Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / sfc / mcdi.c
diff --git a/kernel/drivers/net/ethernet/sfc/mcdi.c b/kernel/drivers/net/ethernet/sfc/mcdi.c
new file mode 100644 (file)
index 0000000..d37928f
--- /dev/null
@@ -0,0 +1,1891 @@
+/****************************************************************************
+ * Driver for Solarflare network controllers and boards
+ * Copyright 2008-2013 Solarflare Communications Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation, incorporated herein by reference.
+ */
+
+#include <linux/delay.h>
+#include <asm/cmpxchg.h>
+#include "net_driver.h"
+#include "nic.h"
+#include "io.h"
+#include "farch_regs.h"
+#include "mcdi_pcol.h"
+#include "phy.h"
+
+/**************************************************************************
+ *
+ * Management-Controller-to-Driver Interface
+ *
+ **************************************************************************
+ */
+
+#define MCDI_RPC_TIMEOUT       (10 * HZ)
+
+/* A reboot/assertion causes the MCDI status word to be set after the
+ * command word is set or a REBOOT event is sent. If we notice a reboot
+ * via these mechanisms then wait 250ms for the status word to be set.
+ */
+#define MCDI_STATUS_DELAY_US           100
+#define MCDI_STATUS_DELAY_COUNT                2500
+#define MCDI_STATUS_SLEEP_MS                                           \
+       (MCDI_STATUS_DELAY_US * MCDI_STATUS_DELAY_COUNT / 1000)
+
+#define SEQ_MASK                                                       \
+       EFX_MASK32(EFX_WIDTH(MCDI_HEADER_SEQ))
+
+struct efx_mcdi_async_param {
+       struct list_head list;
+       unsigned int cmd;
+       size_t inlen;
+       size_t outlen;
+       bool quiet;
+       efx_mcdi_async_completer *complete;
+       unsigned long cookie;
+       /* followed by request/response buffer */
+};
+
+static void efx_mcdi_timeout_async(unsigned long context);
+static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
+                              bool *was_attached_out);
+static bool efx_mcdi_poll_once(struct efx_nic *efx);
+static void efx_mcdi_abandon(struct efx_nic *efx);
+
+int efx_mcdi_init(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi;
+       bool already_attached;
+       int rc;
+
+       efx->mcdi = kzalloc(sizeof(*efx->mcdi), GFP_KERNEL);
+       if (!efx->mcdi)
+               return -ENOMEM;
+
+       mcdi = efx_mcdi(efx);
+       mcdi->efx = efx;
+       init_waitqueue_head(&mcdi->wq);
+       spin_lock_init(&mcdi->iface_lock);
+       mcdi->state = MCDI_STATE_QUIESCENT;
+       mcdi->mode = MCDI_MODE_POLL;
+       spin_lock_init(&mcdi->async_lock);
+       INIT_LIST_HEAD(&mcdi->async_list);
+       setup_timer(&mcdi->async_timer, efx_mcdi_timeout_async,
+                   (unsigned long)mcdi);
+
+       (void) efx_mcdi_poll_reboot(efx);
+       mcdi->new_epoch = true;
+
+       /* Recover from a failed assertion before probing */
+       rc = efx_mcdi_handle_assertion(efx);
+       if (rc)
+               return rc;
+
+       /* Let the MC (and BMC, if this is a LOM) know that the driver
+        * is loaded. We should do this before we reset the NIC.
+        */
+       rc = efx_mcdi_drv_attach(efx, true, &already_attached);
+       if (rc) {
+               netif_err(efx, probe, efx->net_dev,
+                         "Unable to register driver with MCPU\n");
+               return rc;
+       }
+       if (already_attached)
+               /* Not a fatal error */
+               netif_err(efx, probe, efx->net_dev,
+                         "Host already registered with MCPU\n");
+
+       if (efx->mcdi->fn_flags &
+           (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
+               efx->primary = efx;
+
+       return 0;
+}
+
+void efx_mcdi_fini(struct efx_nic *efx)
+{
+       if (!efx->mcdi)
+               return;
+
+       BUG_ON(efx->mcdi->iface.state != MCDI_STATE_QUIESCENT);
+
+       /* Relinquish the device (back to the BMC, if this is a LOM) */
+       efx_mcdi_drv_attach(efx, false, NULL);
+
+       kfree(efx->mcdi);
+}
+
+static void efx_mcdi_send_request(struct efx_nic *efx, unsigned cmd,
+                                 const efx_dword_t *inbuf, size_t inlen)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       efx_dword_t hdr[2];
+       size_t hdr_len;
+       u32 xflags, seqno;
+
+       BUG_ON(mcdi->state == MCDI_STATE_QUIESCENT);
+
+       /* Serialise with efx_mcdi_ev_cpl() and efx_mcdi_ev_death() */
+       spin_lock_bh(&mcdi->iface_lock);
+       ++mcdi->seqno;
+       spin_unlock_bh(&mcdi->iface_lock);
+
+       seqno = mcdi->seqno & SEQ_MASK;
+       xflags = 0;
+       if (mcdi->mode == MCDI_MODE_EVENTS)
+               xflags |= MCDI_HEADER_XFLAGS_EVREQ;
+
+       if (efx->type->mcdi_max_ver == 1) {
+               /* MCDI v1 */
+               EFX_POPULATE_DWORD_7(hdr[0],
+                                    MCDI_HEADER_RESPONSE, 0,
+                                    MCDI_HEADER_RESYNC, 1,
+                                    MCDI_HEADER_CODE, cmd,
+                                    MCDI_HEADER_DATALEN, inlen,
+                                    MCDI_HEADER_SEQ, seqno,
+                                    MCDI_HEADER_XFLAGS, xflags,
+                                    MCDI_HEADER_NOT_EPOCH, !mcdi->new_epoch);
+               hdr_len = 4;
+       } else {
+               /* MCDI v2 */
+               BUG_ON(inlen > MCDI_CTL_SDU_LEN_MAX_V2);
+               EFX_POPULATE_DWORD_7(hdr[0],
+                                    MCDI_HEADER_RESPONSE, 0,
+                                    MCDI_HEADER_RESYNC, 1,
+                                    MCDI_HEADER_CODE, MC_CMD_V2_EXTN,
+                                    MCDI_HEADER_DATALEN, 0,
+                                    MCDI_HEADER_SEQ, seqno,
+                                    MCDI_HEADER_XFLAGS, xflags,
+                                    MCDI_HEADER_NOT_EPOCH, !mcdi->new_epoch);
+               EFX_POPULATE_DWORD_2(hdr[1],
+                                    MC_CMD_V2_EXTN_IN_EXTENDED_CMD, cmd,
+                                    MC_CMD_V2_EXTN_IN_ACTUAL_LEN, inlen);
+               hdr_len = 8;
+       }
+
+       efx->type->mcdi_request(efx, hdr, hdr_len, inbuf, inlen);
+
+       mcdi->new_epoch = false;
+}
+
+static int efx_mcdi_errno(unsigned int mcdi_err)
+{
+       switch (mcdi_err) {
+       case 0:
+               return 0;
+#define TRANSLATE_ERROR(name)                                  \
+       case MC_CMD_ERR_ ## name:                               \
+               return -name;
+       TRANSLATE_ERROR(EPERM);
+       TRANSLATE_ERROR(ENOENT);
+       TRANSLATE_ERROR(EINTR);
+       TRANSLATE_ERROR(EAGAIN);
+       TRANSLATE_ERROR(EACCES);
+       TRANSLATE_ERROR(EBUSY);
+       TRANSLATE_ERROR(EINVAL);
+       TRANSLATE_ERROR(EDEADLK);
+       TRANSLATE_ERROR(ENOSYS);
+       TRANSLATE_ERROR(ETIME);
+       TRANSLATE_ERROR(EALREADY);
+       TRANSLATE_ERROR(ENOSPC);
+#undef TRANSLATE_ERROR
+       case MC_CMD_ERR_ENOTSUP:
+               return -EOPNOTSUPP;
+       case MC_CMD_ERR_ALLOC_FAIL:
+               return -ENOBUFS;
+       case MC_CMD_ERR_MAC_EXIST:
+               return -EADDRINUSE;
+       default:
+               return -EPROTO;
+       }
+}
+
+static void efx_mcdi_read_response_header(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       unsigned int respseq, respcmd, error;
+       efx_dword_t hdr;
+
+       efx->type->mcdi_read_response(efx, &hdr, 0, 4);
+       respseq = EFX_DWORD_FIELD(hdr, MCDI_HEADER_SEQ);
+       respcmd = EFX_DWORD_FIELD(hdr, MCDI_HEADER_CODE);
+       error = EFX_DWORD_FIELD(hdr, MCDI_HEADER_ERROR);
+
+       if (respcmd != MC_CMD_V2_EXTN) {
+               mcdi->resp_hdr_len = 4;
+               mcdi->resp_data_len = EFX_DWORD_FIELD(hdr, MCDI_HEADER_DATALEN);
+       } else {
+               efx->type->mcdi_read_response(efx, &hdr, 4, 4);
+               mcdi->resp_hdr_len = 8;
+               mcdi->resp_data_len =
+                       EFX_DWORD_FIELD(hdr, MC_CMD_V2_EXTN_IN_ACTUAL_LEN);
+       }
+
+       if (error && mcdi->resp_data_len == 0) {
+               netif_err(efx, hw, efx->net_dev, "MC rebooted\n");
+               mcdi->resprc = -EIO;
+       } else if ((respseq ^ mcdi->seqno) & SEQ_MASK) {
+               netif_err(efx, hw, efx->net_dev,
+                         "MC response mismatch tx seq 0x%x rx seq 0x%x\n",
+                         respseq, mcdi->seqno);
+               mcdi->resprc = -EIO;
+       } else if (error) {
+               efx->type->mcdi_read_response(efx, &hdr, mcdi->resp_hdr_len, 4);
+               mcdi->resprc =
+                       efx_mcdi_errno(EFX_DWORD_FIELD(hdr, EFX_DWORD_0));
+       } else {
+               mcdi->resprc = 0;
+       }
+}
+
+static bool efx_mcdi_poll_once(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+
+       rmb();
+       if (!efx->type->mcdi_poll_response(efx))
+               return false;
+
+       spin_lock_bh(&mcdi->iface_lock);
+       efx_mcdi_read_response_header(efx);
+       spin_unlock_bh(&mcdi->iface_lock);
+
+       return true;
+}
+
+static int efx_mcdi_poll(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       unsigned long time, finish;
+       unsigned int spins;
+       int rc;
+
+       /* Check for a reboot atomically with respect to efx_mcdi_copyout() */
+       rc = efx_mcdi_poll_reboot(efx);
+       if (rc) {
+               spin_lock_bh(&mcdi->iface_lock);
+               mcdi->resprc = rc;
+               mcdi->resp_hdr_len = 0;
+               mcdi->resp_data_len = 0;
+               spin_unlock_bh(&mcdi->iface_lock);
+               return 0;
+       }
+
+       /* Poll for completion. Poll quickly (once a us) for the 1st jiffy,
+        * because generally mcdi responses are fast. After that, back off
+        * and poll once a jiffy (approximately)
+        */
+       spins = TICK_USEC;
+       finish = jiffies + MCDI_RPC_TIMEOUT;
+
+       while (1) {
+               if (spins != 0) {
+                       --spins;
+                       udelay(1);
+               } else {
+                       schedule_timeout_uninterruptible(1);
+               }
+
+               time = jiffies;
+
+               if (efx_mcdi_poll_once(efx))
+                       break;
+
+               if (time_after(time, finish))
+                       return -ETIMEDOUT;
+       }
+
+       /* Return rc=0 like wait_event_timeout() */
+       return 0;
+}
+
+/* Test and clear MC-rebooted flag for this port/function; reset
+ * software state as necessary.
+ */
+int efx_mcdi_poll_reboot(struct efx_nic *efx)
+{
+       if (!efx->mcdi)
+               return 0;
+
+       return efx->type->mcdi_poll_reboot(efx);
+}
+
+static bool efx_mcdi_acquire_async(struct efx_mcdi_iface *mcdi)
+{
+       return cmpxchg(&mcdi->state,
+                      MCDI_STATE_QUIESCENT, MCDI_STATE_RUNNING_ASYNC) ==
+               MCDI_STATE_QUIESCENT;
+}
+
+static void efx_mcdi_acquire_sync(struct efx_mcdi_iface *mcdi)
+{
+       /* Wait until the interface becomes QUIESCENT and we win the race
+        * to mark it RUNNING_SYNC.
+        */
+       wait_event(mcdi->wq,
+                  cmpxchg(&mcdi->state,
+                          MCDI_STATE_QUIESCENT, MCDI_STATE_RUNNING_SYNC) ==
+                  MCDI_STATE_QUIESCENT);
+}
+
+static int efx_mcdi_await_completion(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+
+       if (wait_event_timeout(mcdi->wq, mcdi->state == MCDI_STATE_COMPLETED,
+                              MCDI_RPC_TIMEOUT) == 0)
+               return -ETIMEDOUT;
+
+       /* Check if efx_mcdi_set_mode() switched us back to polled completions.
+        * In which case, poll for completions directly. If efx_mcdi_ev_cpl()
+        * completed the request first, then we'll just end up completing the
+        * request again, which is safe.
+        *
+        * We need an smp_rmb() to synchronise with efx_mcdi_mode_poll(), which
+        * wait_event_timeout() implicitly provides.
+        */
+       if (mcdi->mode == MCDI_MODE_POLL)
+               return efx_mcdi_poll(efx);
+
+       return 0;
+}
+
+/* If the interface is RUNNING_SYNC, switch to COMPLETED and wake the
+ * requester.  Return whether this was done.  Does not take any locks.
+ */
+static bool efx_mcdi_complete_sync(struct efx_mcdi_iface *mcdi)
+{
+       if (cmpxchg(&mcdi->state,
+                   MCDI_STATE_RUNNING_SYNC, MCDI_STATE_COMPLETED) ==
+           MCDI_STATE_RUNNING_SYNC) {
+               wake_up(&mcdi->wq);
+               return true;
+       }
+
+       return false;
+}
+
+static void efx_mcdi_release(struct efx_mcdi_iface *mcdi)
+{
+       if (mcdi->mode == MCDI_MODE_EVENTS) {
+               struct efx_mcdi_async_param *async;
+               struct efx_nic *efx = mcdi->efx;
+
+               /* Process the asynchronous request queue */
+               spin_lock_bh(&mcdi->async_lock);
+               async = list_first_entry_or_null(
+                       &mcdi->async_list, struct efx_mcdi_async_param, list);
+               if (async) {
+                       mcdi->state = MCDI_STATE_RUNNING_ASYNC;
+                       efx_mcdi_send_request(efx, async->cmd,
+                                             (const efx_dword_t *)(async + 1),
+                                             async->inlen);
+                       mod_timer(&mcdi->async_timer,
+                                 jiffies + MCDI_RPC_TIMEOUT);
+               }
+               spin_unlock_bh(&mcdi->async_lock);
+
+               if (async)
+                       return;
+       }
+
+       mcdi->state = MCDI_STATE_QUIESCENT;
+       wake_up(&mcdi->wq);
+}
+
+/* If the interface is RUNNING_ASYNC, switch to COMPLETED, call the
+ * asynchronous completion function, and release the interface.
+ * Return whether this was done.  Must be called in bh-disabled
+ * context.  Will take iface_lock and async_lock.
+ */
+static bool efx_mcdi_complete_async(struct efx_mcdi_iface *mcdi, bool timeout)
+{
+       struct efx_nic *efx = mcdi->efx;
+       struct efx_mcdi_async_param *async;
+       size_t hdr_len, data_len, err_len;
+       efx_dword_t *outbuf;
+       MCDI_DECLARE_BUF_OUT_OR_ERR(errbuf, 0);
+       int rc;
+
+       if (cmpxchg(&mcdi->state,
+                   MCDI_STATE_RUNNING_ASYNC, MCDI_STATE_COMPLETED) !=
+           MCDI_STATE_RUNNING_ASYNC)
+               return false;
+
+       spin_lock(&mcdi->iface_lock);
+       if (timeout) {
+               /* Ensure that if the completion event arrives later,
+                * the seqno check in efx_mcdi_ev_cpl() will fail
+                */
+               ++mcdi->seqno;
+               ++mcdi->credits;
+               rc = -ETIMEDOUT;
+               hdr_len = 0;
+               data_len = 0;
+       } else {
+               rc = mcdi->resprc;
+               hdr_len = mcdi->resp_hdr_len;
+               data_len = mcdi->resp_data_len;
+       }
+       spin_unlock(&mcdi->iface_lock);
+
+       /* Stop the timer.  In case the timer function is running, we
+        * must wait for it to return so that there is no possibility
+        * of it aborting the next request.
+        */
+       if (!timeout)
+               del_timer_sync(&mcdi->async_timer);
+
+       spin_lock(&mcdi->async_lock);
+       async = list_first_entry(&mcdi->async_list,
+                                struct efx_mcdi_async_param, list);
+       list_del(&async->list);
+       spin_unlock(&mcdi->async_lock);
+
+       outbuf = (efx_dword_t *)(async + 1);
+       efx->type->mcdi_read_response(efx, outbuf, hdr_len,
+                                     min(async->outlen, data_len));
+       if (!timeout && rc && !async->quiet) {
+               err_len = min(sizeof(errbuf), data_len);
+               efx->type->mcdi_read_response(efx, errbuf, hdr_len,
+                                             sizeof(errbuf));
+               efx_mcdi_display_error(efx, async->cmd, async->inlen, errbuf,
+                                      err_len, rc);
+       }
+       async->complete(efx, async->cookie, rc, outbuf, data_len);
+       kfree(async);
+
+       efx_mcdi_release(mcdi);
+
+       return true;
+}
+
+static void efx_mcdi_ev_cpl(struct efx_nic *efx, unsigned int seqno,
+                           unsigned int datalen, unsigned int mcdi_err)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       bool wake = false;
+
+       spin_lock(&mcdi->iface_lock);
+
+       if ((seqno ^ mcdi->seqno) & SEQ_MASK) {
+               if (mcdi->credits)
+                       /* The request has been cancelled */
+                       --mcdi->credits;
+               else
+                       netif_err(efx, hw, efx->net_dev,
+                                 "MC response mismatch tx seq 0x%x rx "
+                                 "seq 0x%x\n", seqno, mcdi->seqno);
+       } else {
+               if (efx->type->mcdi_max_ver >= 2) {
+                       /* MCDI v2 responses don't fit in an event */
+                       efx_mcdi_read_response_header(efx);
+               } else {
+                       mcdi->resprc = efx_mcdi_errno(mcdi_err);
+                       mcdi->resp_hdr_len = 4;
+                       mcdi->resp_data_len = datalen;
+               }
+
+               wake = true;
+       }
+
+       spin_unlock(&mcdi->iface_lock);
+
+       if (wake) {
+               if (!efx_mcdi_complete_async(mcdi, false))
+                       (void) efx_mcdi_complete_sync(mcdi);
+
+               /* If the interface isn't RUNNING_ASYNC or
+                * RUNNING_SYNC then we've received a duplicate
+                * completion after we've already transitioned back to
+                * QUIESCENT. [A subsequent invocation would increment
+                * seqno, so would have failed the seqno check].
+                */
+       }
+}
+
+static void efx_mcdi_timeout_async(unsigned long context)
+{
+       struct efx_mcdi_iface *mcdi = (struct efx_mcdi_iface *)context;
+
+       efx_mcdi_complete_async(mcdi, true);
+}
+
+static int
+efx_mcdi_check_supported(struct efx_nic *efx, unsigned int cmd, size_t inlen)
+{
+       if (efx->type->mcdi_max_ver < 0 ||
+            (efx->type->mcdi_max_ver < 2 &&
+             cmd > MC_CMD_CMD_SPACE_ESCAPE_7))
+               return -EINVAL;
+
+       if (inlen > MCDI_CTL_SDU_LEN_MAX_V2 ||
+           (efx->type->mcdi_max_ver < 2 &&
+            inlen > MCDI_CTL_SDU_LEN_MAX_V1))
+               return -EMSGSIZE;
+
+       return 0;
+}
+
+static int _efx_mcdi_rpc_finish(struct efx_nic *efx, unsigned cmd, size_t inlen,
+                               efx_dword_t *outbuf, size_t outlen,
+                               size_t *outlen_actual, bool quiet)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       MCDI_DECLARE_BUF_OUT_OR_ERR(errbuf, 0);
+       int rc;
+
+       if (mcdi->mode == MCDI_MODE_POLL)
+               rc = efx_mcdi_poll(efx);
+       else
+               rc = efx_mcdi_await_completion(efx);
+
+       if (rc != 0) {
+               netif_err(efx, hw, efx->net_dev,
+                         "MC command 0x%x inlen %d mode %d timed out\n",
+                         cmd, (int)inlen, mcdi->mode);
+
+               if (mcdi->mode == MCDI_MODE_EVENTS && efx_mcdi_poll_once(efx)) {
+                       netif_err(efx, hw, efx->net_dev,
+                                 "MCDI request was completed without an event\n");
+                       rc = 0;
+               }
+
+               efx_mcdi_abandon(efx);
+
+               /* Close the race with efx_mcdi_ev_cpl() executing just too late
+                * and completing a request we've just cancelled, by ensuring
+                * that the seqno check therein fails.
+                */
+               spin_lock_bh(&mcdi->iface_lock);
+               ++mcdi->seqno;
+               ++mcdi->credits;
+               spin_unlock_bh(&mcdi->iface_lock);
+       }
+
+       if (rc != 0) {
+               if (outlen_actual)
+                       *outlen_actual = 0;
+       } else {
+               size_t hdr_len, data_len, err_len;
+
+               /* At the very least we need a memory barrier here to ensure
+                * we pick up changes from efx_mcdi_ev_cpl(). Protect against
+                * a spurious efx_mcdi_ev_cpl() running concurrently by
+                * acquiring the iface_lock. */
+               spin_lock_bh(&mcdi->iface_lock);
+               rc = mcdi->resprc;
+               hdr_len = mcdi->resp_hdr_len;
+               data_len = mcdi->resp_data_len;
+               err_len = min(sizeof(errbuf), data_len);
+               spin_unlock_bh(&mcdi->iface_lock);
+
+               BUG_ON(rc > 0);
+
+               efx->type->mcdi_read_response(efx, outbuf, hdr_len,
+                                             min(outlen, data_len));
+               if (outlen_actual)
+                       *outlen_actual = data_len;
+
+               efx->type->mcdi_read_response(efx, errbuf, hdr_len, err_len);
+
+               if (cmd == MC_CMD_REBOOT && rc == -EIO) {
+                       /* Don't reset if MC_CMD_REBOOT returns EIO */
+               } else if (rc == -EIO || rc == -EINTR) {
+                       netif_err(efx, hw, efx->net_dev, "MC fatal error %d\n",
+                                 -rc);
+                       efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
+               } else if (rc && !quiet) {
+                       efx_mcdi_display_error(efx, cmd, inlen, errbuf, err_len,
+                                              rc);
+               }
+
+               if (rc == -EIO || rc == -EINTR) {
+                       msleep(MCDI_STATUS_SLEEP_MS);
+                       efx_mcdi_poll_reboot(efx);
+                       mcdi->new_epoch = true;
+               }
+       }
+
+       efx_mcdi_release(mcdi);
+       return rc;
+}
+
+static int _efx_mcdi_rpc(struct efx_nic *efx, unsigned cmd,
+                        const efx_dword_t *inbuf, size_t inlen,
+                        efx_dword_t *outbuf, size_t outlen,
+                        size_t *outlen_actual, bool quiet)
+{
+       int rc;
+
+       rc = efx_mcdi_rpc_start(efx, cmd, inbuf, inlen);
+       if (rc) {
+               if (outlen_actual)
+                       *outlen_actual = 0;
+               return rc;
+       }
+       return _efx_mcdi_rpc_finish(efx, cmd, inlen, outbuf, outlen,
+                                   outlen_actual, quiet);
+}
+
+int efx_mcdi_rpc(struct efx_nic *efx, unsigned cmd,
+                const efx_dword_t *inbuf, size_t inlen,
+                efx_dword_t *outbuf, size_t outlen,
+                size_t *outlen_actual)
+{
+       return _efx_mcdi_rpc(efx, cmd, inbuf, inlen, outbuf, outlen,
+                            outlen_actual, false);
+}
+
+/* Normally, on receiving an error code in the MCDI response,
+ * efx_mcdi_rpc will log an error message containing (among other
+ * things) the raw error code, by means of efx_mcdi_display_error.
+ * This _quiet version suppresses that; if the caller wishes to log
+ * the error conditionally on the return code, it should call this
+ * function and is then responsible for calling efx_mcdi_display_error
+ * as needed.
+ */
+int efx_mcdi_rpc_quiet(struct efx_nic *efx, unsigned cmd,
+                      const efx_dword_t *inbuf, size_t inlen,
+                      efx_dword_t *outbuf, size_t outlen,
+                      size_t *outlen_actual)
+{
+       return _efx_mcdi_rpc(efx, cmd, inbuf, inlen, outbuf, outlen,
+                            outlen_actual, true);
+}
+
+int efx_mcdi_rpc_start(struct efx_nic *efx, unsigned cmd,
+                      const efx_dword_t *inbuf, size_t inlen)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       int rc;
+
+       rc = efx_mcdi_check_supported(efx, cmd, inlen);
+       if (rc)
+               return rc;
+
+       if (efx->mc_bist_for_other_fn)
+               return -ENETDOWN;
+
+       if (mcdi->mode == MCDI_MODE_FAIL)
+               return -ENETDOWN;
+
+       efx_mcdi_acquire_sync(mcdi);
+       efx_mcdi_send_request(efx, cmd, inbuf, inlen);
+       return 0;
+}
+
+static int _efx_mcdi_rpc_async(struct efx_nic *efx, unsigned int cmd,
+                              const efx_dword_t *inbuf, size_t inlen,
+                              size_t outlen,
+                              efx_mcdi_async_completer *complete,
+                              unsigned long cookie, bool quiet)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+       struct efx_mcdi_async_param *async;
+       int rc;
+
+       rc = efx_mcdi_check_supported(efx, cmd, inlen);
+       if (rc)
+               return rc;
+
+       if (efx->mc_bist_for_other_fn)
+               return -ENETDOWN;
+
+       async = kmalloc(sizeof(*async) + ALIGN(max(inlen, outlen), 4),
+                       GFP_ATOMIC);
+       if (!async)
+               return -ENOMEM;
+
+       async->cmd = cmd;
+       async->inlen = inlen;
+       async->outlen = outlen;
+       async->quiet = quiet;
+       async->complete = complete;
+       async->cookie = cookie;
+       memcpy(async + 1, inbuf, inlen);
+
+       spin_lock_bh(&mcdi->async_lock);
+
+       if (mcdi->mode == MCDI_MODE_EVENTS) {
+               list_add_tail(&async->list, &mcdi->async_list);
+
+               /* If this is at the front of the queue, try to start it
+                * immediately
+                */
+               if (mcdi->async_list.next == &async->list &&
+                   efx_mcdi_acquire_async(mcdi)) {
+                       efx_mcdi_send_request(efx, cmd, inbuf, inlen);
+                       mod_timer(&mcdi->async_timer,
+                                 jiffies + MCDI_RPC_TIMEOUT);
+               }
+       } else {
+               kfree(async);
+               rc = -ENETDOWN;
+       }
+
+       spin_unlock_bh(&mcdi->async_lock);
+
+       return rc;
+}
+
+/**
+ * efx_mcdi_rpc_async - Schedule an MCDI command to run asynchronously
+ * @efx: NIC through which to issue the command
+ * @cmd: Command type number
+ * @inbuf: Command parameters
+ * @inlen: Length of command parameters, in bytes
+ * @outlen: Length to allocate for response buffer, in bytes
+ * @complete: Function to be called on completion or cancellation.
+ * @cookie: Arbitrary value to be passed to @complete.
+ *
+ * This function does not sleep and therefore may be called in atomic
+ * context.  It will fail if event queues are disabled or if MCDI
+ * event completions have been disabled due to an error.
+ *
+ * If it succeeds, the @complete function will be called exactly once
+ * in atomic context, when one of the following occurs:
+ * (a) the completion event is received (in NAPI context)
+ * (b) event queues are disabled (in the process that disables them)
+ * (c) the request times-out (in timer context)
+ */
+int
+efx_mcdi_rpc_async(struct efx_nic *efx, unsigned int cmd,
+                  const efx_dword_t *inbuf, size_t inlen, size_t outlen,
+                  efx_mcdi_async_completer *complete, unsigned long cookie)
+{
+       return _efx_mcdi_rpc_async(efx, cmd, inbuf, inlen, outlen, complete,
+                                  cookie, false);
+}
+
+int efx_mcdi_rpc_async_quiet(struct efx_nic *efx, unsigned int cmd,
+                            const efx_dword_t *inbuf, size_t inlen,
+                            size_t outlen, efx_mcdi_async_completer *complete,
+                            unsigned long cookie)
+{
+       return _efx_mcdi_rpc_async(efx, cmd, inbuf, inlen, outlen, complete,
+                                  cookie, true);
+}
+
+int efx_mcdi_rpc_finish(struct efx_nic *efx, unsigned cmd, size_t inlen,
+                       efx_dword_t *outbuf, size_t outlen,
+                       size_t *outlen_actual)
+{
+       return _efx_mcdi_rpc_finish(efx, cmd, inlen, outbuf, outlen,
+                                   outlen_actual, false);
+}
+
+int efx_mcdi_rpc_finish_quiet(struct efx_nic *efx, unsigned cmd, size_t inlen,
+                             efx_dword_t *outbuf, size_t outlen,
+                             size_t *outlen_actual)
+{
+       return _efx_mcdi_rpc_finish(efx, cmd, inlen, outbuf, outlen,
+                                   outlen_actual, true);
+}
+
+void efx_mcdi_display_error(struct efx_nic *efx, unsigned cmd,
+                           size_t inlen, efx_dword_t *outbuf,
+                           size_t outlen, int rc)
+{
+       int code = 0, err_arg = 0;
+
+       if (outlen >= MC_CMD_ERR_CODE_OFST + 4)
+               code = MCDI_DWORD(outbuf, ERR_CODE);
+       if (outlen >= MC_CMD_ERR_ARG_OFST + 4)
+               err_arg = MCDI_DWORD(outbuf, ERR_ARG);
+       netif_err(efx, hw, efx->net_dev,
+                 "MC command 0x%x inlen %d failed rc=%d (raw=%d) arg=%d\n",
+                 cmd, (int)inlen, rc, code, err_arg);
+}
+
+/* Switch to polled MCDI completions.  This can be called in various
+ * error conditions with various locks held, so it must be lockless.
+ * Caller is responsible for flushing asynchronous requests later.
+ */
+void efx_mcdi_mode_poll(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi;
+
+       if (!efx->mcdi)
+               return;
+
+       mcdi = efx_mcdi(efx);
+       /* If already in polling mode, nothing to do.
+        * If in fail-fast state, don't switch to polled completion.
+        * FLR recovery will do that later.
+        */
+       if (mcdi->mode == MCDI_MODE_POLL || mcdi->mode == MCDI_MODE_FAIL)
+               return;
+
+       /* We can switch from event completion to polled completion, because
+        * mcdi requests are always completed in shared memory. We do this by
+        * switching the mode to POLL'd then completing the request.
+        * efx_mcdi_await_completion() will then call efx_mcdi_poll().
+        *
+        * We need an smp_wmb() to synchronise with efx_mcdi_await_completion(),
+        * which efx_mcdi_complete_sync() provides for us.
+        */
+       mcdi->mode = MCDI_MODE_POLL;
+
+       efx_mcdi_complete_sync(mcdi);
+}
+
+/* Flush any running or queued asynchronous requests, after event processing
+ * is stopped
+ */
+void efx_mcdi_flush_async(struct efx_nic *efx)
+{
+       struct efx_mcdi_async_param *async, *next;
+       struct efx_mcdi_iface *mcdi;
+
+       if (!efx->mcdi)
+               return;
+
+       mcdi = efx_mcdi(efx);
+
+       /* We must be in poll or fail mode so no more requests can be queued */
+       BUG_ON(mcdi->mode == MCDI_MODE_EVENTS);
+
+       del_timer_sync(&mcdi->async_timer);
+
+       /* If a request is still running, make sure we give the MC
+        * time to complete it so that the response won't overwrite our
+        * next request.
+        */
+       if (mcdi->state == MCDI_STATE_RUNNING_ASYNC) {
+               efx_mcdi_poll(efx);
+               mcdi->state = MCDI_STATE_QUIESCENT;
+       }
+
+       /* Nothing else will access the async list now, so it is safe
+        * to walk it without holding async_lock.  If we hold it while
+        * calling a completer then lockdep may warn that we have
+        * acquired locks in the wrong order.
+        */
+       list_for_each_entry_safe(async, next, &mcdi->async_list, list) {
+               async->complete(efx, async->cookie, -ENETDOWN, NULL, 0);
+               list_del(&async->list);
+               kfree(async);
+       }
+}
+
+void efx_mcdi_mode_event(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi;
+
+       if (!efx->mcdi)
+               return;
+
+       mcdi = efx_mcdi(efx);
+       /* If already in event completion mode, nothing to do.
+        * If in fail-fast state, don't switch to event completion.  FLR
+        * recovery will do that later.
+        */
+       if (mcdi->mode == MCDI_MODE_EVENTS || mcdi->mode == MCDI_MODE_FAIL)
+               return;
+
+       /* We can't switch from polled to event completion in the middle of a
+        * request, because the completion method is specified in the request.
+        * So acquire the interface to serialise the requestors. We don't need
+        * to acquire the iface_lock to change the mode here, but we do need a
+        * write memory barrier ensure that efx_mcdi_rpc() sees it, which
+        * efx_mcdi_acquire() provides.
+        */
+       efx_mcdi_acquire_sync(mcdi);
+       mcdi->mode = MCDI_MODE_EVENTS;
+       efx_mcdi_release(mcdi);
+}
+
+static void efx_mcdi_ev_death(struct efx_nic *efx, int rc)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+
+       /* If there is an outstanding MCDI request, it has been terminated
+        * either by a BADASSERT or REBOOT event. If the mcdi interface is
+        * in polled mode, then do nothing because the MC reboot handler will
+        * set the header correctly. However, if the mcdi interface is waiting
+        * for a CMDDONE event it won't receive it [and since all MCDI events
+        * are sent to the same queue, we can't be racing with
+        * efx_mcdi_ev_cpl()]
+        *
+        * If there is an outstanding asynchronous request, we can't
+        * complete it now (efx_mcdi_complete() would deadlock).  The
+        * reset process will take care of this.
+        *
+        * There's a race here with efx_mcdi_send_request(), because
+        * we might receive a REBOOT event *before* the request has
+        * been copied out. In polled mode (during startup) this is
+        * irrelevant, because efx_mcdi_complete_sync() is ignored. In
+        * event mode, this condition is just an edge-case of
+        * receiving a REBOOT event after posting the MCDI
+        * request. Did the mc reboot before or after the copyout? The
+        * best we can do always is just return failure.
+        */
+       spin_lock(&mcdi->iface_lock);
+       if (efx_mcdi_complete_sync(mcdi)) {
+               if (mcdi->mode == MCDI_MODE_EVENTS) {
+                       mcdi->resprc = rc;
+                       mcdi->resp_hdr_len = 0;
+                       mcdi->resp_data_len = 0;
+                       ++mcdi->credits;
+               }
+       } else {
+               int count;
+
+               /* Consume the status word since efx_mcdi_rpc_finish() won't */
+               for (count = 0; count < MCDI_STATUS_DELAY_COUNT; ++count) {
+                       if (efx_mcdi_poll_reboot(efx))
+                               break;
+                       udelay(MCDI_STATUS_DELAY_US);
+               }
+               mcdi->new_epoch = true;
+
+               /* Nobody was waiting for an MCDI request, so trigger a reset */
+               efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
+       }
+
+       spin_unlock(&mcdi->iface_lock);
+}
+
+/* The MC is going down in to BIST mode. set the BIST flag to block
+ * new MCDI, cancel any outstanding MCDI and and schedule a BIST-type reset
+ * (which doesn't actually execute a reset, it waits for the controlling
+ * function to reset it).
+ */
+static void efx_mcdi_ev_bist(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+
+       spin_lock(&mcdi->iface_lock);
+       efx->mc_bist_for_other_fn = true;
+       if (efx_mcdi_complete_sync(mcdi)) {
+               if (mcdi->mode == MCDI_MODE_EVENTS) {
+                       mcdi->resprc = -EIO;
+                       mcdi->resp_hdr_len = 0;
+                       mcdi->resp_data_len = 0;
+                       ++mcdi->credits;
+               }
+       }
+       mcdi->new_epoch = true;
+       efx_schedule_reset(efx, RESET_TYPE_MC_BIST);
+       spin_unlock(&mcdi->iface_lock);
+}
+
+/* MCDI timeouts seen, so make all MCDI calls fail-fast and issue an FLR to try
+ * to recover.
+ */
+static void efx_mcdi_abandon(struct efx_nic *efx)
+{
+       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+
+       if (xchg(&mcdi->mode, MCDI_MODE_FAIL) == MCDI_MODE_FAIL)
+               return; /* it had already been done */
+       netif_dbg(efx, hw, efx->net_dev, "MCDI is timing out; trying to recover\n");
+       efx_schedule_reset(efx, RESET_TYPE_MCDI_TIMEOUT);
+}
+
+/* Called from  falcon_process_eventq for MCDI events */
+void efx_mcdi_process_event(struct efx_channel *channel,
+                           efx_qword_t *event)
+{
+       struct efx_nic *efx = channel->efx;
+       int code = EFX_QWORD_FIELD(*event, MCDI_EVENT_CODE);
+       u32 data = EFX_QWORD_FIELD(*event, MCDI_EVENT_DATA);
+
+       switch (code) {
+       case MCDI_EVENT_CODE_BADSSERT:
+               netif_err(efx, hw, efx->net_dev,
+                         "MC watchdog or assertion failure at 0x%x\n", data);
+               efx_mcdi_ev_death(efx, -EINTR);
+               break;
+
+       case MCDI_EVENT_CODE_PMNOTICE:
+               netif_info(efx, wol, efx->net_dev, "MCDI PM event.\n");
+               break;
+
+       case MCDI_EVENT_CODE_CMDDONE:
+               efx_mcdi_ev_cpl(efx,
+                               MCDI_EVENT_FIELD(*event, CMDDONE_SEQ),
+                               MCDI_EVENT_FIELD(*event, CMDDONE_DATALEN),
+                               MCDI_EVENT_FIELD(*event, CMDDONE_ERRNO));
+               break;
+
+       case MCDI_EVENT_CODE_LINKCHANGE:
+               efx_mcdi_process_link_change(efx, event);
+               break;
+       case MCDI_EVENT_CODE_SENSOREVT:
+               efx_mcdi_sensor_event(efx, event);
+               break;
+       case MCDI_EVENT_CODE_SCHEDERR:
+               netif_dbg(efx, hw, efx->net_dev,
+                         "MC Scheduler alert (0x%x)\n", data);
+               break;
+       case MCDI_EVENT_CODE_REBOOT:
+       case MCDI_EVENT_CODE_MC_REBOOT:
+               netif_info(efx, hw, efx->net_dev, "MC Reboot\n");
+               efx_mcdi_ev_death(efx, -EIO);
+               break;
+       case MCDI_EVENT_CODE_MC_BIST:
+               netif_info(efx, hw, efx->net_dev, "MC entered BIST mode\n");
+               efx_mcdi_ev_bist(efx);
+               break;
+       case MCDI_EVENT_CODE_MAC_STATS_DMA:
+               /* MAC stats are gather lazily.  We can ignore this. */
+               break;
+       case MCDI_EVENT_CODE_FLR:
+               efx_siena_sriov_flr(efx, MCDI_EVENT_FIELD(*event, FLR_VF));
+               break;
+       case MCDI_EVENT_CODE_PTP_RX:
+       case MCDI_EVENT_CODE_PTP_FAULT:
+       case MCDI_EVENT_CODE_PTP_PPS:
+               efx_ptp_event(efx, event);
+               break;
+       case MCDI_EVENT_CODE_PTP_TIME:
+               efx_time_sync_event(channel, event);
+               break;
+       case MCDI_EVENT_CODE_TX_FLUSH:
+       case MCDI_EVENT_CODE_RX_FLUSH:
+               /* Two flush events will be sent: one to the same event
+                * queue as completions, and one to event queue 0.
+                * In the latter case the {RX,TX}_FLUSH_TO_DRIVER
+                * flag will be set, and we should ignore the event
+                * because we want to wait for all completions.
+                */
+               BUILD_BUG_ON(MCDI_EVENT_TX_FLUSH_TO_DRIVER_LBN !=
+                            MCDI_EVENT_RX_FLUSH_TO_DRIVER_LBN);
+               if (!MCDI_EVENT_FIELD(*event, TX_FLUSH_TO_DRIVER))
+                       efx_ef10_handle_drain_event(efx);
+               break;
+       case MCDI_EVENT_CODE_TX_ERR:
+       case MCDI_EVENT_CODE_RX_ERR:
+               netif_err(efx, hw, efx->net_dev,
+                         "%s DMA error (event: "EFX_QWORD_FMT")\n",
+                         code == MCDI_EVENT_CODE_TX_ERR ? "TX" : "RX",
+                         EFX_QWORD_VAL(*event));
+               efx_schedule_reset(efx, RESET_TYPE_DMA_ERROR);
+               break;
+       default:
+               netif_err(efx, hw, efx->net_dev, "Unknown MCDI event 0x%x\n",
+                         code);
+       }
+}
+
+/**************************************************************************
+ *
+ * Specific request functions
+ *
+ **************************************************************************
+ */
+
+void efx_mcdi_print_fwver(struct efx_nic *efx, char *buf, size_t len)
+{
+       MCDI_DECLARE_BUF(outbuf,
+                        max(MC_CMD_GET_VERSION_OUT_LEN,
+                            MC_CMD_GET_CAPABILITIES_OUT_LEN));
+       size_t outlength;
+       const __le16 *ver_words;
+       size_t offset;
+       int rc;
+
+       BUILD_BUG_ON(MC_CMD_GET_VERSION_IN_LEN != 0);
+       rc = efx_mcdi_rpc(efx, MC_CMD_GET_VERSION, NULL, 0,
+                         outbuf, sizeof(outbuf), &outlength);
+       if (rc)
+               goto fail;
+       if (outlength < MC_CMD_GET_VERSION_OUT_LEN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       ver_words = (__le16 *)MCDI_PTR(outbuf, GET_VERSION_OUT_VERSION);
+       offset = snprintf(buf, len, "%u.%u.%u.%u",
+                         le16_to_cpu(ver_words[0]), le16_to_cpu(ver_words[1]),
+                         le16_to_cpu(ver_words[2]), le16_to_cpu(ver_words[3]));
+
+       /* EF10 may have multiple datapath firmware variants within a
+        * single version.  Report which variants are running.
+        */
+       if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) {
+               BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
+               rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
+                                 outbuf, sizeof(outbuf), &outlength);
+               if (rc || outlength < MC_CMD_GET_CAPABILITIES_OUT_LEN)
+                       offset += snprintf(
+                               buf + offset, len - offset, " rx? tx?");
+               else
+                       offset += snprintf(
+                               buf + offset, len - offset, " rx%x tx%x",
+                               MCDI_WORD(outbuf,
+                                         GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID),
+                               MCDI_WORD(outbuf,
+                                         GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID));
+
+               /* It's theoretically possible for the string to exceed 31
+                * characters, though in practice the first three version
+                * components are short enough that this doesn't happen.
+                */
+               if (WARN_ON(offset >= len))
+                       buf[0] = 0;
+       }
+
+       return;
+
+fail:
+       netif_err(efx, probe, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+       buf[0] = 0;
+}
+
+static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
+                              bool *was_attached)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_DRV_ATTACH_IN_LEN);
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_DRV_ATTACH_EXT_OUT_LEN);
+       size_t outlen;
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_NEW_STATE,
+                      driver_operating ? 1 : 0);
+       MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_UPDATE, 1);
+       MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_FIRMWARE_ID, MC_CMD_FW_LOW_LATENCY);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_DRV_ATTACH, inbuf, sizeof(inbuf),
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               goto fail;
+       if (outlen < MC_CMD_DRV_ATTACH_OUT_LEN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       if (driver_operating) {
+               if (outlen >= MC_CMD_DRV_ATTACH_EXT_OUT_LEN) {
+                       efx->mcdi->fn_flags =
+                               MCDI_DWORD(outbuf,
+                                          DRV_ATTACH_EXT_OUT_FUNC_FLAGS);
+               } else {
+                       /* Synthesise flags for Siena */
+                       efx->mcdi->fn_flags =
+                               1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL |
+                               1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED |
+                               (efx_port_num(efx) == 0) <<
+                               MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY;
+               }
+       }
+
+       /* We currently assume we have control of the external link
+        * and are completely trusted by firmware.  Abort probing
+        * if that's not true for this function.
+        */
+       if (driver_operating &&
+           (efx->mcdi->fn_flags &
+            (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL |
+             1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED)) !=
+           (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL |
+            1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED)) {
+               netif_err(efx, probe, efx->net_dev,
+                         "This driver version only supports one function per port\n");
+               return -ENODEV;
+       }
+
+       if (was_attached != NULL)
+               *was_attached = MCDI_DWORD(outbuf, DRV_ATTACH_OUT_OLD_STATE);
+       return 0;
+
+fail:
+       netif_err(efx, probe, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+       return rc;
+}
+
+int efx_mcdi_get_board_cfg(struct efx_nic *efx, u8 *mac_address,
+                          u16 *fw_subtype_list, u32 *capabilities)
+{
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_BOARD_CFG_OUT_LENMAX);
+       size_t outlen, i;
+       int port_num = efx_port_num(efx);
+       int rc;
+
+       BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_IN_LEN != 0);
+       /* we need __aligned(2) for ether_addr_copy */
+       BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT0_OFST & 1);
+       BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT1_OFST & 1);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_GET_BOARD_CFG, NULL, 0,
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               goto fail;
+
+       if (outlen < MC_CMD_GET_BOARD_CFG_OUT_LENMIN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       if (mac_address)
+               ether_addr_copy(mac_address,
+                               port_num ?
+                               MCDI_PTR(outbuf, GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT1) :
+                               MCDI_PTR(outbuf, GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT0));
+       if (fw_subtype_list) {
+               for (i = 0;
+                    i < MCDI_VAR_ARRAY_LEN(outlen,
+                                           GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST);
+                    i++)
+                       fw_subtype_list[i] = MCDI_ARRAY_WORD(
+                               outbuf, GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST, i);
+               for (; i < MC_CMD_GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST_MAXNUM; i++)
+                       fw_subtype_list[i] = 0;
+       }
+       if (capabilities) {
+               if (port_num)
+                       *capabilities = MCDI_DWORD(outbuf,
+                                       GET_BOARD_CFG_OUT_CAPABILITIES_PORT1);
+               else
+                       *capabilities = MCDI_DWORD(outbuf,
+                                       GET_BOARD_CFG_OUT_CAPABILITIES_PORT0);
+       }
+
+       return 0;
+
+fail:
+       netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d len=%d\n",
+                 __func__, rc, (int)outlen);
+
+       return rc;
+}
+
+int efx_mcdi_log_ctrl(struct efx_nic *efx, bool evq, bool uart, u32 dest_evq)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_LOG_CTRL_IN_LEN);
+       u32 dest = 0;
+       int rc;
+
+       if (uart)
+               dest |= MC_CMD_LOG_CTRL_IN_LOG_DEST_UART;
+       if (evq)
+               dest |= MC_CMD_LOG_CTRL_IN_LOG_DEST_EVQ;
+
+       MCDI_SET_DWORD(inbuf, LOG_CTRL_IN_LOG_DEST, dest);
+       MCDI_SET_DWORD(inbuf, LOG_CTRL_IN_LOG_DEST_EVQ, dest_evq);
+
+       BUILD_BUG_ON(MC_CMD_LOG_CTRL_OUT_LEN != 0);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_LOG_CTRL, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+int efx_mcdi_nvram_types(struct efx_nic *efx, u32 *nvram_types_out)
+{
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_TYPES_OUT_LEN);
+       size_t outlen;
+       int rc;
+
+       BUILD_BUG_ON(MC_CMD_NVRAM_TYPES_IN_LEN != 0);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_TYPES, NULL, 0,
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               goto fail;
+       if (outlen < MC_CMD_NVRAM_TYPES_OUT_LEN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       *nvram_types_out = MCDI_DWORD(outbuf, NVRAM_TYPES_OUT_TYPES);
+       return 0;
+
+fail:
+       netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n",
+                 __func__, rc);
+       return rc;
+}
+
+int efx_mcdi_nvram_info(struct efx_nic *efx, unsigned int type,
+                       size_t *size_out, size_t *erase_size_out,
+                       bool *protected_out)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_INFO_IN_LEN);
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_INFO_OUT_LEN);
+       size_t outlen;
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_INFO_IN_TYPE, type);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_INFO, inbuf, sizeof(inbuf),
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               goto fail;
+       if (outlen < MC_CMD_NVRAM_INFO_OUT_LEN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       *size_out = MCDI_DWORD(outbuf, NVRAM_INFO_OUT_SIZE);
+       *erase_size_out = MCDI_DWORD(outbuf, NVRAM_INFO_OUT_ERASESIZE);
+       *protected_out = !!(MCDI_DWORD(outbuf, NVRAM_INFO_OUT_FLAGS) &
+                               (1 << MC_CMD_NVRAM_INFO_OUT_PROTECTED_LBN));
+       return 0;
+
+fail:
+       netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+       return rc;
+}
+
+static int efx_mcdi_nvram_test(struct efx_nic *efx, unsigned int type)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_TEST_IN_LEN);
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_TEST_OUT_LEN);
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_TEST_IN_TYPE, type);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_TEST, inbuf, sizeof(inbuf),
+                         outbuf, sizeof(outbuf), NULL);
+       if (rc)
+               return rc;
+
+       switch (MCDI_DWORD(outbuf, NVRAM_TEST_OUT_RESULT)) {
+       case MC_CMD_NVRAM_TEST_PASS:
+       case MC_CMD_NVRAM_TEST_NOTSUPP:
+               return 0;
+       default:
+               return -EIO;
+       }
+}
+
+int efx_mcdi_nvram_test_all(struct efx_nic *efx)
+{
+       u32 nvram_types;
+       unsigned int type;
+       int rc;
+
+       rc = efx_mcdi_nvram_types(efx, &nvram_types);
+       if (rc)
+               goto fail1;
+
+       type = 0;
+       while (nvram_types != 0) {
+               if (nvram_types & 1) {
+                       rc = efx_mcdi_nvram_test(efx, type);
+                       if (rc)
+                               goto fail2;
+               }
+               type++;
+               nvram_types >>= 1;
+       }
+
+       return 0;
+
+fail2:
+       netif_err(efx, hw, efx->net_dev, "%s: failed type=%u\n",
+                 __func__, type);
+fail1:
+       netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+       return rc;
+}
+
+static int efx_mcdi_read_assertion(struct efx_nic *efx)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_ASSERTS_IN_LEN);
+       MCDI_DECLARE_BUF_OUT_OR_ERR(outbuf, MC_CMD_GET_ASSERTS_OUT_LEN);
+       unsigned int flags, index;
+       const char *reason;
+       size_t outlen;
+       int retry;
+       int rc;
+
+       /* Attempt to read any stored assertion state before we reboot
+        * the mcfw out of the assertion handler. Retry twice, once
+        * because a boot-time assertion might cause this command to fail
+        * with EINTR. And once again because GET_ASSERTS can race with
+        * MC_CMD_REBOOT running on the other port. */
+       retry = 2;
+       do {
+               MCDI_SET_DWORD(inbuf, GET_ASSERTS_IN_CLEAR, 1);
+               rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_ASSERTS,
+                                       inbuf, MC_CMD_GET_ASSERTS_IN_LEN,
+                                       outbuf, sizeof(outbuf), &outlen);
+       } while ((rc == -EINTR || rc == -EIO) && retry-- > 0);
+
+       if (rc) {
+               efx_mcdi_display_error(efx, MC_CMD_GET_ASSERTS,
+                                      MC_CMD_GET_ASSERTS_IN_LEN, outbuf,
+                                      outlen, rc);
+               return rc;
+       }
+       if (outlen < MC_CMD_GET_ASSERTS_OUT_LEN)
+               return -EIO;
+
+       /* Print out any recorded assertion state */
+       flags = MCDI_DWORD(outbuf, GET_ASSERTS_OUT_GLOBAL_FLAGS);
+       if (flags == MC_CMD_GET_ASSERTS_FLAGS_NO_FAILS)
+               return 0;
+
+       reason = (flags == MC_CMD_GET_ASSERTS_FLAGS_SYS_FAIL)
+               ? "system-level assertion"
+               : (flags == MC_CMD_GET_ASSERTS_FLAGS_THR_FAIL)
+               ? "thread-level assertion"
+               : (flags == MC_CMD_GET_ASSERTS_FLAGS_WDOG_FIRED)
+               ? "watchdog reset"
+               : "unknown assertion";
+       netif_err(efx, hw, efx->net_dev,
+                 "MCPU %s at PC = 0x%.8x in thread 0x%.8x\n", reason,
+                 MCDI_DWORD(outbuf, GET_ASSERTS_OUT_SAVED_PC_OFFS),
+                 MCDI_DWORD(outbuf, GET_ASSERTS_OUT_THREAD_OFFS));
+
+       /* Print out the registers */
+       for (index = 0;
+            index < MC_CMD_GET_ASSERTS_OUT_GP_REGS_OFFS_NUM;
+            index++)
+               netif_err(efx, hw, efx->net_dev, "R%.2d (?): 0x%.8x\n",
+                         1 + index,
+                         MCDI_ARRAY_DWORD(outbuf, GET_ASSERTS_OUT_GP_REGS_OFFS,
+                                          index));
+
+       return 0;
+}
+
+static void efx_mcdi_exit_assertion(struct efx_nic *efx)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_REBOOT_IN_LEN);
+
+       /* If the MC is running debug firmware, it might now be
+        * waiting for a debugger to attach, but we just want it to
+        * reboot.  We set a flag that makes the command a no-op if it
+        * has already done so.  We don't know what return code to
+        * expect (0 or -EIO), so ignore it.
+        */
+       BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0);
+       MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS,
+                      MC_CMD_REBOOT_FLAGS_AFTER_ASSERTION);
+       (void) efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf, MC_CMD_REBOOT_IN_LEN,
+                           NULL, 0, NULL);
+}
+
+int efx_mcdi_handle_assertion(struct efx_nic *efx)
+{
+       int rc;
+
+       rc = efx_mcdi_read_assertion(efx);
+       if (rc)
+               return rc;
+
+       efx_mcdi_exit_assertion(efx);
+
+       return 0;
+}
+
+void efx_mcdi_set_id_led(struct efx_nic *efx, enum efx_led_mode mode)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_ID_LED_IN_LEN);
+       int rc;
+
+       BUILD_BUG_ON(EFX_LED_OFF != MC_CMD_LED_OFF);
+       BUILD_BUG_ON(EFX_LED_ON != MC_CMD_LED_ON);
+       BUILD_BUG_ON(EFX_LED_DEFAULT != MC_CMD_LED_DEFAULT);
+
+       BUILD_BUG_ON(MC_CMD_SET_ID_LED_OUT_LEN != 0);
+
+       MCDI_SET_DWORD(inbuf, SET_ID_LED_IN_STATE, mode);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_SET_ID_LED, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+}
+
+static int efx_mcdi_reset_func(struct efx_nic *efx)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_ENTITY_RESET_IN_LEN);
+       int rc;
+
+       BUILD_BUG_ON(MC_CMD_ENTITY_RESET_OUT_LEN != 0);
+       MCDI_POPULATE_DWORD_1(inbuf, ENTITY_RESET_IN_FLAG,
+                             ENTITY_RESET_IN_FUNCTION_RESOURCE_RESET, 1);
+       rc = efx_mcdi_rpc(efx, MC_CMD_ENTITY_RESET, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+static int efx_mcdi_reset_mc(struct efx_nic *efx)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_REBOOT_IN_LEN);
+       int rc;
+
+       BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0);
+       MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS, 0);
+       rc = efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       /* White is black, and up is down */
+       if (rc == -EIO)
+               return 0;
+       if (rc == 0)
+               rc = -EIO;
+       return rc;
+}
+
+enum reset_type efx_mcdi_map_reset_reason(enum reset_type reason)
+{
+       return RESET_TYPE_RECOVER_OR_ALL;
+}
+
+int efx_mcdi_reset(struct efx_nic *efx, enum reset_type method)
+{
+       int rc;
+
+       /* If MCDI is down, we can't handle_assertion */
+       if (method == RESET_TYPE_MCDI_TIMEOUT) {
+               rc = pci_reset_function(efx->pci_dev);
+               if (rc)
+                       return rc;
+               /* Re-enable polled MCDI completion */
+               if (efx->mcdi) {
+                       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+                       mcdi->mode = MCDI_MODE_POLL;
+               }
+               return 0;
+       }
+
+       /* Recover from a failed assertion pre-reset */
+       rc = efx_mcdi_handle_assertion(efx);
+       if (rc)
+               return rc;
+
+       if (method == RESET_TYPE_WORLD)
+               return efx_mcdi_reset_mc(efx);
+       else
+               return efx_mcdi_reset_func(efx);
+}
+
+static int efx_mcdi_wol_filter_set(struct efx_nic *efx, u32 type,
+                                  const u8 *mac, int *id_out)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_WOL_FILTER_SET_IN_LEN);
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_WOL_FILTER_SET_OUT_LEN);
+       size_t outlen;
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, WOL_FILTER_SET_IN_WOL_TYPE, type);
+       MCDI_SET_DWORD(inbuf, WOL_FILTER_SET_IN_FILTER_MODE,
+                      MC_CMD_FILTER_MODE_SIMPLE);
+       ether_addr_copy(MCDI_PTR(inbuf, WOL_FILTER_SET_IN_MAGIC_MAC), mac);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_SET, inbuf, sizeof(inbuf),
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               goto fail;
+
+       if (outlen < MC_CMD_WOL_FILTER_SET_OUT_LEN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       *id_out = (int)MCDI_DWORD(outbuf, WOL_FILTER_SET_OUT_FILTER_ID);
+
+       return 0;
+
+fail:
+       *id_out = -1;
+       netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+       return rc;
+
+}
+
+
+int
+efx_mcdi_wol_filter_set_magic(struct efx_nic *efx,  const u8 *mac, int *id_out)
+{
+       return efx_mcdi_wol_filter_set(efx, MC_CMD_WOL_TYPE_MAGIC, mac, id_out);
+}
+
+
+int efx_mcdi_wol_filter_get_magic(struct efx_nic *efx, int *id_out)
+{
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_WOL_FILTER_GET_OUT_LEN);
+       size_t outlen;
+       int rc;
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_GET, NULL, 0,
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               goto fail;
+
+       if (outlen < MC_CMD_WOL_FILTER_GET_OUT_LEN) {
+               rc = -EIO;
+               goto fail;
+       }
+
+       *id_out = (int)MCDI_DWORD(outbuf, WOL_FILTER_GET_OUT_FILTER_ID);
+
+       return 0;
+
+fail:
+       *id_out = -1;
+       netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
+       return rc;
+}
+
+
+int efx_mcdi_wol_filter_remove(struct efx_nic *efx, int id)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_WOL_FILTER_REMOVE_IN_LEN);
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, WOL_FILTER_REMOVE_IN_FILTER_ID, (u32)id);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_REMOVE, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+int efx_mcdi_flush_rxqs(struct efx_nic *efx)
+{
+       struct efx_channel *channel;
+       struct efx_rx_queue *rx_queue;
+       MCDI_DECLARE_BUF(inbuf,
+                        MC_CMD_FLUSH_RX_QUEUES_IN_LEN(EFX_MAX_CHANNELS));
+       int rc, count;
+
+       BUILD_BUG_ON(EFX_MAX_CHANNELS >
+                    MC_CMD_FLUSH_RX_QUEUES_IN_QID_OFST_MAXNUM);
+
+       count = 0;
+       efx_for_each_channel(channel, efx) {
+               efx_for_each_channel_rx_queue(rx_queue, channel) {
+                       if (rx_queue->flush_pending) {
+                               rx_queue->flush_pending = false;
+                               atomic_dec(&efx->rxq_flush_pending);
+                               MCDI_SET_ARRAY_DWORD(
+                                       inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
+                                       count, efx_rx_queue_index(rx_queue));
+                               count++;
+                       }
+               }
+       }
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_FLUSH_RX_QUEUES, inbuf,
+                         MC_CMD_FLUSH_RX_QUEUES_IN_LEN(count), NULL, 0, NULL);
+       WARN_ON(rc < 0);
+
+       return rc;
+}
+
+int efx_mcdi_wol_filter_reset(struct efx_nic *efx)
+{
+       int rc;
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_RESET, NULL, 0, NULL, 0, NULL);
+       return rc;
+}
+
+int efx_mcdi_set_workaround(struct efx_nic *efx, u32 type, bool enabled)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_WORKAROUND_IN_LEN);
+
+       BUILD_BUG_ON(MC_CMD_WORKAROUND_OUT_LEN != 0);
+       MCDI_SET_DWORD(inbuf, WORKAROUND_IN_TYPE, type);
+       MCDI_SET_DWORD(inbuf, WORKAROUND_IN_ENABLED, enabled);
+       return efx_mcdi_rpc(efx, MC_CMD_WORKAROUND, inbuf, sizeof(inbuf),
+                           NULL, 0, NULL);
+}
+
+#ifdef CONFIG_SFC_MTD
+
+#define EFX_MCDI_NVRAM_LEN_MAX 128
+
+static int efx_mcdi_nvram_update_start(struct efx_nic *efx, unsigned int type)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_UPDATE_START_IN_LEN);
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_UPDATE_START_IN_TYPE, type);
+
+       BUILD_BUG_ON(MC_CMD_NVRAM_UPDATE_START_OUT_LEN != 0);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_UPDATE_START, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+static int efx_mcdi_nvram_read(struct efx_nic *efx, unsigned int type,
+                              loff_t offset, u8 *buffer, size_t length)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_READ_IN_LEN);
+       MCDI_DECLARE_BUF(outbuf,
+                        MC_CMD_NVRAM_READ_OUT_LEN(EFX_MCDI_NVRAM_LEN_MAX));
+       size_t outlen;
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_TYPE, type);
+       MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_OFFSET, offset);
+       MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_LENGTH, length);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_READ, inbuf, sizeof(inbuf),
+                         outbuf, sizeof(outbuf), &outlen);
+       if (rc)
+               return rc;
+
+       memcpy(buffer, MCDI_PTR(outbuf, NVRAM_READ_OUT_READ_BUFFER), length);
+       return 0;
+}
+
+static int efx_mcdi_nvram_write(struct efx_nic *efx, unsigned int type,
+                               loff_t offset, const u8 *buffer, size_t length)
+{
+       MCDI_DECLARE_BUF(inbuf,
+                        MC_CMD_NVRAM_WRITE_IN_LEN(EFX_MCDI_NVRAM_LEN_MAX));
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_TYPE, type);
+       MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_OFFSET, offset);
+       MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_LENGTH, length);
+       memcpy(MCDI_PTR(inbuf, NVRAM_WRITE_IN_WRITE_BUFFER), buffer, length);
+
+       BUILD_BUG_ON(MC_CMD_NVRAM_WRITE_OUT_LEN != 0);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_WRITE, inbuf,
+                         ALIGN(MC_CMD_NVRAM_WRITE_IN_LEN(length), 4),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+static int efx_mcdi_nvram_erase(struct efx_nic *efx, unsigned int type,
+                               loff_t offset, size_t length)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_ERASE_IN_LEN);
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_TYPE, type);
+       MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_OFFSET, offset);
+       MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_LENGTH, length);
+
+       BUILD_BUG_ON(MC_CMD_NVRAM_ERASE_OUT_LEN != 0);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_ERASE, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+static int efx_mcdi_nvram_update_finish(struct efx_nic *efx, unsigned int type)
+{
+       MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_UPDATE_FINISH_IN_LEN);
+       int rc;
+
+       MCDI_SET_DWORD(inbuf, NVRAM_UPDATE_FINISH_IN_TYPE, type);
+
+       BUILD_BUG_ON(MC_CMD_NVRAM_UPDATE_FINISH_OUT_LEN != 0);
+
+       rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_UPDATE_FINISH, inbuf, sizeof(inbuf),
+                         NULL, 0, NULL);
+       return rc;
+}
+
+int efx_mcdi_mtd_read(struct mtd_info *mtd, loff_t start,
+                     size_t len, size_t *retlen, u8 *buffer)
+{
+       struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
+       struct efx_nic *efx = mtd->priv;
+       loff_t offset = start;
+       loff_t end = min_t(loff_t, start + len, mtd->size);
+       size_t chunk;
+       int rc = 0;
+
+       while (offset < end) {
+               chunk = min_t(size_t, end - offset, EFX_MCDI_NVRAM_LEN_MAX);
+               rc = efx_mcdi_nvram_read(efx, part->nvram_type, offset,
+                                        buffer, chunk);
+               if (rc)
+                       goto out;
+               offset += chunk;
+               buffer += chunk;
+       }
+out:
+       *retlen = offset - start;
+       return rc;
+}
+
+int efx_mcdi_mtd_erase(struct mtd_info *mtd, loff_t start, size_t len)
+{
+       struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
+       struct efx_nic *efx = mtd->priv;
+       loff_t offset = start & ~((loff_t)(mtd->erasesize - 1));
+       loff_t end = min_t(loff_t, start + len, mtd->size);
+       size_t chunk = part->common.mtd.erasesize;
+       int rc = 0;
+
+       if (!part->updating) {
+               rc = efx_mcdi_nvram_update_start(efx, part->nvram_type);
+               if (rc)
+                       goto out;
+               part->updating = true;
+       }
+
+       /* The MCDI interface can in fact do multiple erase blocks at once;
+        * but erasing may be slow, so we make multiple calls here to avoid
+        * tripping the MCDI RPC timeout. */
+       while (offset < end) {
+               rc = efx_mcdi_nvram_erase(efx, part->nvram_type, offset,
+                                         chunk);
+               if (rc)
+                       goto out;
+               offset += chunk;
+       }
+out:
+       return rc;
+}
+
+int efx_mcdi_mtd_write(struct mtd_info *mtd, loff_t start,
+                      size_t len, size_t *retlen, const u8 *buffer)
+{
+       struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
+       struct efx_nic *efx = mtd->priv;
+       loff_t offset = start;
+       loff_t end = min_t(loff_t, start + len, mtd->size);
+       size_t chunk;
+       int rc = 0;
+
+       if (!part->updating) {
+               rc = efx_mcdi_nvram_update_start(efx, part->nvram_type);
+               if (rc)
+                       goto out;
+               part->updating = true;
+       }
+
+       while (offset < end) {
+               chunk = min_t(size_t, end - offset, EFX_MCDI_NVRAM_LEN_MAX);
+               rc = efx_mcdi_nvram_write(efx, part->nvram_type, offset,
+                                         buffer, chunk);
+               if (rc)
+                       goto out;
+               offset += chunk;
+               buffer += chunk;
+       }
+out:
+       *retlen = offset - start;
+       return rc;
+}
+
+int efx_mcdi_mtd_sync(struct mtd_info *mtd)
+{
+       struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
+       struct efx_nic *efx = mtd->priv;
+       int rc = 0;
+
+       if (part->updating) {
+               part->updating = false;
+               rc = efx_mcdi_nvram_update_finish(efx, part->nvram_type);
+       }
+
+       return rc;
+}
+
+void efx_mcdi_mtd_rename(struct efx_mtd_partition *part)
+{
+       struct efx_mcdi_mtd_partition *mcdi_part =
+               container_of(part, struct efx_mcdi_mtd_partition, common);
+       struct efx_nic *efx = part->mtd.priv;
+
+       snprintf(part->name, sizeof(part->name), "%s %s:%02x",
+                efx->name, part->type_name, mcdi_part->fw_subtype);
+}
+
+#endif /* CONFIG_SFC_MTD */