Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / sfc / nic.c
diff --git a/kernel/drivers/net/ethernet/sfc/nic.c b/kernel/drivers/net/ethernet/sfc/nic.c
new file mode 100644 (file)
index 0000000..89b83e5
--- /dev/null
@@ -0,0 +1,534 @@
+/****************************************************************************
+ * Driver for Solarflare network controllers and boards
+ * Copyright 2005-2006 Fen Systems Ltd.
+ * Copyright 2006-2013 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/bitops.h>
+#include <linux/delay.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/module.h>
+#include <linux/seq_file.h>
+#include <linux/cpu_rmap.h>
+#include "net_driver.h"
+#include "bitfield.h"
+#include "efx.h"
+#include "nic.h"
+#include "ef10_regs.h"
+#include "farch_regs.h"
+#include "io.h"
+#include "workarounds.h"
+
+/**************************************************************************
+ *
+ * Generic buffer handling
+ * These buffers are used for interrupt status, MAC stats, etc.
+ *
+ **************************************************************************/
+
+int efx_nic_alloc_buffer(struct efx_nic *efx, struct efx_buffer *buffer,
+                        unsigned int len, gfp_t gfp_flags)
+{
+       buffer->addr = dma_zalloc_coherent(&efx->pci_dev->dev, len,
+                                          &buffer->dma_addr, gfp_flags);
+       if (!buffer->addr)
+               return -ENOMEM;
+       buffer->len = len;
+       return 0;
+}
+
+void efx_nic_free_buffer(struct efx_nic *efx, struct efx_buffer *buffer)
+{
+       if (buffer->addr) {
+               dma_free_coherent(&efx->pci_dev->dev, buffer->len,
+                                 buffer->addr, buffer->dma_addr);
+               buffer->addr = NULL;
+       }
+}
+
+/* Check whether an event is present in the eventq at the current
+ * read pointer.  Only useful for self-test.
+ */
+bool efx_nic_event_present(struct efx_channel *channel)
+{
+       return efx_event_present(efx_event(channel, channel->eventq_read_ptr));
+}
+
+void efx_nic_event_test_start(struct efx_channel *channel)
+{
+       channel->event_test_cpu = -1;
+       smp_wmb();
+       channel->efx->type->ev_test_generate(channel);
+}
+
+void efx_nic_irq_test_start(struct efx_nic *efx)
+{
+       efx->last_irq_cpu = -1;
+       smp_wmb();
+       efx->type->irq_test_generate(efx);
+}
+
+/* Hook interrupt handler(s)
+ * Try MSI and then legacy interrupts.
+ */
+int efx_nic_init_interrupt(struct efx_nic *efx)
+{
+       struct efx_channel *channel;
+       unsigned int n_irqs;
+       int rc;
+
+       if (!EFX_INT_MODE_USE_MSI(efx)) {
+               rc = request_irq(efx->legacy_irq,
+                                efx->type->irq_handle_legacy, IRQF_SHARED,
+                                efx->name, efx);
+               if (rc) {
+                       netif_err(efx, drv, efx->net_dev,
+                                 "failed to hook legacy IRQ %d\n",
+                                 efx->pci_dev->irq);
+                       goto fail1;
+               }
+               return 0;
+       }
+
+#ifdef CONFIG_RFS_ACCEL
+       if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
+               efx->net_dev->rx_cpu_rmap =
+                       alloc_irq_cpu_rmap(efx->n_rx_channels);
+               if (!efx->net_dev->rx_cpu_rmap) {
+                       rc = -ENOMEM;
+                       goto fail1;
+               }
+       }
+#endif
+
+       /* Hook MSI or MSI-X interrupt */
+       n_irqs = 0;
+       efx_for_each_channel(channel, efx) {
+               rc = request_irq(channel->irq, efx->type->irq_handle_msi,
+                                IRQF_PROBE_SHARED, /* Not shared */
+                                efx->msi_context[channel->channel].name,
+                                &efx->msi_context[channel->channel]);
+               if (rc) {
+                       netif_err(efx, drv, efx->net_dev,
+                                 "failed to hook IRQ %d\n", channel->irq);
+                       goto fail2;
+               }
+               ++n_irqs;
+
+#ifdef CONFIG_RFS_ACCEL
+               if (efx->interrupt_mode == EFX_INT_MODE_MSIX &&
+                   channel->channel < efx->n_rx_channels) {
+                       rc = irq_cpu_rmap_add(efx->net_dev->rx_cpu_rmap,
+                                             channel->irq);
+                       if (rc)
+                               goto fail2;
+               }
+#endif
+       }
+
+       return 0;
+
+ fail2:
+#ifdef CONFIG_RFS_ACCEL
+       free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
+       efx->net_dev->rx_cpu_rmap = NULL;
+#endif
+       efx_for_each_channel(channel, efx) {
+               if (n_irqs-- == 0)
+                       break;
+               free_irq(channel->irq, &efx->msi_context[channel->channel]);
+       }
+ fail1:
+       return rc;
+}
+
+void efx_nic_fini_interrupt(struct efx_nic *efx)
+{
+       struct efx_channel *channel;
+
+#ifdef CONFIG_RFS_ACCEL
+       free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
+       efx->net_dev->rx_cpu_rmap = NULL;
+#endif
+
+       if (EFX_INT_MODE_USE_MSI(efx)) {
+               /* Disable MSI/MSI-X interrupts */
+               efx_for_each_channel(channel, efx)
+                       free_irq(channel->irq,
+                                &efx->msi_context[channel->channel]);
+       } else {
+               /* Disable legacy interrupt */
+               free_irq(efx->legacy_irq, efx);
+       }
+}
+
+/* Register dump */
+
+#define REGISTER_REVISION_FA   1
+#define REGISTER_REVISION_FB   2
+#define REGISTER_REVISION_FC   3
+#define REGISTER_REVISION_FZ   3       /* last Falcon arch revision */
+#define REGISTER_REVISION_ED   4
+#define REGISTER_REVISION_EZ   4       /* latest EF10 revision */
+
+struct efx_nic_reg {
+       u32 offset:24;
+       u32 min_revision:3, max_revision:3;
+};
+
+#define REGISTER(name, arch, min_rev, max_rev) {                       \
+       arch ## R_ ## min_rev ## max_rev ## _ ## name,                  \
+       REGISTER_REVISION_ ## arch ## min_rev,                          \
+       REGISTER_REVISION_ ## arch ## max_rev                           \
+}
+#define REGISTER_AA(name) REGISTER(name, F, A, A)
+#define REGISTER_AB(name) REGISTER(name, F, A, B)
+#define REGISTER_AZ(name) REGISTER(name, F, A, Z)
+#define REGISTER_BB(name) REGISTER(name, F, B, B)
+#define REGISTER_BZ(name) REGISTER(name, F, B, Z)
+#define REGISTER_CZ(name) REGISTER(name, F, C, Z)
+#define REGISTER_DZ(name) REGISTER(name, E, D, Z)
+
+static const struct efx_nic_reg efx_nic_regs[] = {
+       REGISTER_AZ(ADR_REGION),
+       REGISTER_AZ(INT_EN_KER),
+       REGISTER_BZ(INT_EN_CHAR),
+       REGISTER_AZ(INT_ADR_KER),
+       REGISTER_BZ(INT_ADR_CHAR),
+       /* INT_ACK_KER is WO */
+       /* INT_ISR0 is RC */
+       REGISTER_AZ(HW_INIT),
+       REGISTER_CZ(USR_EV_CFG),
+       REGISTER_AB(EE_SPI_HCMD),
+       REGISTER_AB(EE_SPI_HADR),
+       REGISTER_AB(EE_SPI_HDATA),
+       REGISTER_AB(EE_BASE_PAGE),
+       REGISTER_AB(EE_VPD_CFG0),
+       /* EE_VPD_SW_CNTL and EE_VPD_SW_DATA are not used */
+       /* PMBX_DBG_IADDR and PBMX_DBG_IDATA are indirect */
+       /* PCIE_CORE_INDIRECT is indirect */
+       REGISTER_AB(NIC_STAT),
+       REGISTER_AB(GPIO_CTL),
+       REGISTER_AB(GLB_CTL),
+       /* FATAL_INTR_KER and FATAL_INTR_CHAR are partly RC */
+       REGISTER_BZ(DP_CTRL),
+       REGISTER_AZ(MEM_STAT),
+       REGISTER_AZ(CS_DEBUG),
+       REGISTER_AZ(ALTERA_BUILD),
+       REGISTER_AZ(CSR_SPARE),
+       REGISTER_AB(PCIE_SD_CTL0123),
+       REGISTER_AB(PCIE_SD_CTL45),
+       REGISTER_AB(PCIE_PCS_CTL_STAT),
+       /* DEBUG_DATA_OUT is not used */
+       /* DRV_EV is WO */
+       REGISTER_AZ(EVQ_CTL),
+       REGISTER_AZ(EVQ_CNT1),
+       REGISTER_AZ(EVQ_CNT2),
+       REGISTER_AZ(BUF_TBL_CFG),
+       REGISTER_AZ(SRM_RX_DC_CFG),
+       REGISTER_AZ(SRM_TX_DC_CFG),
+       REGISTER_AZ(SRM_CFG),
+       /* BUF_TBL_UPD is WO */
+       REGISTER_AZ(SRM_UPD_EVQ),
+       REGISTER_AZ(SRAM_PARITY),
+       REGISTER_AZ(RX_CFG),
+       REGISTER_BZ(RX_FILTER_CTL),
+       /* RX_FLUSH_DESCQ is WO */
+       REGISTER_AZ(RX_DC_CFG),
+       REGISTER_AZ(RX_DC_PF_WM),
+       REGISTER_BZ(RX_RSS_TKEY),
+       /* RX_NODESC_DROP is RC */
+       REGISTER_AA(RX_SELF_RST),
+       /* RX_DEBUG, RX_PUSH_DROP are not used */
+       REGISTER_CZ(RX_RSS_IPV6_REG1),
+       REGISTER_CZ(RX_RSS_IPV6_REG2),
+       REGISTER_CZ(RX_RSS_IPV6_REG3),
+       /* TX_FLUSH_DESCQ is WO */
+       REGISTER_AZ(TX_DC_CFG),
+       REGISTER_AA(TX_CHKSM_CFG),
+       REGISTER_AZ(TX_CFG),
+       /* TX_PUSH_DROP is not used */
+       REGISTER_AZ(TX_RESERVED),
+       REGISTER_BZ(TX_PACE),
+       /* TX_PACE_DROP_QID is RC */
+       REGISTER_BB(TX_VLAN),
+       REGISTER_BZ(TX_IPFIL_PORTEN),
+       REGISTER_AB(MD_TXD),
+       REGISTER_AB(MD_RXD),
+       REGISTER_AB(MD_CS),
+       REGISTER_AB(MD_PHY_ADR),
+       REGISTER_AB(MD_ID),
+       /* MD_STAT is RC */
+       REGISTER_AB(MAC_STAT_DMA),
+       REGISTER_AB(MAC_CTRL),
+       REGISTER_BB(GEN_MODE),
+       REGISTER_AB(MAC_MC_HASH_REG0),
+       REGISTER_AB(MAC_MC_HASH_REG1),
+       REGISTER_AB(GM_CFG1),
+       REGISTER_AB(GM_CFG2),
+       /* GM_IPG and GM_HD are not used */
+       REGISTER_AB(GM_MAX_FLEN),
+       /* GM_TEST is not used */
+       REGISTER_AB(GM_ADR1),
+       REGISTER_AB(GM_ADR2),
+       REGISTER_AB(GMF_CFG0),
+       REGISTER_AB(GMF_CFG1),
+       REGISTER_AB(GMF_CFG2),
+       REGISTER_AB(GMF_CFG3),
+       REGISTER_AB(GMF_CFG4),
+       REGISTER_AB(GMF_CFG5),
+       REGISTER_BB(TX_SRC_MAC_CTL),
+       REGISTER_AB(XM_ADR_LO),
+       REGISTER_AB(XM_ADR_HI),
+       REGISTER_AB(XM_GLB_CFG),
+       REGISTER_AB(XM_TX_CFG),
+       REGISTER_AB(XM_RX_CFG),
+       REGISTER_AB(XM_MGT_INT_MASK),
+       REGISTER_AB(XM_FC),
+       REGISTER_AB(XM_PAUSE_TIME),
+       REGISTER_AB(XM_TX_PARAM),
+       REGISTER_AB(XM_RX_PARAM),
+       /* XM_MGT_INT_MSK (note no 'A') is RC */
+       REGISTER_AB(XX_PWR_RST),
+       REGISTER_AB(XX_SD_CTL),
+       REGISTER_AB(XX_TXDRV_CTL),
+       /* XX_PRBS_CTL, XX_PRBS_CHK and XX_PRBS_ERR are not used */
+       /* XX_CORE_STAT is partly RC */
+       REGISTER_DZ(BIU_HW_REV_ID),
+       REGISTER_DZ(MC_DB_LWRD),
+       REGISTER_DZ(MC_DB_HWRD),
+};
+
+struct efx_nic_reg_table {
+       u32 offset:24;
+       u32 min_revision:3, max_revision:3;
+       u32 step:6, rows:21;
+};
+
+#define REGISTER_TABLE_DIMENSIONS(_, offset, arch, min_rev, max_rev, step, rows) { \
+       offset,                                                         \
+       REGISTER_REVISION_ ## arch ## min_rev,                          \
+       REGISTER_REVISION_ ## arch ## max_rev,                          \
+       step, rows                                                      \
+}
+#define REGISTER_TABLE(name, arch, min_rev, max_rev)                   \
+       REGISTER_TABLE_DIMENSIONS(                                      \
+               name, arch ## R_ ## min_rev ## max_rev ## _ ## name,    \
+               arch, min_rev, max_rev,                                 \
+               arch ## R_ ## min_rev ## max_rev ## _ ## name ## _STEP, \
+               arch ## R_ ## min_rev ## max_rev ## _ ## name ## _ROWS)
+#define REGISTER_TABLE_AA(name) REGISTER_TABLE(name, F, A, A)
+#define REGISTER_TABLE_AZ(name) REGISTER_TABLE(name, F, A, Z)
+#define REGISTER_TABLE_BB(name) REGISTER_TABLE(name, F, B, B)
+#define REGISTER_TABLE_BZ(name) REGISTER_TABLE(name, F, B, Z)
+#define REGISTER_TABLE_BB_CZ(name)                                     \
+       REGISTER_TABLE_DIMENSIONS(name, FR_BZ_ ## name, F, B, B,        \
+                                 FR_BZ_ ## name ## _STEP,              \
+                                 FR_BB_ ## name ## _ROWS),             \
+       REGISTER_TABLE_DIMENSIONS(name, FR_BZ_ ## name, F, C, Z,        \
+                                 FR_BZ_ ## name ## _STEP,              \
+                                 FR_CZ_ ## name ## _ROWS)
+#define REGISTER_TABLE_CZ(name) REGISTER_TABLE(name, F, C, Z)
+#define REGISTER_TABLE_DZ(name) REGISTER_TABLE(name, E, D, Z)
+
+static const struct efx_nic_reg_table efx_nic_reg_tables[] = {
+       /* DRIVER is not used */
+       /* EVQ_RPTR, TIMER_COMMAND, USR_EV and {RX,TX}_DESC_UPD are WO */
+       REGISTER_TABLE_BB(TX_IPFIL_TBL),
+       REGISTER_TABLE_BB(TX_SRC_MAC_TBL),
+       REGISTER_TABLE_AA(RX_DESC_PTR_TBL_KER),
+       REGISTER_TABLE_BB_CZ(RX_DESC_PTR_TBL),
+       REGISTER_TABLE_AA(TX_DESC_PTR_TBL_KER),
+       REGISTER_TABLE_BB_CZ(TX_DESC_PTR_TBL),
+       REGISTER_TABLE_AA(EVQ_PTR_TBL_KER),
+       REGISTER_TABLE_BB_CZ(EVQ_PTR_TBL),
+       /* We can't reasonably read all of the buffer table (up to 8MB!).
+        * However this driver will only use a few entries.  Reading
+        * 1K entries allows for some expansion of queue count and
+        * size before we need to change the version. */
+       REGISTER_TABLE_DIMENSIONS(BUF_FULL_TBL_KER, FR_AA_BUF_FULL_TBL_KER,
+                                 F, A, A, 8, 1024),
+       REGISTER_TABLE_DIMENSIONS(BUF_FULL_TBL, FR_BZ_BUF_FULL_TBL,
+                                 F, B, Z, 8, 1024),
+       REGISTER_TABLE_CZ(RX_MAC_FILTER_TBL0),
+       REGISTER_TABLE_BB_CZ(TIMER_TBL),
+       REGISTER_TABLE_BB_CZ(TX_PACE_TBL),
+       REGISTER_TABLE_BZ(RX_INDIRECTION_TBL),
+       /* TX_FILTER_TBL0 is huge and not used by this driver */
+       REGISTER_TABLE_CZ(TX_MAC_FILTER_TBL0),
+       REGISTER_TABLE_CZ(MC_TREG_SMEM),
+       /* MSIX_PBA_TABLE is not mapped */
+       /* SRM_DBG is not mapped (and is redundant with BUF_FLL_TBL) */
+       REGISTER_TABLE_BZ(RX_FILTER_TBL0),
+       REGISTER_TABLE_DZ(BIU_MC_SFT_STATUS),
+};
+
+size_t efx_nic_get_regs_len(struct efx_nic *efx)
+{
+       const struct efx_nic_reg *reg;
+       const struct efx_nic_reg_table *table;
+       size_t len = 0;
+
+       for (reg = efx_nic_regs;
+            reg < efx_nic_regs + ARRAY_SIZE(efx_nic_regs);
+            reg++)
+               if (efx->type->revision >= reg->min_revision &&
+                   efx->type->revision <= reg->max_revision)
+                       len += sizeof(efx_oword_t);
+
+       for (table = efx_nic_reg_tables;
+            table < efx_nic_reg_tables + ARRAY_SIZE(efx_nic_reg_tables);
+            table++)
+               if (efx->type->revision >= table->min_revision &&
+                   efx->type->revision <= table->max_revision)
+                       len += table->rows * min_t(size_t, table->step, 16);
+
+       return len;
+}
+
+void efx_nic_get_regs(struct efx_nic *efx, void *buf)
+{
+       const struct efx_nic_reg *reg;
+       const struct efx_nic_reg_table *table;
+
+       for (reg = efx_nic_regs;
+            reg < efx_nic_regs + ARRAY_SIZE(efx_nic_regs);
+            reg++) {
+               if (efx->type->revision >= reg->min_revision &&
+                   efx->type->revision <= reg->max_revision) {
+                       efx_reado(efx, (efx_oword_t *)buf, reg->offset);
+                       buf += sizeof(efx_oword_t);
+               }
+       }
+
+       for (table = efx_nic_reg_tables;
+            table < efx_nic_reg_tables + ARRAY_SIZE(efx_nic_reg_tables);
+            table++) {
+               size_t size, i;
+
+               if (!(efx->type->revision >= table->min_revision &&
+                     efx->type->revision <= table->max_revision))
+                       continue;
+
+               size = min_t(size_t, table->step, 16);
+
+               for (i = 0; i < table->rows; i++) {
+                       switch (table->step) {
+                       case 4: /* 32-bit SRAM */
+                               efx_readd(efx, buf, table->offset + 4 * i);
+                               break;
+                       case 8: /* 64-bit SRAM */
+                               efx_sram_readq(efx,
+                                              efx->membase + table->offset,
+                                              buf, i);
+                               break;
+                       case 16: /* 128-bit-readable register */
+                               efx_reado_table(efx, buf, table->offset, i);
+                               break;
+                       case 32: /* 128-bit register, interleaved */
+                               efx_reado_table(efx, buf, table->offset, 2 * i);
+                               break;
+                       default:
+                               WARN_ON(1);
+                               return;
+                       }
+                       buf += size;
+               }
+       }
+}
+
+/**
+ * efx_nic_describe_stats - Describe supported statistics for ethtool
+ * @desc: Array of &struct efx_hw_stat_desc describing the statistics
+ * @count: Length of the @desc array
+ * @mask: Bitmask of which elements of @desc are enabled
+ * @names: Buffer to copy names to, or %NULL.  The names are copied
+ *     starting at intervals of %ETH_GSTRING_LEN bytes.
+ *
+ * Returns the number of visible statistics, i.e. the number of set
+ * bits in the first @count bits of @mask for which a name is defined.
+ */
+size_t efx_nic_describe_stats(const struct efx_hw_stat_desc *desc, size_t count,
+                             const unsigned long *mask, u8 *names)
+{
+       size_t visible = 0;
+       size_t index;
+
+       for_each_set_bit(index, mask, count) {
+               if (desc[index].name) {
+                       if (names) {
+                               strlcpy(names, desc[index].name,
+                                       ETH_GSTRING_LEN);
+                               names += ETH_GSTRING_LEN;
+                       }
+                       ++visible;
+               }
+       }
+
+       return visible;
+}
+
+/**
+ * efx_nic_update_stats - Convert statistics DMA buffer to array of u64
+ * @desc: Array of &struct efx_hw_stat_desc describing the DMA buffer
+ *     layout.  DMA widths of 0, 16, 32 and 64 are supported; where
+ *     the width is specified as 0 the corresponding element of
+ *     @stats is not updated.
+ * @count: Length of the @desc array
+ * @mask: Bitmask of which elements of @desc are enabled
+ * @stats: Buffer to update with the converted statistics.  The length
+ *     of this array must be at least @count.
+ * @dma_buf: DMA buffer containing hardware statistics
+ * @accumulate: If set, the converted values will be added rather than
+ *     directly stored to the corresponding elements of @stats
+ */
+void efx_nic_update_stats(const struct efx_hw_stat_desc *desc, size_t count,
+                         const unsigned long *mask,
+                         u64 *stats, const void *dma_buf, bool accumulate)
+{
+       size_t index;
+
+       for_each_set_bit(index, mask, count) {
+               if (desc[index].dma_width) {
+                       const void *addr = dma_buf + desc[index].offset;
+                       u64 val;
+
+                       switch (desc[index].dma_width) {
+                       case 16:
+                               val = le16_to_cpup((__le16 *)addr);
+                               break;
+                       case 32:
+                               val = le32_to_cpup((__le32 *)addr);
+                               break;
+                       case 64:
+                               val = le64_to_cpup((__le64 *)addr);
+                               break;
+                       default:
+                               WARN_ON(1);
+                               val = 0;
+                               break;
+                       }
+
+                       if (accumulate)
+                               stats[index] += val;
+                       else
+                               stats[index] = val;
+               }
+       }
+}
+
+void efx_nic_fix_nodesc_drop_stat(struct efx_nic *efx, u64 *rx_nodesc_drops)
+{
+       /* if down, or this is the first update after coming up */
+       if (!(efx->net_dev->flags & IFF_UP) || !efx->rx_nodesc_drops_prev_state)
+               efx->rx_nodesc_drops_while_down +=
+                       *rx_nodesc_drops - efx->rx_nodesc_drops_total;
+       efx->rx_nodesc_drops_total = *rx_nodesc_drops;
+       efx->rx_nodesc_drops_prev_state = !!(efx->net_dev->flags & IFF_UP);
+       *rx_nodesc_drops -= efx->rx_nodesc_drops_while_down;
+}