[PATCH v3] memory-hotplug: fix store_mem_state() return value

[PATCH v3] memory-hotplug: fix store_mem_state() return value

[Reza Arbab]

If store_mem_state() is called to online memory which is already online,
it will return 1, the value it got from device_online().

This is wrong because store_mem_state() is a device_attribute .store
function. Thus a non-negative return value represents input bytes read.

Set the return value to -EINVAL in this case.

Signed-off-by: Reza Arbab <arbab@linux.vnet.ibm.com>
---
v2 -> v3:
* David Rientjes pointed out that the backwards-compatible return 
  value in this situation is -EINVAL, not success. I had mistakenly
  thought the behavior should be the same as online_store().

 drivers/base/memory.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/base/memory.c b/drivers/base/memory.c
index 1cea0ba..bb69e58 100644
--- a/drivers/base/memory.c
+++ b/drivers/base/memory.c
@@ -359,8 +359,11 @@ store_mem_state(struct device *dev,
 err:
 	unlock_device_hotplug();
 
-	if (ret)
+	if (ret < 0)
 		return ret;
+	if (ret)
+		return -EINVAL;
+
 	return count;
 }
 
-- 
1.8.3.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH v3] memory-hotplug: fix store_mem_state() return value

[Andrew Morton]

On Thu,  1 Sep 2016 10:29:37 -0500 Reza Arbab <arbab@linux.vnet.ibm.com> wrote:

> If store_mem_state() is called to online memory which is already online,
> it will return 1, the value it got from device_online().
> 
> This is wrong because store_mem_state() is a device_attribute .store
> function. Thus a non-negative return value represents input bytes read.
> 
> Set the return value to -EINVAL in this case.
> 

I actually made the mistake of reading this code.

What the heck are the return value semantics of bus_type.online? 
Sometimes 0, sometimes 1 and apparently sometimes -Efoo values.  What
are these things trying to tell the caller and why is "1" ever useful
and why doesn't anyone document anything.  grr.

And now I don't understand this patch.  Because:

static int memory_subsys_online(struct device *dev)
{
	struct memory_block *mem = to_memory_block(dev);
	int ret;

	if (mem->state == MEM_ONLINE)
		return 0;

Doesn't that "return 0" contradict the changelog?

Also, is store_mem_state() the correct place to fix this?  Instead,
should memory_block_change_state() detect an attempt to online
already-online memory and itself return -EINVAL, and permit that to be
propagated back?  Well, that depends on the bus_type.online rules which
appear to be undocumented.  What is the bus implementation supposed to
do when a request is made to online an already-online device?


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH v3] memory-hotplug: fix store_mem_state() return value

[Reza Arbab]

On Thu, Sep 01, 2016 at 01:37:17PM -0700, Andrew Morton wrote:
>What the heck are the return value semantics of bus_type.online?
>Sometimes 0, sometimes 1 and apparently sometimes -Efoo values.  What
>are these things trying to tell the caller and why is "1" ever useful
>and why doesn't anyone document anything.  grr.

You might be getting tangled in the two codepaths the way I was.

If you do 'echo 1 > online':
	dev_attr_store
		online_store
			device_online
				memory_subsys_online
					memory_block_change_state

If you do 'echo online > state':
	dev_attr_store
		store_mem_state
			device_online
				memory_subsys_online
					memory_block_change_state

>static int memory_subsys_online(struct device *dev)
>{
>	struct memory_block *mem = to_memory_block(dev);
>	int ret;
>
>	if (mem->state == MEM_ONLINE)
>		return 0;
>
>Doesn't that "return 0" contradict the changelog?

The online-to-online check being used is higher in the call chain:

int device_online(struct device *dev)
{
	if (device_supports_offline(dev)) {
		if (dev->offline) {
			...
		} else {
			ret = 1;
		}
	}

>Also, is store_mem_state() the correct place to fix this?  Instead,
>should memory_block_change_state() detect an attempt to online
>already-online memory and itself return -EINVAL, and permit that to be
>propagated back?

Doing that would affect both codepaths, and as David made clear, would 
break backwards compatibility because their established behaviors are 
different.

'echo 1 > online' returns 0 if the device is already online
'echo online > state' returns -EINVAL if the device is already online

-- 
Reza Arbab

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH v3] memory-hotplug: fix store_mem_state() return value

[Xishi Qiu]

On 2016/9/2 4:37, Andrew Morton wrote:

> On Thu,  1 Sep 2016 10:29:37 -0500 Reza Arbab <arbab@linux.vnet.ibm.com> wrote:
> 
>> If store_mem_state() is called to online memory which is already online,
>> it will return 1, the value it got from device_online().
>>
>> This is wrong because store_mem_state() is a device_attribute .store
>> function. Thus a non-negative return value represents input bytes read.
>>
>> Set the return value to -EINVAL in this case.
>>
> 
> I actually made the mistake of reading this code.
> 
> What the heck are the return value semantics of bus_type.online? 
> Sometimes 0, sometimes 1 and apparently sometimes -Efoo values.  What
> are these things trying to tell the caller and why is "1" ever useful
> and why doesn't anyone document anything.  grr.
> 
> And now I don't understand this patch.  Because:
> 
> static int memory_subsys_online(struct device *dev)
> {
> 	struct memory_block *mem = to_memory_block(dev);
> 	int ret;
> 
> 	if (mem->state == MEM_ONLINE)
> 		return 0;
> 

I think we will not execute here, it will return from device_online(),
because "if (dev->offline)" is false and return 1.

But the two return vaules are different if we do online-to-online.
memory_subsys_online() return 0, and device_online() return 1,
this is a little confusion.

When device_online() return 1, online_store() return 1 and store_mem_state()
return -EINVAL even without this patch, as Reza described in v2.

1. store_mem_state() called with buf="online"
2. device_online() returns 1 because device is already online
3. store_mem_state() returns 1
4. calling code interprets this as 1-byte buffer read
5. store_mem_state() called again with buf="nline"
6. store_mem_state() returns -EINVAL

Thanks,
Xishi Qiu

> Doesn't that "return 0" contradict the changelog?
> 
> Also, is store_mem_state() the correct place to fix this?  Instead,
> should memory_block_change_state() detect an attempt to online
> already-online memory and itself return -EINVAL, and permit that to be
> propagated back?  Well, that depends on the bus_type.online rules which
> appear to be undocumented.  What is the bus implementation supposed to
> do when a request is made to online an already-online device?
> 
> 
> 
> .
> 



--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>