Re: [PATCH v3 00/15] Bring suspend to RAM support to PCIe Aardvark driver

[Lorenzo Pieralisi]

On Fri, Jan 25, 2019 at 01:57:57PM +0100, Miquel Raynal wrote:
> Hi Lorenzo,
> 
> Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx> wrote on Fri, 25 Jan 2019
> 12:40:11 +0000:
> 
> > On Fri, Jan 25, 2019 at 11:05:30AM +0100, Miquel Raynal wrote:
> > > Hi Lorenzo,
> > > 
> > > Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx> wrote on Wed, 23 Jan 2019
> > > 17:05:09 +0000:
> > >   
> > > > On Tue, Jan 08, 2019 at 05:24:25PM +0100, Miquel Raynal wrote:  
> > > > > Hello,
> > > > > 
> > > > > As part of an effort to bring suspend to RAM support to Armada 3700
> > > > > SoCs (main target: ESPRESSObin), this series handles the work around
> > > > > the PCIe IP.
> > > > > 
> > > > > First, more configuration is done in the 'setup' helper as inspired
> > > > > from the U-Boot driver. This is needed to entirely initialize the IP
> > > > > during future resume operation (patch 1).
> > > > > 
> > > > > Then, reset GPIO, PHY and clock support are introduced (patch 2-4). As
> > > > > current device trees do not provide the corresponding properties, not
> > > > > finding one of these properties is not an error and just produces a
> > > > > warning. However, if the property is present, an error during PHY
> > > > > initialization will fail the probe of the driver.
> > > > > 
> > > > > Note: To be sure the clock will be resumed before this driver, a first
> > > > > series adding links between clocks and consumers has been submitted,
> > > > > see [1]. Anyway, having the clock series applied first is not needed.    
> > > > 
> > > > I do not understand what this means, in particular in relation
> > > > to the blocking clock calls in the suspend/resume NOIRQ hooks.  
> > > 
> > > I am not sure to understand your question.
> > > 
> > > As there are multiple points in this sentence I will detail each of
> > > them so please comment on the one which is bothering you:
> > > * I am working in parallel on a series adding device links to the clock
> > >   framework. This way when a driver consumes a clock, the clock
> > >   provider driver will be resumed first.
> > > * If the clock series I am talking about is applied after this one,
> > >   there is no build issue. Of course suspending the platform may
> > >   not work but this is a new feature so nothing will be broken.  
> > 
> > Suspend to RAM will be broken if the clock is suspended and no
> > notification will happen in the NOIRQ phase, it is a new-broken-feature.
> >
> > 
> > > * Device links do not enforce any priority if the suspend/resume phase
> > >   between two drivers is not the same. The PCIe driver suspends in the
> > >   NOIRQ phase. If we want the clock driver to suspend *after* PCIe, its
> > >   suspend/resume callbacks must be promoted to the NOIRQ phase as well
> > >   (and this is part of another series). As of today there is
> > >   no alternative.  
> > 
> > I will merge this series when it works, I have no evidence that it does
> > given what you are writing above, if the series you mention are
> > *necessary* for suspend-to-RAM to work they ought to be merged first.
> 
> I am working actively to bring A3700 SoC suspend to RAM support almost
> from scratch. 
> 
> As of today I have contributed 65 patches spread in 8 series for the
> PHY, clk, PCIe, SATA, USB, pinctrl and net subsystems. Some of them
> have been merged, but the vast majority has not, yet.
> 
> I mentioned this run-time dependency because it exists for people who
> would like to test just the PCIe IP. But S2RAM on A3700 will be a
> new-broken-feature until all patches are merged. While there is
> still one missing, the feature is broken. If everybody waits for
> the other patches to be merged first, it is gonna be a long process :)
> 
> However, if you want to wait for the clock core series to be applied
> first I respect this choice and I will update you when it will be the
> case.

I do not want to merge new code with known issues, it makes no sense.

There are other drivers with a similar problem in the mainline, we
ought to fix those but certainly adding more won't help in that respect.

There is nothing I am saying here that goes against your work, I just
want to have working code in the mainlike kernel, we will merge this
series when it is ready, thanks for bearing with me.

Feel free to ping me/rebase/repost when dependencies are sorted.

Thanks,
Lorenzo