Re: [PATCH 1/1] PM: fix oops in suspend/hibernate code

On 01/06/2011 04:57 PM, Rafael J. Wysocki wrote:

On Thursday, January 06, 2011, Jiri Slaby wrote:

quoted

When ioremap fails (which might happen for some reason),

If it happens, something is seriously wrong (see below).

I agree that something is broken, however ioremap may fail for dozen of
reasons. Ignoring the retval is a *bad* idea and it took me a while to
sort out what is wrong. Especially if one has no console like throughout
suspend. If it was handled properly, I would know immediately. (There
should be a message printed out which I forgot to add.)

BTW, to keep things in context, please post fixes like this in the same thread
in which you reported the problem.  At lease please retain the CC list from
there.

I actually did, there is:
In-Reply-To: [off-list ref]
and it successfully threaded to the conversation for me in TB.

quoted

we nicely oops in suspend_nvs_save due to NULL dereference by memcpy in
there.  Fail gracefully instead.

Signed-off-by: Jiri Slaby <redacted>
Cc: "Rafael J. Wysocki" <redacted>
---
 drivers/acpi/sleep.c    |    5 ++---
 include/linux/suspend.h |    4 ++--
 kernel/power/nvs.c      |    8 +++++++-
 3 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/drivers/acpi/sleep.c b/drivers/acpi/sleep.c
index c423231..f94c9a9 100644
--- a/drivers/acpi/sleep.c
+++ b/drivers/acpi/sleep.c

@@ -124,8 +124,7 @@ static int acpi_pm_freeze(void)
 static int acpi_pm_pre_suspend(void)
 {
 	acpi_pm_freeze();
-	suspend_nvs_save();
-	return 0;
+	return suspend_nvs_save();
 }
 
 /**

@@ -151,7 +150,7 @@ static int acpi_pm_prepare(void)
 {
 	int error = __acpi_pm_prepare();
 	if (!error)
-		acpi_pm_pre_suspend();
+		error = acpi_pm_pre_suspend();
 
 	return error;
 }

diff --git a/include/linux/suspend.h b/include/linux/suspend.h
index c1f4998..3ac2551 100644
--- a/include/linux/suspend.h
+++ b/include/linux/suspend.h

@@ -262,7 +262,7 @@ static inline bool system_entering_hibernation(void) { return false; }
 extern int suspend_nvs_register(unsigned long start, unsigned long size);
 extern int suspend_nvs_alloc(void);
 extern void suspend_nvs_free(void);
-extern void suspend_nvs_save(void);
+extern int suspend_nvs_save(void);
 extern void suspend_nvs_restore(void);
 #else /* CONFIG_SUSPEND_NVS */
 static inline int suspend_nvs_register(unsigned long a, unsigned long b)

@@ -271,7 +271,7 @@ static inline int suspend_nvs_register(unsigned long a, unsigned long b)
 }
 static inline int suspend_nvs_alloc(void) { return 0; }
 static inline void suspend_nvs_free(void) {}
-static inline void suspend_nvs_save(void) {}
+static inline int suspend_nvs_save(void) {}
 static inline void suspend_nvs_restore(void) {}
 #endif /* CONFIG_SUSPEND_NVS */
 

diff --git a/kernel/power/nvs.c b/kernel/power/nvs.c
index 1836db6..57c6fab 100644
--- a/kernel/power/nvs.c
+++ b/kernel/power/nvs.c

@@ -105,7 +105,7 @@ int suspend_nvs_alloc(void)
 /**
  *	suspend_nvs_save - save NVS memory regions
  */
-void suspend_nvs_save(void)
+int suspend_nvs_save(void)
 {
 	struct nvs_page *entry;
 

@@ -114,8 +114,14 @@ void suspend_nvs_save(void)
 	list_for_each_entry(entry, &nvs_list, node)
 		if (entry->data) {
 			entry->kaddr = ioremap(entry->phys_start, entry->size);

I wonder what happens if you simply change the ioremap() here to
ioremap_nocache() without any other modifications?

ioremap *is* ioremap_nocache on x86. And that's the conflict it
complains about I guess? Don't you mean ioremap_cache?

It _really_ shouldn't fail here, because the NVS pages are known to be present.

It fails because of conflicting maps as can be seen in the photo. At
least I think so.

regards,
-- 
js