Re: [Qemu-devel] [PATCH v2] s390x/pci: Set the iommu region size mpcifc request

[Pierre Morel <pmorel@linux.ibm.com>]

On 16/01/2019 13:40, Halil Pasic wrote:
> On Tue, 15 Jan 2019 10:35:42 -0500
> Collin Walling <walling@linux.ibm.com> wrote:
> 
>> On 1/10/19 8:00 AM, Pierre Morel wrote:
>>> The size of the accessible iommu memory region in the guest
>>> is given to the IOMMU by the guest through the mpcifc request
>>> specifying the PCI Base Address and the PCI Address Limit.
>>>
>>> Let set the size of the IOMMU region to:
>>>       (PCI Address Limit) - (PCI Base Address) + 1.
>>>
>>> Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
>>> ---
>>>    hw/s390x/s390-pci-bus.c | 2 +-
>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
>>> index 69e0671..e97696a 100644
>>> --- a/hw/s390x/s390-pci-bus.c
>>> +++ b/hw/s390x/s390-pci-bus.c
>>> @@ -660,7 +660,7 @@ void s390_pci_iommu_enable(S390PCIIOMMU *iommu)
>>>        char *name = g_strdup_printf("iommu-s390-%04x", iommu->pbdev->uid);
>>>        memory_region_init_iommu(&iommu->iommu_mr, sizeof(iommu->iommu_mr),
>>>                                 TYPE_S390_IOMMU_MEMORY_REGION, OBJECT(&iommu->mr),
>>> -                             name, iommu->pal + 1);
>>> +                             name, iommu->pal - iommu->pba + 1);
> 
>  From the the look of this, I would say we basically used the address
> denoting the end of the region as the size of the region. This smells
> like a bug to me, but the commit message and the title ain't clear about
> this, and there is no fixes tag. Because of the latter I did some digging
> and came to commit f7c40aa "s390x/pci: fix failures of dma
> map/unmap" (Yi Min Zhao, 2016-06-19) which basically did the inverse of
> this commit!
> 
> My initial motivation was to check if this is stable material. But now
> I'm very confused. I'm admittedly zPCI incompetent. Could some of the
> people that understand what is going on help me feel better about this
> patch?
> 
> Regards,
> Halil


The patch you speak about corrected the problem described in its comment 
by setting the offset address of the subregion to 0, making sure 
VFIO_PCI works for Z but introduced a bug we did not see at that time by 
making the subregion too large.

This patch correct the bug, I can add a reference to this with:
fixing: commit f7c40aa1e7feb50bc4d4bc171fa811bdd9a93e51

Regards,
Pierre

> 
> 
> 
>>>        iommu->enabled = true;
>>>        memory_region_add_subregion(&iommu->mr, 0, MEMORY_REGION(&iommu->iommu_mr));
>>>        g_free(name);
>>>
>>
>> Acked-by: Collin Walling <walling@linux.ibm.com>
>>
>>
> 


-- 
Pierre Morel
Linux/KVM/QEMU in Böblingen - Germany