On 02/27/2018 07:36 AM, Miroslav Benes wrote:
> On Fri, 23 Feb 2018, Joe Lawrence wrote:
> 
>> [ ... snip ... ]
>>  
>> +If a livepatch is replaced by a cumulative patch, then only the
>> +callbacks belonging to the cumulative patch will be executed.  This
>> +simplifies the livepatching core for it is the responsibility of the
>> +cumulative patch to safely revert whatever needs to be reverted.  See
>> +Documentation/livepatch/cumulative.txt for more information on such
>> +patches.
> 
> s/cumulative/atomic replace/ almost everywhere?
> 
> 'Documentation/livepatch/cumulative.txt' should be 
> 'Documentation/livepatch/cumulative-patches.txt' and we may rename it 
> atomic-replace-patches.txt. I don't know. Cumulative patches forms a 
> subset of atomic replace patches in my understanding. The feature itself 
> is more general. Even if practically used for cumulative patches only. But 
> it is for you and Petr to decide.

Hi Miroslav,

Thanks for reviewing!

I guess I'm a little confused about the distinction here.

I understood a "cumulative-patch" to mean that it would contain the sum
of all changes.  So instead of this:

  patch 1 = A
+ patch 2 =     B
+ patch 3 =         C
-----------------------
  net     = A + B + C

We can group all of the changes together into a single cumulative-patch
for the same net effect:

  patch 1 = A               -replaced by-
  patch 2 = A + B             -replaced by-
  patch 3 = A + B + C

I assumed this would also mean to include any reverted changes as well.
So in the example above, if change C needed to be reverted, then:

  patch 4 = A + B

and that would still be considered a "cumulative-patch".

In my mind, atomic replace is the mechanism that forces patching to be
cumulative.  Perhaps this is too strict?  Are there other use-cases for
atomic-replace?

>>  Example Use-cases
>>  =================
>>
>> [ ... snip ... ]
>>
>> +Test 11
>> +-------
>> +
>> +A similar test as the previous one, except this time load the second
>> +callback demo module as a cumulative (ie, replacement) patch.  The
>> +livepatching core will only execute klp_object callbacks for the latest
>> +cumulative patch on the patch stack.
>> +
>> +- load livepatch
>> +- load second livepatch (atomic replace)
>> +- disable livepatch
> 
> Not needed.

Good catch.

-- Joe

Reply via email to