Date:   Tue, 24 Sep 2019 11:13:31 +0200
From:   David Hildenbrand <david@...hat.com>
To:     Michal Hocko <mhocko@...nel.org>,
        Alastair D'Silva <alastair@...ilva.org>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Oscar Salvador <osalvador@...e.com>,
        Pavel Tatashin <pasha.tatashin@...een.com>,
        Dan Williams <dan.j.williams@...el.com>,
        Wei Yang <richard.weiyang@...il.com>, Qian Cai <cai@....pw>,
        Jason Gunthorpe <jgg@...pe.ca>,
        Logan Gunthorpe <logang@...tatee.com>,
        Ira Weiny <ira.weiny@...el.com>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 1/2] memory_hotplug: Add a bounds check to
 check_hotplug_memory_range()

On 24.09.19 11:09, Michal Hocko wrote:
> On Tue 24-09-19 11:31:05, Alastair D'Silva wrote:
>> On Mon, 2019-09-23 at 14:25 +0200, Michal Hocko wrote:
>>> On Tue 17-09-19 11:07:47, Alastair D'Silva wrote:
>>>> From: Alastair D'Silva <alastair@...ilva.org>
>>>>
>>>> On PowerPC, the address ranges allocated to OpenCAPI LPC memory
>>>> are allocated from firmware. These address ranges may be higher
>>>> than what older kernels permit, as we increased the maximum
>>>> permissable address in commit 4ffe713b7587
>>>> ("powerpc/mm: Increase the max addressable memory to 2PB"). It is
>>>> possible that the addressable range may change again in the
>>>> future.
>>>>
>>>> In this scenario, we end up with a bogus section returned from
>>>> __section_nr (see the discussion on the thread "mm: Trigger bug on
>>>> if a section is not found in __section_nr").
>>>>
>>>> Adding a check here means that we fail early and have an
>>>> opportunity to handle the error gracefully, rather than rumbling
>>>> on and potentially accessing an incorrect section.
>>>>
>>>> Further discussion is also on the thread ("powerpc: Perform a
>>>> bounds
>>>> check in arch_add_memory").
>>>
>>> It would be nicer to refer to this by a message-id based url.
>>> E.g. 
>>> http://lkml.kernel.org/r/20190827052047.31547-1-alastair@au1.ibm.com
>>>
>>
>> Ok.
>>
>>>> Signed-off-by: Alastair D'Silva <alastair@...ilva.org>
>>>> ---
>>>>  include/linux/memory_hotplug.h |  1 +
>>>>  mm/memory_hotplug.c            | 13 ++++++++++++-
>>>>  2 files changed, 13 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/include/linux/memory_hotplug.h
>>>> b/include/linux/memory_hotplug.h
>>>> index f46ea71b4ffd..bc477e98a310 100644
>>>> --- a/include/linux/memory_hotplug.h
>>>> +++ b/include/linux/memory_hotplug.h
>>>> @@ -110,6 +110,7 @@ extern void
>>>> __online_page_increment_counters(struct page *page);
>>>>  extern void __online_page_free(struct page *page);
>>>>  
>>>>  extern int try_online_node(int nid);
>>>> +int check_hotplug_memory_addressable(u64 start, u64 size);
>>>>  
>>>>  extern int arch_add_memory(int nid, u64 start, u64 size,
>>>>  			struct mhp_restrictions *restrictions);
>>>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>>>> index c73f09913165..02cb9a74f561 100644
>>>> --- a/mm/memory_hotplug.c
>>>> +++ b/mm/memory_hotplug.c
>>>> @@ -1030,6 +1030,17 @@ int try_online_node(int nid)
>>>>  	return ret;
>>>>  }
>>>>  
>>>> +int check_hotplug_memory_addressable(u64 start, u64 size)
>>>> +{
>>>> +#ifdef MAX_PHYSMEM_BITS
>>>> +	if ((start + size - 1) >> MAX_PHYSMEM_BITS)
>>>> +		return -E2BIG;
>>>> +#endif
>>>
>>> Is there any arch which doesn't define this? We seemed to be using
>>> this
>>> in sparsemem code without any ifdefs.
>>
>> A few, but none of them would be enabling hotplug (which depends on
>> sparsemem), so you're right, the ifdef could be removed.
>>
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +EXPORT_SYMBOL_GPL(check_hotplug_memory_addressable);
>>>
>>> If you squashed the patch 2 then it would become clear why this needs
>>> to
>>> be exported because you would have a driver user. I find it a bit
>>> unfortunate to expect that any driver which uses the hotplug code is
>>> expected to know that this check should be called. This sounds too
>>> error
>>> prone. Why hasn't been this done at __add_pages layer?
>>>
>>
>> It seemed that is should be a peer of check_hotplug_memory_range(), as
>> it gives similar feedback (whether the provided range is suitable).
> 
> Well, that one seems to do a similar yet a different kind of check. It
> imposes a constrain to the alignment of the memory that is hotplugable
> via add_memory_resource - aka memory with user visible sysfs interface
> and that really has some restrictions on the memory block sizes now.
>  
>> If we did the check in __add_pages, wouldn't we potentially lose bits
>> from the LSBs of start & size, or is there some other requirement that
>> already ensures start & size are always page aligned?
> 
> I do not really think we have to care about page unaligned addresses.
> Callers down the road usually work with pfns. 
> 
>> It appears this patch has been accepted - if we were to make this
>> change, does it go as another spin on this series or a new series?
> 
> yes, the patch has been rushed to Linus unfortunatelly. Although I do
> not really see any reason why. Sigh...
> Anyway, now that it is in Linus' tree then we can only do a follow up on
> top.
> 
>>>> +
>>>>  static int check_hotplug_memory_range(u64 start, u64 size)
>>>>  {
>>>>  	/* memory range must be block size aligned */
>>>> @@ -1040,7 +1051,7 @@ static int check_hotplug_memory_range(u64
>>>> start, u64 size)
>>>>  		return -EINVAL;
>>>>  	}
>>>>  
>>>> -	return 0;
>>>> +	return check_hotplug_memory_addressable(start, size);
>>>
>>> This will result in a silent failure (unlike misaligned case). Is
>>> this
>>> what we want?
>>
>> Good point - I guess it comes down to, is there anything we expect an
>> end user to do about it? I'm not sure there is, in which case the bad
>> RC, which is reported up every call chain that I can see, should be
>> sufficient.
> 
> It seems like a clear HW/platform bug to me. And that should better be
> reported loudly to the log to make sure people do complain to their FW
> friends and have it fixed.
> 

I don't agree in virtual environment. On s390x, MAX_PHYSMEM_BITS is
configurable. For example, if you have paravirtualized memory hotplug
(e.g., virtio-mem), you could add memory to the system that violates
this constraint.

virtio-mem, however, does properly check for MAX_PHYSMEM_BITS itself -
at least in the current RFC v3.

-- 

Thanks,

David / dhildenb

Powered by blists - more mailing lists