pds_vdpa: add support for vdpa and vdpamgmt interfaces
authorShannon Nelson <shannon.nelson@amd.com>
Fri, 19 May 2023 21:56:30 +0000 (14:56 -0700)
committerMichael S. Tsirkin <mst@redhat.com>
Tue, 27 Jun 2023 14:47:09 +0000 (10:47 -0400)
This is the vDPA device support, where we advertise that we can
support the virtio queues and deal with the configuration work
through the pds_core's adminq.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Message-Id: <20230519215632.12343-10-shannon.nelson@amd.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
drivers/vdpa/pds/aux_drv.c
drivers/vdpa/pds/aux_drv.h
drivers/vdpa/pds/debugfs.c
drivers/vdpa/pds/debugfs.h
drivers/vdpa/pds/vdpa_dev.c
drivers/vdpa/pds/vdpa_dev.h

index 0c4a135b148457719b1262968ba644cb1578f8fc..186e9ee22eb1126127a829f221d67dd76e8b3252 100644 (file)
@@ -63,8 +63,21 @@ static int pds_vdpa_probe(struct auxiliary_device *aux_dev,
                goto err_free_mgmt_info;
        }
 
+       /* Let vdpa know that we can provide devices */
+       err = vdpa_mgmtdev_register(&vdpa_aux->vdpa_mdev);
+       if (err) {
+               dev_err(dev, "%s: Failed to initialize vdpa_mgmt interface: %pe\n",
+                       __func__, ERR_PTR(err));
+               goto err_free_virtio;
+       }
+
+       pds_vdpa_debugfs_add_pcidev(vdpa_aux);
+       pds_vdpa_debugfs_add_ident(vdpa_aux);
+
        return 0;
 
+err_free_virtio:
+       vp_modern_remove(&vdpa_aux->vd_mdev);
 err_free_mgmt_info:
        pci_free_irq_vectors(padev->vf_pdev);
 err_free_mem:
@@ -79,9 +92,11 @@ static void pds_vdpa_remove(struct auxiliary_device *aux_dev)
        struct pds_vdpa_aux *vdpa_aux = auxiliary_get_drvdata(aux_dev);
        struct device *dev = &aux_dev->dev;
 
+       vdpa_mgmtdev_unregister(&vdpa_aux->vdpa_mdev);
        vp_modern_remove(&vdpa_aux->vd_mdev);
        pci_free_irq_vectors(vdpa_aux->padev->vf_pdev);
 
+       pds_vdpa_debugfs_del_vdpadev(vdpa_aux);
        kfree(vdpa_aux);
        auxiliary_set_drvdata(aux_dev, NULL);
 
index 99e0ff340bfa48b093facb561d11c01e22ca5b68..26b75344156e4a6b6f6033d195ad583311464e8d 100644 (file)
@@ -13,6 +13,7 @@ struct pds_vdpa_aux {
        struct pds_auxiliary_dev *padev;
 
        struct vdpa_mgmt_dev vdpa_mdev;
+       struct pds_vdpa_device *pdsv;
 
        struct pds_vdpa_ident ident;
 
index d91dceb07380367c7e7c56e4f9c85083215540fe..21a0dc0cb607ca5768dab0bf982dbee1cab8b863 100644 (file)
@@ -10,6 +10,7 @@
 #include <linux/pds/pds_auxbus.h>
 
 #include "aux_drv.h"
+#include "vdpa_dev.h"
 #include "debugfs.h"
 
 static struct dentry *dbfs_dir;
@@ -24,3 +25,265 @@ void pds_vdpa_debugfs_destroy(void)
        debugfs_remove_recursive(dbfs_dir);
        dbfs_dir = NULL;
 }
+
+#define PRINT_SBIT_NAME(__seq, __f, __name)                     \
+       do {                                                    \
+               if ((__f) & (__name))                               \
+                       seq_printf(__seq, " %s", &#__name[16]); \
+       } while (0)
+
+static void print_status_bits(struct seq_file *seq, u8 status)
+{
+       seq_puts(seq, "status:");
+       PRINT_SBIT_NAME(seq, status, VIRTIO_CONFIG_S_ACKNOWLEDGE);
+       PRINT_SBIT_NAME(seq, status, VIRTIO_CONFIG_S_DRIVER);
+       PRINT_SBIT_NAME(seq, status, VIRTIO_CONFIG_S_DRIVER_OK);
+       PRINT_SBIT_NAME(seq, status, VIRTIO_CONFIG_S_FEATURES_OK);
+       PRINT_SBIT_NAME(seq, status, VIRTIO_CONFIG_S_NEEDS_RESET);
+       PRINT_SBIT_NAME(seq, status, VIRTIO_CONFIG_S_FAILED);
+       seq_puts(seq, "\n");
+}
+
+static void print_feature_bits_all(struct seq_file *seq, u64 features)
+{
+       int i;
+
+       seq_puts(seq, "features:");
+
+       for (i = 0; i < (sizeof(u64) * 8); i++) {
+               u64 mask = BIT_ULL(i);
+
+               switch (features & mask) {
+               case BIT_ULL(VIRTIO_NET_F_CSUM):
+                       seq_puts(seq, " VIRTIO_NET_F_CSUM");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_GUEST_CSUM):
+                       seq_puts(seq, " VIRTIO_NET_F_GUEST_CSUM");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_CTRL_GUEST_OFFLOADS):
+                       seq_puts(seq, " VIRTIO_NET_F_CTRL_GUEST_OFFLOADS");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_MTU):
+                       seq_puts(seq, " VIRTIO_NET_F_MTU");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_MAC):
+                       seq_puts(seq, " VIRTIO_NET_F_MAC");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_GUEST_TSO4):
+                       seq_puts(seq, " VIRTIO_NET_F_GUEST_TSO4");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_GUEST_TSO6):
+                       seq_puts(seq, " VIRTIO_NET_F_GUEST_TSO6");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_GUEST_ECN):
+                       seq_puts(seq, " VIRTIO_NET_F_GUEST_ECN");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_GUEST_UFO):
+                       seq_puts(seq, " VIRTIO_NET_F_GUEST_UFO");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_HOST_TSO4):
+                       seq_puts(seq, " VIRTIO_NET_F_HOST_TSO4");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_HOST_TSO6):
+                       seq_puts(seq, " VIRTIO_NET_F_HOST_TSO6");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_HOST_ECN):
+                       seq_puts(seq, " VIRTIO_NET_F_HOST_ECN");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_HOST_UFO):
+                       seq_puts(seq, " VIRTIO_NET_F_HOST_UFO");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_MRG_RXBUF):
+                       seq_puts(seq, " VIRTIO_NET_F_MRG_RXBUF");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_STATUS):
+                       seq_puts(seq, " VIRTIO_NET_F_STATUS");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_CTRL_VQ):
+                       seq_puts(seq, " VIRTIO_NET_F_CTRL_VQ");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_CTRL_RX):
+                       seq_puts(seq, " VIRTIO_NET_F_CTRL_RX");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_CTRL_VLAN):
+                       seq_puts(seq, " VIRTIO_NET_F_CTRL_VLAN");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_CTRL_RX_EXTRA):
+                       seq_puts(seq, " VIRTIO_NET_F_CTRL_RX_EXTRA");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_GUEST_ANNOUNCE):
+                       seq_puts(seq, " VIRTIO_NET_F_GUEST_ANNOUNCE");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_MQ):
+                       seq_puts(seq, " VIRTIO_NET_F_MQ");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR):
+                       seq_puts(seq, " VIRTIO_NET_F_CTRL_MAC_ADDR");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_HASH_REPORT):
+                       seq_puts(seq, " VIRTIO_NET_F_HASH_REPORT");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_RSS):
+                       seq_puts(seq, " VIRTIO_NET_F_RSS");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_RSC_EXT):
+                       seq_puts(seq, " VIRTIO_NET_F_RSC_EXT");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_STANDBY):
+                       seq_puts(seq, " VIRTIO_NET_F_STANDBY");
+                       break;
+               case BIT_ULL(VIRTIO_NET_F_SPEED_DUPLEX):
+                       seq_puts(seq, " VIRTIO_NET_F_SPEED_DUPLEX");
+                       break;
+               case BIT_ULL(VIRTIO_F_NOTIFY_ON_EMPTY):
+                       seq_puts(seq, " VIRTIO_F_NOTIFY_ON_EMPTY");
+                       break;
+               case BIT_ULL(VIRTIO_F_ANY_LAYOUT):
+                       seq_puts(seq, " VIRTIO_F_ANY_LAYOUT");
+                       break;
+               case BIT_ULL(VIRTIO_F_VERSION_1):
+                       seq_puts(seq, " VIRTIO_F_VERSION_1");
+                       break;
+               case BIT_ULL(VIRTIO_F_ACCESS_PLATFORM):
+                       seq_puts(seq, " VIRTIO_F_ACCESS_PLATFORM");
+                       break;
+               case BIT_ULL(VIRTIO_F_RING_PACKED):
+                       seq_puts(seq, " VIRTIO_F_RING_PACKED");
+                       break;
+               case BIT_ULL(VIRTIO_F_ORDER_PLATFORM):
+                       seq_puts(seq, " VIRTIO_F_ORDER_PLATFORM");
+                       break;
+               case BIT_ULL(VIRTIO_F_SR_IOV):
+                       seq_puts(seq, " VIRTIO_F_SR_IOV");
+                       break;
+               case 0:
+                       break;
+               default:
+                       seq_printf(seq, " bit_%d", i);
+                       break;
+               }
+       }
+
+       seq_puts(seq, "\n");
+}
+
+void pds_vdpa_debugfs_add_pcidev(struct pds_vdpa_aux *vdpa_aux)
+{
+       vdpa_aux->dentry = debugfs_create_dir(pci_name(vdpa_aux->padev->vf_pdev), dbfs_dir);
+}
+
+static int identity_show(struct seq_file *seq, void *v)
+{
+       struct pds_vdpa_aux *vdpa_aux = seq->private;
+       struct vdpa_mgmt_dev *mgmt;
+
+       seq_printf(seq, "aux_dev:            %s\n",
+                  dev_name(&vdpa_aux->padev->aux_dev.dev));
+
+       mgmt = &vdpa_aux->vdpa_mdev;
+       seq_printf(seq, "max_vqs:            %d\n", mgmt->max_supported_vqs);
+       seq_printf(seq, "config_attr_mask:   %#llx\n", mgmt->config_attr_mask);
+       seq_printf(seq, "supported_features: %#llx\n", mgmt->supported_features);
+       print_feature_bits_all(seq, mgmt->supported_features);
+
+       return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(identity);
+
+void pds_vdpa_debugfs_add_ident(struct pds_vdpa_aux *vdpa_aux)
+{
+       debugfs_create_file("identity", 0400, vdpa_aux->dentry,
+                           vdpa_aux, &identity_fops);
+}
+
+static int config_show(struct seq_file *seq, void *v)
+{
+       struct pds_vdpa_device *pdsv = seq->private;
+       struct virtio_net_config vc;
+       u64 driver_features;
+       u8 status;
+
+       memcpy_fromio(&vc, pdsv->vdpa_aux->vd_mdev.device,
+                     sizeof(struct virtio_net_config));
+
+       seq_printf(seq, "mac:                  %pM\n", vc.mac);
+       seq_printf(seq, "max_virtqueue_pairs:  %d\n",
+                  __virtio16_to_cpu(true, vc.max_virtqueue_pairs));
+       seq_printf(seq, "mtu:                  %d\n", __virtio16_to_cpu(true, vc.mtu));
+       seq_printf(seq, "speed:                %d\n", le32_to_cpu(vc.speed));
+       seq_printf(seq, "duplex:               %d\n", vc.duplex);
+       seq_printf(seq, "rss_max_key_size:     %d\n", vc.rss_max_key_size);
+       seq_printf(seq, "rss_max_indirection_table_length: %d\n",
+                  le16_to_cpu(vc.rss_max_indirection_table_length));
+       seq_printf(seq, "supported_hash_types: %#x\n",
+                  le32_to_cpu(vc.supported_hash_types));
+       seq_printf(seq, "vn_status:            %#x\n",
+                  __virtio16_to_cpu(true, vc.status));
+
+       status = vp_modern_get_status(&pdsv->vdpa_aux->vd_mdev);
+       seq_printf(seq, "dev_status:           %#x\n", status);
+       print_status_bits(seq, status);
+
+       seq_printf(seq, "req_features:         %#llx\n", pdsv->req_features);
+       print_feature_bits_all(seq, pdsv->req_features);
+       driver_features = vp_modern_get_driver_features(&pdsv->vdpa_aux->vd_mdev);
+       seq_printf(seq, "driver_features:      %#llx\n", driver_features);
+       print_feature_bits_all(seq, driver_features);
+       seq_printf(seq, "vdpa_index:           %d\n", pdsv->vdpa_index);
+       seq_printf(seq, "num_vqs:              %d\n", pdsv->num_vqs);
+
+       return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(config);
+
+static int vq_show(struct seq_file *seq, void *v)
+{
+       struct pds_vdpa_vq_info *vq = seq->private;
+
+       seq_printf(seq, "ready:      %d\n", vq->ready);
+       seq_printf(seq, "desc_addr:  %#llx\n", vq->desc_addr);
+       seq_printf(seq, "avail_addr: %#llx\n", vq->avail_addr);
+       seq_printf(seq, "used_addr:  %#llx\n", vq->used_addr);
+       seq_printf(seq, "q_len:      %d\n", vq->q_len);
+       seq_printf(seq, "qid:        %d\n", vq->qid);
+
+       seq_printf(seq, "doorbell:   %#llx\n", vq->doorbell);
+       seq_printf(seq, "avail_idx:  %d\n", vq->avail_idx);
+       seq_printf(seq, "used_idx:   %d\n", vq->used_idx);
+       seq_printf(seq, "irq:        %d\n", vq->irq);
+       seq_printf(seq, "irq-name:   %s\n", vq->irq_name);
+
+       return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(vq);
+
+void pds_vdpa_debugfs_add_vdpadev(struct pds_vdpa_aux *vdpa_aux)
+{
+       int i;
+
+       debugfs_create_file("config", 0400, vdpa_aux->dentry, vdpa_aux->pdsv, &config_fops);
+
+       for (i = 0; i < vdpa_aux->pdsv->num_vqs; i++) {
+               char name[8];
+
+               snprintf(name, sizeof(name), "vq%02d", i);
+               debugfs_create_file(name, 0400, vdpa_aux->dentry,
+                                   &vdpa_aux->pdsv->vqs[i], &vq_fops);
+       }
+}
+
+void pds_vdpa_debugfs_del_vdpadev(struct pds_vdpa_aux *vdpa_aux)
+{
+       debugfs_remove_recursive(vdpa_aux->dentry);
+       vdpa_aux->dentry = NULL;
+}
+
+void pds_vdpa_debugfs_reset_vdpadev(struct pds_vdpa_aux *vdpa_aux)
+{
+       /* we don't keep track of the entries, so remove it all
+        * then rebuild the basics
+        */
+       pds_vdpa_debugfs_del_vdpadev(vdpa_aux);
+       pds_vdpa_debugfs_add_pcidev(vdpa_aux);
+       pds_vdpa_debugfs_add_ident(vdpa_aux);
+}
index 658849591a99d5f213d1566df42c3c7ff390731d..c088a4e8f1e92949f5db7811e2f773c79251147d 100644 (file)
@@ -8,5 +8,10 @@
 
 void pds_vdpa_debugfs_create(void);
 void pds_vdpa_debugfs_destroy(void);
+void pds_vdpa_debugfs_add_pcidev(struct pds_vdpa_aux *vdpa_aux);
+void pds_vdpa_debugfs_add_ident(struct pds_vdpa_aux *vdpa_aux);
+void pds_vdpa_debugfs_add_vdpadev(struct pds_vdpa_aux *vdpa_aux);
+void pds_vdpa_debugfs_del_vdpadev(struct pds_vdpa_aux *vdpa_aux);
+void pds_vdpa_debugfs_reset_vdpadev(struct pds_vdpa_aux *vdpa_aux);
 
 #endif /* _PDS_VDPA_DEBUGFS_H_ */
index 0f0f0ab8b8114c600a3a8a53b05b5d4e6990b045..5690ee676ad9c4e84803e33642ad5723f13ff8bc 100644 (file)
@@ -4,6 +4,7 @@
 #include <linux/pci.h>
 #include <linux/vdpa.h>
 #include <uapi/linux/vdpa.h>
+#include <linux/virtio_pci_modern.h>
 
 #include <linux/pds/pds_common.h>
 #include <linux/pds/pds_core_if.h>
 
 #include "vdpa_dev.h"
 #include "aux_drv.h"
+#include "cmds.h"
+#include "debugfs.h"
 
+static u64 pds_vdpa_get_driver_features(struct vdpa_device *vdpa_dev);
+
+static struct pds_vdpa_device *vdpa_to_pdsv(struct vdpa_device *vdpa_dev)
+{
+       return container_of(vdpa_dev, struct pds_vdpa_device, vdpa_dev);
+}
+
+static int pds_vdpa_set_vq_address(struct vdpa_device *vdpa_dev, u16 qid,
+                                  u64 desc_addr, u64 driver_addr, u64 device_addr)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       pdsv->vqs[qid].desc_addr = desc_addr;
+       pdsv->vqs[qid].avail_addr = driver_addr;
+       pdsv->vqs[qid].used_addr = device_addr;
+
+       return 0;
+}
+
+static void pds_vdpa_set_vq_num(struct vdpa_device *vdpa_dev, u16 qid, u32 num)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       pdsv->vqs[qid].q_len = num;
+}
+
+static void pds_vdpa_kick_vq(struct vdpa_device *vdpa_dev, u16 qid)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       iowrite16(qid, pdsv->vqs[qid].notify);
+}
+
+static void pds_vdpa_set_vq_cb(struct vdpa_device *vdpa_dev, u16 qid,
+                              struct vdpa_callback *cb)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       pdsv->vqs[qid].event_cb = *cb;
+}
+
+static irqreturn_t pds_vdpa_isr(int irq, void *data)
+{
+       struct pds_vdpa_vq_info *vq;
+
+       vq = data;
+       if (vq->event_cb.callback)
+               vq->event_cb.callback(vq->event_cb.private);
+
+       return IRQ_HANDLED;
+}
+
+static void pds_vdpa_release_irq(struct pds_vdpa_device *pdsv, int qid)
+{
+       if (pdsv->vqs[qid].irq == VIRTIO_MSI_NO_VECTOR)
+               return;
+
+       free_irq(pdsv->vqs[qid].irq, &pdsv->vqs[qid]);
+       pdsv->vqs[qid].irq = VIRTIO_MSI_NO_VECTOR;
+}
+
+static void pds_vdpa_set_vq_ready(struct vdpa_device *vdpa_dev, u16 qid, bool ready)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct pci_dev *pdev = pdsv->vdpa_aux->padev->vf_pdev;
+       struct device *dev = &pdsv->vdpa_dev.dev;
+       u64 driver_features;
+       u16 invert_idx = 0;
+       int irq;
+       int err;
+
+       dev_dbg(dev, "%s: qid %d ready %d => %d\n",
+               __func__, qid, pdsv->vqs[qid].ready, ready);
+       if (ready == pdsv->vqs[qid].ready)
+               return;
+
+       driver_features = pds_vdpa_get_driver_features(vdpa_dev);
+       if (driver_features & BIT_ULL(VIRTIO_F_RING_PACKED))
+               invert_idx = PDS_VDPA_PACKED_INVERT_IDX;
+
+       if (ready) {
+               irq = pci_irq_vector(pdev, qid);
+               snprintf(pdsv->vqs[qid].irq_name, sizeof(pdsv->vqs[qid].irq_name),
+                        "vdpa-%s-%d", dev_name(dev), qid);
+
+               err = request_irq(irq, pds_vdpa_isr, 0,
+                                 pdsv->vqs[qid].irq_name, &pdsv->vqs[qid]);
+               if (err) {
+                       dev_err(dev, "%s: no irq for qid %d: %pe\n",
+                               __func__, qid, ERR_PTR(err));
+                       return;
+               }
+               pdsv->vqs[qid].irq = irq;
+
+               /* Pass vq setup info to DSC using adminq to gather up and
+                * send all info at once so FW can do its full set up in
+                * one easy operation
+                */
+               err = pds_vdpa_cmd_init_vq(pdsv, qid, invert_idx, &pdsv->vqs[qid]);
+               if (err) {
+                       dev_err(dev, "Failed to init vq %d: %pe\n",
+                               qid, ERR_PTR(err));
+                       pds_vdpa_release_irq(pdsv, qid);
+                       ready = false;
+               }
+       } else {
+               err = pds_vdpa_cmd_reset_vq(pdsv, qid, invert_idx, &pdsv->vqs[qid]);
+               if (err)
+                       dev_err(dev, "%s: reset_vq failed qid %d: %pe\n",
+                               __func__, qid, ERR_PTR(err));
+               pds_vdpa_release_irq(pdsv, qid);
+       }
+
+       pdsv->vqs[qid].ready = ready;
+}
+
+static bool pds_vdpa_get_vq_ready(struct vdpa_device *vdpa_dev, u16 qid)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       return pdsv->vqs[qid].ready;
+}
+
+static int pds_vdpa_set_vq_state(struct vdpa_device *vdpa_dev, u16 qid,
+                                const struct vdpa_vq_state *state)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct pds_auxiliary_dev *padev = pdsv->vdpa_aux->padev;
+       struct device *dev = &padev->aux_dev.dev;
+       u64 driver_features;
+       u16 avail;
+       u16 used;
+
+       if (pdsv->vqs[qid].ready) {
+               dev_err(dev, "Setting device position is denied while vq is enabled\n");
+               return -EINVAL;
+       }
+
+       driver_features = pds_vdpa_get_driver_features(vdpa_dev);
+       if (driver_features & BIT_ULL(VIRTIO_F_RING_PACKED)) {
+               avail = state->packed.last_avail_idx |
+                       (state->packed.last_avail_counter << 15);
+               used = state->packed.last_used_idx |
+                      (state->packed.last_used_counter << 15);
+
+               /* The avail and used index are stored with the packed wrap
+                * counter bit inverted.  This way, in case set_vq_state is
+                * not called, the initial value can be set to zero prior to
+                * feature negotiation, and it is good for both packed and
+                * split vq.
+                */
+               avail ^= PDS_VDPA_PACKED_INVERT_IDX;
+               used ^= PDS_VDPA_PACKED_INVERT_IDX;
+       } else {
+               avail = state->split.avail_index;
+               /* state->split does not provide a used_index:
+                * the vq will be set to "empty" here, and the vq will read
+                * the current used index the next time the vq is kicked.
+                */
+               used = avail;
+       }
+
+       if (used != avail) {
+               dev_dbg(dev, "Setting used equal to avail, for interoperability\n");
+               used = avail;
+       }
+
+       pdsv->vqs[qid].avail_idx = avail;
+       pdsv->vqs[qid].used_idx = used;
+
+       return 0;
+}
+
+static int pds_vdpa_get_vq_state(struct vdpa_device *vdpa_dev, u16 qid,
+                                struct vdpa_vq_state *state)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct pds_auxiliary_dev *padev = pdsv->vdpa_aux->padev;
+       struct device *dev = &padev->aux_dev.dev;
+       u64 driver_features;
+       u16 avail;
+       u16 used;
+
+       if (pdsv->vqs[qid].ready) {
+               dev_err(dev, "Getting device position is denied while vq is enabled\n");
+               return -EINVAL;
+       }
+
+       avail = pdsv->vqs[qid].avail_idx;
+       used = pdsv->vqs[qid].used_idx;
+
+       driver_features = pds_vdpa_get_driver_features(vdpa_dev);
+       if (driver_features & BIT_ULL(VIRTIO_F_RING_PACKED)) {
+               avail ^= PDS_VDPA_PACKED_INVERT_IDX;
+               used ^= PDS_VDPA_PACKED_INVERT_IDX;
+
+               state->packed.last_avail_idx = avail & 0x7fff;
+               state->packed.last_avail_counter = avail >> 15;
+               state->packed.last_used_idx = used & 0x7fff;
+               state->packed.last_used_counter = used >> 15;
+       } else {
+               state->split.avail_index = avail;
+               /* state->split does not provide a used_index. */
+       }
+
+       return 0;
+}
+
+static struct vdpa_notification_area
+pds_vdpa_get_vq_notification(struct vdpa_device *vdpa_dev, u16 qid)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct virtio_pci_modern_device *vd_mdev;
+       struct vdpa_notification_area area;
+
+       area.addr = pdsv->vqs[qid].notify_pa;
+
+       vd_mdev = &pdsv->vdpa_aux->vd_mdev;
+       if (!vd_mdev->notify_offset_multiplier)
+               area.size = PDS_PAGE_SIZE;
+       else
+               area.size = vd_mdev->notify_offset_multiplier;
+
+       return area;
+}
+
+static int pds_vdpa_get_vq_irq(struct vdpa_device *vdpa_dev, u16 qid)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       return pdsv->vqs[qid].irq;
+}
+
+static u32 pds_vdpa_get_vq_align(struct vdpa_device *vdpa_dev)
+{
+       return PDS_PAGE_SIZE;
+}
+
+static u32 pds_vdpa_get_vq_group(struct vdpa_device *vdpa_dev, u16 idx)
+{
+       return 0;
+}
+
+static u64 pds_vdpa_get_device_features(struct vdpa_device *vdpa_dev)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       return pdsv->supported_features;
+}
+
+static int pds_vdpa_set_driver_features(struct vdpa_device *vdpa_dev, u64 features)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct device *dev = &pdsv->vdpa_dev.dev;
+       u64 driver_features;
+       u64 nego_features;
+       u64 missing;
+
+       if (!(features & BIT_ULL(VIRTIO_F_ACCESS_PLATFORM)) && features) {
+               dev_err(dev, "VIRTIO_F_ACCESS_PLATFORM is not negotiated\n");
+               return -EOPNOTSUPP;
+       }
+
+       pdsv->req_features = features;
+
+       /* Check for valid feature bits */
+       nego_features = features & le64_to_cpu(pdsv->vdpa_aux->ident.hw_features);
+       missing = pdsv->req_features & ~nego_features;
+       if (missing) {
+               dev_err(dev, "Can't support all requested features in %#llx, missing %#llx features\n",
+                       pdsv->req_features, missing);
+               return -EOPNOTSUPP;
+       }
+
+       driver_features = pds_vdpa_get_driver_features(vdpa_dev);
+       dev_dbg(dev, "%s: %#llx => %#llx\n",
+               __func__, driver_features, nego_features);
+
+       if (driver_features == nego_features)
+               return 0;
+
+       vp_modern_set_features(&pdsv->vdpa_aux->vd_mdev, nego_features);
+
+       return 0;
+}
+
+static u64 pds_vdpa_get_driver_features(struct vdpa_device *vdpa_dev)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       return vp_modern_get_driver_features(&pdsv->vdpa_aux->vd_mdev);
+}
+
+static void pds_vdpa_set_config_cb(struct vdpa_device *vdpa_dev,
+                                  struct vdpa_callback *cb)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       pdsv->config_cb.callback = cb->callback;
+       pdsv->config_cb.private = cb->private;
+}
+
+static u16 pds_vdpa_get_vq_num_max(struct vdpa_device *vdpa_dev)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       /* qemu has assert() that vq_num_max <= VIRTQUEUE_MAX_SIZE (1024) */
+       return min_t(u16, 1024, BIT(le16_to_cpu(pdsv->vdpa_aux->ident.max_qlen)));
+}
+
+static u32 pds_vdpa_get_device_id(struct vdpa_device *vdpa_dev)
+{
+       return VIRTIO_ID_NET;
+}
+
+static u32 pds_vdpa_get_vendor_id(struct vdpa_device *vdpa_dev)
+{
+       return PCI_VENDOR_ID_PENSANDO;
+}
+
+static u8 pds_vdpa_get_status(struct vdpa_device *vdpa_dev)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+
+       return vp_modern_get_status(&pdsv->vdpa_aux->vd_mdev);
+}
+
+static void pds_vdpa_set_status(struct vdpa_device *vdpa_dev, u8 status)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct device *dev = &pdsv->vdpa_dev.dev;
+       u8 old_status;
+       int i;
+
+       old_status = pds_vdpa_get_status(vdpa_dev);
+       dev_dbg(dev, "%s: old %#x new %#x\n", __func__, old_status, status);
+
+       pds_vdpa_cmd_set_status(pdsv, status);
+
+       /* Note: still working with FW on the need for this reset cmd */
+       if (status == 0) {
+               pds_vdpa_cmd_reset(pdsv);
+
+               for (i = 0; i < pdsv->num_vqs; i++) {
+                       pdsv->vqs[i].avail_idx = 0;
+                       pdsv->vqs[i].used_idx = 0;
+               }
+       }
+
+       if (status & ~old_status & VIRTIO_CONFIG_S_FEATURES_OK) {
+               for (i = 0; i < pdsv->num_vqs; i++) {
+                       pdsv->vqs[i].notify =
+                               vp_modern_map_vq_notify(&pdsv->vdpa_aux->vd_mdev,
+                                                       i, &pdsv->vqs[i].notify_pa);
+               }
+       }
+}
+
+static int pds_vdpa_reset(struct vdpa_device *vdpa_dev)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       struct device *dev;
+       int err = 0;
+       u8 status;
+       int i;
+
+       dev = &pdsv->vdpa_aux->padev->aux_dev.dev;
+       status = pds_vdpa_get_status(vdpa_dev);
+
+       if (status == 0)
+               return 0;
+
+       if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
+               /* Reset the vqs */
+               for (i = 0; i < pdsv->num_vqs && !err; i++) {
+                       err = pds_vdpa_cmd_reset_vq(pdsv, i, 0, &pdsv->vqs[i]);
+                       if (err)
+                               dev_err(dev, "%s: reset_vq failed qid %d: %pe\n",
+                                       __func__, i, ERR_PTR(err));
+                       pds_vdpa_release_irq(pdsv, i);
+                       memset(&pdsv->vqs[i], 0, sizeof(pdsv->vqs[0]));
+                       pdsv->vqs[i].ready = false;
+               }
+       }
+
+       pds_vdpa_set_status(vdpa_dev, 0);
+
+       return 0;
+}
+
+static size_t pds_vdpa_get_config_size(struct vdpa_device *vdpa_dev)
+{
+       return sizeof(struct virtio_net_config);
+}
+
+static void pds_vdpa_get_config(struct vdpa_device *vdpa_dev,
+                               unsigned int offset,
+                               void *buf, unsigned int len)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       void __iomem *device;
+
+       if (offset + len > sizeof(struct virtio_net_config)) {
+               WARN(true, "%s: bad read, offset %d len %d\n", __func__, offset, len);
+               return;
+       }
+
+       device = pdsv->vdpa_aux->vd_mdev.device;
+       memcpy_fromio(buf, device + offset, len);
+}
+
+static void pds_vdpa_set_config(struct vdpa_device *vdpa_dev,
+                               unsigned int offset, const void *buf,
+                               unsigned int len)
+{
+       struct pds_vdpa_device *pdsv = vdpa_to_pdsv(vdpa_dev);
+       void __iomem *device;
+
+       if (offset + len > sizeof(struct virtio_net_config)) {
+               WARN(true, "%s: bad read, offset %d len %d\n", __func__, offset, len);
+               return;
+       }
+
+       device = pdsv->vdpa_aux->vd_mdev.device;
+       memcpy_toio(device + offset, buf, len);
+}
+
+static const struct vdpa_config_ops pds_vdpa_ops = {
+       .set_vq_address         = pds_vdpa_set_vq_address,
+       .set_vq_num             = pds_vdpa_set_vq_num,
+       .kick_vq                = pds_vdpa_kick_vq,
+       .set_vq_cb              = pds_vdpa_set_vq_cb,
+       .set_vq_ready           = pds_vdpa_set_vq_ready,
+       .get_vq_ready           = pds_vdpa_get_vq_ready,
+       .set_vq_state           = pds_vdpa_set_vq_state,
+       .get_vq_state           = pds_vdpa_get_vq_state,
+       .get_vq_notification    = pds_vdpa_get_vq_notification,
+       .get_vq_irq             = pds_vdpa_get_vq_irq,
+       .get_vq_align           = pds_vdpa_get_vq_align,
+       .get_vq_group           = pds_vdpa_get_vq_group,
+
+       .get_device_features    = pds_vdpa_get_device_features,
+       .set_driver_features    = pds_vdpa_set_driver_features,
+       .get_driver_features    = pds_vdpa_get_driver_features,
+       .set_config_cb          = pds_vdpa_set_config_cb,
+       .get_vq_num_max         = pds_vdpa_get_vq_num_max,
+       .get_device_id          = pds_vdpa_get_device_id,
+       .get_vendor_id          = pds_vdpa_get_vendor_id,
+       .get_status             = pds_vdpa_get_status,
+       .set_status             = pds_vdpa_set_status,
+       .reset                  = pds_vdpa_reset,
+       .get_config_size        = pds_vdpa_get_config_size,
+       .get_config             = pds_vdpa_get_config,
+       .set_config             = pds_vdpa_set_config,
+};
 static struct virtio_device_id pds_vdpa_id_table[] = {
        {VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID},
        {0},
@@ -21,12 +479,157 @@ static struct virtio_device_id pds_vdpa_id_table[] = {
 static int pds_vdpa_dev_add(struct vdpa_mgmt_dev *mdev, const char *name,
                            const struct vdpa_dev_set_config *add_config)
 {
-       return -EOPNOTSUPP;
+       struct pds_vdpa_aux *vdpa_aux;
+       struct pds_vdpa_device *pdsv;
+       struct vdpa_mgmt_dev *mgmt;
+       u16 fw_max_vqs, vq_pairs;
+       struct device *dma_dev;
+       struct pci_dev *pdev;
+       struct device *dev;
+       u8 mac[ETH_ALEN];
+       int err;
+       int i;
+
+       vdpa_aux = container_of(mdev, struct pds_vdpa_aux, vdpa_mdev);
+       dev = &vdpa_aux->padev->aux_dev.dev;
+       mgmt = &vdpa_aux->vdpa_mdev;
+
+       if (vdpa_aux->pdsv) {
+               dev_warn(dev, "Multiple vDPA devices on a VF is not supported.\n");
+               return -EOPNOTSUPP;
+       }
+
+       pdsv = vdpa_alloc_device(struct pds_vdpa_device, vdpa_dev,
+                                dev, &pds_vdpa_ops, 1, 1, name, false);
+       if (IS_ERR(pdsv)) {
+               dev_err(dev, "Failed to allocate vDPA structure: %pe\n", pdsv);
+               return PTR_ERR(pdsv);
+       }
+
+       vdpa_aux->pdsv = pdsv;
+       pdsv->vdpa_aux = vdpa_aux;
+
+       pdev = vdpa_aux->padev->vf_pdev;
+       dma_dev = &pdev->dev;
+       pdsv->vdpa_dev.dma_dev = dma_dev;
+
+       pdsv->supported_features = mgmt->supported_features;
+
+       if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_FEATURES)) {
+               u64 unsupp_features =
+                       add_config->device_features & ~mgmt->supported_features;
+
+               if (unsupp_features) {
+                       dev_err(dev, "Unsupported features: %#llx\n", unsupp_features);
+                       err = -EOPNOTSUPP;
+                       goto err_unmap;
+               }
+
+               pdsv->supported_features = add_config->device_features;
+       }
+
+       err = pds_vdpa_cmd_reset(pdsv);
+       if (err) {
+               dev_err(dev, "Failed to reset hw: %pe\n", ERR_PTR(err));
+               goto err_unmap;
+       }
+
+       err = pds_vdpa_init_hw(pdsv);
+       if (err) {
+               dev_err(dev, "Failed to init hw: %pe\n", ERR_PTR(err));
+               goto err_unmap;
+       }
+
+       fw_max_vqs = le16_to_cpu(pdsv->vdpa_aux->ident.max_vqs);
+       vq_pairs = fw_max_vqs / 2;
+
+       /* Make sure we have the queues being requested */
+       if (add_config->mask & (1 << VDPA_ATTR_DEV_NET_CFG_MAX_VQP))
+               vq_pairs = add_config->net.max_vq_pairs;
+
+       pdsv->num_vqs = 2 * vq_pairs;
+       if (pdsv->supported_features & BIT_ULL(VIRTIO_NET_F_CTRL_VQ))
+               pdsv->num_vqs++;
+
+       if (pdsv->num_vqs > fw_max_vqs) {
+               dev_err(dev, "%s: queue count requested %u greater than max %u\n",
+                       __func__, pdsv->num_vqs, fw_max_vqs);
+               err = -ENOSPC;
+               goto err_unmap;
+       }
+
+       if (pdsv->num_vqs != fw_max_vqs) {
+               err = pds_vdpa_cmd_set_max_vq_pairs(pdsv, vq_pairs);
+               if (err) {
+                       dev_err(dev, "Failed to set max_vq_pairs: %pe\n",
+                               ERR_PTR(err));
+                       goto err_unmap;
+               }
+       }
+
+       /* Set a mac, either from the user config if provided
+        * or set a random mac if default is 00:..:00
+        */
+       if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MACADDR)) {
+               ether_addr_copy(mac, add_config->net.mac);
+               pds_vdpa_cmd_set_mac(pdsv, mac);
+       } else {
+               struct virtio_net_config __iomem *vc;
+
+               vc = pdsv->vdpa_aux->vd_mdev.device;
+               memcpy_fromio(mac, vc->mac, sizeof(mac));
+               if (is_zero_ether_addr(mac)) {
+                       eth_random_addr(mac);
+                       dev_info(dev, "setting random mac %pM\n", mac);
+                       pds_vdpa_cmd_set_mac(pdsv, mac);
+               }
+       }
+
+       for (i = 0; i < pdsv->num_vqs; i++) {
+               pdsv->vqs[i].qid = i;
+               pdsv->vqs[i].pdsv = pdsv;
+               pdsv->vqs[i].irq = VIRTIO_MSI_NO_VECTOR;
+               pdsv->vqs[i].notify = vp_modern_map_vq_notify(&pdsv->vdpa_aux->vd_mdev,
+                                                             i, &pdsv->vqs[i].notify_pa);
+       }
+
+       pdsv->vdpa_dev.mdev = &vdpa_aux->vdpa_mdev;
+
+       /* We use the _vdpa_register_device() call rather than the
+        * vdpa_register_device() to avoid a deadlock because our
+        * dev_add() is called with the vdpa_dev_lock already set
+        * by vdpa_nl_cmd_dev_add_set_doit()
+        */
+       err = _vdpa_register_device(&pdsv->vdpa_dev, pdsv->num_vqs);
+       if (err) {
+               dev_err(dev, "Failed to register to vDPA bus: %pe\n", ERR_PTR(err));
+               goto err_unmap;
+       }
+
+       pds_vdpa_debugfs_add_vdpadev(vdpa_aux);
+
+       return 0;
+
+err_unmap:
+       put_device(&pdsv->vdpa_dev.dev);
+       vdpa_aux->pdsv = NULL;
+       return err;
 }
 
 static void pds_vdpa_dev_del(struct vdpa_mgmt_dev *mdev,
                             struct vdpa_device *vdpa_dev)
 {
+       struct pds_vdpa_aux *vdpa_aux;
+
+       vdpa_aux = container_of(mdev, struct pds_vdpa_aux, vdpa_mdev);
+       _vdpa_unregister_device(vdpa_dev);
+
+       pds_vdpa_cmd_reset(vdpa_aux->pdsv);
+       pds_vdpa_debugfs_reset_vdpadev(vdpa_aux);
+
+       vdpa_aux->pdsv = NULL;
+
+       dev_info(&vdpa_aux->padev->aux_dev.dev, "Removed vdpa device\n");
 }
 
 static const struct vdpa_mgmtdev_ops pds_vdpa_mgmt_dev_ops = {
@@ -94,6 +697,7 @@ int pds_vdpa_get_mgmt_info(struct pds_vdpa_aux *vdpa_aux)
        mgmt->supported_features = le64_to_cpu(vdpa_aux->ident.hw_features);
        mgmt->config_attr_mask = BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MACADDR);
        mgmt->config_attr_mask |= BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MAX_VQP);
+       mgmt->config_attr_mask |= BIT_ULL(VDPA_ATTR_DEV_FEATURES);
 
        err = pci_alloc_irq_vectors(pdev, vdpa_aux->nintrs, vdpa_aux->nintrs,
                                    PCI_IRQ_MSIX);
index a21596f438c1ddbf75d329e31b5e0146cda16b14..25c1d192f0efb52840956169cad54330a333a003 100644 (file)
@@ -35,12 +35,14 @@ struct pds_vdpa_device {
        struct pds_vdpa_aux *vdpa_aux;
 
        struct pds_vdpa_vq_info vqs[PDS_VDPA_MAX_QUEUES];
+       u64 supported_features;         /* specified device features */
        u64 req_features;               /* features requested by vdpa */
-       u64 actual_features;            /* features negotiated and in use */
        u8 vdpa_index;                  /* rsvd for future subdevice use */
        u8 num_vqs;                     /* num vqs in use */
        struct vdpa_callback config_cb;
 };
 
+#define PDS_VDPA_PACKED_INVERT_IDX     0x8000
+
 int pds_vdpa_get_mgmt_info(struct pds_vdpa_aux *vdpa_aux);
 #endif /* _VDPA_DEV_H_ */