page close to the start of physical memory, we avoid this scenario.

Well... how probable might this be the case in practice?

And wouldn't this issue easily being worked around simply by allocating a second page if the first one is found to alias? Surely the reset code is much smaller than a page.

Quoted 4 lines

On top of that, I'd like to use the bottom of this page as a holding pen for secondary CPUs when doing an SMP kexec. In this case, the new kernel needs to take care not to allocate the page for general use. I suppose we could fix this with a new ATAG / FDT property but that seems a bit overkill.

Is it?

Quoted 4 lines

I was also thinking that it might be useful if bootloaders want to throw all of the CPUs at the kernel during boot, rather than implementing their own SMP pen. In this case, they could make use of this reserved page (I was planning to version the pen implementation).

I get really uncomfortable at the prospect of having more _implicit_ dependencies between the kernel and bootloaders. If ever we want to move things around for whatever reasons in the future then such fixed legacy addresses might really get in the way. So I don't see the passing of that info using an ATAG or a DT node as something not bad at all. This becomes just another resource provided to the kernel, like the ramdisk image, that can be located almost anywhere. And then you have no backward compatibility issues either as the page validity becomes explicit.

Nicolas

#22 lore raw

v4 cover MMU disabling code and kexec fixes

Quoted 13 lines · Nicolas Pitre
Quoted 11 lines · Will Deacon
Quoted 5 lines · Nicolas Pitre

Is this really necessary? Relying on "known physical address" is often going to lead us in situations where that physical address get encoded outside the kernel proper and then we're stuck with it forever. This would be more "reliable" if the interface would be passed that address and if the page was dynamically allocated/freed.

We don't want to allocate this dynamically because we need to avoid the situation where the virtual address of this page aliases with the physical address of the reset code. By keeping the page close to the start of physical memory, we avoid this scenario.

Quoted 5 lines

Well... how probable might this be the case in practice?

And wouldn't this issue easily being worked around simply by allocating a second page if the first one is found to alias? Surely the reset code is much smaller than a page.

Ok, so we can probably get around this corner case by checking for the alias and reallocating in the next section (the ID mapping is done at the section granularity).

Quoted 6 lines
Quoted 4 lines

On top of that, I'd like to use the bottom of this page as a holding pen for secondary CPUs when doing an SMP kexec. In this case, the new kernel needs to take care not to allocate the page for general use. I suppose we could fix this with a new ATAG / FDT property but that seems a bit overkill.

Is it?

Well it means re-encoding a DT blob from within the kernel so that we can tell the next kernel not to allocate from a given page of physical memory. If we already have this infrastructure, then great, but it looks to me like it only lives within dtc (as libfdt) at the moment.

Quoted 5 lines
Quoted 4 lines

I was also thinking that it might be useful if bootloaders want to throw all of the CPUs at the kernel during boot, rather than implementing their own SMP pen. In this case, they could make use of this reserved page (I was planning to version the pen implementation).

Quoted 9 lines

I get really uncomfortable at the prospect of having more _implicit_ dependencies between the kernel and bootloaders. If ever we want to move things around for whatever reasons in the future then such fixed legacy addresses might really get in the way. So I don't see the passing of that info using an ATAG or a DT node as something not bad at all. This becomes just another resource provided to the kernel, like the ramdisk image, that can be located almost anywhere. And then you have no backward compatibility issues either as the page validity becomes explicit.

Ok, point taken on the bootloader interface, I'd just rather have a way to make this unreliable to bootloaders without having to re-encode the .dtb in the kexec code.

Can you think of any clever tricks so that we only look at this pen when we're being kexec'd?

Cheers,

Will

#23 lore raw

v4 cover MMU disabling code and kexec fixes

Quoted 7 lines · Will Deacon
Quoted 6 lines · Nicolas Pitre
Quoted 4 lines · Will Deacon

On top of that, I'd like to use the bottom of this page as a holding pen for secondary CPUs when doing an SMP kexec. In this case, the new kernel needs to take care not to allocate the page for general use. I suppose we could fix this with a new ATAG / FDT property but that seems a bit overkill.

Quoted 6 lines
Quoted 1 line

Is it?

Well it means re-encoding a DT blob from within the kernel so that we can tell the next kernel not to allocate from a given page of physical memory. If we already have this infrastructure, then great, but it looks to me like it only lives within dtc (as libfdt) at the moment.

Hmmm... The kexec user space tools know how to manipulate a device tree for the next kernel, right? So it's only a matter for it to ask the current kernel what the physical address for that page is and stuff it into the FDT it is preparing already.

Nicolas

#24 lore raw

v4 cover MMU disabling code and kexec fixes

Quoted 16 lines · Nicolas Pitre
Quoted 7 lines · Will Deacon
Quoted 6 lines · Nicolas Pitre
Quoted 4 lines · Will Deacon

On top of that, I'd like to use the bottom of this page as a holding pen for secondary CPUs when doing an SMP kexec. In this case, the new kernel needs to take care not to allocate the page for general use. I suppose we could fix this with a new ATAG / FDT property but that seems a bit overkill.

Quoted 6 lines
Quoted 1 line

Is it?

Well it means re-encoding a DT blob from within the kernel so that we can tell the next kernel not to allocate from a given page of physical memory. If we already have this infrastructure, then great, but it looks to me like it only lives within dtc (as libfdt) at the moment.

Quoted 4 lines

Hmmm... The kexec user space tools know how to manipulate a device tree for the next kernel, right? So it's only a matter for it to ask the current kernel what the physical address for that page is and stuff it into the FDT it is preparing already.

oo-er, it seems a bit odd telling userspace about that address but it does fix the problem. I can't think of a better idea, so I'll start looking for a sane interface for passing this address back.

Thanks for the help,

Will

#25 lore raw

v4 cover MMU disabling code and kexec fixes

Hi Nicolas,

Quoted 9 lines · Will Deacon
Quoted 4 lines · Nicolas Pitre

Hmmm... The kexec user space tools know how to manipulate a device tree for the next kernel, right? So it's only a matter for it to ask the current kernel what the physical address for that page is and stuff it into the FDT it is preparing already.

oo-er, it seems a bit odd telling userspace about that address but it does fix the problem. I can't think of a better idea, so I'll start looking for a sane interface for passing this address back.

I had a look at the kexec-tools source for PPC (unfortunately all the fdt stuff is under there, so adding this for ARM is going to be a pain) and here's what they do:

If a .dtb file is passed to kexec, that is passed to the new kernel. It is fixed up by the kexec tool so that things like the command-line can be modified. There is a device tree property called "cpu-release-addr" which specifies the pen address for each CPU and this property is actually read from the device tree and added to the list of reserved regions so that the next kernel doesn't clobber it.

If a .dtb file is not passed, kexec looks at /proc/device-tree/ and constructs a new blob from that.

Note that PPC doesn't make an attempt to allocate the pen dynamically. If we want to do that, I guess we could emit a dummy "cpu-release-addr" property via /proc/, but this has a couple of downsides:

1.) People might try and use this property (and it would allow bootloaders
to specify pens to non-kexec'd kernels, which would likely end in disaster)

2.) We would have to allocate the pen during boot and leave it allocated
for any CONFIG_KEXEC kernels.

Thinking about the concerns you had about bootloaders using the pen if it's at a known physical location, I realised that we don't actually need to populate the pen until a kexec is in progress. Although the pen page will always be reserved, we can just copy the pen code into it on the kexec_prepare path and that way prevent it being abused by anybody else.

So, do you still reckon we should push forward with the dynamic allocation and ATAG/FDT interface or lazily populate at a known address instead?

Thanks,

Will

#26 lore raw

v4 cover MMU disabling code and kexec fixes

In reply to Will Deacon

Hi Will,

Quoted 17 lines · Will Deacon

Hello,

This is version 4 of the patches originally posted here:

v1: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052157.html
v2: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/052559.html
v3: http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/053252.html

There are many changes since v3. Notably:
- Support for Thumb-2
- Based on 3.1-rc3 (now that some of the previous patches have been merged)
- The reserved stack page now sits below swapper and is reserved by memblock during memory init. This has the advantage of being at a known physical address for a platform, which will be helpful for non-hotplug SMP booting.

Quoted 4 lines

Currently, SMP kexec relies on platform_cpu_kill (a CPU hotplug callback) doing the right thing. I'm working on a generic SMP soft reboot implementation but this isn't quite working yet (and my git repo is predictably offline at the moment so I can't provide a link to the patches).

I've tried this on picoxcell and reboot with arm_machine_reset() works nicely, but I couldn't get kexec to work (though it didn't work without your patches...).

With your patches applied I get the following output on booting the new kernel:

[    0.000000] Initializing cgroup subsys cpu
[    0.000000] Linux version 3.1.0-rc3+ (jamiei at pulham) (gcc version 4.5.3 20110223 (prerelease) (GCC) ) #40 PREEMPT Wed Aug 24 13:21:06 BST 2011
[    0.000000] CPU: ARMv6-compatible processor [410fb767] revision 7 (ARMv7), cr=00c5387f
[    0.000000] CPU: VIPT aliasing data cache, unknown instruction cache
[    0.000000] Machine: Picochip picoXcell
[    0.000000] bootconsole [earlycon0] enabled
[    0.000000] Memory policy: ECC disabled, Data cache writeback
[    0.000000] Kernel panic - not syncing: ERROR: Failed to allocate 0x1000 bytes below 0x0.
[    0.000000]
[    0.000000] [<c001368c>] (unwind_backtrace+0x0/0xe0) from [<c02c875c>] (panic+0x68/0x1a0)
[    0.000000] [<c02c875c>] (panic+0x68/0x1a0) from [<c03c5180>] (memblock_alloc_base+0x2c/0x34)
[    0.000000] [<c03c5180>] (memblock_alloc_base+0x2c/0x34) from [<c03be894>] (early_alloc.clone.1+0x10/0x28)
[    0.000000] [<c03be894>] (early_alloc.clone.1+0x10/0x28) from [<c03bf2b4>] (paging_init+0x4c4/0x6d0)
[    0.000000] [<c03bf2b4>] (paging_init+0x4c4/0x6d0) from [<c03bca54>] (setup_arch+0x424/0x780)
[    0.000000] [<c03bca54>] (setup_arch+0x424/0x780) from [<c03bb510>] (start_kernel+0xac/0x2e4)
[    0.000000] [<c03bb510>] (start_kernel+0xac/0x2e4) from [<00008040>] (0x8040)

and the allocation that fails appears to be allocating the vectors page. Without your patches I don't get any output though so it's a definite improvement!

Note that this is a device tree enabled platform so there is no /proc/atags. I'm happy to test any related patches on our platform, but kexec isn't something we really use.

Jamie

#27 lore raw

v4 cover MMU disabling code and kexec fixes

Quoted 1 line · Jamie Iles

Hi Will,

Hello Jamie,

Quoted 9 lines · Will Deacon
Quoted 4 lines

Currently, SMP kexec relies on platform_cpu_kill (a CPU hotplug callback) doing the right thing. I'm working on a generic SMP soft reboot implementation but this isn't quite working yet (and my git repo is predictably offline at the moment so I can't provide a link to the patches).

I've tried this on picoxcell and reboot with arm_machine_reset() works nicely, but I couldn't get kexec to work (though it didn't work without your patches...).

Ok, thanks for giving it a go though!

Quoted 20 lines

With your patches applied I get the following output on booting the new kernel:

[    0.000000] Initializing cgroup subsys cpu
[    0.000000] Linux version 3.1.0-rc3+ (jamiei at pulham) (gcc version 4.5.3 20110223 (prerelease) (GCC) ) #40 PREEMPT Wed Aug 24 13:21:06 BST 2011
[    0.000000] CPU: ARMv6-compatible processor [410fb767] revision 7 (ARMv7), cr=00c5387f
[    0.000000] CPU: VIPT aliasing data cache, unknown instruction cache
[    0.000000] Machine: Picochip picoXcell
[    0.000000] bootconsole [earlycon0] enabled
[    0.000000] Memory policy: ECC disabled, Data cache writeback
[    0.000000] Kernel panic - not syncing: ERROR: Failed to allocate 0x1000 bytes below 0x0.
[    0.000000]
[    0.000000] [<c001368c>] (unwind_backtrace+0x0/0xe0) from [<c02c875c>] (panic+0x68/0x1a0)
[    0.000000] [<c02c875c>] (panic+0x68/0x1a0) from [<c03c5180>] (memblock_alloc_base+0x2c/0x34)
[    0.000000] [<c03c5180>] (memblock_alloc_base+0x2c/0x34) from [<c03be894>] (early_alloc.clone.1+0x10/0x28)
[    0.000000] [<c03be894>] (early_alloc.clone.1+0x10/0x28) from [<c03bf2b4>] (paging_init+0x4c4/0x6d0)
[    0.000000] [<c03bf2b4>] (paging_init+0x4c4/0x6d0) from [<c03bca54>] (setup_arch+0x424/0x780)
[    0.000000] [<c03bca54>] (setup_arch+0x424/0x780) from [<c03bb510>] (start_kernel+0xac/0x2e4)
[    0.000000] [<c03bb510>] (start_kernel+0xac/0x2e4) from [<00008040>] (0x8040)
Quoted 3 lines

and the allocation that fails appears to be allocating the vectors page. Without your patches I don't get any output though so it's a definite improvement!

Wahey, that's not bad. If you're starting to boot the new kernel then lots of the tricky stuff is out of the way.

Quoted 3 lines

Note that this is a device tree enabled platform so there is no /proc/atags. I'm happy to test any related patches on our platform, but kexec isn't something we really use.

This is almost certainly the problem as the ARM kexec code doesn't have FDT support afaict, so the new kernel doesn't think it has any memory. Can you try hardcoding a command line into the kernel image which specifies the memory layout?

I think the way forward with the kexec work is for me to get it working solidly on UP platforms (including Thumb-2 and FDT) and then add SMP support later on.

Will

#28 lore raw

v4 cover MMU disabling code and kexec fixes

Quoted 1 line · Will Deacon

On Wed, Aug 24, 2011 at 01:34:31PM +0100, Jamie Iles wrote:

[...]

Quoted 8 lines
Quoted 3 lines

Note that this is a device tree enabled platform so there is no /proc/atags. I'm happy to test any related patches on our platform, but kexec isn't something we really use.

This is almost certainly the problem as the ARM kexec code doesn't have FDT support afaict, so the new kernel doesn't think it has any memory. Can you try hardcoding a command line into the kernel image which specifies the memory layout?

That does the trick. It now gets to the point where the platform code starts looking for platform specific parts of the device tree (which it can't find) so I'd call that a success.

Feel free to add a Tested-by: Jamie Iles <[email protected]> if you like!

Jamie