Re: [PATCH] powerpc/32: Remove PAGE_KERNEL_TEXT to fix startup failure

On Thu, 2025-09-04 at 18:33 +0200, Christophe Leroy wrote:

PAGE_KERNEL_TEXT is an old macro that is used to tell kernel whether
kernel text has to be mapped read-only or read-write based on build
time options.

But nowadays, with functionnalities like jump_labels, static links,
etc ... more only less all kernels need to be read-write at some
point, and some combinations of configs failed to work due to
innacurate setting of PAGE_KERNEL_TEXT. On the other hand, today
we have CONFIG_STRICT_KERNEL_RWX which implements a more controlled
access to kernel modifications.

Instead of trying to keep PAGE_KERNEL_TEXT accurate with all
possible options that may imply kernel text modification, always
set kernel text read-write at startup and rely on
CONFIG_STRICT_KERNEL_RWX to provide accurate protection.

Reported-by: Erhard Furtner <redacted>
Closes:
https://lore.kernel.org/all/342b4120-911c-4723-82ec-d8c9b03a8aef@mailbox.org/ (local)
Signed-off-by: Christophe Leroy <redacted>

The original issue that Erhard and I were investigating was why the latest
version of the PowerPC page table check series[0] was failing on his G4, when
built as part of a config with many other debugging options enabled.

With further instrumentation, it turns out that this was due to a failed
instruction patch while setting up a jump label for the
page_table_check_disabled static key, which was being checked in
page_table_check_pte_clear(), which was in turn inlined ultimately into
debug_vm_pgtable().

This patch seems to fix the problem, so:

Tested-by: Andrew Donnellan <redacted>

But I'm still curious about why I only see the issue when:

  (a) CONFIG_KFENCE=y (even when disabled using kfence.sample_interval=0) -
noting that changing CONFIG_KFENCE doesn't change the definition of
PAGE_KERNEL_TEXT; and

  (b) when the jump label ends up in a __init function (removing __init from
debug_vm_pgtable() and its associated functions, or changing the code in such a
way that the static key check doesn't get inlined, resolves the issue, and
similarly for test_static_call_init() when CONFIG_STATIC_CALL_SELFTEST=y).

I don't understand the mm code well enough to make sense of this.

[0] https://lore.kernel.org/all/20250813062614.51759-1-ajd@linux.ibm.com/ (local)

-- 
Andrew Donnellan    OzLabs, ADL Canberra
ajd@linux.ibm.com   IBM Australia Limited