Re: powerpc/fadump: CMA init is failing

Mike Rapoport [off-list ref] writes:

quoted

Hi Ritesh,

On Fri, Feb 27, 2026 at 10:39:53PM +0530, Ritesh Harjani wrote:

quoted

Sourabh Jain [off-list ref] writes:

quoted

I noticed CMA init for fadump crashkernel memory is failing.

[    0.000000] cma: pageblock_order not yet initialized. Called during
early boot?
[    0.000000] fadump: Failed to init cma area for firmware-assisted
dump,-22


kernel command-line:
BOOT_IMAGE=(ieee1275//vdevice/v-scsi@30000070/disk@8100000000000000,msdos2)/vmlinuz-7.0.0-rc1+
root=/dev/mapper/rhel_ltcden3--lp12-root ro
rd.lvm.lv=rhel_ltcden3-lp12/root rd.lvm.lv=rhel_ltcden3-lp12/swap
fadump=on crashkernel=3G


Same issue with kdump CMA reservation:

[    0.000000][    T0] cma: pageblock_order not yet initialized. Called
during early boot?

Good that we added those debug prints ;)

I think I know what went wrong, as part of this arch,mm consolidation
patch series [1], I think the order of initialization is changed.

With this patch the new order is ...
start_kernel()
     - setup_arch()
        - xxx_cma_reserve();
     - mm_core_init_early()
        - free_area_init()
           - sparse_init()
              - set_pageblock_order() // this sets the pageblock_order.

Whereas earlier set_pageblock_order() was called from initmem_init(),
just before cma reservations were being made.

start_kernel()
     - setup_arch()
        - initmem_init()
          - sparse_init()
            - set_pageblock_order();  // this sets the pageblock_order
        - xxx_cma_reserve();

So that means, pageblock_order is not initialized before these cma
reservation function calls, hence we are seeing these failures.

setup_arch() {
     ...

	/*
	 * Reserve large chunks of memory for use by CMA for kdump, fadump, KVM and
	 * hugetlb. These must be called after initmem_init(), so that
	 * pageblock_order is initialised.
	 */
	fadump_cma_init();
	kdump_cma_reserve();
	kvm_cma_reserve();

     ...
}


So what if we do..

start_kernel() {
   ...
	setup_arch(&command_line);
	mm_core_init_early();
     setup_arch_post_mm_core_init(); // and here we call CMA reservation functions ?

  
Unless I'm missing something these cma reservations can be moved to
arch_mm_preinit().
It runs after mm_core_init_early() and before memblock moves the free
memory to the buddy.

Right. I think, we should be able to use that...

@Sourabh,

I don't have access to the systems (travelling back...). Could you
please give this a try?

diff --git a/arch/powerpc/kernel/setup-common.c b/arch/powerpc/kernel/setup-common.c
index cb5b73adc250..b1761909c23f 100644
--- a/arch/powerpc/kernel/setup-common.c
+++ b/arch/powerpc/kernel/setup-common.c

@@ -35,7 +35,6 @@
  #include <linux/of_irq.h>
  #include <linux/hugetlb.h>
  #include <linux/pgtable.h>
-#include <asm/kexec.h>
  #include <asm/io.h>
  #include <asm/paca.h>
  #include <asm/processor.h>

@@ -995,15 +994,6 @@ void __init setup_arch(char **cmdline_p)
  
  	initmem_init();
  
-	/*
-	 * Reserve large chunks of memory for use by CMA for kdump, fadump, KVM and
-	 * hugetlb. These must be called after initmem_init(), so that
-	 * pageblock_order is initialised.
-	 */
-	fadump_cma_init();
-	kdump_cma_reserve();
-	kvm_cma_reserve();
-
  	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
  
  	if (ppc_md.setup_arch)

diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
index 29bf347f6012..5ba947e4fe37 100644
--- a/arch/powerpc/mm/mem.c
+++ b/arch/powerpc/mm/mem.c

@@ -30,6 +30,10 @@
  #include <asm/setup.h>
  #include <asm/fixmap.h>
  
+#include <asm/fadump.h>
+#include <asm/kexec.h>
+#include <asm/kvm_ppc.h>
+
  #include <mm/mmu_decl.h>
  
  unsigned long long memory_limit __initdata;

@@ -268,6 +272,16 @@ void __init paging_init(void)
  
  void __init arch_mm_preinit(void)
  {
+
+	/*
+	 * Reserve large chunks of memory for use by CMA for kdump, fadump, KVM
+	 * and hugetlb. These must be called after pageblock_order is
+	 * initialised.
+	 */
+	fadump_cma_init();
+	kdump_cma_reserve();
+	kvm_cma_reserve();
+
  	/*
  	 * book3s is limited to 16 page sizes due to encoding this in
  	 * a 4-bit field for slices.