OpenCloudOS-Kernel/drivers/net/bnx2x/bnx2x_init_ops.h

867 lines
25 KiB
C

/* bnx2x_init_ops.h: Broadcom Everest network driver.
* Static functions needed during the initialization.
* This file is "included" in bnx2x_main.c.
*
* Copyright (c) 2007-2010 Broadcom Corporation
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation.
*
* Maintained by: Eilon Greenstein <eilong@broadcom.com>
* Written by: Vladislav Zolotarov <vladz@broadcom.com>
*/
#ifndef BNX2X_INIT_OPS_H
#define BNX2X_INIT_OPS_H
static int bnx2x_gunzip(struct bnx2x *bp, const u8 *zbuf, int len);
static void bnx2x_reg_wr_ind(struct bnx2x *bp, u32 addr, u32 val);
static void bnx2x_write_dmae_phys_len(struct bnx2x *bp, dma_addr_t phys_addr,
u32 addr, u32 len);
static void bnx2x_init_str_wr(struct bnx2x *bp, u32 addr, const u32 *data,
u32 len)
{
u32 i;
for (i = 0; i < len; i++)
REG_WR(bp, addr + i*4, data[i]);
}
static void bnx2x_init_ind_wr(struct bnx2x *bp, u32 addr, const u32 *data,
u32 len)
{
u32 i;
for (i = 0; i < len; i++)
REG_WR_IND(bp, addr + i*4, data[i]);
}
static void bnx2x_write_big_buf(struct bnx2x *bp, u32 addr, u32 len)
{
if (bp->dmae_ready)
bnx2x_write_dmae_phys_len(bp, GUNZIP_PHYS(bp), addr, len);
else
bnx2x_init_str_wr(bp, addr, GUNZIP_BUF(bp), len);
}
static void bnx2x_init_fill(struct bnx2x *bp, u32 addr, int fill, u32 len)
{
u32 buf_len = (((len*4) > FW_BUF_SIZE) ? FW_BUF_SIZE : (len*4));
u32 buf_len32 = buf_len/4;
u32 i;
memset(GUNZIP_BUF(bp), (u8)fill, buf_len);
for (i = 0; i < len; i += buf_len32) {
u32 cur_len = min(buf_len32, len - i);
bnx2x_write_big_buf(bp, addr + i*4, cur_len);
}
}
static void bnx2x_init_wr_64(struct bnx2x *bp, u32 addr, const u32 *data,
u32 len64)
{
u32 buf_len32 = FW_BUF_SIZE/4;
u32 len = len64*2;
u64 data64 = 0;
u32 i;
/* 64 bit value is in a blob: first low DWORD, then high DWORD */
data64 = HILO_U64((*(data + 1)), (*data));
len64 = min((u32)(FW_BUF_SIZE/8), len64);
for (i = 0; i < len64; i++) {
u64 *pdata = ((u64 *)(GUNZIP_BUF(bp))) + i;
*pdata = data64;
}
for (i = 0; i < len; i += buf_len32) {
u32 cur_len = min(buf_len32, len - i);
bnx2x_write_big_buf(bp, addr + i*4, cur_len);
}
}
/*********************************************************
There are different blobs for each PRAM section.
In addition, each blob write operation is divided into a few operations
in order to decrease the amount of phys. contiguous buffer needed.
Thus, when we select a blob the address may be with some offset
from the beginning of PRAM section.
The same holds for the INT_TABLE sections.
**********************************************************/
#define IF_IS_INT_TABLE_ADDR(base, addr) \
if (((base) <= (addr)) && ((base) + 0x400 >= (addr)))
#define IF_IS_PRAM_ADDR(base, addr) \
if (((base) <= (addr)) && ((base) + 0x40000 >= (addr)))
static const u8 *bnx2x_sel_blob(struct bnx2x *bp, u32 addr, const u8 *data)
{
IF_IS_INT_TABLE_ADDR(TSEM_REG_INT_TABLE, addr)
data = INIT_TSEM_INT_TABLE_DATA(bp);
else
IF_IS_INT_TABLE_ADDR(CSEM_REG_INT_TABLE, addr)
data = INIT_CSEM_INT_TABLE_DATA(bp);
else
IF_IS_INT_TABLE_ADDR(USEM_REG_INT_TABLE, addr)
data = INIT_USEM_INT_TABLE_DATA(bp);
else
IF_IS_INT_TABLE_ADDR(XSEM_REG_INT_TABLE, addr)
data = INIT_XSEM_INT_TABLE_DATA(bp);
else
IF_IS_PRAM_ADDR(TSEM_REG_PRAM, addr)
data = INIT_TSEM_PRAM_DATA(bp);
else
IF_IS_PRAM_ADDR(CSEM_REG_PRAM, addr)
data = INIT_CSEM_PRAM_DATA(bp);
else
IF_IS_PRAM_ADDR(USEM_REG_PRAM, addr)
data = INIT_USEM_PRAM_DATA(bp);
else
IF_IS_PRAM_ADDR(XSEM_REG_PRAM, addr)
data = INIT_XSEM_PRAM_DATA(bp);
return data;
}
static void bnx2x_write_big_buf_wb(struct bnx2x *bp, u32 addr, u32 len)
{
if (bp->dmae_ready)
bnx2x_write_dmae_phys_len(bp, GUNZIP_PHYS(bp), addr, len);
else
bnx2x_init_ind_wr(bp, addr, GUNZIP_BUF(bp), len);
}
static void bnx2x_init_wr_wb(struct bnx2x *bp, u32 addr, const u32 *data,
u32 len)
{
const u32 *old_data = data;
data = (const u32 *)bnx2x_sel_blob(bp, addr, (const u8 *)data);
if (bp->dmae_ready) {
if (old_data != data)
VIRT_WR_DMAE_LEN(bp, data, addr, len, 1);
else
VIRT_WR_DMAE_LEN(bp, data, addr, len, 0);
} else
bnx2x_init_ind_wr(bp, addr, data, len);
}
static void bnx2x_wr_64(struct bnx2x *bp, u32 reg, u32 val_lo, u32 val_hi)
{
u32 wb_write[2];
wb_write[0] = val_lo;
wb_write[1] = val_hi;
REG_WR_DMAE_LEN(bp, reg, wb_write, 2);
}
static void bnx2x_init_wr_zp(struct bnx2x *bp, u32 addr, u32 len, u32 blob_off)
{
const u8 *data = NULL;
int rc;
u32 i;
data = bnx2x_sel_blob(bp, addr, data) + blob_off*4;
rc = bnx2x_gunzip(bp, data, len);
if (rc)
return;
/* gunzip_outlen is in dwords */
len = GUNZIP_OUTLEN(bp);
for (i = 0; i < len; i++)
((u32 *)GUNZIP_BUF(bp))[i] =
cpu_to_le32(((u32 *)GUNZIP_BUF(bp))[i]);
bnx2x_write_big_buf_wb(bp, addr, len);
}
static void bnx2x_init_block(struct bnx2x *bp, u32 block, u32 stage)
{
u16 op_start =
INIT_OPS_OFFSETS(bp)[BLOCK_OPS_IDX(block, stage, STAGE_START)];
u16 op_end =
INIT_OPS_OFFSETS(bp)[BLOCK_OPS_IDX(block, stage, STAGE_END)];
union init_op *op;
int hw_wr;
u32 i, op_type, addr, len;
const u32 *data, *data_base;
/* If empty block */
if (op_start == op_end)
return;
if (CHIP_REV_IS_FPGA(bp))
hw_wr = OP_WR_FPGA;
else if (CHIP_REV_IS_EMUL(bp))
hw_wr = OP_WR_EMUL;
else
hw_wr = OP_WR_ASIC;
data_base = INIT_DATA(bp);
for (i = op_start; i < op_end; i++) {
op = (union init_op *)&(INIT_OPS(bp)[i]);
op_type = op->str_wr.op;
addr = op->str_wr.offset;
len = op->str_wr.data_len;
data = data_base + op->str_wr.data_off;
/* HW/EMUL specific */
if ((op_type > OP_WB) && (op_type == hw_wr))
op_type = OP_WR;
switch (op_type) {
case OP_RD:
REG_RD(bp, addr);
break;
case OP_WR:
REG_WR(bp, addr, op->write.val);
break;
case OP_SW:
bnx2x_init_str_wr(bp, addr, data, len);
break;
case OP_WB:
bnx2x_init_wr_wb(bp, addr, data, len);
break;
case OP_SI:
bnx2x_init_ind_wr(bp, addr, data, len);
break;
case OP_ZR:
bnx2x_init_fill(bp, addr, 0, op->zero.len);
break;
case OP_ZP:
bnx2x_init_wr_zp(bp, addr, len,
op->str_wr.data_off);
break;
case OP_WR_64:
bnx2x_init_wr_64(bp, addr, data, len);
break;
default:
/* happens whenever an op is of a diff HW */
break;
}
}
}
/****************************************************************************
* PXP Arbiter
****************************************************************************/
/*
* This code configures the PCI read/write arbiter
* which implements a weighted round robin
* between the virtual queues in the chip.
*
* The values were derived for each PCI max payload and max request size.
* since max payload and max request size are only known at run time,
* this is done as a separate init stage.
*/
#define NUM_WR_Q 13
#define NUM_RD_Q 29
#define MAX_RD_ORD 3
#define MAX_WR_ORD 2
/* configuration for one arbiter queue */
struct arb_line {
int l;
int add;
int ubound;
};
/* derived configuration for each read queue for each max request size */
static const struct arb_line read_arb_data[NUM_RD_Q][MAX_RD_ORD + 1] = {
/* 1 */ { {8, 64, 25}, {16, 64, 25}, {32, 64, 25}, {64, 64, 41} },
{ {4, 8, 4}, {4, 8, 4}, {4, 8, 4}, {4, 8, 4} },
{ {4, 3, 3}, {4, 3, 3}, {4, 3, 3}, {4, 3, 3} },
{ {8, 3, 6}, {16, 3, 11}, {16, 3, 11}, {16, 3, 11} },
{ {8, 64, 25}, {16, 64, 25}, {32, 64, 25}, {64, 64, 41} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {64, 3, 41} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {64, 3, 41} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {64, 3, 41} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {64, 3, 41} },
/* 10 */{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 64, 6}, {16, 64, 11}, {32, 64, 21}, {32, 64, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
/* 20 */{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 3, 6}, {16, 3, 11}, {32, 3, 21}, {32, 3, 21} },
{ {8, 64, 25}, {16, 64, 41}, {32, 64, 81}, {64, 64, 120} }
};
/* derived configuration for each write queue for each max request size */
static const struct arb_line write_arb_data[NUM_WR_Q][MAX_WR_ORD + 1] = {
/* 1 */ { {4, 6, 3}, {4, 6, 3}, {4, 6, 3} },
{ {4, 2, 3}, {4, 2, 3}, {4, 2, 3} },
{ {8, 2, 6}, {16, 2, 11}, {16, 2, 11} },
{ {8, 2, 6}, {16, 2, 11}, {32, 2, 21} },
{ {8, 2, 6}, {16, 2, 11}, {32, 2, 21} },
{ {8, 2, 6}, {16, 2, 11}, {32, 2, 21} },
{ {8, 64, 25}, {16, 64, 25}, {32, 64, 25} },
{ {8, 2, 6}, {16, 2, 11}, {16, 2, 11} },
{ {8, 2, 6}, {16, 2, 11}, {16, 2, 11} },
/* 10 */{ {8, 9, 6}, {16, 9, 11}, {32, 9, 21} },
{ {8, 47, 19}, {16, 47, 19}, {32, 47, 21} },
{ {8, 9, 6}, {16, 9, 11}, {16, 9, 11} },
{ {8, 64, 25}, {16, 64, 41}, {32, 64, 81} }
};
/* register addresses for read queues */
static const struct arb_line read_arb_addr[NUM_RD_Q-1] = {
/* 1 */ {PXP2_REG_RQ_BW_RD_L0, PXP2_REG_RQ_BW_RD_ADD0,
PXP2_REG_RQ_BW_RD_UBOUND0},
{PXP2_REG_PSWRQ_BW_L1, PXP2_REG_PSWRQ_BW_ADD1,
PXP2_REG_PSWRQ_BW_UB1},
{PXP2_REG_PSWRQ_BW_L2, PXP2_REG_PSWRQ_BW_ADD2,
PXP2_REG_PSWRQ_BW_UB2},
{PXP2_REG_PSWRQ_BW_L3, PXP2_REG_PSWRQ_BW_ADD3,
PXP2_REG_PSWRQ_BW_UB3},
{PXP2_REG_RQ_BW_RD_L4, PXP2_REG_RQ_BW_RD_ADD4,
PXP2_REG_RQ_BW_RD_UBOUND4},
{PXP2_REG_RQ_BW_RD_L5, PXP2_REG_RQ_BW_RD_ADD5,
PXP2_REG_RQ_BW_RD_UBOUND5},
{PXP2_REG_PSWRQ_BW_L6, PXP2_REG_PSWRQ_BW_ADD6,
PXP2_REG_PSWRQ_BW_UB6},
{PXP2_REG_PSWRQ_BW_L7, PXP2_REG_PSWRQ_BW_ADD7,
PXP2_REG_PSWRQ_BW_UB7},
{PXP2_REG_PSWRQ_BW_L8, PXP2_REG_PSWRQ_BW_ADD8,
PXP2_REG_PSWRQ_BW_UB8},
/* 10 */{PXP2_REG_PSWRQ_BW_L9, PXP2_REG_PSWRQ_BW_ADD9,
PXP2_REG_PSWRQ_BW_UB9},
{PXP2_REG_PSWRQ_BW_L10, PXP2_REG_PSWRQ_BW_ADD10,
PXP2_REG_PSWRQ_BW_UB10},
{PXP2_REG_PSWRQ_BW_L11, PXP2_REG_PSWRQ_BW_ADD11,
PXP2_REG_PSWRQ_BW_UB11},
{PXP2_REG_RQ_BW_RD_L12, PXP2_REG_RQ_BW_RD_ADD12,
PXP2_REG_RQ_BW_RD_UBOUND12},
{PXP2_REG_RQ_BW_RD_L13, PXP2_REG_RQ_BW_RD_ADD13,
PXP2_REG_RQ_BW_RD_UBOUND13},
{PXP2_REG_RQ_BW_RD_L14, PXP2_REG_RQ_BW_RD_ADD14,
PXP2_REG_RQ_BW_RD_UBOUND14},
{PXP2_REG_RQ_BW_RD_L15, PXP2_REG_RQ_BW_RD_ADD15,
PXP2_REG_RQ_BW_RD_UBOUND15},
{PXP2_REG_RQ_BW_RD_L16, PXP2_REG_RQ_BW_RD_ADD16,
PXP2_REG_RQ_BW_RD_UBOUND16},
{PXP2_REG_RQ_BW_RD_L17, PXP2_REG_RQ_BW_RD_ADD17,
PXP2_REG_RQ_BW_RD_UBOUND17},
{PXP2_REG_RQ_BW_RD_L18, PXP2_REG_RQ_BW_RD_ADD18,
PXP2_REG_RQ_BW_RD_UBOUND18},
/* 20 */{PXP2_REG_RQ_BW_RD_L19, PXP2_REG_RQ_BW_RD_ADD19,
PXP2_REG_RQ_BW_RD_UBOUND19},
{PXP2_REG_RQ_BW_RD_L20, PXP2_REG_RQ_BW_RD_ADD20,
PXP2_REG_RQ_BW_RD_UBOUND20},
{PXP2_REG_RQ_BW_RD_L22, PXP2_REG_RQ_BW_RD_ADD22,
PXP2_REG_RQ_BW_RD_UBOUND22},
{PXP2_REG_RQ_BW_RD_L23, PXP2_REG_RQ_BW_RD_ADD23,
PXP2_REG_RQ_BW_RD_UBOUND23},
{PXP2_REG_RQ_BW_RD_L24, PXP2_REG_RQ_BW_RD_ADD24,
PXP2_REG_RQ_BW_RD_UBOUND24},
{PXP2_REG_RQ_BW_RD_L25, PXP2_REG_RQ_BW_RD_ADD25,
PXP2_REG_RQ_BW_RD_UBOUND25},
{PXP2_REG_RQ_BW_RD_L26, PXP2_REG_RQ_BW_RD_ADD26,
PXP2_REG_RQ_BW_RD_UBOUND26},
{PXP2_REG_RQ_BW_RD_L27, PXP2_REG_RQ_BW_RD_ADD27,
PXP2_REG_RQ_BW_RD_UBOUND27},
{PXP2_REG_PSWRQ_BW_L28, PXP2_REG_PSWRQ_BW_ADD28,
PXP2_REG_PSWRQ_BW_UB28}
};
/* register addresses for write queues */
static const struct arb_line write_arb_addr[NUM_WR_Q-1] = {
/* 1 */ {PXP2_REG_PSWRQ_BW_L1, PXP2_REG_PSWRQ_BW_ADD1,
PXP2_REG_PSWRQ_BW_UB1},
{PXP2_REG_PSWRQ_BW_L2, PXP2_REG_PSWRQ_BW_ADD2,
PXP2_REG_PSWRQ_BW_UB2},
{PXP2_REG_PSWRQ_BW_L3, PXP2_REG_PSWRQ_BW_ADD3,
PXP2_REG_PSWRQ_BW_UB3},
{PXP2_REG_PSWRQ_BW_L6, PXP2_REG_PSWRQ_BW_ADD6,
PXP2_REG_PSWRQ_BW_UB6},
{PXP2_REG_PSWRQ_BW_L7, PXP2_REG_PSWRQ_BW_ADD7,
PXP2_REG_PSWRQ_BW_UB7},
{PXP2_REG_PSWRQ_BW_L8, PXP2_REG_PSWRQ_BW_ADD8,
PXP2_REG_PSWRQ_BW_UB8},
{PXP2_REG_PSWRQ_BW_L9, PXP2_REG_PSWRQ_BW_ADD9,
PXP2_REG_PSWRQ_BW_UB9},
{PXP2_REG_PSWRQ_BW_L10, PXP2_REG_PSWRQ_BW_ADD10,
PXP2_REG_PSWRQ_BW_UB10},
{PXP2_REG_PSWRQ_BW_L11, PXP2_REG_PSWRQ_BW_ADD11,
PXP2_REG_PSWRQ_BW_UB11},
/* 10 */{PXP2_REG_PSWRQ_BW_L28, PXP2_REG_PSWRQ_BW_ADD28,
PXP2_REG_PSWRQ_BW_UB28},
{PXP2_REG_RQ_BW_WR_L29, PXP2_REG_RQ_BW_WR_ADD29,
PXP2_REG_RQ_BW_WR_UBOUND29},
{PXP2_REG_RQ_BW_WR_L30, PXP2_REG_RQ_BW_WR_ADD30,
PXP2_REG_RQ_BW_WR_UBOUND30}
};
static void bnx2x_init_pxp_arb(struct bnx2x *bp, int r_order, int w_order)
{
u32 val, i;
if (r_order > MAX_RD_ORD) {
DP(NETIF_MSG_HW, "read order of %d order adjusted to %d\n",
r_order, MAX_RD_ORD);
r_order = MAX_RD_ORD;
}
if (w_order > MAX_WR_ORD) {
DP(NETIF_MSG_HW, "write order of %d order adjusted to %d\n",
w_order, MAX_WR_ORD);
w_order = MAX_WR_ORD;
}
if (CHIP_REV_IS_FPGA(bp)) {
DP(NETIF_MSG_HW, "write order adjusted to 1 for FPGA\n");
w_order = 0;
}
DP(NETIF_MSG_HW, "read order %d write order %d\n", r_order, w_order);
for (i = 0; i < NUM_RD_Q-1; i++) {
REG_WR(bp, read_arb_addr[i].l, read_arb_data[i][r_order].l);
REG_WR(bp, read_arb_addr[i].add,
read_arb_data[i][r_order].add);
REG_WR(bp, read_arb_addr[i].ubound,
read_arb_data[i][r_order].ubound);
}
for (i = 0; i < NUM_WR_Q-1; i++) {
if ((write_arb_addr[i].l == PXP2_REG_RQ_BW_WR_L29) ||
(write_arb_addr[i].l == PXP2_REG_RQ_BW_WR_L30)) {
REG_WR(bp, write_arb_addr[i].l,
write_arb_data[i][w_order].l);
REG_WR(bp, write_arb_addr[i].add,
write_arb_data[i][w_order].add);
REG_WR(bp, write_arb_addr[i].ubound,
write_arb_data[i][w_order].ubound);
} else {
val = REG_RD(bp, write_arb_addr[i].l);
REG_WR(bp, write_arb_addr[i].l,
val | (write_arb_data[i][w_order].l << 10));
val = REG_RD(bp, write_arb_addr[i].add);
REG_WR(bp, write_arb_addr[i].add,
val | (write_arb_data[i][w_order].add << 10));
val = REG_RD(bp, write_arb_addr[i].ubound);
REG_WR(bp, write_arb_addr[i].ubound,
val | (write_arb_data[i][w_order].ubound << 7));
}
}
val = write_arb_data[NUM_WR_Q-1][w_order].add;
val += write_arb_data[NUM_WR_Q-1][w_order].ubound << 10;
val += write_arb_data[NUM_WR_Q-1][w_order].l << 17;
REG_WR(bp, PXP2_REG_PSWRQ_BW_RD, val);
val = read_arb_data[NUM_RD_Q-1][r_order].add;
val += read_arb_data[NUM_RD_Q-1][r_order].ubound << 10;
val += read_arb_data[NUM_RD_Q-1][r_order].l << 17;
REG_WR(bp, PXP2_REG_PSWRQ_BW_WR, val);
REG_WR(bp, PXP2_REG_RQ_WR_MBS0, w_order);
REG_WR(bp, PXP2_REG_RQ_WR_MBS1, w_order);
REG_WR(bp, PXP2_REG_RQ_RD_MBS0, r_order);
REG_WR(bp, PXP2_REG_RQ_RD_MBS1, r_order);
if ((CHIP_IS_E1(bp) || CHIP_IS_E1H(bp)) && (r_order == MAX_RD_ORD))
REG_WR(bp, PXP2_REG_RQ_PDR_LIMIT, 0xe00);
if (CHIP_IS_E2(bp))
REG_WR(bp, PXP2_REG_WR_USDMDP_TH, (0x8 << w_order));
else
REG_WR(bp, PXP2_REG_WR_USDMDP_TH, (0x18 << w_order));
if (CHIP_IS_E1H(bp) || CHIP_IS_E2(bp)) {
/* MPS w_order optimal TH presently TH
* 128 0 0 2
* 256 1 1 3
* >=512 2 2 3
*/
/* DMAE is special */
if (CHIP_IS_E2(bp)) {
/* E2 can use optimal TH */
val = w_order;
REG_WR(bp, PXP2_REG_WR_DMAE_MPS, val);
} else {
val = ((w_order == 0) ? 2 : 3);
REG_WR(bp, PXP2_REG_WR_DMAE_MPS, 2);
}
REG_WR(bp, PXP2_REG_WR_HC_MPS, val);
REG_WR(bp, PXP2_REG_WR_USDM_MPS, val);
REG_WR(bp, PXP2_REG_WR_CSDM_MPS, val);
REG_WR(bp, PXP2_REG_WR_TSDM_MPS, val);
REG_WR(bp, PXP2_REG_WR_XSDM_MPS, val);
REG_WR(bp, PXP2_REG_WR_QM_MPS, val);
REG_WR(bp, PXP2_REG_WR_TM_MPS, val);
REG_WR(bp, PXP2_REG_WR_SRC_MPS, val);
REG_WR(bp, PXP2_REG_WR_DBG_MPS, val);
REG_WR(bp, PXP2_REG_WR_CDU_MPS, val);
}
/* Validate number of tags suppoted by device */
#define PCIE_REG_PCIER_TL_HDR_FC_ST 0x2980
val = REG_RD(bp, PCIE_REG_PCIER_TL_HDR_FC_ST);
val &= 0xFF;
if (val <= 0x20)
REG_WR(bp, PXP2_REG_PGL_TAGS_LIMIT, 0x20);
}
/****************************************************************************
* ILT management
****************************************************************************/
/*
* This codes hides the low level HW interaction for ILT management and
* configuration. The API consists of a shadow ILT table which is set by the
* driver and a set of routines to use it to configure the HW.
*
*/
/* ILT HW init operations */
/* ILT memory management operations */
#define ILT_MEMOP_ALLOC 0
#define ILT_MEMOP_FREE 1
/* the phys address is shifted right 12 bits and has an added
* 1=valid bit added to the 53rd bit
* then since this is a wide register(TM)
* we split it into two 32 bit writes
*/
#define ILT_ADDR1(x) ((u32)(((u64)x >> 12) & 0xFFFFFFFF))
#define ILT_ADDR2(x) ((u32)((1 << 20) | ((u64)x >> 44)))
#define ILT_RANGE(f, l) (((l) << 10) | f)
static int bnx2x_ilt_line_mem_op(struct bnx2x *bp, struct ilt_line *line,
u32 size, u8 memop)
{
if (memop == ILT_MEMOP_FREE) {
BNX2X_ILT_FREE(line->page, line->page_mapping, line->size);
return 0;
}
BNX2X_ILT_ZALLOC(line->page, &line->page_mapping, size);
if (!line->page)
return -1;
line->size = size;
return 0;
}
static int bnx2x_ilt_client_mem_op(struct bnx2x *bp, int cli_num, u8 memop)
{
int i, rc;
struct bnx2x_ilt *ilt = BP_ILT(bp);
struct ilt_client_info *ilt_cli = &ilt->clients[cli_num];
if (!ilt || !ilt->lines)
return -1;
if (ilt_cli->flags & (ILT_CLIENT_SKIP_INIT | ILT_CLIENT_SKIP_MEM))
return 0;
for (rc = 0, i = ilt_cli->start; i <= ilt_cli->end && !rc; i++) {
rc = bnx2x_ilt_line_mem_op(bp, &ilt->lines[i],
ilt_cli->page_size, memop);
}
return rc;
}
static int bnx2x_ilt_mem_op(struct bnx2x *bp, u8 memop)
{
int rc = bnx2x_ilt_client_mem_op(bp, ILT_CLIENT_CDU, memop);
if (!rc)
rc = bnx2x_ilt_client_mem_op(bp, ILT_CLIENT_QM, memop);
if (!rc)
rc = bnx2x_ilt_client_mem_op(bp, ILT_CLIENT_SRC, memop);
if (!rc)
rc = bnx2x_ilt_client_mem_op(bp, ILT_CLIENT_TM, memop);
return rc;
}
static void bnx2x_ilt_line_wr(struct bnx2x *bp, int abs_idx,
dma_addr_t page_mapping)
{
u32 reg;
if (CHIP_IS_E1(bp))
reg = PXP2_REG_RQ_ONCHIP_AT + abs_idx*8;
else
reg = PXP2_REG_RQ_ONCHIP_AT_B0 + abs_idx*8;
bnx2x_wr_64(bp, reg, ILT_ADDR1(page_mapping), ILT_ADDR2(page_mapping));
}
static void bnx2x_ilt_line_init_op(struct bnx2x *bp, struct bnx2x_ilt *ilt,
int idx, u8 initop)
{
dma_addr_t null_mapping;
int abs_idx = ilt->start_line + idx;
switch (initop) {
case INITOP_INIT:
/* set in the init-value array */
case INITOP_SET:
bnx2x_ilt_line_wr(bp, abs_idx, ilt->lines[idx].page_mapping);
break;
case INITOP_CLEAR:
null_mapping = 0;
bnx2x_ilt_line_wr(bp, abs_idx, null_mapping);
break;
}
}
static void bnx2x_ilt_boundry_init_op(struct bnx2x *bp,
struct ilt_client_info *ilt_cli,
u32 ilt_start, u8 initop)
{
u32 start_reg = 0;
u32 end_reg = 0;
/* The boundary is either SET or INIT,
CLEAR => SET and for now SET ~~ INIT */
/* find the appropriate regs */
if (CHIP_IS_E1(bp)) {
switch (ilt_cli->client_num) {
case ILT_CLIENT_CDU:
start_reg = PXP2_REG_PSWRQ_CDU0_L2P;
break;
case ILT_CLIENT_QM:
start_reg = PXP2_REG_PSWRQ_QM0_L2P;
break;
case ILT_CLIENT_SRC:
start_reg = PXP2_REG_PSWRQ_SRC0_L2P;
break;
case ILT_CLIENT_TM:
start_reg = PXP2_REG_PSWRQ_TM0_L2P;
break;
}
REG_WR(bp, start_reg + BP_FUNC(bp)*4,
ILT_RANGE((ilt_start + ilt_cli->start),
(ilt_start + ilt_cli->end)));
} else {
switch (ilt_cli->client_num) {
case ILT_CLIENT_CDU:
start_reg = PXP2_REG_RQ_CDU_FIRST_ILT;
end_reg = PXP2_REG_RQ_CDU_LAST_ILT;
break;
case ILT_CLIENT_QM:
start_reg = PXP2_REG_RQ_QM_FIRST_ILT;
end_reg = PXP2_REG_RQ_QM_LAST_ILT;
break;
case ILT_CLIENT_SRC:
start_reg = PXP2_REG_RQ_SRC_FIRST_ILT;
end_reg = PXP2_REG_RQ_SRC_LAST_ILT;
break;
case ILT_CLIENT_TM:
start_reg = PXP2_REG_RQ_TM_FIRST_ILT;
end_reg = PXP2_REG_RQ_TM_LAST_ILT;
break;
}
REG_WR(bp, start_reg, (ilt_start + ilt_cli->start));
REG_WR(bp, end_reg, (ilt_start + ilt_cli->end));
}
}
static void bnx2x_ilt_client_init_op_ilt(struct bnx2x *bp,
struct bnx2x_ilt *ilt,
struct ilt_client_info *ilt_cli,
u8 initop)
{
int i;
if (ilt_cli->flags & ILT_CLIENT_SKIP_INIT)
return;
for (i = ilt_cli->start; i <= ilt_cli->end; i++)
bnx2x_ilt_line_init_op(bp, ilt, i, initop);
/* init/clear the ILT boundries */
bnx2x_ilt_boundry_init_op(bp, ilt_cli, ilt->start_line, initop);
}
static void bnx2x_ilt_client_init_op(struct bnx2x *bp,
struct ilt_client_info *ilt_cli, u8 initop)
{
struct bnx2x_ilt *ilt = BP_ILT(bp);
bnx2x_ilt_client_init_op_ilt(bp, ilt, ilt_cli, initop);
}
static void bnx2x_ilt_client_id_init_op(struct bnx2x *bp,
int cli_num, u8 initop)
{
struct bnx2x_ilt *ilt = BP_ILT(bp);
struct ilt_client_info *ilt_cli = &ilt->clients[cli_num];
bnx2x_ilt_client_init_op(bp, ilt_cli, initop);
}
static void bnx2x_ilt_init_op(struct bnx2x *bp, u8 initop)
{
bnx2x_ilt_client_id_init_op(bp, ILT_CLIENT_CDU, initop);
bnx2x_ilt_client_id_init_op(bp, ILT_CLIENT_QM, initop);
bnx2x_ilt_client_id_init_op(bp, ILT_CLIENT_SRC, initop);
bnx2x_ilt_client_id_init_op(bp, ILT_CLIENT_TM, initop);
}
static void bnx2x_ilt_init_client_psz(struct bnx2x *bp, int cli_num,
u32 psz_reg, u8 initop)
{
struct bnx2x_ilt *ilt = BP_ILT(bp);
struct ilt_client_info *ilt_cli = &ilt->clients[cli_num];
if (ilt_cli->flags & ILT_CLIENT_SKIP_INIT)
return;
switch (initop) {
case INITOP_INIT:
/* set in the init-value array */
case INITOP_SET:
REG_WR(bp, psz_reg, ILOG2(ilt_cli->page_size >> 12));
break;
case INITOP_CLEAR:
break;
}
}
/*
* called during init common stage, ilt clients should be initialized
* prioir to calling this function
*/
static void bnx2x_ilt_init_page_size(struct bnx2x *bp, u8 initop)
{
bnx2x_ilt_init_client_psz(bp, ILT_CLIENT_CDU,
PXP2_REG_RQ_CDU_P_SIZE, initop);
bnx2x_ilt_init_client_psz(bp, ILT_CLIENT_QM,
PXP2_REG_RQ_QM_P_SIZE, initop);
bnx2x_ilt_init_client_psz(bp, ILT_CLIENT_SRC,
PXP2_REG_RQ_SRC_P_SIZE, initop);
bnx2x_ilt_init_client_psz(bp, ILT_CLIENT_TM,
PXP2_REG_RQ_TM_P_SIZE, initop);
}
/****************************************************************************
* QM initializations
****************************************************************************/
#define QM_QUEUES_PER_FUNC 16 /* E1 has 32, but only 16 are used */
#define QM_INIT_MIN_CID_COUNT 31
#define QM_INIT(cid_cnt) (cid_cnt > QM_INIT_MIN_CID_COUNT)
/* called during init port stage */
static void bnx2x_qm_init_cid_count(struct bnx2x *bp, int qm_cid_count,
u8 initop)
{
int port = BP_PORT(bp);
if (QM_INIT(qm_cid_count)) {
switch (initop) {
case INITOP_INIT:
/* set in the init-value array */
case INITOP_SET:
REG_WR(bp, QM_REG_CONNNUM_0 + port*4,
qm_cid_count/16 - 1);
break;
case INITOP_CLEAR:
break;
}
}
}
static void bnx2x_qm_set_ptr_table(struct bnx2x *bp, int qm_cid_count)
{
int i;
u32 wb_data[2];
wb_data[0] = wb_data[1] = 0;
for (i = 0; i < 4 * QM_QUEUES_PER_FUNC; i++) {
REG_WR(bp, QM_REG_BASEADDR + i*4,
qm_cid_count * 4 * (i % QM_QUEUES_PER_FUNC));
bnx2x_init_ind_wr(bp, QM_REG_PTRTBL + i*8,
wb_data, 2);
if (CHIP_IS_E1H(bp)) {
REG_WR(bp, QM_REG_BASEADDR_EXT_A + i*4,
qm_cid_count * 4 * (i % QM_QUEUES_PER_FUNC));
bnx2x_init_ind_wr(bp, QM_REG_PTRTBL_EXT_A + i*8,
wb_data, 2);
}
}
}
/* called during init common stage */
static void bnx2x_qm_init_ptr_table(struct bnx2x *bp, int qm_cid_count,
u8 initop)
{
if (!QM_INIT(qm_cid_count))
return;
switch (initop) {
case INITOP_INIT:
/* set in the init-value array */
case INITOP_SET:
bnx2x_qm_set_ptr_table(bp, qm_cid_count);
break;
case INITOP_CLEAR:
break;
}
}
/****************************************************************************
* SRC initializations
****************************************************************************/
#ifdef BCM_CNIC
/* called during init func stage */
static void bnx2x_src_init_t2(struct bnx2x *bp, struct src_ent *t2,
dma_addr_t t2_mapping, int src_cid_count)
{
int i;
int port = BP_PORT(bp);
/* Initialize T2 */
for (i = 0; i < src_cid_count-1; i++)
t2[i].next = (u64)(t2_mapping + (i+1)*sizeof(struct src_ent));
/* tell the searcher where the T2 table is */
REG_WR(bp, SRC_REG_COUNTFREE0 + port*4, src_cid_count);
bnx2x_wr_64(bp, SRC_REG_FIRSTFREE0 + port*16,
U64_LO(t2_mapping), U64_HI(t2_mapping));
bnx2x_wr_64(bp, SRC_REG_LASTFREE0 + port*16,
U64_LO((u64)t2_mapping +
(src_cid_count-1) * sizeof(struct src_ent)),
U64_HI((u64)t2_mapping +
(src_cid_count-1) * sizeof(struct src_ent)));
}
#endif
#endif /* BNX2X_INIT_OPS_H */