[PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[Sherry Yang]

After enabling CONFIG_GCOV_KERNEL and CONFIG_GCOV_PROFILE_ALL, following
build failure is observed:

In function 'amdv1pt_install_leaf_entry',
    inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:650:3,
    inlined from '__map_single_page0' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:661:1,
    inlined from 'pt_descend' at drivers/iommu/generic_pt/fmt/../pt_iter.h:391:9,
    inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:657:10,
    inlined from '__map_single_page1.constprop' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:661:1:
././include/linux/compiler_types.h:706:45: error: call to '__compiletime_assert_71' declared with attribute error: FIELD_PREP: value too large for the field
  706 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
      |

......

drivers/iommu/generic_pt/fmt/amdv1.h:220:26: note: in expansion of macro 'FIELD_PREP'
  220 |                          FIELD_PREP(AMDV1PT_FMT_OA,
      |                          ^~~~~~~~~~

In the path '__do_map_single_page()', level 0 always invokes
'pt_install_leaf_entry(&pts, map->oa, PAGE_SHIFT, …)'. At runtime that
lands in the 'if (oasz_lg2 == isz_lg2)' arm of 'amdv1pt_install_leaf_entry()';
the contiguous-only 'else' block is unreachable for 4 KiB pages.

With CONFIG_GCOV_KERNEL + CONFIG_GCOV_PROFILE_ALL, the extra
instrumentation changes GCC's inlining so that the "dead" 'else' branch
still gets instantiated. The compiler constant-folds the contiguous OA
expression, runs the 'FIELD_PREP()' compile-time check, and produces:

    FIELD_PREP: value too large for the field

gcov-enabled builds therefore fail even though the code path never executes.

Fix this by keeping the compile-time guard on the NEXT_LEVEL field but
encoding the OA field manually after a runtime 'FIELD_FIT()' check. We
shift-and-mask the value ourselves, so the compiler no longer evaluates
'FIELD_PREP(AMDV1PT_FMT_OA, …)' with constant inputs, while we preserve
the range validation and resulting bit pattern.

Fixes: 879ced2bab1b ("iommupt: Add the AMD IOMMU v1 page table format")
Signed-off-by: Sherry Yang <sherry.yang@oracle.com>
---
 drivers/iommu/generic_pt/fmt/amdv1.h | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/iommu/generic_pt/fmt/amdv1.h b/drivers/iommu/generic_pt/fmt/amdv1.h
index aa8e1a8ec95f..c0a200afbe79 100644
--- a/drivers/iommu/generic_pt/fmt/amdv1.h
+++ b/drivers/iommu/generic_pt/fmt/amdv1.h
@@ -214,13 +214,14 @@ amdv1pt_install_leaf_entry(struct pt_state *pts, pt_oaddr_t oa,
 	} else {
 		unsigned int num_contig_lg2 = oasz_lg2 - isz_lg2;
 		u64 *end = tablep + log2_to_int(num_contig_lg2);
+		pt_oaddr_t contig_oa = oalog2_to_int(oasz_lg2 - PT_GRANULE_LG2SZ - 1) - 1;
+
+		if (PT_WARN_ON(!FIELD_FIT(AMDV1PT_FMT_OA, contig_oa)))
+			return;
 
 		entry |= FIELD_PREP(AMDV1PT_FMT_NEXT_LEVEL,
 				    AMDV1PT_FMT_NL_SIZE) |
-			 FIELD_PREP(AMDV1PT_FMT_OA,
-				    oalog2_to_int(oasz_lg2 - PT_GRANULE_LG2SZ -
-						  1) -
-					    1);
+			 (((u64)contig_oa << __bf_shf(AMDV1PT_FMT_OA)) & AMDV1PT_FMT_OA);
 
 		/* See amdv1pt_clear_entries() */
 		if (num_contig_lg2 <= ilog2(32)) {
-- 
2.50.1

Re: [PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[Jason Gunthorpe]

On Tue, Mar 10, 2026 at 09:31:36AM -0700, Sherry Yang wrote:
> After enabling CONFIG_GCOV_KERNEL and CONFIG_GCOV_PROFILE_ALL, following
> build failure is observed:
> 
> In function 'amdv1pt_install_leaf_entry',
>     inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:650:3,
>     inlined from '__map_single_page0' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:661:1,
>     inlined from 'pt_descend' at drivers/iommu/generic_pt/fmt/../pt_iter.h:391:9,
>     inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:657:10,
>     inlined from '__map_single_page1.constprop' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:661:1:
> ././include/linux/compiler_types.h:706:45: error: call to '__compiletime_assert_71' declared with attribute error: FIELD_PREP: value too large for the field
>   706 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
>       |
> 
> ......
> 
> drivers/iommu/generic_pt/fmt/amdv1.h:220:26: note: in expansion of macro 'FIELD_PREP'
>   220 |                          FIELD_PREP(AMDV1PT_FMT_OA,
>       |                          ^~~~~~~~~~
> 
> In the path '__do_map_single_page()', level 0 always invokes
> 'pt_install_leaf_entry(&pts, map->oa, PAGE_SHIFT, …)'. At runtime that
> lands in the 'if (oasz_lg2 == isz_lg2)' arm of 'amdv1pt_install_leaf_entry()';
> the contiguous-only 'else' block is unreachable for 4 KiB pages.

I think this is fixed already by

commit 98d5110f90ae0dbc5f2f13f033e06f6d57009e0d
Author: Jason Gunthorpe <jgg@ziepe.ca>
Date:   Mon Jan 19 20:19:49 2026 -0400

    iommupt: Make it clearer to the compiler that pts.level == 0 for single page
    
    Older versions of gcc and clang sometimes get tripped up by the build time
    assertion in FIELD_PREP because they can see that the argument to
    FIELD_PREP is constant but can't see that the if condition protecting it
    is also a constant false.

       In file included from <command-line>:
       In function 'amdv1pt_install_leaf_entry',
           inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:651:3,
           inlined from '__map_single_page0' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:662:1,
           inlined from 'pt_descend' at drivers/iommu/generic_pt/fmt/../pt_iter.h:391:9,
           inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:658:10,
           inlined from '__map_single_page1.constprop' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:662:1:


??

Jason

Re: [PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[Sherry Yang]

> On Mar 10, 2026, at 9:41 AM, Jason Gunthorpe <jgg@nvidia.com> wrote:
> 
> On Tue, Mar 10, 2026 at 09:31:36AM -0700, Sherry Yang wrote:
>> After enabling CONFIG_GCOV_KERNEL and CONFIG_GCOV_PROFILE_ALL, following
>> build failure is observed:
>> 
>> In function 'amdv1pt_install_leaf_entry',
>>    inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:650:3,
>>    inlined from '__map_single_page0' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:661:1,
>>    inlined from 'pt_descend' at drivers/iommu/generic_pt/fmt/../pt_iter.h:391:9,
>>    inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:657:10,
>>    inlined from '__map_single_page1.constprop' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:661:1:
>> ././include/linux/compiler_types.h:706:45: error: call to '__compiletime_assert_71' declared with attribute error: FIELD_PREP: value too large for the field
>>  706 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
>>      |
>> 
>> ......
>> 
>> drivers/iommu/generic_pt/fmt/amdv1.h:220:26: note: in expansion of macro 'FIELD_PREP'
>>  220 |                          FIELD_PREP(AMDV1PT_FMT_OA,
>>      |                          ^~~~~~~~~~
>> 
>> In the path '__do_map_single_page()', level 0 always invokes
>> 'pt_install_leaf_entry(&pts, map->oa, PAGE_SHIFT, …)'. At runtime that
>> lands in the 'if (oasz_lg2 == isz_lg2)' arm of 'amdv1pt_install_leaf_entry()';
>> the contiguous-only 'else' block is unreachable for 4 KiB pages.
> 
> I think this is fixed already by
> 
> commit 98d5110f90ae0dbc5f2f13f033e06f6d57009e0d
> Author: Jason Gunthorpe <jgg@ziepe.ca>
> Date:   Mon Jan 19 20:19:49 2026 -0400
> 
>    iommupt: Make it clearer to the compiler that pts.level == 0 for single page
> 
>    Older versions of gcc and clang sometimes get tripped up by the build time
>    assertion in FIELD_PREP because they can see that the argument to
>    FIELD_PREP is constant but can't see that the if condition protecting it
>    is also a constant false.
> 
>       In file included from <command-line>:
>       In function 'amdv1pt_install_leaf_entry',
>           inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:651:3,
>           inlined from '__map_single_page0' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:662:1,
>           inlined from 'pt_descend' at drivers/iommu/generic_pt/fmt/../pt_iter.h:391:9,
>           inlined from '__do_map_single_page' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:658:10,
>           inlined from '__map_single_page1.constprop' at drivers/iommu/generic_pt/fmt/../iommu_pt.h:662:1:
> 
> 
> ??
> 
> Jason

Hi Jason,

Unfortunately, I ran into the FIELD_PREP build failure with 98d5110f90ae (“iommupt: Make it clearer to the compiler that pts.level == 0 for single page”) already merged, which means it doesn’t work for me. I noticed you mentioned clang 18 hit the same issue and fixed by the patch. We’re using GCC 14.2.1, looks like GCC still constant-folds the contiguous branch despite the change to pts.level.

Thanks,
Sherry

Re: [PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[Jason Gunthorpe]

On Tue, Mar 10, 2026 at 05:17:31PM +0000, Sherry Yang wrote:

> Unfortunately, I ran into the FIELD_PREP build failure with
> 98d5110f90ae (“iommupt: Make it clearer to the compiler that
> pts.level == 0 for single page”) already merged, which means it
> doesn’t work for me. I noticed you mentioned clang 18 hit the same
> issue and fixed by the patch. We’re using GCC 14.2.1, looks like GCC
> still constant-folds the contiguous branch despite the change to
> pts.level.

This is all a performance path, I'm reluctant to add more code for
everyone just to hide things from GCC bugs.

Maybe you can rework this to be only when gcov is on, or maybe you can
fix it by futher enhancing what the other commit does. Perhaps an
always inline annotation is all that is missing?

Or maybe we should remove the use of FIELD_PREP?

Jason

Re: [PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[Sherry Yang]

> On Mar 10, 2026, at 10:57 AM, Jason Gunthorpe <jgg@nvidia.com> wrote:
> 
> On Tue, Mar 10, 2026 at 05:17:31PM +0000, Sherry Yang wrote:
> 
>> Unfortunately, I ran into the FIELD_PREP build failure with
>> 98d5110f90ae (“iommupt: Make it clearer to the compiler that
>> pts.level == 0 for single page”) already merged, which means it
>> doesn’t work for me. I noticed you mentioned clang 18 hit the same
>> issue and fixed by the patch. We’re using GCC 14.2.1, looks like GCC
>> still constant-folds the contiguous branch despite the change to
>> pts.level.
> 
> This is all a performance path, I'm reluctant to add more code for
> everyone just to hide things from GCC bugs.
> 
> Maybe you can rework this to be only when gcov is on, or maybe you can
> fix it by futher enhancing what the other commit does. Perhaps an
> always inline annotation is all that is missing?


Thanks for the pointer — marking amdv1pt_install_leaf_entry() as
__always_inline fixes the gcc/gcov build failure for me while keeping
the original FIELD_PREP() path intact.

If you’re happy with this approach, I'll send the patch out.

Thanks,
Sherry

Re: [PATCH] iommupt/amdv1: avoid GCOV builds triggering FIELD_PREP build failure

[Jason Gunthorpe]

On Wed, Mar 25, 2026 at 11:39:37PM +0000, Sherry Yang wrote:
> 
> > On Mar 10, 2026, at 10:57 AM, Jason Gunthorpe <jgg@nvidia.com> wrote:
> > 
> > On Tue, Mar 10, 2026 at 05:17:31PM +0000, Sherry Yang wrote:
> > 
> >> Unfortunately, I ran into the FIELD_PREP build failure with
> >> 98d5110f90ae (“iommupt: Make it clearer to the compiler that
> >> pts.level == 0 for single page”) already merged, which means it
> >> doesn’t work for me. I noticed you mentioned clang 18 hit the same
> >> issue and fixed by the patch. We’re using GCC 14.2.1, looks like GCC
> >> still constant-folds the contiguous branch despite the change to
> >> pts.level.
> > 
> > This is all a performance path, I'm reluctant to add more code for
> > everyone just to hide things from GCC bugs.
> > 
> > Maybe you can rework this to be only when gcov is on, or maybe you can
> > fix it by futher enhancing what the other commit does. Perhaps an
> > always inline annotation is all that is missing?
> 
> 
> Thanks for the pointer — marking amdv1pt_install_leaf_entry() as
> __always_inline fixes the gcc/gcov build failure for me while keeping
> the original FIELD_PREP() path intact.
> 
> If you’re happy with this approach, I'll send the patch out.

That sounds good thanks

Jason