[PATCH] powerpc: iommu: fix build when neither PCI or IBMVIO is set

[PATCH] powerpc: iommu: fix build when neither PCI or IBMVIO is set

[Randy Dunlap]

When neither CONFIG_PCI nor CONFIG_IBMVIO is enabled:

../arch/powerpc/kernel/iommu.c:178:30: error: 'fail_iommu_bus_notifier' defined but not used [-Werror=unused-variable]
  178 | static struct notifier_block fail_iommu_bus_notifier = {

If only that struct is bounded by 2 #if defined() phrases (PCI && IBMVIO):

../arch/powerpc/kernel/iommu.c:162:12: error: 'fail_iommu_bus_notify' defined but not used [-Werror=unused-function]
  162 | static int fail_iommu_bus_notify(struct notifier_block *nb,

If that function is also guarded by 2 #if defined() phrases:

In file included from ../include/linux/dma-mapping.h:7,
                 from ../arch/powerpc/kernel/iommu.c:19:
../include/linux/device.h:131:26: error: 'dev_attr_fail_iommu' defined but not used [-Werror=unused-variable]
  131 |  struct device_attribute dev_attr_##_name = __ATTR_RW(_name)
../arch/powerpc/kernel/iommu.c:160:8: note: in expansion of macro 'DEVICE_ATTR_RW'
  160 | static DEVICE_ATTR_RW(fail_iommu);

and the snowball continues to grow.
Next I got this one:

../arch/powerpc/kernel/iommu.c: In function 'iommu_range_alloc':
../arch/powerpc/kernel/iommu.c:234:6: error: implicit declaration of function 'should_fail_iommu'; did you mean 'should_failslab'? [-Werror=implicit-function-declaration]
  234 |  if (should_fail_iommu(dev))

and

../arch/powerpc/kernel/iommu.c: In function 'should_fail_iommu':
../arch/powerpc/kernel/iommu.c:122:50: error: 'fail_iommu' undeclared (first use in this function)
  122 |  return dev->archdata.fail_iommu && should_fail(&fail_iommu, 1);

So combine CONFIG_FAIL_IOMMU && (CONFIG_PCI || CONFIG_IBMVIO)
to decide on building some of this code/data.

This came from a .config file from the kernel test robot, but it was
not for this particular build problem.

Fixes: d6b9a81b2a45 ("powerpc: IOMMU fault injection")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Anton Blanchard <anton@samba.org>
---
Found/seen in v5.12-rc1.

 arch/powerpc/kernel/iommu.c |   13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

--- lnx-512-rc1.orig/arch/powerpc/kernel/iommu.c
+++ lnx-512-rc1/arch/powerpc/kernel/iommu.c
@@ -115,7 +115,13 @@ static int __init setup_iommu_pool_hash(
 }
 subsys_initcall(setup_iommu_pool_hash);
 
-#ifdef CONFIG_FAIL_IOMMU
+#if defined(CONFIG_FAIL_IOMMU) && \
+   (defined(CONFIG_PCI) || defined(CONFIG_IBMVIO))
+
+static bool should_fail_iommu(struct device *dev)
+{
+	return dev->archdata.fail_iommu && should_fail(&fail_iommu, 1);
+}
 
 static DECLARE_FAULT_ATTR(fail_iommu);
 
@@ -125,11 +131,6 @@ static int __init setup_fail_iommu(char
 }
 __setup("fail_iommu=", setup_fail_iommu);
 
-static bool should_fail_iommu(struct device *dev)
-{
-	return dev->archdata.fail_iommu && should_fail(&fail_iommu, 1);
-}
-
 static int __init fail_iommu_debugfs(void)
 {
 	struct dentry *dir = fault_create_debugfs_attr("fail_iommu",

Re: [PATCH] powerpc: iommu: fix build when neither PCI or IBMVIO is set

[Michael Ellerman]

Randy Dunlap <rdunlap@infradead.org> writes:
> When neither CONFIG_PCI nor CONFIG_IBMVIO is enabled:
>
> ../arch/powerpc/kernel/iommu.c:178:30: error: 'fail_iommu_bus_notifier' defined but not used [-Werror=unused-variable]
>   178 | static struct notifier_block fail_iommu_bus_notifier = {
>
> If only that struct is bounded by 2 #if defined() phrases (PCI && IBMVIO):
>
> ../arch/powerpc/kernel/iommu.c:162:12: error: 'fail_iommu_bus_notify' defined but not used [-Werror=unused-function]
>   162 | static int fail_iommu_bus_notify(struct notifier_block *nb,
>
> If that function is also guarded by 2 #if defined() phrases:
>
> In file included from ../include/linux/dma-mapping.h:7,
>                  from ../arch/powerpc/kernel/iommu.c:19:
> ../include/linux/device.h:131:26: error: 'dev_attr_fail_iommu' defined but not used [-Werror=unused-variable]
>   131 |  struct device_attribute dev_attr_##_name = __ATTR_RW(_name)
> ../arch/powerpc/kernel/iommu.c:160:8: note: in expansion of macro 'DEVICE_ATTR_RW'
>   160 | static DEVICE_ATTR_RW(fail_iommu);
>
> and the snowball continues to grow.
> Next I got this one:
>
> ../arch/powerpc/kernel/iommu.c: In function 'iommu_range_alloc':
> ../arch/powerpc/kernel/iommu.c:234:6: error: implicit declaration of function 'should_fail_iommu'; did you mean 'should_failslab'? [-Werror=implicit-function-declaration]
>   234 |  if (should_fail_iommu(dev))
>
> and
>
> ../arch/powerpc/kernel/iommu.c: In function 'should_fail_iommu':
> ../arch/powerpc/kernel/iommu.c:122:50: error: 'fail_iommu' undeclared (first use in this function)
>   122 |  return dev->archdata.fail_iommu && should_fail(&fail_iommu, 1);
>
> So combine CONFIG_FAIL_IOMMU && (CONFIG_PCI || CONFIG_IBMVIO)
> to decide on building some of this code/data.

Couldn't we just make FAIL_IOMMU depend on PCI || IBMVIO?

cheers

Re: [PATCH] powerpc: iommu: fix build when neither PCI or IBMVIO is set

[Randy Dunlap]

On March 2, 2021 3:08:43 AM PST, Michael Ellerman <mpe@ellerman.id.au> wrote:
>Randy Dunlap <rdunlap@infradead.org> writes:
>> When neither CONFIG_PCI nor CONFIG_IBMVIO is enabled:
>>
>> ../arch/powerpc/kernel/iommu.c:178:30: error:
>'fail_iommu_bus_notifier' defined but not used
>[-Werror=unused-variable]
>>   178 | static struct notifier_block fail_iommu_bus_notifier = {
>>
>> If only that struct is bounded by 2 #if defined() phrases (PCI &&
>IBMVIO):
>>
>> ../arch/powerpc/kernel/iommu.c:162:12: error: 'fail_iommu_bus_notify'
>defined but not used [-Werror=unused-function]
>>   162 | static int fail_iommu_bus_notify(struct notifier_block *nb,
>>
>> If that function is also guarded by 2 #if defined() phrases:
>>
>> In file included from ../include/linux/dma-mapping.h:7,
>>                  from ../arch/powerpc/kernel/iommu.c:19:
>> ../include/linux/device.h:131:26: error: 'dev_attr_fail_iommu'
>defined but not used [-Werror=unused-variable]
>>   131 |  struct device_attribute dev_attr_##_name = __ATTR_RW(_name)
>> ../arch/powerpc/kernel/iommu.c:160:8: note: in expansion of macro
>'DEVICE_ATTR_RW'
>>   160 | static DEVICE_ATTR_RW(fail_iommu);
>>
>> and the snowball continues to grow.
>> Next I got this one:
>>
>> ../arch/powerpc/kernel/iommu.c: In function 'iommu_range_alloc':
>> ../arch/powerpc/kernel/iommu.c:234:6: error: implicit declaration of
>function 'should_fail_iommu'; did you mean 'should_failslab'?
>[-Werror=implicit-function-declaration]
>>   234 |  if (should_fail_iommu(dev))
>>
>> and
>>
>> ../arch/powerpc/kernel/iommu.c: In function 'should_fail_iommu':
>> ../arch/powerpc/kernel/iommu.c:122:50: error: 'fail_iommu' undeclared
>(first use in this function)
>>   122 |  return dev->archdata.fail_iommu && should_fail(&fail_iommu,
>1);
>>
>> So combine CONFIG_FAIL_IOMMU && (CONFIG_PCI || CONFIG_IBMVIO)
>> to decide on building some of this code/data.
>
>Couldn't we just make FAIL_IOMMU depend on PCI || IBMVIO?
>
>cheers

Yes, I thought of that about 5 seconds after hitting Send. But I can't do it just now -- am away from computer.


-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.