On 14.10.24 20:20, Heiko Carstens wrote:
On Mon, Oct 14, 2024 at 04:46:13PM +0200, David Hildenbrand wrote:
s390 currently always results in is_kdump_kernel() == false, because
it sets "elfcorehdr_addr = ELFCORE_ADDR_MAX;" early during setup_arch to
deactivate the elfcorehdr= kernel parameter.
Let's follow the powerpc example and implement our own logic.
This is required for virtio-mem to reliably identify a kdump
environment to not try hotplugging memory.
Tested-by: Mario Casquero <mcasquer@xxxxxxxxxx>
Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
---
arch/s390/include/asm/kexec.h | 4 ++++
arch/s390/kernel/crash_dump.c | 6 ++++++
2 files changed, 10 insertions(+)
Looks like this could work. But the comment in smp.c above
dump_available() needs to be updated.
A right, I remember that there was some outdated documentation.
Are you willing to do that, or should I provide an addon patch?
I can squash the following:
diff --git a/arch/s390/kernel/smp.c b/arch/s390/kernel/smp.c
index 4df56fdb2488..a4f538876462 100644
--- a/arch/s390/kernel/smp.c
+++ b/arch/s390/kernel/smp.c
@@ -587,16 +587,16 @@ int smp_store_status(int cpu)
* with sigp stop-and-store-status. The firmware or the boot-loader
* stored the registers of the boot CPU in the absolute lowcore in the
* memory of the old system.
- * 3) kdump and the old kernel did not store the CPU state,
- * or stand-alone kdump for DASD
- * condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
+ * 3) kdump or stand-alone kdump for DASD
+ * condition: OLDMEM_BASE != NULL && !is_ipl_type_dump() == false
* The state for all CPUs except the boot CPU needs to be collected
* with sigp stop-and-store-status. The kexec code or the boot-loader
* stored the registers of the boot CPU in the memory of the old system.
- * 4) kdump and the old kernel stored the CPU state
- * condition: OLDMEM_BASE != NULL && is_kdump_kernel()
- * This case does not exist for s390 anymore, setup_arch explicitly
- * deactivates the elfcorehdr= kernel parameter
+ *
+ * Note that the old Kdump mode where the old kernel stored the CPU state
+ * does no longer exist: setup_arch explicitly deactivates the elfcorehdr=
+ * kernel parameter. The is_kudmp_kernel() implementation on s390 is independent
+ * of the elfcorehdr= parameter.
*/
static bool dump_available(void)
{
Does that sound reasonable? I'm not so sure about the "2) stand-alone kdump for
SCSI/NVMe (zfcp/nvme dump with swapped memory)": is that really "kdump" ?
--
Cheers,
David / dhildenb
