The second operand passed to slot_addr() is declared as int or unsigned int
in all call sites. The left-shift to get the offset of a slot can overflow
if swiotlb size is larger than 4G.
Convert the macro to an inline function and declare the second argument as
phys_addr_t to avoid the potential overflow.
Fixes: 26a7e094783d ("swiotlb: refactor swiotlb_tbl_map_single")
Signed-off-by: Chao Gao <[email protected]>
---
kernel/dma/swiotlb.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
index c5a9190b218f..41ea9fb3efe1 100644
--- a/kernel/dma/swiotlb.c
+++ b/kernel/dma/swiotlb.c
@@ -579,7 +579,10 @@ static void swiotlb_bounce(struct device *dev, phys_addr_t tlb_addr, size_t size
}
}
-#define slot_addr(start, idx) ((start) + ((idx) << IO_TLB_SHIFT))
+static inline phys_addr_t slot_addr(phys_addr_t start, phys_addr_t idx)
+{
+ return start + (idx << IO_TLB_SHIFT);
+}
/*
* Carefully handle integer overflow which can occur when boundary_mask == ~0UL.
base-commit: 5c850d31880e00f063fa2a3746ba212c4bcc510f
--
2.25.1
I also encountered this when sending out another version of the 64-bit swiotlb.
https://lore.kernel.org/all/[email protected]/
Unfortunately, I could not find an environment (e.g., powerpc) to allocate more
than 4G until swiotlb supports 64-bit.
Although xen supports 64-bit, but the hypervisor side limits the max to < 4G.
Dongli Zhang
On 8/19/22 1:45 AM, Chao Gao wrote:
> The second operand passed to slot_addr() is declared as int or unsigned int
> in all call sites. The left-shift to get the offset of a slot can overflow
> if swiotlb size is larger than 4G.
>
> Convert the macro to an inline function and declare the second argument as
> phys_addr_t to avoid the potential overflow.
>
> Fixes: 26a7e094783d ("swiotlb: refactor swiotlb_tbl_map_single")
> Signed-off-by: Chao Gao <[email protected]>
> ---
> kernel/dma/swiotlb.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> index c5a9190b218f..41ea9fb3efe1 100644
> --- a/kernel/dma/swiotlb.c
> +++ b/kernel/dma/swiotlb.c
> @@ -579,7 +579,10 @@ static void swiotlb_bounce(struct device *dev, phys_addr_t tlb_addr, size_t size
> }
> }
>
> -#define slot_addr(start, idx) ((start) + ((idx) << IO_TLB_SHIFT))
> +static inline phys_addr_t slot_addr(phys_addr_t start, phys_addr_t idx)
> +{
> + return start + (idx << IO_TLB_SHIFT);
> +}
>
> /*
> * Carefully handle integer overflow which can occur when boundary_mask == ~0UL.
>
> base-commit: 5c850d31880e00f063fa2a3746ba212c4bcc510f
>
On Fri, Aug 19, 2022 at 06:44:05AM -0700, Dongli Zhang wrote:
>I also encountered this when sending out another version of the 64-bit swiotlb.
>
>https://lore.kernel.org/all/[email protected]/
>
>Unfortunately, I could not find an environment (e.g., powerpc) to allocate more
>than 4G until swiotlb supports 64-bit.
>
>Although xen supports 64-bit, but the hypervisor side limits the max to < 4G.
Sorry. I didn't notice your series before. I agree that the overflow
isn't an issue if swiotlb size cannot be larger than 4GB. That's why
I said the overflow is a potential issue.
In an internal effort to measure the impact of swiotlb size to IO
performance of confidential VM (e.g., TDX VM), we simply added
SWIOTLB_ANY to the default io_tlb_mem to lift the restriction on swiotlb
size. Then we hit this issue and worked out this fix. I posted this
fix because I think the fix by itself is helpful because it removes the
implicit dependency of the left-shift in slot_addr() on swiotlb size and
then someone trying to lift the size limitation won't hit the same issue.
On 8/19/22 6:42 PM, Chao Gao wrote:
> On Fri, Aug 19, 2022 at 06:44:05AM -0700, Dongli Zhang wrote:
>> I also encountered this when sending out another version of the 64-bit swiotlb.
>>
>> https://urldefense.com/v3/__https://lore.kernel.org/all/[email protected]/__;!!ACWV5N9M2RV99hQ!O-2m8d_6yG-OJx1eoiF-wmpJy13HaSz884huJjbeRA9tUdXnUbWsD34MAoY21pSYMdS8tKOM0_7teFvOa3w$
>>
>> Unfortunately, I could not find an environment (e.g., powerpc) to allocate more
>> than 4G until swiotlb supports 64-bit.
>>
>> Although xen supports 64-bit, but the hypervisor side limits the max to < 4G.
>
> Sorry. I didn't notice your series before. I agree that the overflow
> isn't an issue if swiotlb size cannot be larger than 4GB. That's why
> I said the overflow is a potential issue.
>
> In an internal effort to measure the impact of swiotlb size to IO
> performance of confidential VM (e.g., TDX VM), we simply added
> SWIOTLB_ANY to the default io_tlb_mem to lift the restriction on swiotlb
> size. Then we hit this issue and worked out this fix. I posted this
> fix because I think the fix by itself is helpful because it removes the
> implicit dependency of the left-shift in slot_addr() on swiotlb size and
> then someone trying to lift the size limitation won't hit the same issue.
>
Thank you very much for the explanation! I was just curious how to test this
without code modification or powerpc hardware.
Although my RB may not count much:
Reviewed-by: Dongli Zhang <[email protected]>
Dongli Zhang
On Sat, Aug 20, 2022 at 09:42:38AM +0800, Chao Gao wrote:
> In an internal effort to measure the impact of swiotlb size to IO
> performance of confidential VM (e.g., TDX VM), we simply added
> SWIOTLB_ANY to the default io_tlb_mem to lift the restriction on swiotlb
> size. Then we hit this issue and worked out this fix. I posted this
> fix because I think the fix by itself is helpful because it removes the
> implicit dependency of the left-shift in slot_addr() on swiotlb size and
> then someone trying to lift the size limitation won't hit the same issue.
SWIOTLB_ANY is used for real, so I think this is a legitimate fix.
I'll apply it.