Fix ppc64 build error introduced in v6.6.161

Hello,

Drgn builds stable kernels regularly in order to run its test suite against. We
encountered an error compiling v6.6.161 for ppc64.

The 6.6.161 update contained "powerpc/iommu: Move pSeries specific functions to
pseries/iommu.c" [1]. On ppc64 configurations without CONFIG_IOMMU_API, this
breaks the build (details in the patch).

Thank you!
Stephen

https://lore.kernel.org/all/20261017160448.281572781@linuxfoundation.org/

Stephen Brennan (1):
  powerpc/pseries/iommu: Restore CONFIG_IOMMU_API dependency

 arch/powerpc/platforms/pseries/iommu.c | 6 ++++++
 1 file changed, 6 insertions(+)

-- 
2.52.0

In stable v6.6.161 there is a build error with ppc64 when
CONFIG_IOMMU_API is not enabled:

    arch/powerpc/platforms/pseries/iommu.c: In function 'iommu_take_ownership':
    arch/powerpc/platforms/pseries/iommu.c:69:25: error: 'struct iommu_table_ops' has no member named 'xchg_no_kill'
       69 |         if (!tbl->it_ops->xchg_no_kill)
    	  |                         ^~
      AS      arch/powerpc/lib/copy_mc_64.o
    arch/powerpc/platforms/pseries/iommu.c: At top level:
    arch/powerpc/platforms/pseries/iommu.c:1787:37: error: 'spapr_tce_table_group_ops' defined but not used [-Werror=unused-variable]
     1787 | static struct iommu_table_group_ops spapr_tce_table_group_ops = {
    	  |                                     ^~~~~~~~~~~~~~~~~~~~~~~~~
    cc1: all warnings being treated as errors

This is due to stable v6.6.161 commit 1e9dfe15404d ("powerpc/iommu: Move
pSeries specific functions to pseries/iommu.c"). It was included as a
stable-dep-of commit 8f70caad82e9 ("powerpc/pseries/iommu: IOMMU
incorrectly marks MMIO range in DDW").

It was backported without its associated fix: commit af199e6ca29c
("powerpc/pseries/iommu: Define spapr_tce_table_group_ops only with
CONFIG_IOMMU_API").

The code was supposed to only build under CONFIG_IOMMU_API, but the new
location doesn't have this condition. Unfortunately, the fixing commit
comes after a fairly large refactor, and doesn't apply on v6.16.161.
Instead, apply the ifdefs where they would belong prior to the refactor.

Signed-off-by: Stephen Brennan <redacted>
---
 arch/powerpc/platforms/pseries/iommu.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/arch/powerpc/platforms/pseries/iommu.c b/arch/powerpc/platforms/pseries/iommu.c
index 6a45e3e86f79b..a79927e870a87 100644
--- a/arch/powerpc/platforms/pseries/iommu.c
+++ b/arch/powerpc/platforms/pseries/iommu.c

@@ -54,6 +54,7 @@ enum {
 	DDW_EXT_QUERY_OUT_SIZE = 2
 };
 
+#ifdef CONFIG_IOMMU_API
 static int iommu_take_ownership(struct iommu_table *tbl)
 {
 	unsigned long flags, i, sz = (tbl->it_size + 7) >> 3;

@@ -104,6 +105,7 @@ static void iommu_release_ownership(struct iommu_table *tbl)
 		spin_unlock(&tbl->pools[i].lock);
 	spin_unlock_irqrestore(&tbl->large_pool.lock, flags);
 }
+#endif
 
 static struct iommu_table *iommu_pseries_alloc_table(int node)
 {

@@ -118,7 +120,9 @@ static struct iommu_table *iommu_pseries_alloc_table(int node)
 	return tbl;
 }
 
+#ifdef CONFIG_IOMMU_API
 static struct iommu_table_group_ops spapr_tce_table_group_ops;
+#endif
 
 static struct iommu_table_group *iommu_pseries_alloc_group(int node)
 {

@@ -1702,6 +1706,7 @@ static bool iommu_bypass_supported_pSeriesLP(struct pci_dev *pdev, u64 dma_mask)
 	return false;
 }
 
+#ifdef CONFIG_IOMMU_API
 /*
  * A simple iommu_table_group_ops which only allows reusing the existing
  * iommu_table. This handles VFIO for POWER7 or the nested KVM.

@@ -1792,6 +1797,7 @@ static struct iommu_table_group_ops spapr_tce_table_group_ops = {
 	.take_ownership = spapr_tce_take_ownership,
 	.release_ownership = spapr_tce_release_ownership,
 };
+#endif
 
 static int iommu_mem_notifier(struct notifier_block *nb, unsigned long action,
 		void *data)

-- 
2.52.0

powerpc/pseries/iommu: Restore CONFIG_IOMMU_API dependency

This is a hand-adapted replacement for af199e6ca29c ("powerpc/pseries/iommu:
Define spapr_tce_table_group_ops only with CONFIG_IOMMU_API") rather than a
clean cherry-pick, and it has no ack from the powerpc side.

Michael, Madhavan: could one of you review it? I will queue it for 6.6 once you
are happy with it.

-- 
Thanks,
Sasha

On 11/6/26 4:54 AM, Stephen Brennan wrote:

In stable v6.6.161 there is a build error with ppc64 when
CONFIG_IOMMU_API is not enabled:

     arch/powerpc/platforms/pseries/iommu.c: In function 'iommu_take_ownership':
     arch/powerpc/platforms/pseries/iommu.c:69:25: error: 'struct iommu_table_ops' has no member named 'xchg_no_kill'
        69 |         if (!tbl->it_ops->xchg_no_kill)
     	  |                         ^~
       AS      arch/powerpc/lib/copy_mc_64.o
     arch/powerpc/platforms/pseries/iommu.c: At top level:
     arch/powerpc/platforms/pseries/iommu.c:1787:37: error: 'spapr_tce_table_group_ops' defined but not used [-Werror=unused-variable]
      1787 | static struct iommu_table_group_ops spapr_tce_table_group_ops = {
     	  |                                     ^~~~~~~~~~~~~~~~~~~~~~~~~
     cc1: all warnings being treated as errors

This is due to stable v6.6.161 commit 1e9dfe15404d ("powerpc/iommu: Move
pSeries specific functions to pseries/iommu.c"). It was included as a
stable-dep-of commit 8f70caad82e9 ("powerpc/pseries/iommu: IOMMU
incorrectly marks MMIO range in DDW").

It was backported without its associated fix: commit af199e6ca29c
("powerpc/pseries/iommu: Define spapr_tce_table_group_ops only with
CONFIG_IOMMU_API").

The code was supposed to only build under CONFIG_IOMMU_API, but the new
location doesn't have this condition. Unfortunately, the fixing commit
comes after a fairly large refactor, and doesn't apply on v6.16.161.
Instead, apply the ifdefs where they would belong prior to the refactor.

Looks fine to me
Acked-by: Madhavan Srinivasan <maddy@linux.ibm.com>

11 more quoted lines