Re: [PATCH] mm/memory-failure: Disable soft offline for HugeTLB pages by default

[Kyle Meyer <kyle.meyer@hpe.com>]

On Thu, Sep 11, 2025 at 10:56:36AM -0700, Luck, Tony wrote:
> On Thu, Sep 11, 2025 at 10:46:10AM +0200, David Hildenbrand wrote:
> > On 10.09.25 18:15, Kyle Meyer wrote:
> > > Soft offlining a HugeTLB page reduces the available HugeTLB page pool.
> > > Since HugeTLB pages are preallocated, reducing the available HugeTLB
> > > page pool can cause allocation failures.
> > > 
> > > /proc/sys/vm/enable_soft_offline provides a sysctl interface to
> > > disable/enable soft offline:
> > > 
> > > 0 - Soft offline is disabled.
> > > 1 - Soft offline is enabled.
> > > 
> > > The current sysctl interface does not distinguish between HugeTLB pages
> > > and other page types.
> > > 
> > > Disable soft offline for HugeTLB pages by default (1) and extend the
> > > sysctl interface to preserve existing behavior (2):
> > > 
> > > 0 - Soft offline is disabled.
> > > 1 - Soft offline is enabled (excluding HugeTLB pages).
> > > 2 - Soft offline is enabled (including HugeTLB pages).
> > > 
> > > Update documentation for the sysctl interface, reference the sysctl
> > > interface in the sysfs ABI documentation, and update HugeTLB soft
> > > offline selftests.
> > 
> > I'm sure you spotted that the documentation for
> > "/sys/devices/system/memory/soft_offline_pag" resides under "testing".
> 
> But that is only one of several places in the kernel that
> feed into the page offline code.
> 
> This patch was motivated by the GHES path where BIOS indicates
> a corrected error threshold was exceeded. There's also the
> drivers/ras/cec.c path where Linux does it's own threshold
> counting.
> > 
> > If your read about MADV_SOFT_OFFLINE in the man page it clearly says:
> > 
> > "This feature is intended for testing of memory error-handling code; it is
> > available  only if the kernel was configured with CONFIG_MEMORY_FAILURE."
> 
> Agreed that this all depends on CONFIG_MEMORY_FAILURE ... so if any
> part of the flow is compiled in when that is "=n" then some
> changes are needed to fix that.
> 
> > 
> > So I'm sorry to say: I miss why we should add all this complexity to make a
> > feature used for testing soft-offlining work differently for hugetlb folios
> > -- with a testing interface.

I would also like to note that the current sysctl interface already affects
testing interfaces. Please see the following commit:

56374430c5dfc ("mm/memory-failure: userspace controls soft-offlining pages")

The sysctl interface should probably be mentioned in
sysfs-memory-page-offline with or without this patch.

Thanks,
Kyle Meyer