@@ -8,6 +8,7 @@
#include <linux/local_lock.h>
#include <linux/of_address.h>
#include <linux/of_platform.h>
+#include <linux/pm_runtime.h>
#include <kvm/arm_smmu_v3.h>
@@ -18,6 +19,7 @@ struct host_arm_smmu_device {
pkvm_handle_t id;
u32 boot_gbpa;
unsigned int pgd_order;
+ atomic_t initialized;
};
#define smmu_to_host(_smmu) \
@@ -134,8 +136,10 @@ static struct iommu_ops kvm_arm_smmu_ops;
static struct iommu_device *kvm_arm_smmu_probe_device(struct device *dev)
{
+ int ret;
struct arm_smmu_device *smmu;
struct kvm_arm_smmu_master *master;
+ struct host_arm_smmu_device *host_smmu;
struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
if (!fwspec || fwspec->ops != &kvm_arm_smmu_ops)
@@ -156,7 +160,28 @@ static struct iommu_device *kvm_arm_smmu_probe_device(struct device *dev)
master->smmu = smmu;
dev_iommu_priv_set(dev, master);
+ if (!device_link_add(dev, smmu->dev,
+ DL_FLAG_PM_RUNTIME | DL_FLAG_RPM_ACTIVE |
+ DL_FLAG_AUTOREMOVE_SUPPLIER)) {
+ ret = -ENOLINK;
+ goto err_free;
+ }
+
+ /*
+ * If the SMMU has just been initialized by the hypervisor, release the
+ * extra PM reference taken by kvm_arm_smmu_probe(). Not sure yet how
+ * to improve this. Maybe have KVM call us back when it finished
+ * initializing?
+ */
+ host_smmu = smmu_to_host(smmu);
+ if (atomic_add_unless(&host_smmu->initialized, 1, 1))
+ pm_runtime_put_noidle(smmu->dev);
+
return &smmu->iommu;
+
+err_free:
+ kfree(master);
+ return ERR_PTR(ret);
}
static void kvm_arm_smmu_release_device(struct device *dev)
@@ -685,6 +710,30 @@ static int kvm_arm_smmu_probe(struct platform_device *pdev)
kvm_arm_smmu_cur++;
+ /*
+ * The state of endpoints dictates when the SMMU is powered off. To turn
+ * the SMMU on and off, a genpd driver uses SCMI over the SMC transport,
+ * or some other platform-specific SMC. Those power requests are caught
+ * by the hypervisor, so that the hyp driver doesn't touch the hardware
+ * state while it is off.
+ *
+ * We are making a big assumption here, that TLBs and caches are invalid
+ * on power on, and therefore we don't need to wake the SMMU when
+ * modifying page tables, stream tables and context tables. If this
+ * assumption does not hold on some systems, then we'll need to grab RPM
+ * reference in map(), attach(), etc, so the hyp driver can send
+ * invalidations.
+ */
+ hyp_smmu->caches_clean_on_power_on = true;
+
+ pm_runtime_set_active(dev);
+ pm_runtime_enable(dev);
+ /*
+ * Take a reference to keep the SMMU powered on while the hypervisor
+ * initializes it.
+ */
+ pm_runtime_resume_and_get(dev);
+
return 0;
}
@@ -697,6 +746,11 @@ static int kvm_arm_smmu_remove(struct platform_device *pdev)
* There was an error during hypervisor setup. The hyp driver may
* have already enabled the device, so disable it.
*/
+
+ if (!atomic_read(&host_smmu->initialized))
+ pm_runtime_put_noidle(&pdev->dev);
+ pm_runtime_disable(&pdev->dev);
+ pm_runtime_set_suspended(&pdev->dev);
arm_smmu_unregister_iommu(smmu);
arm_smmu_device_disable(smmu);
arm_smmu_update_gbpa(smmu, host_smmu->boot_gbpa, GBPA_ABORT);
Enable runtime PM for the KVM SMMUv3 driver. The PM link to DMA masters dictates when the SMMU should be powered on. Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org> --- .../iommu/arm/arm-smmu-v3/arm-smmu-v3-kvm.c | 54 +++++++++++++++++++ 1 file changed, 54 insertions(+)