Date:   Wed, 31 Aug 2016 18:05:07 +0200
From:   loic pallardy <loic.pallardy@...com>
To:     Suman Anna <s-anna@...com>,
        Bjorn Andersson <bjorn.andersson@...aro.org>
CC:     <ohad@...ery.com>, <lee.jones@...aro.org>,
        <linux-remoteproc@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 2/2] remoteproc: core: Add fixed memory region support



On 08/31/2016 01:13 AM, Suman Anna wrote:
>
Hi Suman,
Hi Loic, Bjorn,
>
> On 08/29/2016 03:09 AM, loic pallardy wrote:
>>
>>
>> On 08/27/2016 02:32 AM, Bjorn Andersson wrote:
>>> On Fri 26 Aug 13:19 PDT 2016, Loic Pallardy wrote:
>>>
>>>> Some coprocessors request fixed memory mapping for firmware execution
>>>> and associated communication linked.
>>>> Memory resources are defined in firmware resource table.
>>>> Resource address different from 0x0 and 0xFFFFFFFF is considered as
>>>> predefined
>>>
>>> Do you think we're required to support both 0 and -1 for this?
>> Hi Bjorn,
>> You're right, only -1 is needed. SoC can have internal RAM in 0x0 for
>> example.
>> I'll update in a V2.
>>>
>>>> and already reserved at system level.
>>>> In that case, remoteproc core doesn't need to perform any allocation.
>>>> Memory region access can be managed using memremap/memunmap functions
>>>>
>>>> Signed-off-by: Loic Pallardy <loic.pallardy@...com>
>>>> ---
>>>>  drivers/remoteproc/remoteproc_core.c | 61
>>>> ++++++++++++++++++++++++++----------
>>>>  include/linux/remoteproc.h           |  4 +++
>>>>  2 files changed, 49 insertions(+), 16 deletions(-)
>>>>
>>>> diff --git a/drivers/remoteproc/remoteproc_core.c
>>>> b/drivers/remoteproc/remoteproc_core.c
>>>> index 18f4286..0ddbb92 100644
>>>> --- a/drivers/remoteproc/remoteproc_core.c
>>>> +++ b/drivers/remoteproc/remoteproc_core.c
>>>> @@ -213,13 +213,25 @@ int rproc_alloc_vring(struct rproc_vdev *rvdev,
>>>> int i)
>>>>      /* actual size of vring (in bytes) */
>>>>      size = PAGE_ALIGN(vring_size(rvring->len, rvring->align));
>>>>
>>>> -    /*
>>>> -     * Allocate non-cacheable memory for the vring. In the future
>>>> -     * this call will also configure the IOMMU for us
>>>> -     */
>>>> -    va = dma_alloc_coherent(dev->parent, size, &dma, GFP_KERNEL);
>>>> +    rsc = (void *)rproc->table_ptr + rvdev->rsc_offset;
>>>> +
>>>> +    /* check if specific memory region requested by firmware */
>>>> +    if (rsc->vring[i].da != 0 && rsc->vring[i].da != FW_RSC_ADDR_ANY) {
>>>
>>> I think we should convert that reserved field in the vring to a "pa";
>>> allowing this resource to not be 1:1 mapped into the remote. And if
>>> nothing else just to be consistent with the carveouts and devmem.
>> In fact vring doesn't have pa because coprocessor diretly access it
>> without help of hardware accelerator. On both carveout and devmem,
>> hardware accelerators may be used.
>> That's true having pa field will be more consistent from host pov.
>
> I agree, and I actually have a need for the pa/dma address without
> disturbing the da as well.
>>
>> Regards,
>> Loic
>>
>>>
>>> @Suman, do you have any input on this?
>
> I was thinking about this as well, and the way I actually envisioned
> this is to add additional rproc_ops with the default behavior falling
> back to the dma_alloc API. I had two use-cases in mind for that - one is
> the same as what Loic is trying to resolve here, and the other is a case
> where I want to allocate these memories not through DMA API, but like
> say allocate from an remote processor internal RAM or an on-chip
> internal memory. This is the case atleast for vrings and vring buffers.
> I think these decisions are best made in the individual platform drivers
> as the integration can definitely vary from one SoC to another.
>
Same use cases on ST side.
Indeed we can create some new ops to alloc and release memory chunk at 
driver level.
Driver will be free to use reserved memeory framework or simply memremap 
chunks.
I'll propose a patch for that.

> The other thing this series makes an assumption is that with a fixed da,
> it is assuming the device is not behind an MMU, and whatever da is
> pointing to is a bus accessible address. We have traditional meant the
> da as "device address" so it translated as bus address on devices that
> are not behind an MMU, or actual virtual addresses as seen by the device
> if behind an MMU. On TI SoCs on some devices, we do have an MMU and so
> we have a non (-1) da, but it is not valid for memremapping.
> At the same time, we would also need any allocated address to be filled in.
>
Today da is used because there is no other field which can be used to 
fix this resource. But agree da is only for device address and pa is 
missing.
If all agree I'll convert reserved field in pa.

Thanks for your review.
Regards,
Loic
> regards
> Suman
>

Powered by blists - more mailing lists