insert_crashkernel_resources() adds crash memory to iomem_resource if
generic crashkernel reservation is enabled on an architecture.

On PowerPC, system RAM is added to iomem_resource. See commit
c40dd2f766440 ("powerpc: Add System RAM to /proc/iomem").

Enabling generic crashkernel reservation on PowerPC leads to a conflict
when system RAM is added to iomem_resource because a part of the system
RAM, the crashkernel memory, has already been added to iomem_resource.

The next commit in the series "powerpc/crash: use generic crashkernel
reservation" enables generic crashkernel reservation on PowerPC. If the
crashkernel is added to iomem_resource, the kernel fails to add
system RAM to /proc/iomem and prints the following traces:

CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.13.0-rc2+
snip...
NIP [c000000002016b3c] add_system_ram_resources+0xf0/0x15c
LR [c000000002016b34] add_system_ram_resources+0xe8/0x15c
Call Trace:
[c00000000484bbc0] [c000000002016b34] add_system_ram_resources+0xe8/0x15c
[c00000000484bc20] [c000000000010a4c] do_one_initcall+0x7c/0x39c
[c00000000484bd00] [c000000002005418] do_initcalls+0x144/0x18c
[c00000000484bd90] [c000000002005714] kernel_init_freeable+0x21c/0x290
[c00000000484bdf0] [c0000000000110f4] kernel_init+0x2c/0x1b8
[c00000000484be50] [c00000000000dd3c] ret_from_kernel_user_thread+0x14/0x1c

To avoid this, an architecture hook is added in
insert_crashkernel_resources(), allowing the architecture to decide
whether crashkernel memory should be added to iomem_resource.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Baoquan he <bhe@redhat.com>
Cc: Hari Bathini <hbathini@linux.ibm.com>
CC: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: kexec@lists.infradead.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Sourabh Jain <sourabhjain@linux.ibm.com>
---
 include/linux/crash_reserve.h | 11 +++++++++++
 kernel/crash_reserve.c        |  3 +++
 2 files changed, 14 insertions(+)

diff --git a/include/linux/crash_reserve.h b/include/linux/crash_reserve.h
index 1fe7e7d1b214..f1205d044dae 100644
--- a/include/linux/crash_reserve.h
+++ b/include/linux/crash_reserve.h
@@ -18,6 +18,17 @@ int __init parse_crashkernel(char *cmdline, unsigned long long system_ram,
 		unsigned long long *crash_size, unsigned long long *crash_base,
 		unsigned long long *low_size, bool *high);
 
+#ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
+
+#ifndef arch_add_crash_res_to_iomem
+static inline bool arch_add_crash_res_to_iomem(void)
+{
+	return true;
+}
+#endif
+
+#endif
+
 #ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
 #ifndef DEFAULT_CRASH_KERNEL_LOW_SIZE
 #define DEFAULT_CRASH_KERNEL_LOW_SIZE	(128UL << 20)
diff --git a/kernel/crash_reserve.c b/kernel/crash_reserve.c
index aff7c0fdbefa..190104f32fe1 100644
--- a/kernel/crash_reserve.c
+++ b/kernel/crash_reserve.c
@@ -460,6 +460,9 @@ void __init reserve_crashkernel_generic(unsigned long long crash_size,
 #ifndef HAVE_ARCH_ADD_CRASH_RES_TO_IOMEM_EARLY
 static __init int insert_crashkernel_resources(void)
 {
+	if (!arch_add_crash_res_to_iomem())
+		return 0;
+
 	if (crashk_res.start < crashk_res.end)
 		insert_resource(&iomem_resource, &crashk_res);
 
-- 
2.47.1

Hi,

On 01/08/25 at 03:44pm, Sourabh Jain wrote:
> insert_crashkernel_resources() adds crash memory to iomem_resource if
> generic crashkernel reservation is enabled on an architecture.
> 
> On PowerPC, system RAM is added to iomem_resource. See commit
> c40dd2f766440 ("powerpc: Add System RAM to /proc/iomem").
> 
> Enabling generic crashkernel reservation on PowerPC leads to a conflict
> when system RAM is added to iomem_resource because a part of the system
> RAM, the crashkernel memory, has already been added to iomem_resource.
> 
> The next commit in the series "powerpc/crash: use generic crashkernel
> reservation" enables generic crashkernel reservation on PowerPC. If the
> crashkernel is added to iomem_resource, the kernel fails to add
> system RAM to /proc/iomem and prints the following traces:
> 
> CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.13.0-rc2+
> snip...
> NIP [c000000002016b3c] add_system_ram_resources+0xf0/0x15c
> LR [c000000002016b34] add_system_ram_resources+0xe8/0x15c
> Call Trace:
> [c00000000484bbc0] [c000000002016b34] add_system_ram_resources+0xe8/0x15c
> [c00000000484bc20] [c000000000010a4c] do_one_initcall+0x7c/0x39c
> [c00000000484bd00] [c000000002005418] do_initcalls+0x144/0x18c
> [c00000000484bd90] [c000000002005714] kernel_init_freeable+0x21c/0x290
> [c00000000484bdf0] [c0000000000110f4] kernel_init+0x2c/0x1b8
> [c00000000484be50] [c00000000000dd3c] ret_from_kernel_user_thread+0x14/0x1c
> 
> To avoid this, an architecture hook is added in
> insert_crashkernel_resources(), allowing the architecture to decide
> whether crashkernel memory should be added to iomem_resource.

Have you tried defining HAVE_ARCH_ADD_CRASH_RES_TO_IOMEM_EARLY in ppc to
add crashkernel region to iomem early? Now there are two branches in the
existing code, adding a hook will make three ways.

Hello Baoquan

On 08/01/25 16:55, Baoquan he wrote:
> Hi,
>
> On 01/08/25 at 03:44pm, Sourabh Jain wrote:
>> insert_crashkernel_resources() adds crash memory to iomem_resource if
>> generic crashkernel reservation is enabled on an architecture.
>>
>> On PowerPC, system RAM is added to iomem_resource. See commit
>> c40dd2f766440 ("powerpc: Add System RAM to /proc/iomem").
>>
>> Enabling generic crashkernel reservation on PowerPC leads to a conflict
>> when system RAM is added to iomem_resource because a part of the system
>> RAM, the crashkernel memory, has already been added to iomem_resource.
>>
>> The next commit in the series "powerpc/crash: use generic crashkernel
>> reservation" enables generic crashkernel reservation on PowerPC. If the
>> crashkernel is added to iomem_resource, the kernel fails to add
>> system RAM to /proc/iomem and prints the following traces:
>>
>> CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.13.0-rc2+
>> snip...
>> NIP [c000000002016b3c] add_system_ram_resources+0xf0/0x15c
>> LR [c000000002016b34] add_system_ram_resources+0xe8/0x15c
>> Call Trace:
>> [c00000000484bbc0] [c000000002016b34] add_system_ram_resources+0xe8/0x15c
>> [c00000000484bc20] [c000000000010a4c] do_one_initcall+0x7c/0x39c
>> [c00000000484bd00] [c000000002005418] do_initcalls+0x144/0x18c
>> [c00000000484bd90] [c000000002005714] kernel_init_freeable+0x21c/0x290
>> [c00000000484bdf0] [c0000000000110f4] kernel_init+0x2c/0x1b8
>> [c00000000484be50] [c00000000000dd3c] ret_from_kernel_user_thread+0x14/0x1c
>>
>> To avoid this, an architecture hook is added in
>> insert_crashkernel_resources(), allowing the architecture to decide
>> whether crashkernel memory should be added to iomem_resource.
> Have you tried defining HAVE_ARCH_ADD_CRASH_RES_TO_IOMEM_EARLY in ppc to
> add crashkernel region to iomem early?


I didn’t try, but I think even if I do, the kernel will run into the 
same issue
because as soon as crashkernel is added to iomem, there will be a resource
conflict when the PowerPC code tries to add system RAM to iomem. Which
happens during subsys_initcall.

Regardless, I will give it a try.

> Now there are two branches in the
> existing code, adding a hook will make three ways.

I agree. I can try updating powerpc code to not consider crashkernel 
memory as
iomem conflict.

Thanks for the review.

- Sourabh Jain