[RFC v3 09/10] iommu/arm-smmu: Implement reserved region get/put callbacks
Auger Eric
eric.auger at redhat.com
Wed Dec 7 07:02:57 PST 2016
Hi Robin,
On 06/12/2016 19:55, Robin Murphy wrote:
> On 15/11/16 13:09, Eric Auger wrote:
>> The get() populates the list with the PCI host bridge windows
>> and the MSI IOVA range.
>>
>> At the moment an arbitray MSI IOVA window is set at 0x8000000
>> of size 1MB. This will allow to report those info in iommu-group
>> sysfs?
First thank you for reviewing the series. This is definitively helpful!
>>
>> Signed-off-by: Eric Auger <eric.auger at redhat.com>
>>
>> ---
>>
>> RFC v2 -> v3:
>> - use existing get/put_resv_regions
>>
>> RFC v1 -> v2:
>> - use defines for MSI IOVA base and length
>> ---
>> drivers/iommu/arm-smmu.c | 52 ++++++++++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 52 insertions(+)
>>
>> diff --git a/drivers/iommu/arm-smmu.c b/drivers/iommu/arm-smmu.c
>> index 8f72814..81f1a83 100644
>> --- a/drivers/iommu/arm-smmu.c
>> +++ b/drivers/iommu/arm-smmu.c
>> @@ -278,6 +278,9 @@ enum arm_smmu_s2cr_privcfg {
>>
>> #define FSYNR0_WNR (1 << 4)
>>
>> +#define MSI_IOVA_BASE 0x8000000
>> +#define MSI_IOVA_LENGTH 0x100000
>> +
>> static int force_stage;
>> module_param(force_stage, int, S_IRUGO);
>> MODULE_PARM_DESC(force_stage,
>> @@ -1545,6 +1548,53 @@ static int arm_smmu_of_xlate(struct device *dev, struct of_phandle_args *args)
>> return iommu_fwspec_add_ids(dev, &fwid, 1);
>> }
>>
>> +static void arm_smmu_get_resv_regions(struct device *dev,
>> + struct list_head *head)
>> +{
>> + struct iommu_resv_region *region;
>> + struct pci_host_bridge *bridge;
>> + struct resource_entry *window;
>> +
>> + /* MSI region */
>> + region = iommu_alloc_resv_region(MSI_IOVA_BASE, MSI_IOVA_LENGTH,
>> + IOMMU_RESV_MSI);
>> + if (!region)
>> + return;
>> +
>> + list_add_tail(®ion->list, head);
>> +
>> + if (!dev_is_pci(dev))
>> + return;
>> +
>> + bridge = pci_find_host_bridge(to_pci_dev(dev)->bus);
>> +
>> + resource_list_for_each_entry(window, &bridge->windows) {
>> + phys_addr_t start;
>> + size_t length;
>> +
>> + if (resource_type(window->res) != IORESOURCE_MEM &&
>> + resource_type(window->res) != IORESOURCE_IO)
>
> As Joerg commented elsewhere, considering anything other than memory
> resources isn't right (I appreciate you've merely copied my own mistake
> here). We need some other way to handle root complexes where the CPU
> MMIO views of PCI windows appear in PCI memory space - using the I/O
> address of I/O resources only works by chance on Juno, and it still
> doesn't account for config space. I suggest we just leave that out for
> the time being to make life easier (does it even apply to anything other
> than Juno?) and figure it out later.
OK so I understand I should remove IORESOURCE_IO check.
>
>> + continue;
>> +
>> + start = window->res->start - window->offset;
>> + length = window->res->end - window->res->start + 1;
>> + region = iommu_alloc_resv_region(start, length,
>> + IOMMU_RESV_NOMAP);
>> + if (!region)
>> + return;
>> + list_add_tail(®ion->list, head);
>> + }
>> +}
>
> Either way, there's nothing SMMU-specific about PCI windows. The fact
> that we'd have to copy-paste all of this into the SMMUv3 driver
> unchanged suggests it should go somewhere common (although I would be
> inclined to leave the insertion of the fake MSI region to driver-private
> wrappers). As I said before, the current iova_reserve_pci_windows()
> simply wants splitting into appropriate public callbacks for
> get_resv_regions and apply_resv_regions.
Do you mean somewhere common in the arm-smmu subsystem (new file) or in
another subsystem (pci?)
More generally the current implementation does not handle the case where
any of those PCIe host bridge window collide with the MSI window. To me
this is a flaw.
1) Either we take into account the PCIe windows and prevent any
collision when allocating the MSI window.
2) or we do not care about PCIe host bridge windows at kernel level.
If 1) we are back to the original issue of where do we put the MSI
window. Obviously at a place which might not be QEMU friendly anymore.
What allocation policy shall we use?
Second option - sorry I may look stubborn - which I definitively prefer
and which was also advocated by Alex, we handle PCI host bridge windows
at user level. MSI window is reported through the iommu group sysfs.
PCIe host bridge windows can be enumerated through /proc/iomem. Both x86
iommu and arm smmu would report an MSI reserved window. ARM MSI window
would become a de facto reserved window for guests.
Thoughts?
Eric
>
> Robin.
>
>> +static void arm_smmu_put_resv_regions(struct device *dev,
>> + struct list_head *head)
>> +{
>> + struct iommu_resv_region *entry, *next;
>> +
>> + list_for_each_entry_safe(entry, next, head, list)
>> + kfree(entry);
>> +}
>> +
>> static struct iommu_ops arm_smmu_ops = {
>> .capable = arm_smmu_capable,
>> .domain_alloc = arm_smmu_domain_alloc,
>> @@ -1560,6 +1610,8 @@ static int arm_smmu_of_xlate(struct device *dev, struct of_phandle_args *args)
>> .domain_get_attr = arm_smmu_domain_get_attr,
>> .domain_set_attr = arm_smmu_domain_set_attr,
>> .of_xlate = arm_smmu_of_xlate,
>> + .get_resv_regions = arm_smmu_get_resv_regions,
>> + .put_resv_regions = arm_smmu_put_resv_regions,
>> .pgsize_bitmap = -1UL, /* Restricted during device attach */
>> };
>>
>>
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>
More information about the linux-arm-kernel
mailing list