[Problem]
efi_pstore creates sysfs files when logging kernel messages to NVRAM.
Currently, the sysfs files are updated in a workqueue which is registered in a write callback.
On the other hand, a situation which users need the sysfs files is just oops case because a system will
be down and they can't access to sysfs files in other cases like panic, reboot or emergency_restart.
Also, if kernel panics due to a bug of workqueue operations and a write callback of efi_pstore is called in
panic case, efi_pstore may fail due to a failure of schedule_work().
And panic_notifier_chain()/emergency_restart() is not kicked if efi_pstore fails.
This may cause user's unwanted result.
[Patch Description]
This patch skips a registration of a workqueue in a write callback in cases other than oops.
Also, a logic of erase callback is freshly created instead of sharing with a write callback because the sysfs
operation is kicked in just oops case in a write callback and an erase callback doesn't work without any change.
By separating a logic to a write callback, an erase callback works correctly.
Signed-off-by: Seiji Aguchi <[email protected]>
---
drivers/firmware/efivars.c | 53 ++++++++++++++++++++++++++++++++++++++++++-
1 files changed, 51 insertions(+), 2 deletions(-)
diff --git a/drivers/firmware/efivars.c b/drivers/firmware/efivars.c
index 8717ea9..4811abb 100644
--- a/drivers/firmware/efivars.c
+++ b/drivers/firmware/efivars.c
@@ -747,7 +747,11 @@ static int efi_pstore_write(enum pstore_type_id type,
spin_unlock_irqrestore(&efivars->lock, flags);
- schedule_work(&efivar_work);
+ /*
+ * The user may want to see an entry for this write in sysfs.
+ */
+ if (reason == KMSG_DUMP_OOPS)
+ schedule_work(&efivar_work);
*id = part;
return ret;
@@ -756,7 +760,52 @@ static int efi_pstore_write(enum pstore_type_id type,
static int efi_pstore_erase(enum pstore_type_id type, u64 id,
struct pstore_info *psi)
{
- efi_pstore_write(type, 0, &id, (unsigned int)id, 0, psi);
+ char stub_name[DUMP_NAME_LEN];
+ efi_char16_t efi_name[DUMP_NAME_LEN];
+ efi_guid_t vendor = LINUX_EFI_CRASH_GUID;
+ struct efivars *efivars = psi->data;
+ struct efivar_entry *entry, *found = NULL;
+ int i;
+ unsigned long flags;
+
+ sprintf(stub_name, "dump-type%u-%llu-", type, id);
+
+ spin_lock_irqsave(&efivars->lock, flags);
+
+ for (i = 0; i < DUMP_NAME_LEN; i++)
+ efi_name[i] = stub_name[i];
+
+ /*
+ * Clean up any entries with the same name
+ */
+
+ list_for_each_entry(entry, &efivars->list, list) {
+ get_var_data_locked(efivars, &entry->var);
+
+ if (efi_guidcmp(entry->var.VendorGuid, vendor))
+ continue;
+ if (utf16_strncmp(entry->var.VariableName, efi_name,
+ utf16_strlen(efi_name)))
+ continue;
+ /* Needs to be a prefix */
+ if (entry->var.VariableName[utf16_strlen(efi_name)] == 0)
+ continue;
+
+ /* found */
+ found = entry;
+ efivars->ops->set_variable(entry->var.VariableName,
+ &entry->var.VendorGuid,
+ PSTORE_EFI_ATTRIBUTES,
+ 0, NULL);
+ }
+
+ if (found)
+ list_del(&found->list);
+
+ spin_unlock_irqrestore(&efivars->lock, flags);
+
+ if (found)
+ efivar_unregister(found);
return 0;
}
-- 1.7.1
On Mon, Aug 20, 2012 at 12:15 PM, Seiji Aguchi <[email protected]> wrote:
> [Problem]
> efi_pstore creates sysfs files when logging kernel messages to NVRAM.
> Currently, the sysfs files are updated in a workqueue which is registered in a write callback.
>
> On the other hand, a situation which users need the sysfs files is just oops case because a system will
> be down and they can't access to sysfs files in other cases like panic, reboot or emergency_restart.
>
> Also, if kernel panics due to a bug of workqueue operations and a write callback of efi_pstore is called in
> panic case, efi_pstore may fail due to a failure of schedule_work().
> And panic_notifier_chain()/emergency_restart() is not kicked if efi_pstore fails.
> This may cause user's unwanted result.
>
> [Patch Description]
> This patch skips a registration of a workqueue in a write callback in cases other than oops.
>
> Also, a logic of erase callback is freshly created instead of sharing with a write callback because the sysfs
> operation is kicked in just oops case in a write callback and an erase callback doesn't work without any change.
> By separating a logic to a write callback, an erase callback works correctly.
I don't understand why this part of the patch is required.
>
> Signed-off-by: Seiji Aguchi <[email protected]>
> ---
> drivers/firmware/efivars.c | 53 ++++++++++++++++++++++++++++++++++++++++++-
> 1 files changed, 51 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/firmware/efivars.c b/drivers/firmware/efivars.c
> index 8717ea9..4811abb 100644
> --- a/drivers/firmware/efivars.c
> +++ b/drivers/firmware/efivars.c
> @@ -747,7 +747,11 @@ static int efi_pstore_write(enum pstore_type_id type,
>
> spin_unlock_irqrestore(&efivars->lock, flags);
>
> - schedule_work(&efivar_work);
> + /*
> + * The user may want to see an entry for this write in sysfs.
> + */
> + if (reason == KMSG_DUMP_OOPS)
> + schedule_work(&efivar_work);
>
> *id = part;
> return ret;
> @@ -756,7 +760,52 @@ static int efi_pstore_write(enum pstore_type_id type,
> static int efi_pstore_erase(enum pstore_type_id type, u64 id,
> struct pstore_info *psi)
> {
> - efi_pstore_write(type, 0, &id, (unsigned int)id, 0, psi);
> + char stub_name[DUMP_NAME_LEN];
> + efi_char16_t efi_name[DUMP_NAME_LEN];
> + efi_guid_t vendor = LINUX_EFI_CRASH_GUID;
> + struct efivars *efivars = psi->data;
> + struct efivar_entry *entry, *found = NULL;
> + int i;
> + unsigned long flags;
> +
> + sprintf(stub_name, "dump-type%u-%llu-", type, id);
> +
> + spin_lock_irqsave(&efivars->lock, flags);
> +
> + for (i = 0; i < DUMP_NAME_LEN; i++)
> + efi_name[i] = stub_name[i];
> +
> + /*
> + * Clean up any entries with the same name
> + */
> +
> + list_for_each_entry(entry, &efivars->list, list) {
> + get_var_data_locked(efivars, &entry->var);
> +
> + if (efi_guidcmp(entry->var.VendorGuid, vendor))
> + continue;
> + if (utf16_strncmp(entry->var.VariableName, efi_name,
> + utf16_strlen(efi_name)))
> + continue;
> + /* Needs to be a prefix */
> + if (entry->var.VariableName[utf16_strlen(efi_name)] == 0)
> + continue;
> +
> + /* found */
> + found = entry;
> + efivars->ops->set_variable(entry->var.VariableName,
> + &entry->var.VendorGuid,
> + PSTORE_EFI_ATTRIBUTES,
> + 0, NULL);
> + }
> +
> + if (found)
> + list_del(&found->list);
> +
> + spin_unlock_irqrestore(&efivars->lock, flags);
> +
> + if (found)
> + efivar_unregister(found);
>
> return 0;
> }
> -- 1.7.1
> > Also, a logic of erase callback is freshly created instead of sharing
> > with a write callback because the sysfs operation is kicked in just oops case in a write callback and an erase callback doesn't work
> without any change.
> > By separating a logic to a write callback, an erase callback works correctly.
>
> I don't understand why this part of the patch is required.
If I have to keep sharing the code of an erase callback with a write callback,
I can change my patch as follows.
I just thought the code would be ieasy to read by separating them.
<snip>
efi_pstore_write()
{
+ /*
+ * The user may want to update when they erases an entry or see it for this write in sysfs in oops case.
+ */
+ if (!size || reason == KMSG_DUMP_OOPS)
+ schedule_work(&efivar_work);
}
<snip>
Seiji