kexec fixes and soft restart code

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, call_with_stack:

? ? ? ?void call_with_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.

Reviewed-by Nicolas Pitre <[email protected]>

Signed-off-by Dave Martin <[email protected]>

Signed-off-by Will Deacon <[email protected]>

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 fleshes out the soft_restart implementation so that it branches to the reset code using the identity mapping. This requires us to change to a temporary stack, held within the kernel image as a static array, to avoid conflicting with the new view of memory.

Signed-off-by Will Deacon <[email protected]>

---
?arch/arm/kernel/process.c | ? 51 ++++++++++++++++++++++++++++++++++---------- ?1 files changed, 39 insertions(+), 12 deletions(-)

arch/arm/kernel/process.c +2 -1

index a92ca50..577d092 100644
@@ -92,29 +92,56 @@ static int __init hlt_setup(char *__unused)
 ?__setup("nohlt", nohlt_setup);
 ?__setup("hlt", hlt_setup);
 
-void soft_restart(unsigned long addr)
+extern void call_with_stack(void (*fn)(void *), void *arg, void *sp);
+typedef void (*phys_reset_t)(unsigned long);

arch/arm/kernel/process.c +2 -8

+static void __soft_restart(void *addr)
 ?{
- ? ? ? /* Disable interrupts first */
- ? ? ? local_irq_disable();
- ? ? ? local_fiq_disable();
+ ? ? ? phys_reset_t phys_reset;
 
- ? ? ? /*
- ? ? ? ?* Tell the mm system that we are going to reboot -
- ? ? ? ?* we may need it to insert some 1:1 mappings so that
- ? ? ? ?* soft boot works.
- ? ? ? ?*/

? ? ? ?/* Push out any further dirty data, and ensure cache is empty */ ? ? ? ?flush_cache_all();

- ? ? ? cpu_reset(addr);
+ ? ? ? /* Switch to the identity mapping. */
+ ? ? ? phys_reset = (phys_reset_t)(unsigned long)virt_to_phys(cpu_reset);
+ ? ? ? phys_reset((unsigned long)addr);

Quoted 23 lines · 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 fleshes out the soft_restart implementation so that it branches to the reset code using the identity mapping. This requires us to change to a temporary stack, held within the kernel image as a static array, to avoid conflicting with the new view of memory.

Signed-off-by Will Deacon <[email protected]>

---
?arch/arm/kernel/process.c | ? 51 ++++++++++++++++++++++++++++++++++---------- ?1 files changed, 39 insertions(+), 12 deletions(-)

arch/arm/kernel/process.c +0 -0

index a92ca50..577d092 100644
@@ -92,29 +92,56 @@ static int __init hlt_setup(char *__unused)
 ?__setup("nohlt", nohlt_setup);
 ?__setup("hlt", hlt_setup);
 

arch/arm/kernel/process.c static int __init hlt_setup(char *__unused) · line 106

Quoted 13 lines

arch/arm/kernel/process.c +12 -1

-void soft_restart(unsigned long addr)
+extern void call_with_stack(void (*fn)(void *), void *arg, void *sp);
+typedef void (*phys_reset_t)(unsigned long);
+
+/*
+ * A temporary stack to use for CPU reset. This is static so that we
+ * don't clobber it with the identity mapping. When running with this
+ * stack, any references to the current task *will not work* so you
+ * should really do as little as possible before jumping to your reset
+ * code.
+ */
+#define SOFT_RESTART_STACK_WORDS ? ? ? 32
+static u32 soft_restart_stack[SOFT_RESTART_STACK_WORDS];

Just for ABI stack alignment requirements, do we get the right alignment of this array in memory?

Quoted 8 lines · Will Deacon

arch/arm/kernel/process.c +0 -0

index a92ca50..577d092 100644
@@ -92,29 +92,56 @@ static int __init hlt_setup(char *__unused)
 ?__setup("nohlt", nohlt_setup);
 ?__setup("hlt", hlt_setup);
 

arch/arm/kernel/process.c static int __init hlt_setup(char *__unused) · line 106

Quoted 13 lines

arch/arm/kernel/process.c +12 -1

-void soft_restart(unsigned long addr)
+extern void call_with_stack(void (*fn)(void *), void *arg, void *sp);
+typedef void (*phys_reset_t)(unsigned long);
+
+/*
+ * A temporary stack to use for CPU reset. This is static so that we
+ * don't clobber it with the identity mapping. When running with this
+ * stack, any references to the current task *will not work* so you
+ * should really do as little as possible before jumping to your reset
+ * code.
+ */
+#define SOFT_RESTART_STACK_WORDS ? ? ? 32
+static u32 soft_restart_stack[SOFT_RESTART_STACK_WORDS];

Just for ABI stack alignment requirements, do we get the right alignment of this array in memory?

Quoted 19 lines

arch/arm/kernel/process.c +4 -9

+static void __soft_restart(void *addr)
 ?{
- ? ? ? /* Disable interrupts first */
- ? ? ? local_irq_disable();
- ? ? ? local_fiq_disable();
+ ? ? ? phys_reset_t phys_reset;
 
- ? ? ? /*
- ? ? ? ?* Tell the mm system that we are going to reboot -
- ? ? ? ?* we may need it to insert some 1:1 mappings so that
- ? ? ? ?* soft boot works.
- ? ? ? ?*/
+ ? ? ? /* Take out a flat memory mapping. */
 ? ? ? ?setup_mm_for_reboot();
 
- ? ? ? /* Clean and invalidate caches */
+ ? ? ? /* Clean and invalidate caches. */
 ? ? ? ?flush_cache_all();
 

arch/arm/kernel/process.c static int __init hlt_setup(char *__unused) · line 118

Quoted 3 lines

arch/arm/kernel/process.c +1 -1

- ? ? ? /* Turn off caching */
+ ? ? ? /* Turn off caching. */
 ? ? ? ?cpu_proc_fin();

A few unnecessary comment changes :) (it's not clear in the Linux coding style but many single-line comments don't have a full-stop.

arch/arm/kernel/process.c static int __init hlt_setup(char *__unused)

Quoted 3 lines

arch/arm/kernel/process.c +3 -0

+void soft_restart(unsigned long addr)
+{
+ ? ? ? u32 *stack = soft_restart_stack + SOFT_RESTART_STACK_WORDS;

We could use ARRAY_SIZE() here and just avoid defining SOFT_RESTART_STACK_WORDS.

Quoted 8 lines · Will Deacon

arch/arm/kernel/process.c +0 -0

index 577d092..664d351 100644
@@ -137,6 +137,10 @@ void soft_restart(unsigned long addr)
 	local_irq_disable();
 	local_fiq_disable();
 

arch/arm/kernel/process.c void soft_restart(unsigned long addr) · line 143

Quoted 4 lines

arch/arm/kernel/process.c +4 -0

+	/* Disable the L2 if we're the last man standing. */
+	if (num_online_cpus() == 1)
+		outer_disable();
+

When does this code execute on another cpu? My understanding is that soft_restart() is only called on one CPU so I would think checking the number of online cpus isn't necessary.

Hi Stephen,

Quoted 3 lines · Stephen Boyd

When does this code execute on another cpu? My understanding is that soft_restart() is only called on one CPU so I would think checking the number of online cpus isn't necessary.

Currently, yes, this only executes on one CPU. The code is there in case we want to implement a SMP kexec solution that doesn't require CPU hotplug at some point.

I suppose you could end up in a situation where the reset path leads to a soft reboot and you haven't dealt with the secondaries at this point, but then you have bigger problems to worry about (namely that whatever you jump to after turning the MMU off needs to deal with the secondaries somehow).