[PATCH 0/2] Fix incorrect warning from dma-debug

Wed Jan 25 12:32:34 PST 2017

Hi Robin,

On Wed, Jan 25, 2017 at 6:27 PM, Robin Murphy <robin.murphy at arm.com> wrote:
> On 25/01/17 16:23, Geert Uytterhoeven wrote:
>> On Mon, May 9, 2016 at 11:37 AM, Robin Murphy <robin.murphy at arm.com> wrote:
>>> On 08/05/16 11:59, Niklas Söderlund wrote:
>>>> While using CONFIG_DMA_API_DEBUG i came across this warning which I
>>>> think is a false positive. As shown dma_sync_single_for_device() are
>>>> called from the dma_map_single() call path. This triggers the warning
>>>> since the dma-debug code have not yet been made aware of the mapping.
>>>
>>> Almost right ;) The thing being mapped (the SPI device's buffer) and the
>>> thing being synced (the IOMMU's PTE) are entirely unrelated. Due to the
>>> current of_iommu_init() setup, the IOMMU is probed long before
>>> dma_debug_init() gets called, therefore DMA debug is missing entries for
>>> some of the initial page table mappings and gets confused when we update
>>> them later.
>>
>> I think I've been seeing the same as Niklas since quite a while.
>> Finally I had a deeper look, and it looks like there is a bug somewhere,
>> causing the wrong IOMMU PTE to be synced.
>>
>>>> I try to solve this by introducing __dma_sync_single_for_device() which
>>>> do not call into the dma-debug code. I'm no expert and this might be a
>>>> bad way of solving the problem but it allowed me to keep working.
>>>
>>> The simple fix should be to just call dma_debug_init() from a sufficiently
>>> earlier initcall level. The best would be to sort out a proper device
>>> dependency order to avoid the whole early-IOMMU-creation thing entirely.
>>
>> And so I did. After disabling the call to dma_debug_fs_init(), you can call
>> dma_debug_init() quite early. But the warning didn't go away:
>
> Yet the underlying reason has subtly changed!
>
>>     ipmmu-vmsa e67b0000.mmu: DMA-API: device driver tries to sync DMA memory
>>         it has not allocated [device address=0x000000067bab2ff8] [size=8 bytes]
>>     ------------[ cut here ]------------
>>     WARNING: CPU: 0 PID: 174 at lib/dma-debug.c:1235 check_sync+0xcc/0x568
>>     ...
>>     [<ffffff800823a3a4>] check_sync+0xcc/0x568
>>     [<ffffff800823a8d0>] debug_dma_sync_single_for_device+0x44/0x4c
>>     [<ffffff80082b8d34>] __arm_lpae_set_pte.isra.3+0x6c/0x78
>>     [<ffffff80082b977c>] __arm_lpae_map+0x318/0x384
>>     [<ffffff80082b9c58>] arm_lpae_map+0xb0/0xc4
>>     [<ffffff80082bbc58>] ipmmu_map+0x48/0x58
>>     [<ffffff80082b6754>] iommu_map+0x120/0x1fc
>>     [<ffffff80082b7bc8>] __iommu_dma_map+0xb8/0xec
>>     [<ffffff80082b8514>] iommu_dma_map_page+0x50/0x58
>>     [<ffffff8008092d28>] __iommu_map_page+0x54/0x98
>>
>> So, who allocated that memory?
>>
>> During early kernel init (before fs_initcall(dma_debug_init)):
>>
>>     arm-lpae io-pgtable: arm_lpae_alloc_pgtable:652: cfg->ias = 32
>
> Note that you have a 32-bit IAS...

Ah, so that's what it means...

>>         data->pg_shift = 12 va_bits = 20
>>     arm-lpae io-pgtable: arm_lpae_alloc_pgtable:657: data->bits_per_level = 9
>>         data->levels = 3 pgd_bits = 2
>>     ipmmu-vmsa e67b0000.mmu: __arm_lpae_alloc_pages:224
>>         dma_map_single(0xffffffc63bab2000, 32) returned 0x000000067bab2000
>>
>> Hence 0x67bab2000 is the PGD, which has only 4 entries (32 bytes).
>> Call stack:
>>
>>     [<ffffff80082b9240>] __arm_lpae_alloc_pages.isra.11+0x144/0x1e8
>>     [<ffffff80082b93c0>] arm_64_lpae_alloc_pgtable_s1+0xdc/0x118
>>     [<ffffff80082b9440>] arm_32_lpae_alloc_pgtable_s1+0x44/0x68
>>     [<ffffff80082b8b1c>] alloc_io_pgtable_ops+0x4c/0x80
>>     [<ffffff80082bbf28>] ipmmu_attach_device+0xd0/0x3b0
>>
>> When starting DMA from the device:
>>
>>     iommu: map: iova 0xfffffff000 pa 0x000000067a555000 size 0x1000 pgsize 4096
>
> ...then count the f's carefully.

Indeed, 40-bits.

>>     arm-lpae io-pgtable: __arm_lpae_map:318: iova 0xfffffff000
>>         phys 0x000000067a555000 size 4096 lvl 1 ptep 0xffffffc63bab2000
>>     arm-lpae io-pgtable: __arm_lpae_map:320: incr. ptep to 0xffffffc63bab2ff8
>>     ipmmu-vmsa e67b0000.mmu: __arm_lpae_alloc_pages:224
>>         dma_map_single(0xffffffc63a490000, 4096) returned 0x000000067a490000
>>     ipmmu-vmsa e67b0000.mmu: DMA-API: device driver tries to sync DMA memory
>>         it has not allocated [device address=0x000000067bab2ff8] [size=8 bytes]
>>
>> __arm_lpae_map() added "ARM_LPAE_LVL_IDX(iova, lvl, data)" == 0xff8 to ptep
>> (the PGD base address), but the PGD has only 32 bytes, leading to the warning.
>>
>> Does my analysis make sense?
>> Do you have a clue?
>
> The initial false positive misleads from the fact that this is actually
> DMA-debug doing its job admirably. The bug lies in however you ended up
> trying to map a 40-bit IOVA in a 32-bit pagetable.

Right, I completely forgot that I was still running with a patch to
allow the DMAC to use 40-bit addresses:

    dma_set_mask_and_coherent(dmac->dev, DMA_BIT_MASK(40));

which works fine without IOMMU, but fails with, as the IPMMU-VMSA driver
doesn't support that yet.

Thanks a lot, and sorry for the noise...

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds