[PATCH v4] dma-debug: fix incorrect pfn calculation

[PATCH v4] dma-debug: fix incorrect pfn calculation

[miles.chen]

From: Miles Chen <miles.chen@mediatek.com>

dma-debug reports the following warning:

[name:panic&]WARNING: CPU: 3 PID: 298 at kernel-4.4/lib/dma-debug.c:604
debug _dma_assert_idle+0x1a8/0x230()
DMA-API: cpu touching an active dma mapped cacheline [cln=0x00000882300]
CPU: 3 PID: 298 Comm: vold Tainted: G        W  O    4.4.22+ #1
Hardware name: MT6739 (DT)
Call trace:
[<ffffff800808acd0>] dump_backtrace+0x0/0x1d4
[<ffffff800808affc>] show_stack+0x14/0x1c
[<ffffff800838019c>] dump_stack+0xa8/0xe0
[<ffffff80080a0594>] warn_slowpath_common+0xf4/0x11c
[<ffffff80080a061c>] warn_slowpath_fmt+0x60/0x80
[<ffffff80083afe24>] debug_dma_assert_idle+0x1a8/0x230
[<ffffff80081dca9c>] wp_page_copy.isra.96+0x118/0x520
[<ffffff80081de114>] do_wp_page+0x4fc/0x534
[<ffffff80081e0a14>] handle_mm_fault+0xd4c/0x1310
[<ffffff8008098798>] do_page_fault+0x1c8/0x394
[<ffffff800808231c>] do_mem_abort+0x50/0xec

I found that debug_dma_alloc_coherent() and debug_dma_free_coherent()
assume that dma_alloc_coherent() always returns a linear address.  However
it's possible that dma_alloc_coherent() returns a non-linear address.  In
this case, page_to_pfn(virt_to_page(virt)) will return an incorrect pfn.
If the pfn is valid and mapped as a COW page, we will hit the warning when
doing wp_page_copy().

Fix this by calculating pfn for linear and non-linear addresses.

Signed-off-by: Miles Chen <miles.chen@mediatek.com>
---
 lib/dma-debug.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/lib/dma-debug.c b/lib/dma-debug.c
index ea4cc3d..1b34d21 100644
--- a/lib/dma-debug.c
+++ b/lib/dma-debug.c
@@ -1495,14 +1495,22 @@ void debug_dma_alloc_coherent(struct device *dev, size_t size,
 	if (!entry)
 		return;
 
+	/* handle vmalloc and linear addresses */
+	if (!is_vmalloc_addr(virt) && !virt_to_page(virt))
+		return;
+
 	entry->type      = dma_debug_coherent;
 	entry->dev       = dev;
-	entry->pfn	 = page_to_pfn(virt_to_page(virt));
 	entry->offset	 = offset_in_page(virt);
 	entry->size      = size;
 	entry->dev_addr  = dma_addr;
 	entry->direction = DMA_BIDIRECTIONAL;
 
+	if (is_vmalloc_addr(virt))
+		entry->pfn = vmalloc_to_pfn(virt);
+	else
+		entry->pfn = page_to_pfn(virt_to_page(virt));
+
 	add_dma_entry(entry);
 }
 EXPORT_SYMBOL(debug_dma_alloc_coherent);
@@ -1513,13 +1521,21 @@ void debug_dma_free_coherent(struct device *dev, size_t size,
 	struct dma_debug_entry ref = {
 		.type           = dma_debug_coherent,
 		.dev            = dev,
-		.pfn		= page_to_pfn(virt_to_page(virt)),
 		.offset		= offset_in_page(virt),
 		.dev_addr       = addr,
 		.size           = size,
 		.direction      = DMA_BIDIRECTIONAL,
 	};
 
+	/* handle vmalloc and linear addresses */
+	if (!is_vmalloc_addr(virt) && !virt_to_page(virt))
+		return;
+
+	if (is_vmalloc_addr(virt))
+		ref.pfn = vmalloc_to_pfn(virt);
+	else
+		ref.pfn = page_to_pfn(virt_to_page(virt));
+
 	if (unlikely(dma_debug_disabled()))
 		return;
 
-- 
1.9.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH v4] dma-debug: fix incorrect pfn calculation

[Andrew Morton]

On Fri, 17 Nov 2017 06:56:12 +0800 <miles.chen-NuS5LvNUpcJWk0Htik3J/w@public.gmane.org> wrote:

> From: Miles Chen <miles.chen-NuS5LvNUpcJWk0Htik3J/w@public.gmane.org>
> 
> dma-debug reports the following warning:
> 
> [name:panic&]WARNING: CPU: 3 PID: 298 at kernel-4.4/lib/dma-debug.c:604
> debug _dma_assert_idle+0x1a8/0x230()
> DMA-API: cpu touching an active dma mapped cacheline [cln=0x00000882300]
> CPU: 3 PID: 298 Comm: vold Tainted: G        W  O    4.4.22+ #1
> Hardware name: MT6739 (DT)
> Call trace:
> [<ffffff800808acd0>] dump_backtrace+0x0/0x1d4
> [<ffffff800808affc>] show_stack+0x14/0x1c
> [<ffffff800838019c>] dump_stack+0xa8/0xe0
> [<ffffff80080a0594>] warn_slowpath_common+0xf4/0x11c
> [<ffffff80080a061c>] warn_slowpath_fmt+0x60/0x80
> [<ffffff80083afe24>] debug_dma_assert_idle+0x1a8/0x230
> [<ffffff80081dca9c>] wp_page_copy.isra.96+0x118/0x520
> [<ffffff80081de114>] do_wp_page+0x4fc/0x534
> [<ffffff80081e0a14>] handle_mm_fault+0xd4c/0x1310
> [<ffffff8008098798>] do_page_fault+0x1c8/0x394
> [<ffffff800808231c>] do_mem_abort+0x50/0xec
> 
> I found that debug_dma_alloc_coherent() and debug_dma_free_coherent()
> assume that dma_alloc_coherent() always returns a linear address.  However
> it's possible that dma_alloc_coherent() returns a non-linear address.  In
> this case, page_to_pfn(virt_to_page(virt)) will return an incorrect pfn.
> If the pfn is valid and mapped as a COW page, we will hit the warning when
> doing wp_page_copy().
> 
> Fix this by calculating pfn for linear and non-linear addresses.
> 

It's a shame you didn't Cc Christoph, who was the sole reviewer of the
earlier version.

And it's a shame you didn't capture the result of that review
discussion in the v3 changelog.

And it's a shame that you didn't describe how this patch differs from
earlier versions.


Oh well, here's the incremental patch:

--- a/lib/dma-debug.c~dma-debug-fix-incorrect-pfn-calculation-v4
+++ a/lib/dma-debug.c
@@ -1495,15 +1495,22 @@ void debug_dma_alloc_coherent(struct dev
 	if (!entry)
 		return;
 
+	/* handle vmalloc and linear addresses */
+	if (!is_vmalloc_addr(virt) && !virt_to_page(virt))
+		return;
+
 	entry->type      = dma_debug_coherent;
 	entry->dev       = dev;
-	entry->pfn	 = is_vmalloc_addr(virt) ? vmalloc_to_pfn(virt) :
-						page_to_pfn(virt_to_page(virt));
 	entry->offset	 = offset_in_page(virt);
 	entry->size      = size;
 	entry->dev_addr  = dma_addr;
 	entry->direction = DMA_BIDIRECTIONAL;
 
+	if (is_vmalloc_addr(virt))
+		entry->pfn = vmalloc_to_pfn(virt);
+	else
+		entry->pfn = page_to_pfn(virt_to_page(virt));
+
 	add_dma_entry(entry);
 }
 EXPORT_SYMBOL(debug_dma_alloc_coherent);
@@ -1514,14 +1521,21 @@ void debug_dma_free_coherent(struct devi
 	struct dma_debug_entry ref = {
 		.type           = dma_debug_coherent,
 		.dev            = dev,
-		.pfn		= is_vmalloc_addr(virt) ? vmalloc_to_pfn(virt) :
-						page_to_pfn(virt_to_page(virt)),
 		.offset		= offset_in_page(virt),
 		.dev_addr       = addr,
 		.size           = size,
 		.direction      = DMA_BIDIRECTIONAL,
 	};
 
+	/* handle vmalloc and linear addresses */
+	if (!is_vmalloc_addr(virt) && !virt_to_page(virt))
+		return;
+
+	if (is_vmalloc_addr(virt))
+		ref.pfn = vmalloc_to_pfn(virt);
+	else
+		ref.pfn = page_to_pfn(virt_to_page(virt));
+
 	if (unlikely(dma_debug_disabled()))
 		return;
 
_

Re: [PATCH v4] dma-debug: fix incorrect pfn calculation

[Miles Chen]

On Thu, 2017-11-16 at 16:13 -0800, Andrew Morton wrote:
> On Fri, 17 Nov 2017 06:56:12 +0800 <miles.chen@mediatek.com> wrote:
> 
> > From: Miles Chen <miles.chen@mediatek.com>
> > 
> > dma-debug reports the following warning:
> > 
> > [name:panic&]WARNING: CPU: 3 PID: 298 at kernel-4.4/lib/dma-debug.c:604
> > debug _dma_assert_idle+0x1a8/0x230()
> > DMA-API: cpu touching an active dma mapped cacheline [cln=0x00000882300]
> > CPU: 3 PID: 298 Comm: vold Tainted: G        W  O    4.4.22+ #1
> > Hardware name: MT6739 (DT)
> > Call trace:
> > [<ffffff800808acd0>] dump_backtrace+0x0/0x1d4
> > [<ffffff800808affc>] show_stack+0x14/0x1c
> > [<ffffff800838019c>] dump_stack+0xa8/0xe0
> > [<ffffff80080a0594>] warn_slowpath_common+0xf4/0x11c
> > [<ffffff80080a061c>] warn_slowpath_fmt+0x60/0x80
> > [<ffffff80083afe24>] debug_dma_assert_idle+0x1a8/0x230
> > [<ffffff80081dca9c>] wp_page_copy.isra.96+0x118/0x520
> > [<ffffff80081de114>] do_wp_page+0x4fc/0x534
> > [<ffffff80081e0a14>] handle_mm_fault+0xd4c/0x1310
> > [<ffffff8008098798>] do_page_fault+0x1c8/0x394
> > [<ffffff800808231c>] do_mem_abort+0x50/0xec
> > 
> > I found that debug_dma_alloc_coherent() and debug_dma_free_coherent()
> > assume that dma_alloc_coherent() always returns a linear address.  However
> > it's possible that dma_alloc_coherent() returns a non-linear address.  In
> > this case, page_to_pfn(virt_to_page(virt)) will return an incorrect pfn.
> > If the pfn is valid and mapped as a COW page, we will hit the warning when
> > doing wp_page_copy().
> > 
> > Fix this by calculating pfn for linear and non-linear addresses.
> > 
> 
> It's a shame you didn't Cc Christoph, who was the sole reviewer of the
> earlier version.
> 
> And it's a shame you didn't capture the result of that review
> discussion in the v3 changelog.
> 
> And it's a shame that you didn't describe how this patch differs from
> earlier versions.


I am truly sorry about this. I was not sure if I can submit a patch
based on a linux-next patch, so I submit a new patch based on the latest
mainline kernel again.

I know how to do this now. I will do it correctly next time.

Is there anyway to fix this? (send another patch with v3 discussion and
the difference from earlier versions to the commit message).

> Oh well, here's the incremental patch:
> 
> --- a/lib/dma-debug.c~dma-debug-fix-incorrect-pfn-calculation-v4
> +++ a/lib/dma-debug.c
> @@ -1495,15 +1495,22 @@ void debug_dma_alloc_coherent(struct dev
>  	if (!entry)
>  		return;
>  
> +	/* handle vmalloc and linear addresses */
> +	if (!is_vmalloc_addr(virt) && !virt_to_page(virt))
> +		return;
> +
>  	entry->type      = dma_debug_coherent;
>  	entry->dev       = dev;
> -	entry->pfn	 = is_vmalloc_addr(virt) ? vmalloc_to_pfn(virt) :
> -						page_to_pfn(virt_to_page(virt));
>  	entry->offset	 = offset_in_page(virt);
>  	entry->size      = size;
>  	entry->dev_addr  = dma_addr;
>  	entry->direction = DMA_BIDIRECTIONAL;
>  
> +	if (is_vmalloc_addr(virt))
> +		entry->pfn = vmalloc_to_pfn(virt);
> +	else
> +		entry->pfn = page_to_pfn(virt_to_page(virt));
> +
>  	add_dma_entry(entry);
>  }
>  EXPORT_SYMBOL(debug_dma_alloc_coherent);
> @@ -1514,14 +1521,21 @@ void debug_dma_free_coherent(struct devi
>  	struct dma_debug_entry ref = {
>  		.type           = dma_debug_coherent,
>  		.dev            = dev,
> -		.pfn		= is_vmalloc_addr(virt) ? vmalloc_to_pfn(virt) :
> -						page_to_pfn(virt_to_page(virt)),
>  		.offset		= offset_in_page(virt),
>  		.dev_addr       = addr,
>  		.size           = size,
>  		.direction      = DMA_BIDIRECTIONAL,
>  	};
>  
> +	/* handle vmalloc and linear addresses */
> +	if (!is_vmalloc_addr(virt) && !virt_to_page(virt))
> +		return;
> +
> +	if (is_vmalloc_addr(virt))
> +		ref.pfn = vmalloc_to_pfn(virt);
> +	else
> +		ref.pfn = page_to_pfn(virt_to_page(virt));
> +
>  	if (unlikely(dma_debug_disabled()))
>  		return;
>  
> _
> 
> 


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH v4] dma-debug: fix incorrect pfn calculation

[Andrew Morton]

On Fri, 17 Nov 2017 09:23:18 +0800 Miles Chen <miles.chen@mediatek.com> wrote:

> On Thu, 2017-11-16 at 16:13 -0800, Andrew Morton wrote:
> > On Fri, 17 Nov 2017 06:56:12 +0800 <miles.chen@mediatek.com> wrote:
> > 
> > > From: Miles Chen <miles.chen@mediatek.com>
> > > 
> > > dma-debug reports the following warning:
> > > 
> > > [name:panic&]WARNING: CPU: 3 PID: 298 at kernel-4.4/lib/dma-debug.c:604
> > > debug _dma_assert_idle+0x1a8/0x230()
> > > DMA-API: cpu touching an active dma mapped cacheline [cln=0x00000882300]
> > > CPU: 3 PID: 298 Comm: vold Tainted: G        W  O    4.4.22+ #1
> > > Hardware name: MT6739 (DT)
> > > Call trace:
> > > [<ffffff800808acd0>] dump_backtrace+0x0/0x1d4
> > > [<ffffff800808affc>] show_stack+0x14/0x1c
> > > [<ffffff800838019c>] dump_stack+0xa8/0xe0
> > > [<ffffff80080a0594>] warn_slowpath_common+0xf4/0x11c
> > > [<ffffff80080a061c>] warn_slowpath_fmt+0x60/0x80
> > > [<ffffff80083afe24>] debug_dma_assert_idle+0x1a8/0x230
> > > [<ffffff80081dca9c>] wp_page_copy.isra.96+0x118/0x520
> > > [<ffffff80081de114>] do_wp_page+0x4fc/0x534
> > > [<ffffff80081e0a14>] handle_mm_fault+0xd4c/0x1310
> > > [<ffffff8008098798>] do_page_fault+0x1c8/0x394
> > > [<ffffff800808231c>] do_mem_abort+0x50/0xec
> > > 
> > > I found that debug_dma_alloc_coherent() and debug_dma_free_coherent()
> > > assume that dma_alloc_coherent() always returns a linear address.  However
> > > it's possible that dma_alloc_coherent() returns a non-linear address.  In
> > > this case, page_to_pfn(virt_to_page(virt)) will return an incorrect pfn.
> > > If the pfn is valid and mapped as a COW page, we will hit the warning when
> > > doing wp_page_copy().
> > > 
> > > Fix this by calculating pfn for linear and non-linear addresses.
> > > 
> > 
> > It's a shame you didn't Cc Christoph, who was the sole reviewer of the
> > earlier version.
> > 
> > And it's a shame you didn't capture the result of that review
> > discussion in the v3 changelog.
> > 
> > And it's a shame that you didn't describe how this patch differs from
> > earlier versions.
> 
> 
> I am truly sorry about this. I was not sure if I can submit a patch
> based on a linux-next patch, so I submit a new patch based on the latest
> mainline kernel again.
> 
> I know how to do this now. I will do it correctly next time.
> 
> Is there anyway to fix this? (send another patch with v3 discussion and
> the difference from earlier versions to the commit message).

A complete resend is perfectly OK - I will handle the changelog
modifications, etc.

My point is that the Cc: line was incomplete and that the changelog is
missing information, as described above.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>