Re: PSCI version 1.1 and SYSTEM_RESET2

[Sodagudi Prasad]

On 2019-05-01 02:49, Sudeep Holla wrote:
> On Tue, Apr 30, 2019 at 05:07:31PM -0700, Sodagudi Prasad wrote:
> > On 2019-04-30 14:44, Sodagudi Prasad wrote:
> > +Sudeep
> >
> > > Hi Mark/Will,
> > >
> > > I would like to understand whether ARM linux community have plans to
> > > support PSCI version 1.1 or not.
> > > PSCI_1_1 specification introduced support for SYSTEM_RESET2 command
> > > and this new command helps mobile devices to SYSTEM_WARM_RESET
> > > support. Rebooting devices with warm reboot helps to capture the
> > > snapshot of the ram contents for post-mortem analysis.
> >
> > I think, there is a recent discussion from Sudeep for the 
> > SYSTEM_RESET2
> > support.
> > https://patchwork.kernel.org/patch/10884345/
>
> This has landed in -next, and hopefully must appear in v5.2
>
> > Hi Sudeep,
> >
> > I was going through your discussion in the below list -
> > https://lore.kernel.org/lkml/d73d3580-4ec1-a281-4585-5c776fc08c79@xxxxxxxxxx/
> >
> > There is no provision to set up reboot mode dynamically instead kernel
> > command line parameter.
> > Looking for options to reboot device with warm reboot option when 
> > kernel
> > crashed.
> >
> > panic() --> emergency_restart() --> machine_emergency_restart() -->
> > machine_restart(NULL);
> >
> > It would nice if there is a config option to reboot the device either 
> > in
> > warm or cold in the case of kernel panic.
>
> I presume you prefer to do warm boot in case of panic to get a dump of
> the memory to inspect ? If so, is kexec/kdump not the mechanism to
> achieve that ?

Hi Sudeep,

Thanks for your response and sharing details about your patch.
<Sudeep>  If so, is kexec/kdump not the mechanism to achieve that?
Qualcomm is having vendor specific solution to capture ram contents and 
for offline analysis.

> I am just trying to understand the use case. Xilinx asked for the same
> but never got to understand their use case.

Here is the background -
Usually, power off drivers are overriding arm_pm_restart and 
pm_power_off callbacks and registering with reboot notifier with  some 
priority for the reboot operations.  Here is the Qualcomm poweroff 
driver for reference.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/power/reset/msm-poweroff.c

Before vendor chip set specific power off driver is probed, 
arm_pm_restart functions pointer holds the psci_sys_reset function. Once 
vendor power off driver is probed,  vendor drivers can override the 
arm_pm_restart function pointer.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/firmware/psci.c#n562

Once vendor driver is probed, drivers can take care of devices warm or 
hard reset configuration part properly.  But there is a window from 
start_kernel() to vendor specific driver probed, devices are getting 
cold resets even if kernel crashed.  This is due to arm_pm_restart 
points to psci_sys_reset function by default.  Is this problem clear 
now?

Qualcomm downstream kernel has a lot of use cases with respect device 
reset sequence and the downstream driver is much different from upstream 
drivers. I think, the above-mentioned problem is common for all the 
chipset vendors and it is not specific Qualcomm use cases.  I have one 
downstream solution to this problem but thought to bring up this problem 
to the upstream community for a common solution, so that all the vendors 
can use it.

I have modified below flow to avoid cold restart in the case of early 
kernel panic.
panic() --> emergency_restart() --> machine_emergency_restart() --> 
machine_restart(NULL);

-Thanks, Prasad

> --
> Regards,
> Sudeep

--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora 
Forum,
Linux Foundation Collaborative Project