[PATCH] s390/kdump: write protect the crash kernel memory by its virtual address

From: Mike Rapoport (Microsoft)

Date: Tue Nov 17 2026 - 09:01:17 EST


crash_protect_pages() passes crashk_res.start to set_memory_ro() and
set_memory_rw(). That is a physical address. Since the identity mapping
may start above zero, it is not the address of the crash kernel memory
in the kernel page tables.

With CONFIG_RANDOMIZE_IDENTITY_BASE the page table walk finds no mapping
at that address and fails. The error is ignored, and the crash kernel
memory stays writable after "kexec -p" loads it.

Use the identity mapping address of the crash kernel memory.

Fixes: c98d2ecae08f ("s390/mm: Uncouple physical vs virtual address spaces")
Signed-off-by: Mike Rapoport (Microsoft) <rppt@xxxxxxxxxx>
---
I found this while working on a generic implementation of the set_memory
APIs. Only build tested.
---
arch/s390/kernel/machine_kexec.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/arch/s390/kernel/machine_kexec.c b/arch/s390/kernel/machine_kexec.c
index baeb3dcfc1c8..20ea3018462f 100644
--- a/arch/s390/kernel/machine_kexec.c
+++ b/arch/s390/kernel/machine_kexec.c
@@ -151,15 +151,16 @@ void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)

static void crash_protect_pages(int protect)
{
- unsigned long size;
+ unsigned long addr, size;

if (!crashk_res.end)
return;
+ addr = (unsigned long)__va(crashk_res.start);
size = resource_size(&crashk_res);
if (protect)
- set_memory_ro(crashk_res.start, size >> PAGE_SHIFT);
+ set_memory_ro(addr, size >> PAGE_SHIFT);
else
- set_memory_rw(crashk_res.start, size >> PAGE_SHIFT);
+ set_memory_rw(addr, size >> PAGE_SHIFT);
}

void arch_kexec_protect_crashkres(void)

---
base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4
change-id: 20261117-fixes-s390-82943938787a

--
Sincerely yours,
Mike.