[PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Geert Uytterhoeven]

When unbinding the (IOMMU-enabled) R-Car SATA device on Salvator-XS
(R-Car H3 ES2.0), in preparation of rebinding against vfio-platform for
device pass-through for virtualization:

    echo ee300000.sata > /sys/bus/platform/drivers/sata_rcar/unbind

the kernel crashes with:

    Unable to handle kernel paging request at virtual address ffffffbf029ffffc
    Mem abort info:
      ESR = 0x96000006
      Exception class = DABT (current EL), IL = 32 bits
      SET = 0, FnV = 0
      EA = 0, S1PTW = 0
    Data abort info:
      ISV = 0, ISS = 0x00000006
      CM = 0, WnR = 0
    swapper pgtable: 4k pages, 39-bit VAs, pgdp = 000000007e8c586c
    [ffffffbf029ffffc] pgd=000000073bfc6003, pud=000000073bfc6003, pmd=0000000000000000
    Internal error: Oops: 96000006 [#1] SMP
    Modules linked in:
    CPU: 0 PID: 1098 Comm: bash Not tainted 5.0.0-rc5-salvator-x-00452-g37596f884f4318ef #287
    Hardware name: Renesas Salvator-X 2nd version board based on r8a7795 ES2.0+ (DT)
    pstate: 60400005 (nZCv daif +PAN -UAO)
    pc : __free_pages+0x8/0x58
    lr : __dma_direct_free_pages+0x50/0x5c
    sp : ffffff801268baa0
    x29: ffffff801268baa0 x28: 0000000000000000
    x27: ffffffc6f9c60bf0 x26: ffffffc6f9c60bf0
    x25: ffffffc6f9c60810 x24: 0000000000000000
    x23: 00000000fffff000 x22: ffffff8012145000
    x21: 0000000000000800 x20: ffffffbf029fffc8
    x19: 0000000000000000 x18: ffffffc6f86c42c8
    x17: 0000000000000000 x16: 0000000000000070
    x15: 0000000000000003 x14: 0000000000000000
    x13: ffffff801103d7f8 x12: 0000000000000028
    x11: ffffff8011117604 x10: 0000000000009ad8
    x9 : ffffff80110126d0 x8 : ffffffc6f7563000
    x7 : 6b6b6b6b6b6b6b6b x6 : 0000000000000018
    x5 : ffffff8011cf3cc8 x4 : 0000000000004000
    x3 : 0000000000080000 x2 : 0000000000000001
    x1 : 0000000000000000 x0 : ffffffbf029fffc8
    Process bash (pid: 1098, stack limit = 0x00000000c38e3e32)
    Call trace:
     __free_pages+0x8/0x58
     __dma_direct_free_pages+0x50/0x5c
     arch_dma_free+0x1c/0x98
     dma_direct_free+0x14/0x24
     dma_free_attrs+0x9c/0xdc
     dmam_release+0x18/0x20
     release_nodes+0x25c/0x28c
     devres_release_all+0x48/0x4c
     device_release_driver_internal+0x184/0x1f0
     device_release_driver+0x14/0x1c
     unbind_store+0x70/0xb8
     drv_attr_store+0x24/0x34
     sysfs_kf_write+0x4c/0x64
     kernfs_fop_write+0x154/0x1c4
     __vfs_write+0x34/0x164
     vfs_write+0xb4/0x16c
     ksys_write+0x5c/0xbc
     __arm64_sys_write+0x14/0x1c
     el0_svc_common+0x98/0x114
     el0_svc_handler+0x1c/0x24
     el0_svc+0x8/0xc
    Code: d51b4234 17fffffa a9bf7bfd 910003fd (b9403404)
    ---[ end trace 8c564cdd3a1a840f ]---

While I've bisected this to commit e8e683ae9a736407 ("iommu/of: Fix
probe-deferral"), and reverting that commit on post-v5.0-rc4 kernels
does fix the problem, this turned out to be a red herring.

On arm64, arch_teardown_dma_ops() resets dev->dma_ops to NULL.
Hence if a driver has used a managed DMA allocation API, the allocated
DMA memory will be freed using the direct DMA ops, while it may have
been allocated using a custom DMA ops (iommu_dma_ops in this case).

Fix this by reversing the order of the calls to devres_release_all() and
arch_teardown_dma_ops().

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
Question:
Is this safe on arm32, which calls arm_teardown_iommu_dma_ops() instead
of resetting dev->dma_ops?

---
Adding some debug code, and comparing before/after commit
e8e683ae9a736407:

     sata_rcar ee300000.sata: of_iommu_configure:227: err = -517
    -sata_rcar ee300000.sata: of_iommu_configure:230: calling iommu_probe_device()
    -sata_rcar ee300000.sata: iommu_probe_device:122: Calling ipmmu_add_device
    -sata_rcar ee300000.sata: ipmmu_add_device:893
    -sata_rcar ee300000.sata: of_iommu_configure:232: iommu_probe_device() returned -19
    -sata_rcar ee300000.sata: dma_alloc_attrs:257: size 2048, ops =           (null)
    -sata_rcar ee300000.sata: __dma_direct_alloc_pages:104: size 4096
    -sata_rcar ee300000.sata: __dma_direct_alloc_pages:115: trying dma_alloc_from_contiguous()
    -sata_rcar ee300000.sata: dma_alloc_from_contiguous:202: cma_alloc(1) returned page ffffffbf00d20e00
    -sata_rcar ee300000.sata: dma_alloc_attrs:271: allocated using dma_direct_alloc()
    -sata_rcar ee300000.sata: dmam_alloc_attrs:98: size 2048 vaddr ffffff8011e95000 dma_handle 0x0x000000007c040000 attrs 0x0
    -scsi host0: sata_rcar
    -ata1: SATA max UDMA/133 irq 172
    -ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)

DMA memory used to be allocated from CMA, dma_map_ops = NULL.
The SATA driver was probed and initialized.

    +sata_rcar ee300000.sata: of_iommu_configure:234: skipping iommu_probe_device()

SATA driver initialization is now deferred.
Later, it is retried:

    +sata_rcar ee300000.sata: of_iommu_configure:227: err = 0
    +sata_rcar ee300000.sata: of_iommu_configure:230: calling iommu_probe_device()
    +sata_rcar ee300000.sata: iommu_probe_device:122: Calling ipmmu_add_device
    +sata_rcar ee300000.sata: ipmmu_add_device:893
    +sata_rcar ee300000.sata: Using IPMMU context 1
    +iommu: Adding device ee300000.sata to group 0
    +sata_rcar ee300000.sata: ipmmu_add_device:908: Success
    +sata_rcar ee300000.sata: of_iommu_configure:232: iommu_probe_device() returned 0
    +sata_rcar ee300000.sata: dma_alloc_attrs:257: size 2048, ops = iommu_dma_ops
    +sata_rcar ee300000.sata: dma_alloc_attrs:274: allocated using __iommu_alloc_attrs()
    +sata_rcar ee300000.sata: dmam_alloc_attrs:98: size 2048 vaddr ffffff8012135000 dma_handle 0x0x00000000fffff000 attrs 0x0
    +scsi host0: sata_rcar
    +ata1: SATA max UDMA/133 irq 172
    +ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)

DMA memory is now allocated using __iommu_alloc_attrs(), as dma_map_ops
= iommu_dma_ops.

Unbind:

    -sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8011e95000 dma_handle 0x0x000000007c040000 attrs 0x0
    +sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8012135000 dma_handle 0x0x00000000fffff000 attrs 0x0
     sata_rcar ee300000.sata: dma_free_attrs:289: size 2048, ops =           (null)

dma_map_ops = NULL, both before/after!

     sata_rcar ee300000.sata: dma_free_attrs:308: calling dma_direct_free()
    -sata_rcar ee300000.sata: __dma_direct_free_pages:189: size 2048 page ffffffbf00d20e00
    -sata_rcar ee300000.sata: dma_release_from_contiguous:223: cma_release(1) freed
    -sata_rcar ee300000.sata: __dma_direct_free_pages:194: skipping __free_pages()

DMA memory was correctly freed using CMA.

    +sata_rcar ee300000.sata: __dma_direct_free_pages:189: size 2048 page ffffffbf029fffc8
    +sata_rcar ee300000.sata: dma_release_from_contiguous:225: cma_release(1) failed, cma area ffffff8011cf5530
    +sata_rcar ee300000.sata: __dma_direct_free_pages:191: calling __free_pages()
    +Unable to handle kernel paging request at virtual address ffffffbf029ffffc

Memory is incorrectly freed using the direct ops, as dma_map_ops = NULL.
Oops...

After reversing the order of the calls to arch_teardown_dma_ops() and
devres_release_all(), dma_map_ops is still valid, and the DMA memory is
now released using __iommu_free_attrs():

    +sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8012145000 dma_handle 0x0x00000000fffff000 attrs 0x0
    +sata_rcar ee300000.sata: dma_free_attrs:289: size 2048, ops = iommu_dma_ops
    +sata_rcar ee300000.sata: dma_free_attrs:311: calling __iommu_free_attrs()
---
 drivers/base/dd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/base/dd.c b/drivers/base/dd.c
index 8ac10af17c0043a3..d62487d024559620 100644
--- a/drivers/base/dd.c
+++ b/drivers/base/dd.c
@@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
 			drv->remove(dev);
 
 		device_links_driver_cleanup(dev);
-		arch_teardown_dma_ops(dev);
 
 		devres_release_all(dev);
+		arch_teardown_dma_ops(dev);
 		dev->driver = NULL;
 		dev_set_drvdata(dev, NULL);
 		if (dev->pm_domain && dev->pm_domain->dismiss)
-- 
2.17.1

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Christoph Hellwig]

Yes, we should not reset the DMA ops before releasing all resources:

Acked-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Rafael J. Wysocki]

On Thu, Feb 7, 2019 at 8:42 PM Geert Uytterhoeven
<geert+renesas@glider.be> wrote:
>
> When unbinding the (IOMMU-enabled) R-Car SATA device on Salvator-XS
> (R-Car H3 ES2.0), in preparation of rebinding against vfio-platform for
> device pass-through for virtualization:
>
>     echo ee300000.sata > /sys/bus/platform/drivers/sata_rcar/unbind
>
> the kernel crashes with:
>
>     Unable to handle kernel paging request at virtual address ffffffbf029ffffc
>     Mem abort info:
>       ESR = 0x96000006
>       Exception class = DABT (current EL), IL = 32 bits
>       SET = 0, FnV = 0
>       EA = 0, S1PTW = 0
>     Data abort info:
>       ISV = 0, ISS = 0x00000006
>       CM = 0, WnR = 0
>     swapper pgtable: 4k pages, 39-bit VAs, pgdp = 000000007e8c586c
>     [ffffffbf029ffffc] pgd=000000073bfc6003, pud=000000073bfc6003, pmd=0000000000000000
>     Internal error: Oops: 96000006 [#1] SMP
>     Modules linked in:
>     CPU: 0 PID: 1098 Comm: bash Not tainted 5.0.0-rc5-salvator-x-00452-g37596f884f4318ef #287
>     Hardware name: Renesas Salvator-X 2nd version board based on r8a7795 ES2.0+ (DT)
>     pstate: 60400005 (nZCv daif +PAN -UAO)
>     pc : __free_pages+0x8/0x58
>     lr : __dma_direct_free_pages+0x50/0x5c
>     sp : ffffff801268baa0
>     x29: ffffff801268baa0 x28: 0000000000000000
>     x27: ffffffc6f9c60bf0 x26: ffffffc6f9c60bf0
>     x25: ffffffc6f9c60810 x24: 0000000000000000
>     x23: 00000000fffff000 x22: ffffff8012145000
>     x21: 0000000000000800 x20: ffffffbf029fffc8
>     x19: 0000000000000000 x18: ffffffc6f86c42c8
>     x17: 0000000000000000 x16: 0000000000000070
>     x15: 0000000000000003 x14: 0000000000000000
>     x13: ffffff801103d7f8 x12: 0000000000000028
>     x11: ffffff8011117604 x10: 0000000000009ad8
>     x9 : ffffff80110126d0 x8 : ffffffc6f7563000
>     x7 : 6b6b6b6b6b6b6b6b x6 : 0000000000000018
>     x5 : ffffff8011cf3cc8 x4 : 0000000000004000
>     x3 : 0000000000080000 x2 : 0000000000000001
>     x1 : 0000000000000000 x0 : ffffffbf029fffc8
>     Process bash (pid: 1098, stack limit = 0x00000000c38e3e32)
>     Call trace:
>      __free_pages+0x8/0x58
>      __dma_direct_free_pages+0x50/0x5c
>      arch_dma_free+0x1c/0x98
>      dma_direct_free+0x14/0x24
>      dma_free_attrs+0x9c/0xdc
>      dmam_release+0x18/0x20
>      release_nodes+0x25c/0x28c
>      devres_release_all+0x48/0x4c
>      device_release_driver_internal+0x184/0x1f0
>      device_release_driver+0x14/0x1c
>      unbind_store+0x70/0xb8
>      drv_attr_store+0x24/0x34
>      sysfs_kf_write+0x4c/0x64
>      kernfs_fop_write+0x154/0x1c4
>      __vfs_write+0x34/0x164
>      vfs_write+0xb4/0x16c
>      ksys_write+0x5c/0xbc
>      __arm64_sys_write+0x14/0x1c
>      el0_svc_common+0x98/0x114
>      el0_svc_handler+0x1c/0x24
>      el0_svc+0x8/0xc
>     Code: d51b4234 17fffffa a9bf7bfd 910003fd (b9403404)
>     ---[ end trace 8c564cdd3a1a840f ]---
>
> While I've bisected this to commit e8e683ae9a736407 ("iommu/of: Fix
> probe-deferral"), and reverting that commit on post-v5.0-rc4 kernels
> does fix the problem, this turned out to be a red herring.
>
> On arm64, arch_teardown_dma_ops() resets dev->dma_ops to NULL.
> Hence if a driver has used a managed DMA allocation API, the allocated
> DMA memory will be freed using the direct DMA ops, while it may have
> been allocated using a custom DMA ops (iommu_dma_ops in this case).
>
> Fix this by reversing the order of the calls to devres_release_all() and
> arch_teardown_dma_ops().
>
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>

Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

> ---
> Question:
> Is this safe on arm32, which calls arm_teardown_iommu_dma_ops() instead
> of resetting dev->dma_ops?
>
> ---
> Adding some debug code, and comparing before/after commit
> e8e683ae9a736407:
>
>      sata_rcar ee300000.sata: of_iommu_configure:227: err = -517
>     -sata_rcar ee300000.sata: of_iommu_configure:230: calling iommu_probe_device()
>     -sata_rcar ee300000.sata: iommu_probe_device:122: Calling ipmmu_add_device
>     -sata_rcar ee300000.sata: ipmmu_add_device:893
>     -sata_rcar ee300000.sata: of_iommu_configure:232: iommu_probe_device() returned -19
>     -sata_rcar ee300000.sata: dma_alloc_attrs:257: size 2048, ops =           (null)
>     -sata_rcar ee300000.sata: __dma_direct_alloc_pages:104: size 4096
>     -sata_rcar ee300000.sata: __dma_direct_alloc_pages:115: trying dma_alloc_from_contiguous()
>     -sata_rcar ee300000.sata: dma_alloc_from_contiguous:202: cma_alloc(1) returned page ffffffbf00d20e00
>     -sata_rcar ee300000.sata: dma_alloc_attrs:271: allocated using dma_direct_alloc()
>     -sata_rcar ee300000.sata: dmam_alloc_attrs:98: size 2048 vaddr ffffff8011e95000 dma_handle 0x0x000000007c040000 attrs 0x0
>     -scsi host0: sata_rcar
>     -ata1: SATA max UDMA/133 irq 172
>     -ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)
>
> DMA memory used to be allocated from CMA, dma_map_ops = NULL.
> The SATA driver was probed and initialized.
>
>     +sata_rcar ee300000.sata: of_iommu_configure:234: skipping iommu_probe_device()
>
> SATA driver initialization is now deferred.
> Later, it is retried:
>
>     +sata_rcar ee300000.sata: of_iommu_configure:227: err = 0
>     +sata_rcar ee300000.sata: of_iommu_configure:230: calling iommu_probe_device()
>     +sata_rcar ee300000.sata: iommu_probe_device:122: Calling ipmmu_add_device
>     +sata_rcar ee300000.sata: ipmmu_add_device:893
>     +sata_rcar ee300000.sata: Using IPMMU context 1
>     +iommu: Adding device ee300000.sata to group 0
>     +sata_rcar ee300000.sata: ipmmu_add_device:908: Success
>     +sata_rcar ee300000.sata: of_iommu_configure:232: iommu_probe_device() returned 0
>     +sata_rcar ee300000.sata: dma_alloc_attrs:257: size 2048, ops = iommu_dma_ops
>     +sata_rcar ee300000.sata: dma_alloc_attrs:274: allocated using __iommu_alloc_attrs()
>     +sata_rcar ee300000.sata: dmam_alloc_attrs:98: size 2048 vaddr ffffff8012135000 dma_handle 0x0x00000000fffff000 attrs 0x0
>     +scsi host0: sata_rcar
>     +ata1: SATA max UDMA/133 irq 172
>     +ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)
>
> DMA memory is now allocated using __iommu_alloc_attrs(), as dma_map_ops
> = iommu_dma_ops.
>
> Unbind:
>
>     -sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8011e95000 dma_handle 0x0x000000007c040000 attrs 0x0
>     +sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8012135000 dma_handle 0x0x00000000fffff000 attrs 0x0
>      sata_rcar ee300000.sata: dma_free_attrs:289: size 2048, ops =           (null)
>
> dma_map_ops = NULL, both before/after!
>
>      sata_rcar ee300000.sata: dma_free_attrs:308: calling dma_direct_free()
>     -sata_rcar ee300000.sata: __dma_direct_free_pages:189: size 2048 page ffffffbf00d20e00
>     -sata_rcar ee300000.sata: dma_release_from_contiguous:223: cma_release(1) freed
>     -sata_rcar ee300000.sata: __dma_direct_free_pages:194: skipping __free_pages()
>
> DMA memory was correctly freed using CMA.
>
>     +sata_rcar ee300000.sata: __dma_direct_free_pages:189: size 2048 page ffffffbf029fffc8
>     +sata_rcar ee300000.sata: dma_release_from_contiguous:225: cma_release(1) failed, cma area ffffff8011cf5530
>     +sata_rcar ee300000.sata: __dma_direct_free_pages:191: calling __free_pages()
>     +Unable to handle kernel paging request at virtual address ffffffbf029ffffc
>
> Memory is incorrectly freed using the direct ops, as dma_map_ops = NULL.
> Oops...
>
> After reversing the order of the calls to arch_teardown_dma_ops() and
> devres_release_all(), dma_map_ops is still valid, and the DMA memory is
> now released using __iommu_free_attrs():
>
>     +sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8012145000 dma_handle 0x0x00000000fffff000 attrs 0x0
>     +sata_rcar ee300000.sata: dma_free_attrs:289: size 2048, ops = iommu_dma_ops
>     +sata_rcar ee300000.sata: dma_free_attrs:311: calling __iommu_free_attrs()
> ---
>  drivers/base/dd.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> index 8ac10af17c0043a3..d62487d024559620 100644
> --- a/drivers/base/dd.c
> +++ b/drivers/base/dd.c
> @@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
>                         drv->remove(dev);
>
>                 device_links_driver_cleanup(dev);
> -               arch_teardown_dma_ops(dev);
>
>                 devres_release_all(dev);
> +               arch_teardown_dma_ops(dev);
>                 dev->driver = NULL;
>                 dev_set_drvdata(dev, NULL);
>                 if (dev->pm_domain && dev->pm_domain->dismiss)
> --
> 2.17.1
>

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Robin Murphy]

On 07/02/2019 19:36, Geert Uytterhoeven wrote:
> When unbinding the (IOMMU-enabled) R-Car SATA device on Salvator-XS
> (R-Car H3 ES2.0), in preparation of rebinding against vfio-platform for
> device pass-through for virtualization:
> 
>      echo ee300000.sata > /sys/bus/platform/drivers/sata_rcar/unbind
> 
> the kernel crashes with:
> 
>      Unable to handle kernel paging request at virtual address ffffffbf029ffffc
>      Mem abort info:
>        ESR = 0x96000006
>        Exception class = DABT (current EL), IL = 32 bits
>        SET = 0, FnV = 0
>        EA = 0, S1PTW = 0
>      Data abort info:
>        ISV = 0, ISS = 0x00000006
>        CM = 0, WnR = 0
>      swapper pgtable: 4k pages, 39-bit VAs, pgdp = 000000007e8c586c
>      [ffffffbf029ffffc] pgd=000000073bfc6003, pud=000000073bfc6003, pmd=0000000000000000
>      Internal error: Oops: 96000006 [#1] SMP
>      Modules linked in:
>      CPU: 0 PID: 1098 Comm: bash Not tainted 5.0.0-rc5-salvator-x-00452-g37596f884f4318ef #287
>      Hardware name: Renesas Salvator-X 2nd version board based on r8a7795 ES2.0+ (DT)
>      pstate: 60400005 (nZCv daif +PAN -UAO)
>      pc : __free_pages+0x8/0x58
>      lr : __dma_direct_free_pages+0x50/0x5c
>      sp : ffffff801268baa0
>      x29: ffffff801268baa0 x28: 0000000000000000
>      x27: ffffffc6f9c60bf0 x26: ffffffc6f9c60bf0
>      x25: ffffffc6f9c60810 x24: 0000000000000000
>      x23: 00000000fffff000 x22: ffffff8012145000
>      x21: 0000000000000800 x20: ffffffbf029fffc8
>      x19: 0000000000000000 x18: ffffffc6f86c42c8
>      x17: 0000000000000000 x16: 0000000000000070
>      x15: 0000000000000003 x14: 0000000000000000
>      x13: ffffff801103d7f8 x12: 0000000000000028
>      x11: ffffff8011117604 x10: 0000000000009ad8
>      x9 : ffffff80110126d0 x8 : ffffffc6f7563000
>      x7 : 6b6b6b6b6b6b6b6b x6 : 0000000000000018
>      x5 : ffffff8011cf3cc8 x4 : 0000000000004000
>      x3 : 0000000000080000 x2 : 0000000000000001
>      x1 : 0000000000000000 x0 : ffffffbf029fffc8
>      Process bash (pid: 1098, stack limit = 0x00000000c38e3e32)
>      Call trace:
>       __free_pages+0x8/0x58
>       __dma_direct_free_pages+0x50/0x5c
>       arch_dma_free+0x1c/0x98
>       dma_direct_free+0x14/0x24
>       dma_free_attrs+0x9c/0xdc
>       dmam_release+0x18/0x20
>       release_nodes+0x25c/0x28c
>       devres_release_all+0x48/0x4c
>       device_release_driver_internal+0x184/0x1f0
>       device_release_driver+0x14/0x1c
>       unbind_store+0x70/0xb8
>       drv_attr_store+0x24/0x34
>       sysfs_kf_write+0x4c/0x64
>       kernfs_fop_write+0x154/0x1c4
>       __vfs_write+0x34/0x164
>       vfs_write+0xb4/0x16c
>       ksys_write+0x5c/0xbc
>       __arm64_sys_write+0x14/0x1c
>       el0_svc_common+0x98/0x114
>       el0_svc_handler+0x1c/0x24
>       el0_svc+0x8/0xc
>      Code: d51b4234 17fffffa a9bf7bfd 910003fd (b9403404)
>      ---[ end trace 8c564cdd3a1a840f ]---
> 
> While I've bisected this to commit e8e683ae9a736407 ("iommu/of: Fix
> probe-deferral"), and reverting that commit on post-v5.0-rc4 kernels
> does fix the problem, this turned out to be a red herring.
> 
> On arm64, arch_teardown_dma_ops() resets dev->dma_ops to NULL.
> Hence if a driver has used a managed DMA allocation API, the allocated
> DMA memory will be freed using the direct DMA ops, while it may have
> been allocated using a custom DMA ops (iommu_dma_ops in this case).
> 
> Fix this by reversing the order of the calls to devres_release_all() and
> arch_teardown_dma_ops().
> 
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
> Question:
> Is this safe on arm32, which calls arm_teardown_iommu_dma_ops() instead
> of resetting dev->dma_ops?

Should be - the principle is the same, and even if it did break that 
would only be indicative of a separate bug being hidden by this one. 
This fix looks entirely valid and correct to me:

Reviewed-by: Robin Murphy <robin.murphy@arm.com>

> 
> ---
> Adding some debug code, and comparing before/after commit
> e8e683ae9a736407:
> 
>       sata_rcar ee300000.sata: of_iommu_configure:227: err = -517
>      -sata_rcar ee300000.sata: of_iommu_configure:230: calling iommu_probe_device()
>      -sata_rcar ee300000.sata: iommu_probe_device:122: Calling ipmmu_add_device
>      -sata_rcar ee300000.sata: ipmmu_add_device:893
>      -sata_rcar ee300000.sata: of_iommu_configure:232: iommu_probe_device() returned -19
>      -sata_rcar ee300000.sata: dma_alloc_attrs:257: size 2048, ops =           (null)
>      -sata_rcar ee300000.sata: __dma_direct_alloc_pages:104: size 4096
>      -sata_rcar ee300000.sata: __dma_direct_alloc_pages:115: trying dma_alloc_from_contiguous()
>      -sata_rcar ee300000.sata: dma_alloc_from_contiguous:202: cma_alloc(1) returned page ffffffbf00d20e00
>      -sata_rcar ee300000.sata: dma_alloc_attrs:271: allocated using dma_direct_alloc()
>      -sata_rcar ee300000.sata: dmam_alloc_attrs:98: size 2048 vaddr ffffff8011e95000 dma_handle 0x0x000000007c040000 attrs 0x0
>      -scsi host0: sata_rcar
>      -ata1: SATA max UDMA/133 irq 172
>      -ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)
> 
> DMA memory used to be allocated from CMA, dma_map_ops = NULL.
> The SATA driver was probed and initialized.
> 
>      +sata_rcar ee300000.sata: of_iommu_configure:234: skipping iommu_probe_device()
> 
> SATA driver initialization is now deferred.
> Later, it is retried:
> 
>      +sata_rcar ee300000.sata: of_iommu_configure:227: err = 0
>      +sata_rcar ee300000.sata: of_iommu_configure:230: calling iommu_probe_device()
>      +sata_rcar ee300000.sata: iommu_probe_device:122: Calling ipmmu_add_device
>      +sata_rcar ee300000.sata: ipmmu_add_device:893
>      +sata_rcar ee300000.sata: Using IPMMU context 1
>      +iommu: Adding device ee300000.sata to group 0
>      +sata_rcar ee300000.sata: ipmmu_add_device:908: Success
>      +sata_rcar ee300000.sata: of_iommu_configure:232: iommu_probe_device() returned 0
>      +sata_rcar ee300000.sata: dma_alloc_attrs:257: size 2048, ops = iommu_dma_ops
>      +sata_rcar ee300000.sata: dma_alloc_attrs:274: allocated using __iommu_alloc_attrs()
>      +sata_rcar ee300000.sata: dmam_alloc_attrs:98: size 2048 vaddr ffffff8012135000 dma_handle 0x0x00000000fffff000 attrs 0x0
>      +scsi host0: sata_rcar
>      +ata1: SATA max UDMA/133 irq 172
>      +ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)
> 
> DMA memory is now allocated using __iommu_alloc_attrs(), as dma_map_ops
> = iommu_dma_ops.
> 
> Unbind:
> 
>      -sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8011e95000 dma_handle 0x0x000000007c040000 attrs 0x0
>      +sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8012135000 dma_handle 0x0x00000000fffff000 attrs 0x0
>       sata_rcar ee300000.sata: dma_free_attrs:289: size 2048, ops =           (null)
> 
> dma_map_ops = NULL, both before/after!
> 
>       sata_rcar ee300000.sata: dma_free_attrs:308: calling dma_direct_free()
>      -sata_rcar ee300000.sata: __dma_direct_free_pages:189: size 2048 page ffffffbf00d20e00
>      -sata_rcar ee300000.sata: dma_release_from_contiguous:223: cma_release(1) freed
>      -sata_rcar ee300000.sata: __dma_direct_free_pages:194: skipping __free_pages()
> 
> DMA memory was correctly freed using CMA.
> 
>      +sata_rcar ee300000.sata: __dma_direct_free_pages:189: size 2048 page ffffffbf029fffc8
>      +sata_rcar ee300000.sata: dma_release_from_contiguous:225: cma_release(1) failed, cma area ffffff8011cf5530
>      +sata_rcar ee300000.sata: __dma_direct_free_pages:191: calling __free_pages()
>      +Unable to handle kernel paging request at virtual address ffffffbf029ffffc
> 
> Memory is incorrectly freed using the direct ops, as dma_map_ops = NULL.
> Oops...
> 
> After reversing the order of the calls to arch_teardown_dma_ops() and
> devres_release_all(), dma_map_ops is still valid, and the DMA memory is
> now released using __iommu_free_attrs():
> 
>      +sata_rcar ee300000.sata: dmam_release:32: size 2048 vaddr ffffff8012145000 dma_handle 0x0x00000000fffff000 attrs 0x0
>      +sata_rcar ee300000.sata: dma_free_attrs:289: size 2048, ops = iommu_dma_ops
>      +sata_rcar ee300000.sata: dma_free_attrs:311: calling __iommu_free_attrs()
> ---
>   drivers/base/dd.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> index 8ac10af17c0043a3..d62487d024559620 100644
> --- a/drivers/base/dd.c
> +++ b/drivers/base/dd.c
> @@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
>   			drv->remove(dev);
>   
>   		device_links_driver_cleanup(dev);
> -		arch_teardown_dma_ops(dev);
>   
>   		devres_release_all(dev);
> +		arch_teardown_dma_ops(dev);
>   		dev->driver = NULL;
>   		dev_set_drvdata(dev, NULL);
>   		if (dev->pm_domain && dev->pm_domain->dismiss)
> 

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Joerg Roedel]

Hi Geert,

On Thu, Feb 07, 2019 at 08:36:53PM +0100, Geert Uytterhoeven wrote:
> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> index 8ac10af17c0043a3..d62487d024559620 100644
> --- a/drivers/base/dd.c
> +++ b/drivers/base/dd.c
> @@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
>  			drv->remove(dev);
>  
>  		device_links_driver_cleanup(dev);
> -		arch_teardown_dma_ops(dev);
>  
>  		devres_release_all(dev);
> +		arch_teardown_dma_ops(dev);
>  		dev->driver = NULL;
>  		dev_set_drvdata(dev, NULL);
>  		if (dev->pm_domain && dev->pm_domain->dismiss)

Thanks for the fix! Should it also be tagged for stable and get a Fixes
tag? I know it only triggers with a fix in v5.0-rc, but still...

Regards,

	Joerg

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Robin Murphy]

On 08/02/2019 16:40, Joerg Roedel wrote:
> Hi Geert,
> 
> On Thu, Feb 07, 2019 at 08:36:53PM +0100, Geert Uytterhoeven wrote:
>> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
>> index 8ac10af17c0043a3..d62487d024559620 100644
>> --- a/drivers/base/dd.c
>> +++ b/drivers/base/dd.c
>> @@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
>>   			drv->remove(dev);
>>   
>>   		device_links_driver_cleanup(dev);
>> -		arch_teardown_dma_ops(dev);
>>   
>>   		devres_release_all(dev);
>> +		arch_teardown_dma_ops(dev);
>>   		dev->driver = NULL;
>>   		dev_set_drvdata(dev, NULL);
>>   		if (dev->pm_domain && dev->pm_domain->dismiss)
> 
> Thanks for the fix! Should it also be tagged for stable and get a Fixes
> tag? I know it only triggers with a fix in v5.0-rc, but still...

I think so:

Fixes: 09515ef5ddad ("of/acpi: Configure dma operations at probe time 
for platform/amba/pci bus devices")

There aren't many drivers using dmam_alloc_*(), let alone which would 
also find themselves behind an IOMMU on an Arm system, but it turns out 
I actually have another one which can reproduce the BUG() with 5.0-rc.

I've tried a 4.12 kernel with a bit of instrumentation[1] and sure 
enough the devres-managed buffer is freed with the wrong ops[2] even 
then. How it manages not to blow up more catastrophically I have no 
idea... I guess at best it just leaks the buffers and IOMMU mappings, 
and at worst quietly frees random other pages instead.

Robin.

--------------
[1]

diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 4f3eecedca2d..f4dbaa5598e3 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -491,6 +491,7 @@ static inline void *dma_alloc_attrs(struct device 
*dev, size_t size,
  		return NULL;

  	cpu_addr = ops->alloc(dev, size, dma_handle, flag, attrs);
+	dev_info(dev, "alloc %lx %lx\n", (unsigned long)cpu_addr, (unsigned 
long)ops);
  	debug_dma_alloc_coherent(dev, size, *dma_handle, cpu_addr);
  	return cpu_addr;
  }
@@ -512,6 +513,7 @@ static inline void dma_free_attrs(struct device 
*dev, size_t size,

  	debug_dma_free_coherent(dev, size, cpu_addr, dma_handle);
  	ops->free(dev, size, cpu_addr, dma_handle, attrs);
+	dev_info(dev, "free %lx %lx\n", (unsigned long)cpu_addr, (unsigned 
long)ops);
  }

  static inline void *dma_alloc_coherent(struct device *dev, size_t size,

-------------
[2]

/ # echo '0000:03:00.0' > /sys/bus/pci/drivers/sata_sil24/bind
[  107.417252] sata_sil24 0000:03:00.0: alloc ffff00000a6f9000 
ffff0000089b8090
[  107.424397] sata_sil24 0000:03:00.0: alloc ffff00000a719000 
ffff0000089b8090
[  107.432216] scsi host0: sata_sil24
[  107.436134] scsi host1: sata_sil24
[  107.439853] ata7: SATA max UDMA/100 host m128@0x50084000 port 
0x50080000 irq 51
[  107.447228] ata8: SATA max UDMA/100 host m128@0x50084000 port 
0x50082000 irq 51
/ # echo '0000:03:00.0' > /sys/bus/pci/drivers/sata_sil24/unbind
...
[  112.048654] sata_sil24 0000:03:00.0: free ffff00000a719000 
ffff0000089b8120
[  112.055579] sata_sil24 0000:03:00.0: free ffff00000a6f9000 
ffff0000089b8120

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Geert Uytterhoeven]

Hi Robin,

On Fri, Feb 8, 2019 at 6:55 PM Robin Murphy <robin.murphy@arm.com> wrote:
> On 08/02/2019 16:40, Joerg Roedel wrote:
> > On Thu, Feb 07, 2019 at 08:36:53PM +0100, Geert Uytterhoeven wrote:
> >> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> >> index 8ac10af17c0043a3..d62487d024559620 100644
> >> --- a/drivers/base/dd.c
> >> +++ b/drivers/base/dd.c
> >> @@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
> >>                      drv->remove(dev);
> >>
> >>              device_links_driver_cleanup(dev);
> >> -            arch_teardown_dma_ops(dev);
> >>
> >>              devres_release_all(dev);
> >> +            arch_teardown_dma_ops(dev);
> >>              dev->driver = NULL;
> >>              dev_set_drvdata(dev, NULL);
> >>              if (dev->pm_domain && dev->pm_domain->dismiss)
> >
> > Thanks for the fix! Should it also be tagged for stable and get a Fixes

FTR, Greg has added it to driver-core-testing, with a CC to stable.

> > tag? I know it only triggers with a fix in v5.0-rc, but still...
>
> I think so:
>
> Fixes: 09515ef5ddad ("of/acpi: Configure dma operations at probe time
> for platform/amba/pci bus devices")

Thanks! It won't backport cleanly due to commit dc3c05504d38849f
("dma-mapping: remove dma_deconfigure") in v4.20, though.

> There aren't many drivers using dmam_alloc_*(), let alone which would
> also find themselves behind an IOMMU on an Arm system, but it turns out
> I actually have another one which can reproduce the BUG() with 5.0-rc.

SATA core uses dmam_alloc_*().

> I've tried a 4.12 kernel with a bit of instrumentation[1] and sure
> enough the devres-managed buffer is freed with the wrong ops[2] even
> then. How it manages not to blow up more catastrophically I have no
> idea... I guess at best it just leaks the buffers and IOMMU mappings,
> and at worst quietly frees random other pages instead.

May depend on the actual ops, and whether CMA is used or not.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@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

Re: [PATCH/RFC] driver core: Postpone DMA tear-down until after devres release

[Robin Murphy]

On 08/02/2019 18:55, Geert Uytterhoeven wrote:
> Hi Robin,
> 
> On Fri, Feb 8, 2019 at 6:55 PM Robin Murphy <robin.murphy@arm.com> wrote:
>> On 08/02/2019 16:40, Joerg Roedel wrote:
>>> On Thu, Feb 07, 2019 at 08:36:53PM +0100, Geert Uytterhoeven wrote:
>>>> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
>>>> index 8ac10af17c0043a3..d62487d024559620 100644
>>>> --- a/drivers/base/dd.c
>>>> +++ b/drivers/base/dd.c
>>>> @@ -968,9 +968,9 @@ static void __device_release_driver(struct device *dev, struct device *parent)
>>>>                       drv->remove(dev);
>>>>
>>>>               device_links_driver_cleanup(dev);
>>>> -            arch_teardown_dma_ops(dev);
>>>>
>>>>               devres_release_all(dev);
>>>> +            arch_teardown_dma_ops(dev);
>>>>               dev->driver = NULL;
>>>>               dev_set_drvdata(dev, NULL);
>>>>               if (dev->pm_domain && dev->pm_domain->dismiss)
>>>
>>> Thanks for the fix! Should it also be tagged for stable and get a Fixes
> 
> FTR, Greg has added it to driver-core-testing, with a CC to stable.

So I see, great!

>>> tag? I know it only triggers with a fix in v5.0-rc, but still...
>>
>> I think so:
>>
>> Fixes: 09515ef5ddad ("of/acpi: Configure dma operations at probe time
>> for platform/amba/pci bus devices")
> 
> Thanks! It won't backport cleanly due to commit dc3c05504d38849f
> ("dma-mapping: remove dma_deconfigure") in v4.20, though.

Ah yes - backports beyond that should simply be a case of moving the 
dma_deconfigure() wrapper in the same manner.

Thanks,
Robin.

>> There aren't many drivers using dmam_alloc_*(), let alone which would
>> also find themselves behind an IOMMU on an Arm system, but it turns out
>> I actually have another one which can reproduce the BUG() with 5.0-rc.
> 
> SATA core uses dmam_alloc_*().
> 
>> I've tried a 4.12 kernel with a bit of instrumentation[1] and sure
>> enough the devres-managed buffer is freed with the wrong ops[2] even
>> then. How it manages not to blow up more catastrophically I have no
>> idea... I guess at best it just leaks the buffers and IOMMU mappings,
>> and at worst quietly frees random other pages instead.
> 
> May depend on the actual ops, and whether CMA is used or not.
> 
> Gr{oetje,eeting}s,
> 
>                          Geert
>