Hi, all:
It's been a long time since my patch v3. I don't known when Laurent's
probe-deferral series will be upstreamed? I really have no time to wait. So
I add patch 4/5, to keep both cases(Laurent's series upstreamed or not),
smmu-v3 can always work well.
Changelog:
v3 -> v4:
1. base upon the lastest arm-smmu-v3.c update patchset.
2. add patch 4/5, keep smmu-v3 can still work normally before Laurent's series
upstreamed.
Describe process flow as below:
All: (A-->B means: function A directly or indirectly call function B)
[1] of_iommu_init-->arm_smmu_of_iommu_init-->of_iommu_set_ops
1) without Laurent's series
[1] of_platform_populate-->of_dma_configure-->of_iommu_configure-->.of_xlate
/* time before step [2], so smmu is NULL yet */
smmu = platform_get_drvdata(pdev);
if (!smmu) {
dev->archdata.iommu = ERR_PTR(-EPROBE_DEFER);
return -EPROBE_DEFER;
}
[2] arm_smmu_device_dt_probe-->platform_set_drvdata
[3] call .add_device
if (dev->archdata.iommu == ERR_PTR(-EPROBE_DEFER)) {
dev->archdata.iommu = NULL;
/* scan master's "iommus" property */
return of_iommu_configure(dev, dev->of_node) ? 0 : -ENODEV;
}
[4] call master's probe function
2) with Laurent's series
[1] of_platform_populate-->of_dma_configure
the call of of_iommu_configure have been putted off, see [4] below
so the value of dev->archdata.iommu have no change
[2] arm_smmu_device_dt_probe-->platform_set_drvdata
[3] call .add_device
if (dev->archdata.iommu == ERR_PTR(-EPROBE_DEFER)) {
/* see [1], dev->archdata.iommu is NULL, so this branch can not
be entered */
}
[4] call .of_xlate
[5] call master's probe function
v2 -> v3:
1. add support for pci device hotplug, which missed in patch v2.
2. only support #iommu-cells = <1>, add corresponding description in
arm,smmu-v3.txt.
3. add function find_smmu_by_device which extracted from find_smmu_by_node, to
resolve
the problem mentioned by Robin Murphy in [PATCH v2 7/9].
Additionally:
+ platform_set_drvdata(pdev, smmu); //Patch v2
+ dev->archdata.iommu = smmu; //Patch v3, dev = &pdev->dev
v1 -> v2:
update the implementation of patch 1/9 according to Will Deacon's suggestion.
update the comment of patch 3/9 and 4/9.
use arm_smmu_options to skip the execution of command CMD_PREFETCH_CONFIG, see
patch 5/9.
patch 6/9 is base on Laurent's series, to support probe deferral.
patch 7/9 according to Robin Murphy's suggestion, remove global variable
arm_smmu_devices, thanks.
patch 9/9 add support for a master with multiple stream IDs.
Zhen Lei (5):
iommu/arm-smmu: to support probe deferral
iommu/arm-smmu: rename __arm_smmu_get_pci_sid
iommu/arm-smmu: add support for non-pci devices
iommu/arm-smmu: to backward compatible with probe non-deferral
iommu/arm-smmu: describe the limitation of #iommu-cells
.../devicetree/bindings/iommu/arm,smmu-v3.txt | 6 +
drivers/iommu/arm-smmu-v3.c | 166 +++++++++++++++++----
2 files changed, 143 insertions(+), 29 deletions(-)
--
2.5.0
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
