[RFC PATCH v4 08/16] coco: host: arm64: Add helpers to unlock and destroy RMM vdev

["Aneesh Kumar K.V (Arm)" <aneesh.kumar@kernel.org>]

- define the SMCCC IDs and inline wrappers for RMI_VDEV_UNLOCK and
  RMI_VDEV_DESTROY
- extend vdev_create() to treat communication failures as fatal and
  tear down the newly created vdev
- provide vdev_unlock_and_destroy() that drives the vdev back to the
  unlocked state, issues the destroy call, and frees the delegated granule
- hook the new helper into the TSM unbind path so host cleanup always
  unlock and destroy RMM vdev and releases cached buffers

Signed-off-by: Aneesh Kumar K.V (Arm) <aneesh.kumar@kernel.org>
---
 arch/arm64/include/asm/rmi_cmds.h        | 20 +++++++
 arch/arm64/include/asm/rmi_smc.h         |  2 +
 drivers/virt/coco/arm-cca-host/arm-cca.c | 25 +++++++++
 drivers/virt/coco/arm-cca-host/rmi-da.c  | 69 ++++++++++++++++++++++--
 drivers/virt/coco/arm-cca-host/rmi-da.h  |  3 ++
 5 files changed, 116 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/include/asm/rmi_cmds.h b/arch/arm64/include/asm/rmi_cmds.h
index 03dffba763e1..aa7ef9f07517 100644
--- a/arch/arm64/include/asm/rmi_cmds.h
+++ b/arch/arm64/include/asm/rmi_cmds.h
@@ -949,4 +949,24 @@ static inline unsigned long rmi_pdev_stream_key_purge(unsigned long pdev1_phys,
 	return res.a0;
 }
 
+static inline unsigned long rmi_vdev_unlock(unsigned long rd,
+		unsigned long pdev_phys, unsigned long vdev_phys)
+{
+	struct arm_smccc_res res;
+
+	arm_smccc_1_1_invoke(SMC_RMI_VDEV_UNLOCK, rd, pdev_phys, vdev_phys, &res);
+
+	return res.a0;
+}
+
+static inline unsigned long rmi_vdev_destroy(unsigned long rd,
+		unsigned long pdev_phys, unsigned long vdev_phys)
+{
+	struct arm_smccc_res res;
+
+	arm_smccc_1_1_invoke(SMC_RMI_VDEV_DESTROY, rd, pdev_phys, vdev_phys, &res);
+
+	return res.a0;
+}
+
 #endif /* __ASM_RMI_CMDS_H */
diff --git a/arch/arm64/include/asm/rmi_smc.h b/arch/arm64/include/asm/rmi_smc.h
index d14d13a9f169..6cd5439f56ec 100644
--- a/arch/arm64/include/asm/rmi_smc.h
+++ b/arch/arm64/include/asm/rmi_smc.h
@@ -649,6 +649,8 @@ enum rmi_vdev_state {
 	RMI_VDEV_LOCKED,
 	RMI_VDEV_STARTED,
 	RMI_VDEV_ERROR,
+	RMI_VDEV_KEY_REFRESH,
+	RMI_VDEV_KEY_PURGE,
 };
 
 #define MAX_VDEV_ADDR_RANGE 8
diff --git a/drivers/virt/coco/arm-cca-host/arm-cca.c b/drivers/virt/coco/arm-cca-host/arm-cca.c
index 5930a30dd16f..b75fa20513a9 100644
--- a/drivers/virt/coco/arm-cca-host/arm-cca.c
+++ b/drivers/virt/coco/arm-cca-host/arm-cca.c
@@ -473,12 +473,37 @@ static struct pci_tdi *cca_tsm_bind(struct pci_dev *pdev, struct kvm *kvm, u32 t
 	return &no_free_ptr(host_tdi)->tdi;
 }
 
+/*
+ * All device memory should be unmapped by now.
+ * 1. A pci device destroy will cause a driver remove (vfio) which will have
+ *    done a dmabuf based unmap
+ * 2. A vdevice/idevice destroy from VMM should have done a unmap_private_range
+ *    vm ioctl before
+ * 3. A guest unlock request should have done a rsi_invalidiate_mem_mapping
+ *    before unlock rhi
+ * 4. vfio_pci_core_close_device() should trigger tsm unbind if vdevice is not
+ *    already distroyed and that path involves vfio_pci_dma_buf_cleanup() which
+ *    should get kvm to unmap the devmap
+ */
+static void cca_tsm_unbind(struct pci_tdi *tdi)
+{
+	struct cca_host_tdi *host_tdi;
+	struct realm *realm = &tdi->kvm->arch.realm;
+
+	host_tdi = container_of(tdi, struct cca_host_tdi, tdi);
+	cca_vdev_unlock_and_destroy(realm, tdi->pdev, tdi->pdev->tsm->dsm_dev);
+	kvfree(host_tdi->interface_report);
+	kvfree(host_tdi->measurements);
+	kfree(host_tdi);
+}
+
 static struct pci_tsm_ops cca_link_pci_ops = {
 	.probe = cca_tsm_pci_probe,
 	.remove = cca_tsm_pci_remove,
 	.connect = cca_tsm_connect,
 	.disconnect = cca_tsm_disconnect,
 	.bind = cca_tsm_bind,
+	.unbind = cca_tsm_unbind,
 };
 
 static void cca_link_tsm_remove(void *tsm_dev)
diff --git a/drivers/virt/coco/arm-cca-host/rmi-da.c b/drivers/virt/coco/arm-cca-host/rmi-da.c
index 128079d5b993..ef25392562e0 100644
--- a/drivers/virt/coco/arm-cca-host/rmi-da.c
+++ b/drivers/virt/coco/arm-cca-host/rmi-da.c
@@ -1018,15 +1018,25 @@ void *cca_vdev_create(struct realm *realm, struct pci_dev *pdev,
 	host_tdi->rmm_vdev = rmm_vdev;
 	host_tdi->realm = realm;
 
-	submit_vdev_state_transition_work(pdev, RMI_VDEV_UNLOCKED);
+	ret = submit_vdev_state_transition_work(pdev, RMI_VDEV_UNLOCKED);
+	/* failure is treated as rmi_vdev_create failure */
+	if (ret)
+		goto err_vdev_comm;
 
-	ret = rmi_vdev_lock(rd_phys, rmm_pdev_phys, rmm_vdev_phys);
+	if (rmi_vdev_lock(rd_phys, rmm_pdev_phys, rmm_vdev_phys)) {
+		ret = -ENXIO;
+		goto err_vdev_comm;
+	}
 
-	submit_vdev_state_transition_work(pdev, RMI_VDEV_LOCKED);
+	ret = submit_vdev_state_transition_work(pdev, RMI_VDEV_LOCKED);
+	if (ret)
+		goto err_vdev_comm;
 
 	free_page((unsigned long)params);
 	return rmm_vdev;
 
+err_vdev_comm:
+	rmi_vdev_destroy(rd_phys, rmm_pdev_phys, rmm_vdev_phys);
 err_vdev_create:
 	free_page((unsigned long)params);
 err_params_alloc:
@@ -1073,3 +1083,56 @@ int cca_pdev_purge_stream_key(struct pci_dev *pdev1,
 
 	return submit_stream_work(pdev1, pdev2, stream_handle);
 }
+
+void cca_vdev_unlock_and_destroy(struct realm *realm,
+		struct pci_dev *pdev, struct pci_dev *pf0_dev)
+{
+	int ret;
+	phys_addr_t rmm_pdev_phys;
+	phys_addr_t rmm_vdev_phys;
+	struct cca_host_pdev_dsc *pdev_dsc;
+	struct cca_host_tdi *host_tdi;
+	phys_addr_t rd_phys = virt_to_phys(realm->rd);
+
+	host_tdi = to_cca_host_tdi(pdev);
+	rmm_vdev_phys = virt_to_phys(host_tdi->rmm_vdev);
+
+	pdev_dsc = to_cca_pdev_dsc(pf0_dev);
+	rmm_pdev_phys = virt_to_phys(pdev_dsc->rmm_pdev);
+	if (rmi_vdev_unlock(rd_phys, rmm_pdev_phys, rmm_vdev_phys)) {
+		pci_err(pdev, "failed to unlock vdev\n");
+		goto unlock_err;
+	}
+
+	if (rmm_has_reg2_feature(RMI_FEATURE_REGISTER_2_VDEV_KROU)) {
+		struct pci_dev *rp = pcie_find_root_port(pf0_dev);
+		struct cca_host_pf0_ep_dsc *pf0_ep_dsc = to_cca_pf0_ep_dsc(pf0_dev);
+
+		ret = submit_vdev_state_transition_work(pdev, RMI_VDEV_KEY_REFRESH);
+		if (ret)
+			pci_err(pdev, "failed to transition vdev to KEY_REFRESH state (%d)\n", ret);
+
+		ret = cca_pdev_refresh_stream_key(pf0_dev, rp, pf0_ep_dsc->stream_handle);
+		if (ret)
+			pci_err(pf0_dev, "failed to refresh pdev stream key (%d)\n", ret);
+
+		ret = cca_pdev_purge_stream_key(pf0_dev, rp, pf0_ep_dsc->stream_handle);
+		if (ret)
+			pci_err(pf0_dev, "failed to purge pdev stream key (%d)\n", ret);
+	}
+
+	ret = submit_vdev_state_transition_work(pdev, RMI_VDEV_UNLOCKED);
+	if (ret)
+		pci_err(pdev, "failed to unlock vdev (%d)\n", ret);
+
+unlock_err:
+	/* Try to destroy even in case of error */
+	if (rmi_vdev_destroy(rd_phys, rmm_pdev_phys, rmm_vdev_phys))
+		pci_err(pdev, "failed to destroy vdev\n");
+
+	if (!rmi_undelegate_page(rmm_vdev_phys))
+		free_page((unsigned long)host_tdi->rmm_vdev);
+
+	host_tdi->rmm_vdev = NULL;
+	host_tdi->realm = NULL;
+}
diff --git a/drivers/virt/coco/arm-cca-host/rmi-da.h b/drivers/virt/coco/arm-cca-host/rmi-da.h
index d6cdbc638d6d..97f7eaf1f779 100644
--- a/drivers/virt/coco/arm-cca-host/rmi-da.h
+++ b/drivers/virt/coco/arm-cca-host/rmi-da.h
@@ -15,6 +15,7 @@
 #include <linux/wait.h>
 #include <asm/rmi_cmds.h>
 #include <asm/rmi_smc.h>
+#include <asm/rhi.h>
 
 #define MAX_CACHE_OBJ_SIZE	SZ_16M
 #define CACHE_CHUNK_SIZE	SZ_4K
@@ -239,5 +240,7 @@ int cca_pdev_refresh_stream_key(struct pci_dev *pdev1,
 		struct pci_dev *pdev2, unsigned long stream_handle);
 int cca_pdev_purge_stream_key(struct pci_dev *pdev1,
 		struct pci_dev *pdev2, unsigned long stream_handle);
+void cca_vdev_unlock_and_destroy(struct realm *realm, struct pci_dev *pdev,
+		struct pci_dev *pf0_dev);
 
 #endif
-- 
2.43.0