2012-02-28 17:59:06

by Tom Mingarelli

Subject: [Intel IOMMU PATCH] Re-evaluate RMRR info for devices removed from si domain

This patch is being submitted to handle the case where a pci device is
placed into the si domain, when booting in iommu passthrough mode, then
removed. The RMRR information for such devices need to be re-processed
to avoid DMA Read errors due to the Present Bit being cleared in the
device's context entry.
Signed-off-by: Thomas Mingarelli <[email protected]>

diff -up linux-3.2.4/drivers/iommu/intel-iommu.c.ORIG linux-3.2.4/drivers/iommu/intel-iommu.c
--- linux-3.2.4/drivers/iommu/intel-iommu.c.ORIG 2012-02-22 16:10:07.105254130 -0600
+++ linux-3.2.4/drivers/iommu/intel-iommu.c 2012-02-28 12:49:38.537694065 -0600
@@ -2676,7 +2676,8 @@ static int iommu_dummy(struct pci_dev *p
static int iommu_no_mapping(struct device *dev)
{
struct pci_dev *pdev;
- int found;
+ struct dmar_rmrr_unit *rmrr;
+ int i, ret, found;

if (unlikely(dev->bus != &pci_bus_type))
return 1;
@@ -2699,7 +2700,24 @@ static int iommu_no_mapping(struct devic
*/
domain_remove_one_dev_info(si_domain, pdev);
printk(KERN_INFO "32bit %s uses non-identity mapping\n",
- pci_name(pdev));
+ pci_name(pdev));
+ printk(KERN_INFO "IOMMU: Re-processing RMRR information for device %s:\n",
+ pci_name(pdev));
+ for_each_rmrr_units(rmrr) {
+ for (i = 0; i < rmrr->devices_cnt; i++) {
+ /*
+ * Here we are just concerned with finding the
+ * one device that was removed from the
+ * si_domain and re-evaluating its RMRR info.
+ */
+ if (rmrr->devices[i] != pdev)
+ continue;
+ ret = iommu_prepare_rmrr_dev(rmrr, pdev);
+ if (ret)
+ printk(KERN_ERR
+ "IOMMU: mapping reserved region failed for device.\n");
+ }
+ }
return 0;
}
} else {

2012-02-28 23:06:28

by Chris Wright

Subject: Re: [Intel IOMMU PATCH] Re-evaluate RMRR info for devices removed from si domain

* Mingarelli, Thomas ([email protected]) wrote:
> This patch is being submitted to handle the case where a pci device is
> placed into the si domain, when booting in iommu passthrough mode, then
> removed. The RMRR information for such devices need to be re-processed
> to avoid DMA Read errors due to the Present Bit being cleared in the
> device's context entry.

What happens if that device is attached to a VM, then detached and
rebound to a host driver? I suspect the reserved memory range won't get
mapped properly for that device in that case.

thanks,
-chris