Re: [PATCH] RFC: Universal scan proposal

[Dmitry Shmidt <dimitrysh@google.com>]

On Wed, Nov 23, 2016 at 12:43 AM, Arend Van Spriel
<arend.vanspriel@broadcom.com> wrote:
> On 22-11-2016 21:54, Dmitry Shmidt wrote:
>> On Tue, Nov 22, 2016 at 12:41 PM, Arend Van Spriel
>> <arend.vanspriel@broadcom.com> wrote:
>>> On 22-11-2016 18:29, Dmitry Shmidt wrote:
>>>> Hi Luca,
>>>>
>>>> On Mon, Nov 21, 2016 at 11:24 PM, Luca Coelho <luca@coelho.fi> wrote:
>>>>> Hi Dmitry,
>>>>> On Wed, 2016-11-16 at 22:47 +0000, dimitrysh@google.com wrote:
>>>>>>  From 68a9d37a4c7e9dc7a90a6e922cdea52737a98d66 Mon Sep 17 00:00:00 2001
>>>>>> From: Dmitry Shmidt <dimitrysh@google.com>
>>>>>> Date: Wed, 16 Nov 2016 14:27:26 -0800
>>>>>> Subject: [PATCH] RFC: Universal scan proposal
>>>>>>
>>>>>>    Currently we have sched scan with possibility of various
>>>>>> intervals. We would like to extend it to support also
>>>>>> different types of scan.
>>>>>>    In case of powerful wlan CPU, all this functionality
>>>>>> can be offloaded.
>>>>>>    In general case FW processes additional scan requests
>>>>>> and puts them into queue based on start time and interval.
>>>>>> Once current request is fulfilled, FW adds it (if interval != 0)
>>>>>> again to the queue with proper interval. If requests are
>>>>>> overlapping, new request can be combined with either one before,
>>>>>> or one after, assuming that requests are not mutually exclusive.
>>>>>>    Combining requests is done by combining scan channels, ssids,
>>>>>> bssids and types of scan result. Once combined request was fulfilled
>>>>>> it will be reinserted as two (or three) different requests based on
>>>>>> their type and interval.
>>>>>>    Each request has attribute:
>>>>>> Type: connectivity / location
>>>>>> Report: none / batch / immediate
>>>>>>    Request may have priority and can be inserted into
>>>>>> the head of the queue.
>>>>>>    Types of scans:
>>>>>> - Normal scan
>>>>>> - Scheduled scan
>>>>>> - Hotlist (BSSID scan)
>>>>>> - Roaming
>>>>>> - AutoJoin
>>>>>>
>>>>>> Change-Id: I9f3e4c975784f1c1c5156887144d80fc5a26bffa
>>>>>> Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
>>>>>> ---
>>>>>
>>>>> I like the initiative and I think this is definitely something that can
>>>>> improve concurrent scanning instances.  But IMHO the most important is
>>>>> to discuss the semantics of this change, such as which scans can be
>>>>> combined, who makes the decisions of combining them, how priorities are
>>>>> sorted out etc.  I think the types of scan are not relevant in the
>>>>> nl80211 API, but the characteristics of the scans are.  For instance,
>>>>> "urgent scan" (for initial connection), best-effort scan for roaming...
>>>>> and latency requirements, such as low-latency for location and initial
>>>>> connection and high-latency for scheduled scan.  Then we decided, in
>>>>> the kernel, how to combine and prioritize them according to their
>>>>> characteristics, instead of having to map scan types to these
>>>>> characteristics.
>>>>>
>>>>> What do you think?
>>>>
>>>> 1. Combining scans.
>>>> There are two scenarios in general: combine scans that can be offload
>>>> and scans that can not be offload due to "weak" FW / wlan SoC. In last
>>>> case this approach maybe not attractive at all - non-mobile device
>>>> may not need all these different types of scan.
>>>> In case of offload - it will be FW code decision - I just wanted to propose
>>>> the way how to do this efficiently.
>>>
>>> I think Luca is looking at it as a way to deal with multiple user-space
>>> entities request the device to perform a scan. Ignoring the scan types
>>> on android it can be wifi-hal and wpa_supplicant.
>>>
>>>> 2. Priority - very good point, we need to have it. I am just not sure
>>>> that we need like scale priority - maybe just flag - urgent / not urgent.
>>>> Urgent one will be inserted in queue as is and conflicting request
>>>> should be postponed or combined.
>>>
>>> What if we get another urgent one?
>>
>> We may restrict it only to one urgent scan - what is the use case for
>> two requests?
>
> We don't know I guess so we can restrict to one for now. Just wondered
> reading this.
>
>>>> 3. Scan types - I am not sure I fully understood your question, but
>>>> if the idea is for kernel to decide about type of scan based on its
>>>> characteristics instead of specific type request performance may
>>>> cause confusion to wifi manager.
>>>> However, it would definitely simplify kernel API. Still I am not sure
>>>> if userspace wifi manager will "like" it.
>>>
>>> Actually the idea is to hide the notion of scan type entirely from the
>>> API and kernel code. If you add the scan type as an attribute in the
>>> API, you still need to provide additional attributes for that scan type
>>> so the question is what the scan type itself provides, ie. how does it
>>> affect the scan itself.
>>
>> Right, some scans types can be easily hidden behind scan parameters
>> like scan for SSID or BSSID or maybe even Roaming with list of
>> SSIDs or BSSIDs, or mix... However, scan history for example should
>> be separate, right?
>
> Not sure what scan history means. cfg80211 currently maintains the
> latest information for a given BSS. So I guess for a "scan history"
> request user-space can specify "history depth" parameter. It may become
> a bit memory intensive to retain all BSS information like that so we may
> need to identify the BSS attributes for which historic values are needed
> by user-space. Now what to do when a "scan history" is in progress and a
> "normal scan" is requested with flush flag attribute set?

Scan history is minimized scan results with time stamp usually restricted
to specific channels and maybe BSSIDs. It is definitely possible to
replace it with "scan depth" as you've suggested. I am not sure it is
worth it due to memory consumption. Also it is mostly offloaded feature.
Now, scan for history or scan for something else is just a request.
They can be combined. Results should be separated, but scan is scan.
This is the whole idea behind generic scan framework. If user wants
to flush current scan results - it is totally fine, it is not related
to history.

>>>> 4. There is an interesting question: to separate scan results for
>>>> connection and location or not? The problem is how to "trust" them.
>>>> I mean scan results are scan results, but for location we may decide
>>>> to look for fewer channels and even maybe specific BSSIDs, i.e. not
>>>> full scan results, and we may need to indicate that right now
>>>> scan results are not full in order not to confuse wifi manager.
>
> Another things is that when combining requests, we are combining results
> as well. So if requester A wants to look for SSIDs X and Y and requester
> B wants to look for SSID Z, both requester A and B will get results for
> X, Y, and Z when requests are combined.

We can use filtering attached to any request, so indeed we can combine
results, but we can answer proper values to request.

> Regards,
> Arend