Re: [PATCH v2 1/1] remoteproc: core: probe subdevices before booting coprocessor

From: Bjorn Andersson <bjorn.andersson@linaro.org>
To: loic pallardy <loic.pallardy@st.com>
Cc: ohad@wizery.com, lee.jones@linaro.org, patrice.chotard@st.com,
	linux-remoteproc@vger.kernel.org, kernel@stlinux.com
Subject: Re: [PATCH v2 1/1] remoteproc: core: probe subdevices before booting coprocessor
Date: Tue, 13 Dec 2016 12:24:16 -0800	[thread overview]
Message-ID: <20161213202416.GJ3439@tuxbot> (raw)
In-Reply-To: <aa670a73-757e-9b7c-3fcf-cef5d4e1acc2@st.com>

On Tue 13 Dec 09:22 PST 2016, loic pallardy wrote:

> 
> 
> On 12/11/2016 09:30 PM, loic pallardy wrote:
> >
> >
> >On 12/08/2016 12:23 AM, Bjorn Andersson wrote:
> >>On Wed 07 Dec 12:49 PST 2016, Loic Pallardy wrote:
> >>
> >>>With subdevice support introduction, coprocessor boot sequence has
> >>>changed. Related coprocessor subdevices are now starting after firmware
> >>>and resource table loading and coprocessor boot.
> >>>
> [..]
> >>
> >>
> >>I still believe that resources that are so strictly tied to the state of
> >>the firmware should be declared as part of the remoteproc, not
> >>registered by other devices that might or might not finish their
> >>registration in time - unless those resources can be associated
> >>to the remote dynamically.
> >Yes please drop out this patch. I'll work on firmware side to be more
> >flexible for initialization sequence.
> >
> Hi Bjorn,
> 
> Reworking coprocessor boot sequence to guarantee struct fw_rsc_vdev of
> resource table is accessed only after first kick generated by virtio_rpmsg,
> I found the issue in coming from a patch present in remoteproc pull request
> for v4.10 [1]: remoteproc: Merge table_ptr and cached_table pointers [2]
> 
> This patch disables the capability of rproc subdev to modifiy dynamically
> the resource table after coprocessor boot. rproc_virtio_set function has no
> more effect after rproc_boot.
> 

You're right, the resource table can't be read-only.

> Patch above was not fixing the right issue but just masking it...
> Reverting patch [2] fix it.
> 

Thanks for testing and reporting this, I will prepare a fix/revert of
it.

> As subdevices are booted after rproc (and I agree with your arguments
> mentioned previously), dynamic resource table modifications should be
> possible (of course HW sync between host and rproc is required for rsc
> access).
>
> Moreover reading patch [2] commit header, I think that security memory lock
> down should be an option and not the default. If memory region has to be
> locked down, I agree that all resource table content will be fixed and no
> dynamic modification possible. But that's not valid all today.

The commit message captures your requirement from our discussions at
Linaro Connect, I'm sorry for missing the fact that it doesn't work.

Regards,
Bjorn