Re: [PATCH v2 2/8] remoteproc: qcom: Introduce driver to store pil info in IMEM

[Bjorn Andersson <bjorn.andersson@linaro.org>]

On Wed 22 Jan 11:04 PST 2020, Mathieu Poirier wrote:

> On Tue, 21 Jan 2020 at 19:02, Bjorn Andersson
> <bjorn.andersson@linaro.org> wrote:
> >
> > On Fri 10 Jan 13:18 PST 2020, Mathieu Poirier wrote:
> > > On Thu, Dec 26, 2019 at 09:32:09PM -0800, Bjorn Andersson wrote:
> > [..]
> > > > diff --git a/drivers/remoteproc/qcom_pil_info.c b/drivers/remoteproc/qcom_pil_info.c
> > > > new file mode 100644
> > > > index 000000000000..b0897ae9eae5
> > > > --- /dev/null
> > > > +++ b/drivers/remoteproc/qcom_pil_info.c
> > > > @@ -0,0 +1,150 @@
> > > > +// SPDX-License-Identifier: GPL-2.0-only
> > > > +/*
> > > > + * Copyright (c) 2019 Linaro Ltd.
> > > > + */
> > > > +#include <linux/module.h>
> > > > +#include <linux/kernel.h>
> > > > +#include <linux/of.h>
> > > > +#include <linux/platform_device.h>
> > > > +#include <linux/mutex.h>
> > > > +#include <linux/regmap.h>
> > > > +#include <linux/mfd/syscon.h>
> > > > +#include <linux/slab.h>
> > >
> > > These should be in alphabetical order if there is no depencencies
> > > between them, something checkpatch complains about.
> > >
> >
> > Of course.
> >
> > > > +
> > > > +struct pil_reloc_entry {
> > > > +   char name[8];
> > >
> > > Please add a #define for the name length and reuse it in qcom_pil_info_store()
> > >
> >
> > Ok
> >
> > [..]
> > > > +void qcom_pil_info_store(const char *image, phys_addr_t base, size_t size)
> > > > +{
> > > > +   struct pil_reloc_entry *entry;
> > > > +   int idx = -1;
> > > > +   int i;
> > > > +
> > > > +   mutex_lock(&reloc_mutex);
> > > > +   if (!_reloc)
> > >
> > > Since it is available, I would use function qcom_pil_info_available().  Also
> > > checkpatch complains about indentation problems related to the 'if' condition
> > > but I can't see what makes it angry.
> > >
> >
> > Sure thing, and I'll double check the indentation.
> >
> > > > +           goto unlock;
> > > > +
> > > > +   for (i = 0; i < PIL_INFO_ENTRIES; i++) {
> > > > +           if (!_reloc->entries[i].name[0]) {
> > > > +                   if (idx == -1)
> > > > +                           idx = i;
> > > > +                   continue;
> > > > +           }
> > > > +
> > > > +           if (!strncmp(_reloc->entries[i].name, image, 8)) {
> > > > +                   idx = i;
> > > > +                   goto found;
> > > > +           }
> > > > +   }
> > > > +
> > > > +   if (idx == -1) {
> > > > +           dev_warn(_reloc->dev, "insufficient PIL info slots\n");
> > > > +           goto unlock;
> > >
> > > Given how this function is used in the next patch I think an error should be
> > > reported to the caller.
> > >
> >
> > Just to clarify, certain global errors will cause the entire device to
> > be reset and allow memory contents to be extracted for analysis in post
> > mortem tools. This patch ensures that this information contains
> > (structured) information about where each remote processor is loaded.
> > Afaict the purpose of propagating errors from this function would be for
> > the caller to abort the launching of a remote processor.
> >
> > I think it's better to take the risk of having insufficient data for the
> > post mortem tools than to fail booting a remote processor for a reason
> > that won't affect normal operation.
> 
> I understand the reasoning.  In that case it is probably best to let
> the caller decide what to do with the returned error than deal with it
> locally, especially since this is an exported function.  When using
> qcom_pil_info_store(), I would write a comment that justifies the
> reason for ignoring the return value (what you have above is quite
> good).  Otherwise it is just a matter of time before automated tools
> pickup on the anomaly and send patches to fix it.
> 

You're right, moving the decision to the remoteproc drivers will result
in the decision being implemented in the right place. I will respin it
accordingly.

Thanks!
Bjorn