[PATCH] iommu: Check return of __iommu_attach_device()

[Shameerali Kolothum Thodi]

> -----Original Message-----
> From: Robin Murphy [mailto:robin.murphy at arm.com]
> Sent: 20 November 2020 14:07
> To: Will Deacon <will at kernel.org>; Shameerali Kolothum Thodi
> <shameerali.kolothum.thodi at huawei.com>
> Cc: linux-arm-kernel at lists.infradead.org; iommu at lists.linux-foundation.org;
> Linuxarm <linuxarm at huawei.com>
> Subject: Re: [PATCH] iommu: Check return of __iommu_attach_device()
> 
> On 2020-11-20 11:15, Will Deacon wrote:
> > On Thu, Nov 19, 2020 at 04:58:46PM +0000, Shameer Kolothum wrote:
> >> Currently iommu_create_device_direct_mappings() is called
> >> without checking the return of __iommu_attach_device(). This
> >> may result in failures in iommu driver if dev attach returns
> >> error.
> >>
> >> Fixes: ce574c27ae27("iommu: Move
> iommu_group_create_direct_mappings() out of iommu_group_add_device()")
> >> Signed-off-by: Shameer Kolothum
> <shameerali.kolothum.thodi at huawei.com>
> >> ---
> >> Crash log:
> >> [   31.353605] hns3 0000:7d:00.3: Adding to iommu group 10
> >> [   31.358822] Unable to handle kernel NULL pointer dereference at virtual
> address 0000000000000018
> >> [   31.367567] Mem abort info:
> >> [   31.370350]   ESR = 0x96000004
> >> [   31.373391]   EC = 0x25: DABT (current EL), IL = 32 bits
> >> [   31.378680]   SET = 0, FnV = 0
> >> [   31.381720]   EA = 0, S1PTW = 0
> >> [   31.384847] Data abort info:
> >> [   31.387716]   ISV = 0, ISS = 0x00000004
> >> [   31.391535]   CM = 0, WnR = 0
> >> [   31.394491] [0000000000000018] user address but active_mm is
> swapper
> >> [   31.400818] Internal error: Oops: 96000004 [#1] PREEMPT SMP
> >> [   31.406365] Modules linked in:
> >> [   31.409409] CPU: 21 PID: 1 Comm: swapper/0 Not tainted
> 5.10.0-rc4-00008-gdd5aba9d719-dirty #79
> >> [   31.417980] Hardware name: Huawei TaiShan 200 (Model
> 2280)/BC82AMDD, BIOS 2280-V2 CS V3.B220.01 03/19/2020
> >> [   31.427588] pstate: 00c00009 (nzcv daif +PAN +UAO -TCO BTYPE=--)
> >> [   31.433566] pc : arm_smmu_tlb_inv_range+0x178/0x1f0
> >> [   31.438422] lr : arm_smmu_tlb_inv_range+0x5c/0x1f0
> >> [   31.443190] sp : ffff80001043b4e0
> >> ...
> >> [   31.531175] Call trace:
> >> [   31.533613]  arm_smmu_tlb_inv_range+0x178/0x1f0
> >> [   31.538122]  arm_smmu_iotlb_sync+0x2c/0x38
> >> [   31.542200]  iommu_unmap+0x60/0x90
> >> [   31.545585]  __iommu_map+0x110/0x1f0
> >> [   31.549144]
> iommu_create_device_direct_mappings.isra.34+0x1ac/0x250
> >> [   31.555468]  iommu_probe_device+0x6c/0x110
> >> [   31.559551]  iort_iommu_configure_id+0x114/0x218
> >> [   31.564148]  acpi_dma_configure_id+0x94/0xe0
> >> [   31.568402]  pci_dma_configure+0xc8/0xf0
> >> [   31.572310]  really_probe+0xd4/0x3e0
> >> [   31.575871]  driver_probe_device+0x5c/0xc0
> >>
> >> ---
> >>   drivers/iommu/iommu.c | 10 ++++++----
> >>   1 file changed, 6 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> >> index b53446bb8c6b..0f4dc25d46c9 100644
> >> --- a/drivers/iommu/iommu.c
> >> +++ b/drivers/iommu/iommu.c
> >> @@ -264,16 +264,18 @@ int iommu_probe_device(struct device *dev)
> >>   	 */
> >>   	iommu_alloc_default_domain(group, dev);
> >>
> >> -	if (group->default_domain)
> >> +	if (group->default_domain) {
> >>   		ret = __iommu_attach_device(group->default_domain, dev);
> >> +		if (ret) {
> >> +			iommu_group_put(group);
> >> +			goto err_release;
> >> +		}
> >> +	}
> >
> > This looks sensible to me, but what I don't understand is where that
> > NULL pointer is coming from in the first place. iommu_map() operates
> > on the domain, so why does it matter if the attach fails? What is being
> > accessed at arm_smmu_tlb_inv_range+0x178/0x1f0 ?
> 
> Probably because the domain is a hollow fake until the first successful
> attach - even TLB maintenance depends on having decided a pagetable format.

I think, in this particular instance, what happens is, dev reports RMR
regions (IOMMU_RESV_DIRECT) but attach_dev() fails early without
setting, smmu_domain->smmu  =  smmu.

iommu_probe_device()
  __iommu_attach_dev()  -->return err, but carries on.
  iommu_create_device_direct_mappings()
    iommu_get_resv_regions()  --> dev has IOMMU_RESV_DIRECT regions
    iommu_map()  
      __iommu_map()   
        arm_smmu_map()  -->return err
      iommu_unmap()    --> unroll on map failure
        __iommu_unmap()  --> size is zero. So returns.
        iommu_iotlb_sync()
          arm_smmu_iotlb_sync()
            arm_smmu_tlb_inv_range()  --> smmu is NULL

Thanks,
Shameer