[PATCH] iommu: Convert unreachable() to BUG()

[PATCH] iommu: Convert unreachable() to BUG()

[Josh Poimboeuf]

Bare unreachable() should be avoided as it generates undefined behavior,
e.g. falling through to the next function.  Use BUG() instead so the
error is defined.

Fixes the following warnings:

  drivers/iommu/dma-iommu.o: warning: objtool: iommu_dma_sw_msi+0x92: can't find jump dest instruction at .text+0x54d5
  vmlinux.o: warning: objtool: iommu_dma_get_msi_page() falls through to next function __iommu_dma_unmap()

Reported-by: Randy Dunlap <rdunlap@infradead.org>
Closes: https://lore.kernel.org/314f8809-cd59-479b-97d7-49356bf1c8d1@infradead.org
Reported-by: Paul E. McKenney <paulmck@kernel.org>
Closes: https://lore.kernel.org/5dd1f35e-8ece-43b7-ad6d-86d02d2718f6@paulmck-laptop
Fixes: 6aa63a4ec947 ("iommu: Sort out domain user data")
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
---
 drivers/iommu/dma-iommu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 7b2734de2ba9..6054d0ab8023 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -1753,7 +1753,7 @@ static size_t cookie_msi_granule(const struct iommu_domain *domain)
 	case IOMMU_COOKIE_DMA_MSI:
 		return PAGE_SIZE;
 	default:
-		unreachable();
+		BUG();
 	};
 }
 
@@ -1765,7 +1765,7 @@ static struct list_head *cookie_msi_pages(const struct iommu_domain *domain)
 	case IOMMU_COOKIE_DMA_MSI:
 		return &domain->msi_cookie->msi_page_list;
 	default:
-		unreachable();
+		BUG();
 	};
 }
 
-- 
2.48.1

Re: [PATCH] iommu: Convert unreachable() to BUG()

[Peter Zijlstra]

On Wed, Mar 26, 2025 at 10:28:46PM -0700, Josh Poimboeuf wrote:
> Bare unreachable() should be avoided as it generates undefined behavior,
> e.g. falling through to the next function.  Use BUG() instead so the
> error is defined.

Right; I did a pass like this a while ago and thought I'd removed all
unreachable() abuse.

Compilers see this as a clue to just stop code-gen. Very bad behaviour
if you ever actually get there. BUG() at the very least stops the kernel
dead, instead of continuing to run random code that comes after.

Re: [PATCH] iommu: Convert unreachable() to BUG()

[Jason Gunthorpe]

On Wed, Mar 26, 2025 at 10:28:46PM -0700, Josh Poimboeuf wrote:
> Bare unreachable() should be avoided as it generates undefined behavior,
> e.g. falling through to the next function.  Use BUG() instead so the
> error is defined.
> 
> Fixes the following warnings:
> 
>   drivers/iommu/dma-iommu.o: warning: objtool: iommu_dma_sw_msi+0x92: can't find jump dest instruction at .text+0x54d5
>   vmlinux.o: warning: objtool: iommu_dma_get_msi_page() falls through to next function __iommu_dma_unmap()
> 
> Reported-by: Randy Dunlap <rdunlap@infradead.org>
> Closes: https://lore.kernel.org/314f8809-cd59-479b-97d7-49356bf1c8d1@infradead.org
> Reported-by: Paul E. McKenney <paulmck@kernel.org>
> Closes: https://lore.kernel.org/5dd1f35e-8ece-43b7-ad6d-86d02d2718f6@paulmck-laptop
> Fixes: 6aa63a4ec947 ("iommu: Sort out domain user data")
> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
> ---
>  drivers/iommu/dma-iommu.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)

The offending patch is in the iommufd tree, so applied thanks for the
quick response

Jason

Re: [PATCH] iommu: Convert unreachable() to BUG()

[Paul E. McKenney]

On Wed, Mar 26, 2025 at 10:28:46PM -0700, Josh Poimboeuf wrote:
> Bare unreachable() should be avoided as it generates undefined behavior,
> e.g. falling through to the next function.  Use BUG() instead so the
> error is defined.
> 
> Fixes the following warnings:
> 
>   drivers/iommu/dma-iommu.o: warning: objtool: iommu_dma_sw_msi+0x92: can't find jump dest instruction at .text+0x54d5
>   vmlinux.o: warning: objtool: iommu_dma_get_msi_page() falls through to next function __iommu_dma_unmap()
> 
> Reported-by: Randy Dunlap <rdunlap@infradead.org>
> Closes: https://lore.kernel.org/314f8809-cd59-479b-97d7-49356bf1c8d1@infradead.org
> Reported-by: Paul E. McKenney <paulmck@kernel.org>
> Closes: https://lore.kernel.org/5dd1f35e-8ece-43b7-ad6d-86d02d2718f6@paulmck-laptop
> Fixes: 6aa63a4ec947 ("iommu: Sort out domain user data")
> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>

Thank you for the quick response!

Tested-by: Paul E. McKenney <paulmck@kernel.org>

> ---
>  drivers/iommu/dma-iommu.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 7b2734de2ba9..6054d0ab8023 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -1753,7 +1753,7 @@ static size_t cookie_msi_granule(const struct iommu_domain *domain)
>  	case IOMMU_COOKIE_DMA_MSI:
>  		return PAGE_SIZE;
>  	default:
> -		unreachable();
> +		BUG();
>  	};
>  }
>  
> @@ -1765,7 +1765,7 @@ static struct list_head *cookie_msi_pages(const struct iommu_domain *domain)
>  	case IOMMU_COOKIE_DMA_MSI:
>  		return &domain->msi_cookie->msi_page_list;
>  	default:
> -		unreachable();
> +		BUG();
>  	};
>  }
>  
> -- 
> 2.48.1
> 

Re: [PATCH] iommu: Convert unreachable() to BUG()

[Josh Poimboeuf]

On Thu, Mar 27, 2025 at 01:37:18PM +0100, Peter Zijlstra wrote:
> On Wed, Mar 26, 2025 at 10:28:46PM -0700, Josh Poimboeuf wrote:
> > Bare unreachable() should be avoided as it generates undefined behavior,
> > e.g. falling through to the next function.  Use BUG() instead so the
> > error is defined.
> 
> Right; I did a pass like this a while ago and thought I'd removed all
> unreachable() abuse.

Any reason not to just "#define unreachable() BUG()" and convert UD2 and
similar to use __builtin_unreachable()?

-- 
Josh

Re: [PATCH] iommu: Convert unreachable() to BUG()

[Peter Zijlstra]

On Tue, Apr 01, 2025 at 05:50:32PM -0700, Josh Poimboeuf wrote:
> On Thu, Mar 27, 2025 at 01:37:18PM +0100, Peter Zijlstra wrote:
> > On Wed, Mar 26, 2025 at 10:28:46PM -0700, Josh Poimboeuf wrote:
> > > Bare unreachable() should be avoided as it generates undefined behavior,
> > > e.g. falling through to the next function.  Use BUG() instead so the
> > > error is defined.
> > 
> > Right; I did a pass like this a while ago and thought I'd removed all
> > unreachable() abuse.
> 
> Any reason not to just "#define unreachable() BUG()" and convert UD2 and
> similar to use __builtin_unreachable()?

Just remove unreachable() entirely at that point. But you're going to
have to update all the various arch code that does use it correctly :/

There are a few sites besides BUG() that need it; eg, long jumps when
bootstrapping etc.

But basically nothing outside of arch code should ever need or want
unreachable.