Re: staging: ion: ION allocation fall back order depends on heap linkage order

[Alexey Skidanov <alexey.skidanov@intel.com>]

On 02/07/2018 05:32 PM, Laura Abbott wrote:
> On 02/07/2018 07:10 AM, Alexey Skidanov wrote:
>>
>>
>> On 02/07/2018 04:58 PM, Laura Abbott wrote:
>>> On 02/06/2018 11:05 PM, Alexey Skidanov wrote:
>>>>
>>>>
>>>>> Yup, you've hit upon a key problem. Having fallbacks be stable
>>>>> was always a problem and the recommendation these days is to
>>>>> not rely on them. You can specify a heap at a time and fallback
>>>>> manually if you want that behavior.
>>>>>
>>>>> If you have a proposal to make fallbacks work reliably without
>>>>> overly complicating the ABI I'm happy to review it.
>>>>>
>>>>> Thanks,
>>>>> Laura
>>>>>
>>>> I think it's possible to "automate" the "manual fallback" behavior. But
>>>> the real issues is using heap id to specify the particular heap object.
>>>>
>>>> Current API (allocation IOCTL) requires to specify the particular heap
>>>> object by using heap id. From the other hand, the user space doesn't
>>>> control the heaps creation order and heap id assignment. So it may be
>>>> tricky, especially when more than one object of the same heap type is
>>>> created automatically.
>>>>
>>>> Thanks,
>>>> Alexey
>>>>
>>>>
>>>
>>> The query ioctl is designed to get the heap ID information without
>>> needing to rely on the linking order or anything else defined in
>>> the kernel.
>>>
>>> Thanks,
>>> Laura
>>
>> That is true. But if we have 2 *automatically created* heaps of the same
>> type, how userspace can distinguish between them?
>>
>> Thanks,
>> Alexey
>>
> 
> The query ioctl also gives the name which should be different
> for each heap. It's not ideal but the name/heap type are the best
> way to differentiate between heaps without resorting to hard
> coding.
> 
> Thanks,
> Laura

You are correct ... It will work (assuming that user space developer
knows where to look for the name :) )

So, the userspace may pass the list of pairs [heap type, name] (as part
of allocation ioctl) defining the fallback order.

Thanks,
Alexey

_______________________________________________
devel mailing list
devel@linuxdriverproject.org
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel