Message-ID: <20160907085038.GA30243@e106950-lin.cambridge.arm.com>
Date:   Wed, 7 Sep 2016 09:50:38 +0100
From:   Brian Starkey <brian.starkey@....com>
To:     Laura Abbott <labbott@...hat.com>
Cc:     Sumit Semwal <sumit.semwal@...aro.org>,
        John Stultz <john.stultz@...aro.org>,
        Arve Hjønnevåg <arve@...roid.com>,
        Riley Andrews <riandrews@...roid.com>,
        Daniel Vetter <daniel.vetter@...ll.ch>,
        linaro-mm-sig@...ts.linaro.org, devel@...verdev.osuosl.org,
        linux-kernel@...r.kernel.org,
        Eun Taik Lee <eun.taik.lee@...sung.com>,
        Liviu Dudau <Liviu.Dudau@....com>,
        Jon Medhurst <tixy@...aro.org>,
        Mitchel Humpherys <mitchelh@...eaurora.org>,
        Jeremy Gebben <jgebben@...eaurora.org>,
        Bryan Huntsman <bryanh@...eaurora.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Android Kernel Team <kernel-team@...roid.com>,
        Chen Feng <puck.chen@...ilicon.com>
Subject: Re: [PATCHv2 1/4] staging: android: ion: Drop heap type masks

On Tue, Sep 06, 2016 at 03:16:52PM -0700, Laura Abbott wrote:
>On 09/05/2016 04:20 AM, Brian Starkey wrote:
>>Hi,
>>
>>On Fri, Sep 02, 2016 at 12:36:25PM -0700, Laura Abbott wrote:
>>>On 09/02/2016 06:41 AM, Brian Starkey wrote:
>>>>Hi Laura,
>>>>
>>>>On Thu, Sep 01, 2016 at 03:40:41PM -0700, Laura Abbott wrote:
>>>>>
>>>>>There is no advantage to having heap types be a mask. The ion client has
>>>>>long since dropped the mask. Drop the notion of heap type masks as well.
>>>>>
>>>>
>>>>I know this is the same patch you sent last time, so sorry for not
>>>>picking this up then - but I'm curious what "The" ion client is here?
>>>>
>>>
>>>ion_client_create used to take a mask to indicate what heap types it
>>>could allocate from. This hasn't been the case since 2bb9f5034ec7
>>>("gpu: ion: Remove heapmask from client"). "The ion client" probably
>>>should have been "struct ion_client"
>>
>>Ah I see, the in-kernel ion_client. Sorry, I completely forgot that
>>even existed (because it's totally useless - how is a driver meant to
>>find the global ion_device?)
>>
>>>
>>>>Our ion client(s) certainly still use these masks, and it's still
>>>>used as a mask within ion itself - even if the relationship between a
>>>>mask and a heap type has been somewhat lost.
>>>
>>>Where is it used in Ion? I don't see it in tree unless I missed something
>>>and I'm not eager to keep this around for out of tree code. What's the
>>>actual use for this?
>>
>>You're certainly right that these heap-ID-to-allocation-mask macros
>>are unused in the kernel, but I don't really see the reason for
>>removing them - they are convenient (for now).
>>
>>Example: I'm using the dummy ion driver, and I want to allocate from
>>the SYSTEM_CONTIG heap - the ION_HEAP_SYSTEM_CONTIG_MASK gives me the
>>exact mask I need for that.
>>
>>It seems your opinion is that heap-IDs are already, and should be,
>>completely decoupled from their type. That sounds like a good idea to
>>me, but it's not true (yet) - again check out the dummy driver.
>>
>
>Good point, I need to clean up the dummy driver to stop using heap
>types as the id ;)
>
>I get that it's convenient but it's a bad practice to conflate the
>namespaces.
>
>>At the moment, heap-IDs are assigned by ion drivers in any way they
>>see fit. For as long as that stays the case there's always going to
>>be heap-masks hard-coded in UAPI kernel headers (in-tree or not), so
>>removing these particular masks seems a bit fruitless.
>>
>
>It's not fruitless, the concept of type as mask makes no sense. They
>are two different name spaces and I've found Ion users have a hard
>time keeping them separate and pass in the heap type mask when using
>non dummy

You can add me to that group :-)

>
>>I'd rather see driver-assigned heap-IDs disappear completely, and have
>>them assigned by ion core from an idr or something. At that point
>>these macros really *are* meaningless, and I'd be totally fine with
>>removing them (and userspace won't be able to depend on hard-coded
>>allocation masks any more - it will have to use the query ioctl,
>>which I assume is the whole point?).
>>
>
>Ideally yes we'd be able to get rid of the hard coded device IDs.
>I consider the query ioctl a stepping stone to that, depending on
>how enthusiastic people are about Ion.
>
>>IMO it's not the right time to remove these macros, because they still
>>have meaning and usefulness.
>>
>
>I still think they should be deleted to avoid namespace polution.
>

If you nuke type-as-ID in dummy, and put a clear comment on the
ion_heap_type enum stating that heap-type and heap-ID are strictly
different, then I'm happy.
I think without that there'll still be confusion.

Thanks!
Brian

>>Cheers,
>>Brian
>>
>>>
>>>>
>>>>Thanks,
>>>>Brian
>>>>
>>>>>Signed-off-by: Laura Abbott <labbott@...hat.com>
>>>>>---
>>>>>drivers/staging/android/uapi/ion.h | 6 ------
>>>>>1 file changed, 6 deletions(-)
>>>>>
>>>>>diff --git a/drivers/staging/android/uapi/ion.h b/drivers/staging/android/uapi/ion.h
>>>>>index 0a8e40f..a9c4e8b 100644
>>>>>--- a/drivers/staging/android/uapi/ion.h
>>>>>+++ b/drivers/staging/android/uapi/ion.h
>>>>>@@ -44,14 +44,8 @@ enum ion_heap_type {
>>>>>                  * must be last so device specific heaps always
>>>>>                  * are at the end of this enum
>>>>>                  */
>>>>>-    ION_NUM_HEAPS = 16,
>>>>>};
>>>>>
>>>>>-#define ION_HEAP_SYSTEM_MASK        (1 << ION_HEAP_TYPE_SYSTEM)
>>>>>-#define ION_HEAP_SYSTEM_CONTIG_MASK    (1 << ION_HEAP_TYPE_SYSTEM_CONTIG)
>>>>>-#define ION_HEAP_CARVEOUT_MASK        (1 << ION_HEAP_TYPE_CARVEOUT)
>>>>>-#define ION_HEAP_TYPE_DMA_MASK        (1 << ION_HEAP_TYPE_DMA)
>>>>>-
>>>>>#define ION_NUM_HEAP_IDS        (sizeof(unsigned int) * 8)
>>>>>
>>>>>/**
>>>>>--
>>>>>2.7.4
>>>>>
>>>
>

Powered by blists - more mailing lists