Re: [PATCH 1/2] remoteproc: imx_dsp_rproc: add mandatory find_loaded_rsc_table op

[Iuliana Prodan]

On 9/5/2023 12:16 AM, Mathieu Poirier wrote:
> On Tue, Jul 18, 2023 at 06:40:53PM +0300, Iuliana Prodan wrote:
> > On 7/18/2023 6:23 PM, Mathieu Poirier wrote:
> > > On Tue, Jul 18, 2023 at 11:25:03AM +0300, Iuliana Prodan wrote:
> > > > On 7/17/2023 8:42 PM, Mathieu Poirier wrote:
> > > > > On Thu, Jul 13, 2023 at 01:42:20AM +0300, Iuliana Prodan (OSS) wrote:
> > > > > > From: Iuliana Prodan <iuliana.prodan@xxxxxxx>
> > > > > >
> > > > > > Add the .find_loaded_rsc_table operation for i.MX DSP.
> > > > > > We need it for inter-process communication between DSP
> > > > > > and main core.
> > > > > >
> > > > > > This callback is used to find the resource table (defined
> > > > > > in remote processor linker script) where the address of the
> > > > > > vrings along with the other allocated resources (carveouts etc)
> > > > > > are stored.
> > > > > > If this is not found, the vrings are not allocated and
> > > > > > the IPC between cores will not work.
> > > > > Is there a constraint on the system memory the M4 can address?  If so there
> > > > > will be a need to declare address ranges for vrings and buffers in reserved
> > > > > memory in the DT.
> > > > >
> > > > > Thanks,
> > > > > Mathieu
> > > > Hi Mathieu,
> > > >
> > > > No, there is no constraint on memory.
> > > >
> > > > We want the Cortex A core to communicate with the HiFi4 DSP.
> > > > The Cortex A is in charge of starting the DSP and loading the firmware in
> > > > HiFi4's memory.
> > > > When using rpmsg for IPC, the Cortex A needs to find the resource table
> > > > (defined in the DSP linker script) and this is done using
> > > > .find_loaded_rsc_table callback.
> > > >
> > > > For the DT, we are using a (not upstream) device tree where we have the
> > > > reserved-memory for dsp_vdev0vring0, dsp_vdev0vring1 and dsp_vdev0buffer.
> > > That is the part I'm interested in.  Don't we need the reserved-memory entries?
> > > Otherwise the M4 may not be able to access the memory chosen by the application
> > > processor, most likely leading to a crash.
> > This kernel module (imx_dsp_rproc) is used only for DSP.
> > For M4 core we use imx_rproc.
> The point here is that if I merge this patch and someone tries to load a
> firmware image that has a resource table, the system will likely crash because
> reserved memories haven't been specified in the DT.
>
> Unless there is a very good reason not to, I would like to see the companion DT
> changes submitted with this patch so that the feature is complete.
TBH, so far, I haven't realized that those reserved-memory nodes are not 
upstream
I see they are, as examples, in fsl,dsp.yaml [1]

Let me check why is that and I'll get back - hope, with the DT changes.

[1] 
https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/dsp/fsl,dsp.yaml#L170

> > Iulia
> >
> > > > Iulia
> > > >
> > > >
> > > > > > Signed-off-by: Iuliana Prodan <iuliana.prodan@xxxxxxx>
> > > > > > ---
> > > > > >     drivers/remoteproc/imx_dsp_rproc.c | 1 +
> > > > > >     1 file changed, 1 insertion(+)
> > > > > >
> > > > > > diff --git a/drivers/remoteproc/imx_dsp_rproc.c b/drivers/remoteproc/imx_dsp_rproc.c
> > > > > > index d95fa5586189..b5634507d953 100644
> > > > > > --- a/drivers/remoteproc/imx_dsp_rproc.c
> > > > > > +++ b/drivers/remoteproc/imx_dsp_rproc.c
> > > > > > @@ -941,6 +941,7 @@ static const struct rproc_ops imx_dsp_rproc_ops = {
> > > > > >     	.kick		= imx_dsp_rproc_kick,
> > > > > >     	.load		= imx_dsp_rproc_elf_load_segments,
> > > > > >     	.parse_fw	= imx_dsp_rproc_parse_fw,
> > > > > > +	.find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
> > > > > >     	.sanity_check	= rproc_elf_sanity_check,
> > > > > >     	.get_boot_addr	= rproc_elf_get_boot_addr,
> > > > > >     };
> > > > > > --
> > > > > > 2.17.1