Re: [RFC PATCH 0/3] watchdog servicing during decompression

[Russell King - ARM Linux admin]

On Thu, Oct 17, 2019 at 02:34:52PM +0200, Rasmus Villemoes wrote:
> On 17/10/2019 14.03, Russell King - ARM Linux admin wrote:
> > We used to have this on ARM - it was called from the decompressor
> > code via an arch_decomp_wdog() hook.
> > 
> > That code got removed because it is entirely unsuitable for a multi-
> > platform kernel.  This looks like it takes an address for the watchdog
> > from the Kconfig, and builds that into the decompressor, making the
> > decompressor specific to that board or platform.
> > 
> > I'm not sure distros are going to like that given where we are with
> > multiplatform kernels.
> 
> This is definitely not for multiplatform kernels or general distros,
> it's for kernels that are built as part of a BSP for a specific board -
> hence the "Say N unless you know you need this.".
> 
> I didn't know it used to exist. But I do know that something like this
> is carried out-of-tree for lots of boards, so I thought I'd try to get
> upstream support.

Sorry, it does still exist, just been moved around a bit.

See lib/inflate.c:

STATIC int INIT inflate(void)
{
...
#ifdef ARCH_HAS_DECOMP_WDOG
    arch_decomp_wdog();
#endif

Given that it still exists, maybe this hook name should be used for
this same issue in the LZ4 code?

> The first two patches, or something like them, would be nice on their
> own, as that would minimize the conflicts when forward-porting the
> board-specific patch. But such a half-implemented feature that requires
> out-of-tree patches to actually do anything useful of course won't fly.
> 
> I'm not really a big fan of the third patch, even though it does work
> for all the cases I've encountered so far - I'm sure there would be
> boards where a much more complicated solution would be needed. Another
> method I thought of was to just supply a __weak no-op
> decompress_keepalive(), and then have a config option to point at an
> extra object file to link into the compressed/vmlinux (similar to the
> initramfs_source option that also points to some external resource).
> 
> But if the mainline kernel doesn't want anything like this
> re-introduced, that's also fine, that just means a bit of job security.

Well, we'll see whether the arm-soc people have anything to add...

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up