cma: powerpc fadump fixes

cma_init_reserved_mem() checks base and size alignment with CMA_MIN_ALIGNMENT_BYTES. However, some users might call this during early boot when pageblock_order is 0. That means if base and size does not have pageblock_order alignment, it can cause functional failures during cma activate area.

So let's enforce pageblock_order to be non-zero during cma_init_reserved_mem().

Signed-off-by Ritesh Harjani (IBM) <[email protected]>

---
  mm/cma.c | 9 +++++++++
  1 file changed, 9 insertions(+)

mm/cma.c +2 -0

index 3e9724716bad..36d753e7a0bf 100644
@@ -182,6 +182,15 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
  	if (!size || !memblock_is_region_reserved(base, size))
  		return -EINVAL;
 
+	/*
+	 * CMA uses CMA_MIN_ALIGNMENT_BYTES as alignment requirement which

This patch refactors all CMA related initialization and alignment code to within fadump_cma_init() which gets called in the end. This also means that we keep [reserve_dump_area_start, boot_memory_size] page aligned during fadump_reserve_mem(). Then later in fadump_cma_init() we extract the aligned chunk and provide it to CMA. This inherently also fixes an issue in the current code where the reserve_dump_area_start is not aligned when the physical memory can have holes and the suitable chunk starts at an unaligned boundary.

After this we should be able to call fadump_cma_init() independently later in setup_arch() where pageblock_order is non-zero.

Suggested-by Sourabh Jain <[email protected]>

Signed-off-by Ritesh Harjani (IBM) <[email protected]>

---
  arch/powerpc/kernel/fadump.c | 34 ++++++++++++++++++++++------------
  1 file changed, 22 insertions(+), 12 deletions(-)

arch/powerpc/kernel/fadump.c +17 -1

index 162327d66982..ffaec625b7a8 100644
@@ -80,7 +80,7 @@ static struct cma *fadump_cma;
   */
  static void __init fadump_cma_init(void)
  {
-	unsigned long long base, size;
+	unsigned long long base, size, end;
  	int rc;
 
  	if (!fw_dump.fadump_supported || !fw_dump.fadump_enabled ||

@@ -92,8 +92,24 @@ static void __init fadump_cma_init(void)
  	if (fw_dump.nocma || !fw_dump.boot_memory_size)
  		return;
 
+	/*
+	 * [base, end) should be reserved during early init in
+	 * fadump_reserve_mem(). No need to check this here as
+	 * cma_init_reserved_mem() already checks for overlap.
+	 * Here we give the aligned chunk of this reserved memory to CMA.
+	 */
  	base = fw_dump.reserve_dump_area_start;
  	size = fw_dump.boot_memory_size;
+	end = base + size;
+
+	base = ALIGN(base, CMA_MIN_ALIGNMENT_BYTES);
+	end = ALIGN_DOWN(end, CMA_MIN_ALIGNMENT_BYTES);
+	size = end - base;
+
+	if (end <= base) {
+		pr_warn("%s: Too less memory to give to CMA\n", __func__);
+		return;
+	}
 
  	rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
  	if (rc) {

@@ -116,11 +132,12 @@ static void __init fadump_cma_init(void)
  	/*
  	 * So we now have successfully initialized cma area for fadump.
  	 */

arch/powerpc/kernel/fadump.c +1 -0

+	return;
  }
  #else

static void __init fadump_cma_init(void) { }

arch/powerpc/kernel/fadump.c +0 -7

@@ -553,13 +570,6 @@ int __init fadump_reserve_mem(void)
  	if (!fw_dump.dump_active) {
  		fw_dump.boot_memory_size =
  			PAGE_ALIGN(fadump_calculate_reserve_size());
-#ifdef CONFIG_CMA
-		if (!fw_dump.nocma) {
-			fw_dump.boot_memory_size =
-				ALIGN(fw_dump.boot_memory_size,
-				      CMA_MIN_ALIGNMENT_BYTES);
-		}
-#endif
 
  		bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
  		if (fw_dump.boot_memory_size < bootmem_min) {

Please find the v2 of cma related powerpc fadump fixes.

Patch-1 is a change in mm/cma.c to make sure we return an error if someone uses cma_init_reserved_mem() before the pageblock_order is initalized.

I guess, it's best if Patch-1 goes via mm tree and since rest of the changes are powerpc fadump fixes hence those should go via powerpc tree. Right?

v1 -> v2:
=========
1. Review comments from David to call fadump_cma_init() after the pageblock_order is initialized. Also to catch usages if someone tries to call cma_init_reserved_mem() before pageblock_order is initialized.

[v1]: https://lore.kernel.org/linuxppc-dev/c1e66d3e69c8d90988c02b84c79db5d9dd93f053.1728386179.git.ritesh.list@gmail.com/

Ritesh Harjani (IBM) (4):
cma: Enforce non-zero pageblock_order during cma_init_reserved_mem()
fadump: Refactor and prepare fadump_cma_init for late init
fadump: Reserve page-aligned boot_memory_size during fadump_reserve_mem
fadump: Move fadump_cma_init to setup_arch() after initmem_init()

 arch/powerpc/include/asm/fadump.h  |  7 ++++
 arch/powerpc/kernel/fadump.c       | 55 +++++++++++++++---------------
 arch/powerpc/kernel/setup-common.c |  6 ++--
 mm/cma.c                           |  9 +++++
 4 files changed, 48 insertions(+), 29 deletions(-)

--
2.46.0

Quoted 7 lines · Ritesh Harjani (IBM)

Please find the v2 of cma related powerpc fadump fixes.

Patch-1 is a change in mm/cma.c to make sure we return an error if someone uses cma_init_reserved_mem() before the pageblock_order is initalized.

I guess, it's best if Patch-1 goes via mm tree and since rest of the changes are powerpc fadump fixes hence those should go via powerpc tree. Right?

Yes I think that will work.

Because there's no actual dependency on patch 1, correct?

Let's see if the mm folks are happy with the approach, and if so you should send patch 1 on its own, and patches 2-4 as a separate series.

Quoted 5 lines · Ritesh Harjani (IBM)

Please find the v2 of cma related powerpc fadump fixes.

Patch-1 is a change in mm/cma.c to make sure we return an error if someone uses cma_init_reserved_mem() before the pageblock_order is initalized.

Quoted 2 lines

I guess, it's best if Patch-1 goes via mm tree and since rest of the changes are powerpc fadump fixes hence those should go via powerpc tree. Right?

Yes I think that will work.

Because there's no actual dependency on patch 1, correct?

Let's see if the mm folks are happy with the approach, and if so you should send patch 1 on its own, and patches 2-4 as a separate series.

Then I can take the series (2-4) as fixes, and patch 1 can go via the mm tree (probably in next, not as a fix).

cheers

Quoted 22 lines

v1 -> v2:
=========
1. Review comments from David to call fadump_cma_init() after the pageblock_order is initialized. Also to catch usages if someone tries to call cma_init_reserved_mem() before pageblock_order is initialized.

[v1]: https://lore.kernel.org/linuxppc-dev/c1e66d3e69c8d90988c02b84c79db5d9dd93f053.1728386179.git.ritesh.list@gmail.com/

Ritesh Harjani (IBM) (4):
cma: Enforce non-zero pageblock_order during cma_init_reserved_mem()
fadump: Refactor and prepare fadump_cma_init for late init
fadump: Reserve page-aligned boot_memory_size during fadump_reserve_mem
fadump: Move fadump_cma_init to setup_arch() after initmem_init()

 arch/powerpc/include/asm/fadump.h  |  7 ++++
 arch/powerpc/kernel/fadump.c       | 55 +++++++++++++++---------------
 arch/powerpc/kernel/setup-common.c |  6 ++--
 mm/cma.c                           |  9 +++++
 4 files changed, 48 insertions(+), 29 deletions(-)

--
2.46.0