OSDN Git Service

ice: Register auxiliary device to provide RDMA
authorDave Ertman <david.m.ertman@intel.com>
Thu, 20 May 2021 14:37:51 +0000 (09:37 -0500)
committerTony Nguyen <anthony.l.nguyen@intel.com>
Sat, 29 May 2021 03:11:13 +0000 (20:11 -0700)
Register ice client auxiliary RDMA device on the auxiliary bus per
PCIe device function for the auxiliary driver (irdma) to attach to.
It allows to realize a single RDMA driver (irdma) capable of working with
multiple netdev drivers over multi-generation Intel HW supporting RDMA.
There is no load ordering dependencies between ice and irdma.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
drivers/net/ethernet/intel/Kconfig
drivers/net/ethernet/intel/ice/ice.h
drivers/net/ethernet/intel/ice/ice_idc.c
drivers/net/ethernet/intel/ice/ice_main.c

index c1d1556..d8a12da 100644 (file)
@@ -294,6 +294,7 @@ config ICE
        tristate "Intel(R) Ethernet Connection E800 Series Support"
        default n
        depends on PCI_MSI
+       select AUXILIARY_BUS
        select DIMLIB
        select NET_DEVLINK
        select PLDMFW
index 225f8a5..228055e 100644 (file)
@@ -34,6 +34,7 @@
 #include <linux/if_bridge.h>
 #include <linux/ctype.h>
 #include <linux/bpf.h>
+#include <linux/auxiliary_bus.h>
 #include <linux/avf/virtchnl.h>
 #include <linux/cpu_rmap.h>
 #include <linux/dim.h>
@@ -647,6 +648,8 @@ int ice_get_rss_key(struct ice_vsi *vsi, u8 *seed);
 void ice_fill_rss_lut(u8 *lut, u16 rss_table_size, u16 rss_size);
 int ice_schedule_reset(struct ice_pf *pf, enum ice_reset_req reset);
 void ice_print_link_msg(struct ice_vsi *vsi, bool isup);
+int ice_plug_aux_dev(struct ice_pf *pf);
+void ice_unplug_aux_dev(struct ice_pf *pf);
 int ice_init_rdma(struct ice_pf *pf);
 const char *ice_stat_str(enum ice_status stat_err);
 const char *ice_aq_str(enum ice_aq_err aq_err);
@@ -678,8 +681,10 @@ void ice_service_task_schedule(struct ice_pf *pf);
  */
 static inline void ice_set_rdma_cap(struct ice_pf *pf)
 {
-       if (pf->hw.func_caps.common_cap.rdma && pf->num_rdma_msix)
+       if (pf->hw.func_caps.common_cap.rdma && pf->num_rdma_msix) {
                set_bit(ICE_FLAG_RDMA_ENA, pf->flags);
+               ice_plug_aux_dev(pf);
+       }
 }
 
 /**
@@ -688,6 +693,7 @@ static inline void ice_set_rdma_cap(struct ice_pf *pf)
  */
 static inline void ice_clear_rdma_cap(struct ice_pf *pf)
 {
+       ice_unplug_aux_dev(pf);
        clear_bit(ICE_FLAG_RDMA_ENA, pf->flags);
 }
 #endif /* _ICE_H_ */
index adc9015..1f2afdf 100644 (file)
@@ -250,6 +250,71 @@ static int ice_reserve_rdma_qvector(struct ice_pf *pf)
 }
 
 /**
+ * ice_adev_release - function to be mapped to AUX dev's release op
+ * @dev: pointer to device to free
+ */
+static void ice_adev_release(struct device *dev)
+{
+       struct iidc_auxiliary_dev *iadev;
+
+       iadev = container_of(dev, struct iidc_auxiliary_dev, adev.dev);
+       kfree(iadev);
+}
+
+/**
+ * ice_plug_aux_dev - allocate and register AUX device
+ * @pf: pointer to pf struct
+ */
+int ice_plug_aux_dev(struct ice_pf *pf)
+{
+       struct iidc_auxiliary_dev *iadev;
+       struct auxiliary_device *adev;
+       int ret;
+
+       iadev = kzalloc(sizeof(*iadev), GFP_KERNEL);
+       if (!iadev)
+               return -ENOMEM;
+
+       adev = &iadev->adev;
+       pf->adev = adev;
+       iadev->pf = pf;
+
+       adev->id = pf->aux_idx;
+       adev->dev.release = ice_adev_release;
+       adev->dev.parent = &pf->pdev->dev;
+       adev->name = IIDC_RDMA_ROCE_NAME;
+
+       ret = auxiliary_device_init(adev);
+       if (ret) {
+               pf->adev = NULL;
+               kfree(iadev);
+               return ret;
+       }
+
+       ret = auxiliary_device_add(adev);
+       if (ret) {
+               pf->adev = NULL;
+               auxiliary_device_uninit(adev);
+               return ret;
+       }
+
+       return 0;
+}
+
+/* ice_unplug_aux_dev - unregister and free AUX device
+ * @pf: pointer to pf struct
+ */
+void ice_unplug_aux_dev(struct ice_pf *pf)
+{
+       if (!pf->adev)
+               return;
+
+       auxiliary_device_delete(pf->adev);
+       auxiliary_device_uninit(pf->adev);
+       pf->adev = NULL;
+}
+
+/**
  * ice_init_rdma - initializes PF for RDMA use
  * @pf: ptr to ice_pf
  */
@@ -260,8 +325,10 @@ int ice_init_rdma(struct ice_pf *pf)
 
        /* Reserve vector resources */
        ret = ice_reserve_rdma_qvector(pf);
-       if (ret < 0)
+       if (ret < 0) {
                dev_err(dev, "failed to reserve vectors for RDMA\n");
+               return ret;
+       }
 
-       return ret;
+       return ice_plug_aux_dev(pf);
 }
index 9d4570b..254cfc1 100644 (file)
@@ -456,6 +456,8 @@ ice_prepare_for_reset(struct ice_pf *pf)
        if (test_bit(ICE_PREPARED_FOR_RESET, pf->state))
                return;
 
+       ice_unplug_aux_dev(pf);
+
        /* Notify VFs of impending reset */
        if (ice_check_sq_alive(hw, &hw->mailboxq))
                ice_vc_notify_reset(pf);
@@ -2120,6 +2122,8 @@ int ice_schedule_reset(struct ice_pf *pf, enum ice_reset_req reset)
                return -EBUSY;
        }
 
+       ice_unplug_aux_dev(pf);
+
        switch (reset) {
        case ICE_RESET_PFR:
                set_bit(ICE_PFR_REQ, pf->state);
@@ -4456,11 +4460,12 @@ static void ice_remove(struct pci_dev *pdev)
                ice_free_vfs(pf);
        }
 
-       set_bit(ICE_DOWN, pf->state);
        ice_service_task_stop(pf);
 
        ice_aq_cancel_waiting_tasks(pf);
+       ice_unplug_aux_dev(pf);
        ida_free(&ice_aux_ida, pf->aux_idx);
+       set_bit(ICE_DOWN, pf->state);
 
        mutex_destroy(&(&pf->hw)->fdir_fltr_lock);
        ice_deinit_lag(pf);
@@ -4616,6 +4621,8 @@ static int __maybe_unused ice_suspend(struct device *dev)
         */
        disabled = ice_service_task_stop(pf);
 
+       ice_unplug_aux_dev(pf);
+
        /* Already suspended?, then there is nothing to do */
        if (test_and_set_bit(ICE_SUSPENDED, pf->state)) {
                if (!disabled)
@@ -6286,6 +6293,8 @@ static void ice_rebuild(struct ice_pf *pf, enum ice_reset_req reset_type)
 
        /* if we get here, reset flow is successful */
        clear_bit(ICE_RESET_FAILED, pf->state);
+
+       ice_plug_aux_dev(pf);
        return;
 
 err_vsi_rebuild: