[PATCH 00/15] efi_loader: Add support for logging to a buffer

[Simon Glass]

Hi,

On Wed, 20 Nov 2024 at 19:19, Simon Glass <sjg at chromium.org> wrote:
>
> Hi,
>
> On Wed, 20 Nov 2024 at 13:49, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
> >
> > Am 20. November 2024 19:06:33 MEZ schrieb Tom Rini <trini at konsulko.com>:
> > >On Wed, Nov 20, 2024 at 05:55:18PM +0200, Ilias Apalodimas wrote:
> > >> Hi Simon,
> > >>
> > >> On Wed, 20 Nov 2024 at 17:37, Simon Glass <sjg at chromium.org> wrote:
> > >> >
> > >> > HI Ilias,
> > >> >
> > >> > On Fri, 1 Nov 2024 at 05:32, Ilias Apalodimas
> > >> > <ilias.apalodimas at linaro.org> wrote:
> > >> > >
> > >> > > Hi Simon,
> > >> > >
> > >> > > On Thu, 31 Oct 2024 at 20:02, Simon Glass <sjg at chromium.org> wrote:
> > >> > > >
> > >> > > > Hi Ilias,
> > >> > > >
> > >> > > > On Tue, 29 Oct 2024 at 19:32, Ilias Apalodimas
> > >> > > > <ilias.apalodimas at linaro.org> wrote:
> > >> > > > >
> > >> > > > > On Tue, 29 Oct 2024 at 17:45, Simon Glass <sjg at chromium.org> wrote:
> > >> > > > > >
> > >> > > > > > Hi Ilias,
> > >> > > > > >
> > >> > > > > > On Tue, 29 Oct 2024 at 10:58, Ilias Apalodimas
> > >> > > > > > <ilias.apalodimas at linaro.org> wrote:
> > >> > > > > > >
> > >> > > > > > > Hi Simon,
> > >> > > > > > >
> > >> > > > > > > On Mon, 28 Oct 2024 at 14:48, Simon Glass <sjg at chromium.org> wrote:
> > >> > > > > > > >
> > >> > > > > > > > It is a bit of a pain to log EFI boot-services calls at present. The
> > >> > > > > > > > output goes to the console so cannot easily be inspected later. Also it
> > >> > > > > > > > would be useful to be able to store the log and review it later, perhaps
> > >> > > > > > > > after something has gone wrong.
> > >> > > > > > > >
> > >> > > > > > > > This series makes a start on implementing a log-to-buffer feature. It
> > >> > > > > > > > provides a simple 'efidebug log' command to inspect the buffer. For now,
> > >> > > > > > > > only memory allocations are logged.
> > >> > > > > > >
> > >> > > > > > > Why is this problem specific to EFI and no U-Boot in general? Do we
> > >> > > > > > > have a similar machinery for malloc()?
> > >> > > > > >
> > >> > > > > > Mostly because an app can make EFI calls and we want to know what they
> > >> > > > > > are, e.g. to debug them and figure out what might be wrong when
> > >> > > > > > something doesn't boot.
> > >> > > > >
> > >> > > > > EFI_PRINT() has been proven pretty useful for this. I don't personally
> > >> > > > > see the point of adding ~1300 lines of code to replace a print.
> > >> > > > > What would make more sense is teach EFI_PRINT to log errors in a buffer.
> > >> > > >
> > >> > > > Is that a NAK? Please be clear if you are reviewing the code or just
> > >> > > > rejecting the whole idea.
> > >> > >
> > >> > > For the idea, no. But I don't think what's implemented here is what we want.
> > >> > >
> > >> > > To track what EFI services are called, we already have EFI_ENTRY and EFI_EXIT.
> > >> > > Why don't we instead, add a logging service (and we already have
> > >> > > ftrace iirc) and plug it in the macros above?
> > >> > > That would make more sense not to mention way less code.
> > >> >
> > >> > I am wanting to programmatically log and manage what EFI_LOADER does,
> > >> > so that bootstd can present a high-level view of what is going on,
> > >> > e.g. which protocols are used, how much memory is allocated and where.
> > >> > So this is not just about logging text output.
> > >>
> > >> Why the EFI_LOADER only? Bootstd is supposed to cover more cases, so
> > >> why not a generic framework for all boot commands?
> > >
> > >This feels similar to the point I've made elsewhere in this overarching
> > >series, why not do this at existing common points in the code path?
> > >
> >
> > The common code point is the log library. Just add an event there for which the test can register a handler.
> >
> > With a log event you get:
> >
> > function name
> > source location
> > message class
> > message text
> > message priority
> >
> > and all of this with minimal invasiveness.
>
> OK guys, I think I got the message :-)
>
> I'm going to apply this to my tree for now. While I'm at it I think it
> is time to go through my backlog and apply some other things that I'd
> like in there.

One point I didn't mention is that this series allows logging of all
calls, not just the ones that come in from the app. At present there
are quite a few efi_allocate_pages() calls which are entirely
internal.

Regards,
Simon