1
0
Fork 0
mirror of synced 2025-03-06 20:59:54 +01:00
linux/drivers/net/ethernet/qlogic/qed/qed_hw.c
Konstantin Khorenko e346e231b4 qed: Fix scheduling in a tasklet while getting stats
Here we've got to a situation when tasklet called usleep_range() in PTT
acquire logic, thus welcome to the "scheduling while atomic" BUG().

  BUG: scheduling while atomic: swapper/24/0/0x00000100

   [<ffffffffb41c6199>] schedule+0x29/0x70
   [<ffffffffb41c5512>] schedule_hrtimeout_range_clock+0xb2/0x150
   [<ffffffffb41c55c3>] schedule_hrtimeout_range+0x13/0x20
   [<ffffffffb41c3bcf>] usleep_range+0x4f/0x70
   [<ffffffffc08d3e58>] qed_ptt_acquire+0x38/0x100 [qed]
   [<ffffffffc08eac48>] _qed_get_vport_stats+0x458/0x580 [qed]
   [<ffffffffc08ead8c>] qed_get_vport_stats+0x1c/0xd0 [qed]
   [<ffffffffc08dffd3>] qed_get_protocol_stats+0x93/0x100 [qed]
                        qed_mcp_send_protocol_stats
            case MFW_DRV_MSG_GET_LAN_STATS:
            case MFW_DRV_MSG_GET_FCOE_STATS:
            case MFW_DRV_MSG_GET_ISCSI_STATS:
            case MFW_DRV_MSG_GET_RDMA_STATS:
   [<ffffffffc08e36d8>] qed_mcp_handle_events+0x2d8/0x890 [qed]
                        qed_int_assertion
                        qed_int_attentions
   [<ffffffffc08d9490>] qed_int_sp_dpc+0xa50/0xdc0 [qed]
   [<ffffffffb3aa7623>] tasklet_action+0x83/0x140
   [<ffffffffb41d9125>] __do_softirq+0x125/0x2bb
   [<ffffffffb41d560c>] call_softirq+0x1c/0x30
   [<ffffffffb3a30645>] do_softirq+0x65/0xa0
   [<ffffffffb3aa78d5>] irq_exit+0x105/0x110
   [<ffffffffb41d8996>] do_IRQ+0x56/0xf0

Fix this by making caller to provide the context whether it could be in
atomic context flow or not when getting stats from QED driver.
QED driver based on the context provided decide to schedule out or not
when acquiring the PTT BAR window.

We faced the BUG_ON() while getting vport stats, but according to the
code same issue could happen for fcoe and iscsi statistics as well, so
fixing them too.

Fixes: 6c75424612 ("qed: Add support for NCSI statistics.")
Fixes: 1e128c8129 ("qed: Add support for hardware offloaded FCoE.")
Fixes: 2f2b2614e8 ("qed: Provide iSCSI statistics to management")
Cc: Sudarsana Kalluru <skalluru@marvell.com>
Cc: David Miller <davem@davemloft.net>
Cc: Manish Chopra <manishc@marvell.com>

Signed-off-by: Konstantin Khorenko <khorenko@virtuozzo.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-07-29 17:09:18 +01:00

932 lines
25 KiB
C

// SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
/* QLogic qed NIC Driver
* Copyright (c) 2015-2017 QLogic Corporation
* Copyright (c) 2019-2020 Marvell International Ltd.
*/
#include <linux/types.h>
#include <linux/io.h>
#include <linux/delay.h>
#include <linux/dma-mapping.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/mutex.h>
#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/string.h>
#include <linux/qed/qed_chain.h>
#include "qed.h"
#include "qed_hsi.h"
#include "qed_hw.h"
#include "qed_reg_addr.h"
#include "qed_sriov.h"
#define QED_BAR_ACQUIRE_TIMEOUT_USLEEP_CNT 1000
#define QED_BAR_ACQUIRE_TIMEOUT_USLEEP 1000
#define QED_BAR_ACQUIRE_TIMEOUT_UDELAY_CNT 100000
#define QED_BAR_ACQUIRE_TIMEOUT_UDELAY 10
/* Invalid values */
#define QED_BAR_INVALID_OFFSET (cpu_to_le32(-1))
struct qed_ptt {
struct list_head list_entry;
unsigned int idx;
struct pxp_ptt_entry pxp;
u8 hwfn_id;
};
struct qed_ptt_pool {
struct list_head free_list;
spinlock_t lock; /* ptt synchronized access */
struct qed_ptt ptts[PXP_EXTERNAL_BAR_PF_WINDOW_NUM];
};
int qed_ptt_pool_alloc(struct qed_hwfn *p_hwfn)
{
struct qed_ptt_pool *p_pool = kmalloc(sizeof(*p_pool), GFP_KERNEL);
int i;
if (!p_pool)
return -ENOMEM;
INIT_LIST_HEAD(&p_pool->free_list);
for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
p_pool->ptts[i].idx = i;
p_pool->ptts[i].pxp.offset = QED_BAR_INVALID_OFFSET;
p_pool->ptts[i].pxp.pretend.control = 0;
p_pool->ptts[i].hwfn_id = p_hwfn->my_id;
if (i >= RESERVED_PTT_MAX)
list_add(&p_pool->ptts[i].list_entry,
&p_pool->free_list);
}
p_hwfn->p_ptt_pool = p_pool;
spin_lock_init(&p_pool->lock);
return 0;
}
void qed_ptt_invalidate(struct qed_hwfn *p_hwfn)
{
struct qed_ptt *p_ptt;
int i;
for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
p_ptt = &p_hwfn->p_ptt_pool->ptts[i];
p_ptt->pxp.offset = QED_BAR_INVALID_OFFSET;
}
}
void qed_ptt_pool_free(struct qed_hwfn *p_hwfn)
{
kfree(p_hwfn->p_ptt_pool);
p_hwfn->p_ptt_pool = NULL;
}
struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn)
{
return qed_ptt_acquire_context(p_hwfn, false);
}
struct qed_ptt *qed_ptt_acquire_context(struct qed_hwfn *p_hwfn, bool is_atomic)
{
struct qed_ptt *p_ptt;
unsigned int i, count;
if (is_atomic)
count = QED_BAR_ACQUIRE_TIMEOUT_UDELAY_CNT;
else
count = QED_BAR_ACQUIRE_TIMEOUT_USLEEP_CNT;
/* Take the free PTT from the list */
for (i = 0; i < count; i++) {
spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
if (!list_empty(&p_hwfn->p_ptt_pool->free_list)) {
p_ptt = list_first_entry(&p_hwfn->p_ptt_pool->free_list,
struct qed_ptt, list_entry);
list_del(&p_ptt->list_entry);
spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
"allocated ptt %d\n", p_ptt->idx);
return p_ptt;
}
spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
if (is_atomic)
udelay(QED_BAR_ACQUIRE_TIMEOUT_UDELAY);
else
usleep_range(QED_BAR_ACQUIRE_TIMEOUT_USLEEP,
QED_BAR_ACQUIRE_TIMEOUT_USLEEP * 2);
}
DP_NOTICE(p_hwfn, "PTT acquire timeout - failed to allocate PTT\n");
return NULL;
}
void qed_ptt_release(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
{
spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
list_add(&p_ptt->list_entry, &p_hwfn->p_ptt_pool->free_list);
spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
}
u32 qed_ptt_get_hw_addr(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
{
/* The HW is using DWORDS and we need to translate it to Bytes */
return le32_to_cpu(p_ptt->pxp.offset) << 2;
}
static u32 qed_ptt_config_addr(struct qed_ptt *p_ptt)
{
return PXP_PF_WINDOW_ADMIN_PER_PF_START +
p_ptt->idx * sizeof(struct pxp_ptt_entry);
}
u32 qed_ptt_get_bar_addr(struct qed_ptt *p_ptt)
{
return PXP_EXTERNAL_BAR_PF_WINDOW_START +
p_ptt->idx * PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE;
}
void qed_ptt_set_win(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, u32 new_hw_addr)
{
u32 prev_hw_addr;
prev_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
if (new_hw_addr == prev_hw_addr)
return;
/* Update PTT entery in admin window */
DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
"Updating PTT entry %d to offset 0x%x\n",
p_ptt->idx, new_hw_addr);
/* The HW is using DWORDS and the address is in Bytes */
p_ptt->pxp.offset = cpu_to_le32(new_hw_addr >> 2);
REG_WR(p_hwfn,
qed_ptt_config_addr(p_ptt) +
offsetof(struct pxp_ptt_entry, offset),
le32_to_cpu(p_ptt->pxp.offset));
}
static u32 qed_set_ptt(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, u32 hw_addr)
{
u32 win_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
u32 offset;
offset = hw_addr - win_hw_addr;
if (p_ptt->hwfn_id != p_hwfn->my_id)
DP_NOTICE(p_hwfn,
"ptt[%d] of hwfn[%02x] is used by hwfn[%02x]!\n",
p_ptt->idx, p_ptt->hwfn_id, p_hwfn->my_id);
/* Verify the address is within the window */
if (hw_addr < win_hw_addr ||
offset >= PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE) {
qed_ptt_set_win(p_hwfn, p_ptt, hw_addr);
offset = 0;
}
return qed_ptt_get_bar_addr(p_ptt) + offset;
}
struct qed_ptt *qed_get_reserved_ptt(struct qed_hwfn *p_hwfn,
enum reserved_ptts ptt_idx)
{
if (ptt_idx >= RESERVED_PTT_MAX) {
DP_NOTICE(p_hwfn,
"Requested PTT %d is out of range\n", ptt_idx);
return NULL;
}
return &p_hwfn->p_ptt_pool->ptts[ptt_idx];
}
void qed_wr(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
u32 hw_addr, u32 val)
{
u32 bar_addr = qed_set_ptt(p_hwfn, p_ptt, hw_addr);
REG_WR(p_hwfn, bar_addr, val);
DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
"bar_addr 0x%x, hw_addr 0x%x, val 0x%x\n",
bar_addr, hw_addr, val);
}
u32 qed_rd(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
u32 hw_addr)
{
u32 bar_addr = qed_set_ptt(p_hwfn, p_ptt, hw_addr);
u32 val = REG_RD(p_hwfn, bar_addr);
DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
"bar_addr 0x%x, hw_addr 0x%x, val 0x%x\n",
bar_addr, hw_addr, val);
return val;
}
static void qed_memcpy_hw(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
void *addr, u32 hw_addr, size_t n, bool to_device)
{
u32 dw_count, *host_addr, hw_offset;
size_t quota, done = 0;
u32 __iomem *reg_addr;
while (done < n) {
quota = min_t(size_t, n - done,
PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE);
if (IS_PF(p_hwfn->cdev)) {
qed_ptt_set_win(p_hwfn, p_ptt, hw_addr + done);
hw_offset = qed_ptt_get_bar_addr(p_ptt);
} else {
hw_offset = hw_addr + done;
}
dw_count = quota / 4;
host_addr = (u32 *)((u8 *)addr + done);
reg_addr = (u32 __iomem *)REG_ADDR(p_hwfn, hw_offset);
if (to_device)
while (dw_count--)
DIRECT_REG_WR(reg_addr++, *host_addr++);
else
while (dw_count--)
*host_addr++ = DIRECT_REG_RD(reg_addr++);
done += quota;
}
}
void qed_memcpy_from(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, void *dest, u32 hw_addr, size_t n)
{
DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
"hw_addr 0x%x, dest %p hw_addr 0x%x, size %lu\n",
hw_addr, dest, hw_addr, (unsigned long)n);
qed_memcpy_hw(p_hwfn, p_ptt, dest, hw_addr, n, false);
}
void qed_memcpy_to(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, u32 hw_addr, void *src, size_t n)
{
DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
"hw_addr 0x%x, hw_addr 0x%x, src %p size %lu\n",
hw_addr, hw_addr, src, (unsigned long)n);
qed_memcpy_hw(p_hwfn, p_ptt, src, hw_addr, n, true);
}
void qed_fid_pretend(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt, u16 fid)
{
u16 control = 0;
SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
/* Every pretend undos previous pretends, including
* previous port pretend.
*/
SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
p_ptt->pxp.pretend.control = cpu_to_le16(control);
p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
REG_WR(p_hwfn,
qed_ptt_config_addr(p_ptt) +
offsetof(struct pxp_ptt_entry, pretend),
*(u32 *)&p_ptt->pxp.pretend);
}
void qed_port_pretend(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, u8 port_id)
{
u16 control = 0;
SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
p_ptt->pxp.pretend.control = cpu_to_le16(control);
REG_WR(p_hwfn,
qed_ptt_config_addr(p_ptt) +
offsetof(struct pxp_ptt_entry, pretend),
*(u32 *)&p_ptt->pxp.pretend);
}
void qed_port_unpretend(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
{
u16 control = 0;
SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
p_ptt->pxp.pretend.control = cpu_to_le16(control);
REG_WR(p_hwfn,
qed_ptt_config_addr(p_ptt) +
offsetof(struct pxp_ptt_entry, pretend),
*(u32 *)&p_ptt->pxp.pretend);
}
void qed_port_fid_pretend(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, u8 port_id, u16 fid)
{
u16 control = 0;
SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
p_ptt->pxp.pretend.control = cpu_to_le16(control);
p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
REG_WR(p_hwfn,
qed_ptt_config_addr(p_ptt) +
offsetof(struct pxp_ptt_entry, pretend),
*(u32 *)&p_ptt->pxp.pretend);
}
u32 qed_vfid_to_concrete(struct qed_hwfn *p_hwfn, u8 vfid)
{
u32 concrete_fid = 0;
SET_FIELD(concrete_fid, PXP_CONCRETE_FID_PFID, p_hwfn->rel_pf_id);
SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFID, vfid);
SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFVALID, 1);
return concrete_fid;
}
/* DMAE */
#define QED_DMAE_FLAGS_IS_SET(params, flag) \
((params) != NULL && GET_FIELD((params)->flags, QED_DMAE_PARAMS_##flag))
static void qed_dmae_opcode(struct qed_hwfn *p_hwfn,
const u8 is_src_type_grc,
const u8 is_dst_type_grc,
struct qed_dmae_params *p_params)
{
u8 src_pfid, dst_pfid, port_id;
u16 opcode_b = 0;
u32 opcode = 0;
/* Whether the source is the PCIe or the GRC.
* 0- The source is the PCIe
* 1- The source is the GRC.
*/
SET_FIELD(opcode, DMAE_CMD_SRC,
(is_src_type_grc ? dmae_cmd_src_grc : dmae_cmd_src_pcie));
src_pfid = QED_DMAE_FLAGS_IS_SET(p_params, SRC_PF_VALID) ?
p_params->src_pfid : p_hwfn->rel_pf_id;
SET_FIELD(opcode, DMAE_CMD_SRC_PF_ID, src_pfid);
/* The destination of the DMA can be: 0-None 1-PCIe 2-GRC 3-None */
SET_FIELD(opcode, DMAE_CMD_DST,
(is_dst_type_grc ? dmae_cmd_dst_grc : dmae_cmd_dst_pcie));
dst_pfid = QED_DMAE_FLAGS_IS_SET(p_params, DST_PF_VALID) ?
p_params->dst_pfid : p_hwfn->rel_pf_id;
SET_FIELD(opcode, DMAE_CMD_DST_PF_ID, dst_pfid);
/* Whether to write a completion word to the completion destination:
* 0-Do not write a completion word
* 1-Write the completion word
*/
SET_FIELD(opcode, DMAE_CMD_COMP_WORD_EN, 1);
SET_FIELD(opcode, DMAE_CMD_SRC_ADDR_RESET, 1);
if (QED_DMAE_FLAGS_IS_SET(p_params, COMPLETION_DST))
SET_FIELD(opcode, DMAE_CMD_COMP_FUNC, 1);
/* swapping mode 3 - big endian */
SET_FIELD(opcode, DMAE_CMD_ENDIANITY_MODE, DMAE_CMD_ENDIANITY);
port_id = (QED_DMAE_FLAGS_IS_SET(p_params, PORT_VALID)) ?
p_params->port_id : p_hwfn->port_id;
SET_FIELD(opcode, DMAE_CMD_PORT_ID, port_id);
/* reset source address in next go */
SET_FIELD(opcode, DMAE_CMD_SRC_ADDR_RESET, 1);
/* reset dest address in next go */
SET_FIELD(opcode, DMAE_CMD_DST_ADDR_RESET, 1);
/* SRC/DST VFID: all 1's - pf, otherwise VF id */
if (QED_DMAE_FLAGS_IS_SET(p_params, SRC_VF_VALID)) {
SET_FIELD(opcode, DMAE_CMD_SRC_VF_ID_VALID, 1);
SET_FIELD(opcode_b, DMAE_CMD_SRC_VF_ID, p_params->src_vfid);
} else {
SET_FIELD(opcode_b, DMAE_CMD_SRC_VF_ID, 0xFF);
}
if (QED_DMAE_FLAGS_IS_SET(p_params, DST_VF_VALID)) {
SET_FIELD(opcode, DMAE_CMD_DST_VF_ID_VALID, 1);
SET_FIELD(opcode_b, DMAE_CMD_DST_VF_ID, p_params->dst_vfid);
} else {
SET_FIELD(opcode_b, DMAE_CMD_DST_VF_ID, 0xFF);
}
p_hwfn->dmae_info.p_dmae_cmd->opcode = cpu_to_le32(opcode);
p_hwfn->dmae_info.p_dmae_cmd->opcode_b = cpu_to_le16(opcode_b);
}
u32 qed_dmae_idx_to_go_cmd(u8 idx)
{
/* All the DMAE 'go' registers form an array in internal memory */
return DMAE_REG_GO_C0 + (idx << 2);
}
static int qed_dmae_post_command(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt)
{
struct dmae_cmd *p_command = p_hwfn->dmae_info.p_dmae_cmd;
u8 idx_cmd = p_hwfn->dmae_info.channel, i;
int qed_status = 0;
/* verify address is not NULL */
if ((((!p_command->dst_addr_lo) && (!p_command->dst_addr_hi)) ||
((!p_command->src_addr_lo) && (!p_command->src_addr_hi)))) {
DP_NOTICE(p_hwfn,
"source or destination address 0 idx_cmd=%d\n"
"opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
idx_cmd,
le32_to_cpu(p_command->opcode),
le16_to_cpu(p_command->opcode_b),
le16_to_cpu(p_command->length_dw),
le32_to_cpu(p_command->src_addr_hi),
le32_to_cpu(p_command->src_addr_lo),
le32_to_cpu(p_command->dst_addr_hi),
le32_to_cpu(p_command->dst_addr_lo));
return -EINVAL;
}
DP_VERBOSE(p_hwfn,
NETIF_MSG_HW,
"Posting DMAE command [idx %d]: opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
idx_cmd,
le32_to_cpu(p_command->opcode),
le16_to_cpu(p_command->opcode_b),
le16_to_cpu(p_command->length_dw),
le32_to_cpu(p_command->src_addr_hi),
le32_to_cpu(p_command->src_addr_lo),
le32_to_cpu(p_command->dst_addr_hi),
le32_to_cpu(p_command->dst_addr_lo));
/* Copy the command to DMAE - need to do it before every call
* for source/dest address no reset.
* The first 9 DWs are the command registers, the 10 DW is the
* GO register, and the rest are result registers
* (which are read only by the client).
*/
for (i = 0; i < DMAE_CMD_SIZE; i++) {
u32 data = (i < DMAE_CMD_SIZE_TO_FILL) ?
*(((u32 *)p_command) + i) : 0;
qed_wr(p_hwfn, p_ptt,
DMAE_REG_CMD_MEM +
(idx_cmd * DMAE_CMD_SIZE * sizeof(u32)) +
(i * sizeof(u32)), data);
}
qed_wr(p_hwfn, p_ptt, qed_dmae_idx_to_go_cmd(idx_cmd), DMAE_GO_VALUE);
return qed_status;
}
int qed_dmae_info_alloc(struct qed_hwfn *p_hwfn)
{
dma_addr_t *p_addr = &p_hwfn->dmae_info.completion_word_phys_addr;
struct dmae_cmd **p_cmd = &p_hwfn->dmae_info.p_dmae_cmd;
u32 **p_buff = &p_hwfn->dmae_info.p_intermediate_buffer;
u32 **p_comp = &p_hwfn->dmae_info.p_completion_word;
*p_comp = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
sizeof(u32), p_addr, GFP_KERNEL);
if (!*p_comp)
goto err;
p_addr = &p_hwfn->dmae_info.dmae_cmd_phys_addr;
*p_cmd = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
sizeof(struct dmae_cmd),
p_addr, GFP_KERNEL);
if (!*p_cmd)
goto err;
p_addr = &p_hwfn->dmae_info.intermediate_buffer_phys_addr;
*p_buff = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
sizeof(u32) * DMAE_MAX_RW_SIZE,
p_addr, GFP_KERNEL);
if (!*p_buff)
goto err;
p_hwfn->dmae_info.channel = p_hwfn->rel_pf_id;
return 0;
err:
qed_dmae_info_free(p_hwfn);
return -ENOMEM;
}
void qed_dmae_info_free(struct qed_hwfn *p_hwfn)
{
dma_addr_t p_phys;
/* Just make sure no one is in the middle */
mutex_lock(&p_hwfn->dmae_info.mutex);
if (p_hwfn->dmae_info.p_completion_word) {
p_phys = p_hwfn->dmae_info.completion_word_phys_addr;
dma_free_coherent(&p_hwfn->cdev->pdev->dev,
sizeof(u32),
p_hwfn->dmae_info.p_completion_word, p_phys);
p_hwfn->dmae_info.p_completion_word = NULL;
}
if (p_hwfn->dmae_info.p_dmae_cmd) {
p_phys = p_hwfn->dmae_info.dmae_cmd_phys_addr;
dma_free_coherent(&p_hwfn->cdev->pdev->dev,
sizeof(struct dmae_cmd),
p_hwfn->dmae_info.p_dmae_cmd, p_phys);
p_hwfn->dmae_info.p_dmae_cmd = NULL;
}
if (p_hwfn->dmae_info.p_intermediate_buffer) {
p_phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
dma_free_coherent(&p_hwfn->cdev->pdev->dev,
sizeof(u32) * DMAE_MAX_RW_SIZE,
p_hwfn->dmae_info.p_intermediate_buffer,
p_phys);
p_hwfn->dmae_info.p_intermediate_buffer = NULL;
}
mutex_unlock(&p_hwfn->dmae_info.mutex);
}
static int qed_dmae_operation_wait(struct qed_hwfn *p_hwfn)
{
u32 wait_cnt_limit = 10000, wait_cnt = 0;
int qed_status = 0;
barrier();
while (*p_hwfn->dmae_info.p_completion_word != DMAE_COMPLETION_VAL) {
udelay(DMAE_MIN_WAIT_TIME);
if (++wait_cnt > wait_cnt_limit) {
DP_NOTICE(p_hwfn->cdev,
"Timed-out waiting for operation to complete. Completion word is 0x%08x expected 0x%08x.\n",
*p_hwfn->dmae_info.p_completion_word,
DMAE_COMPLETION_VAL);
qed_status = -EBUSY;
break;
}
/* to sync the completion_word since we are not
* using the volatile keyword for p_completion_word
*/
barrier();
}
if (qed_status == 0)
*p_hwfn->dmae_info.p_completion_word = 0;
return qed_status;
}
static int qed_dmae_execute_sub_operation(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
u64 src_addr,
u64 dst_addr,
u8 src_type,
u8 dst_type,
u32 length_dw)
{
dma_addr_t phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
int qed_status = 0;
switch (src_type) {
case QED_DMAE_ADDRESS_GRC:
case QED_DMAE_ADDRESS_HOST_PHYS:
cmd->src_addr_hi = cpu_to_le32(upper_32_bits(src_addr));
cmd->src_addr_lo = cpu_to_le32(lower_32_bits(src_addr));
break;
/* for virtual source addresses we use the intermediate buffer. */
case QED_DMAE_ADDRESS_HOST_VIRT:
cmd->src_addr_hi = cpu_to_le32(upper_32_bits(phys));
cmd->src_addr_lo = cpu_to_le32(lower_32_bits(phys));
memcpy(&p_hwfn->dmae_info.p_intermediate_buffer[0],
(void *)(uintptr_t)src_addr,
length_dw * sizeof(u32));
break;
default:
return -EINVAL;
}
switch (dst_type) {
case QED_DMAE_ADDRESS_GRC:
case QED_DMAE_ADDRESS_HOST_PHYS:
cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(dst_addr));
cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(dst_addr));
break;
/* for virtual source addresses we use the intermediate buffer. */
case QED_DMAE_ADDRESS_HOST_VIRT:
cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(phys));
cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(phys));
break;
default:
return -EINVAL;
}
cmd->length_dw = cpu_to_le16((u16)length_dw);
qed_dmae_post_command(p_hwfn, p_ptt);
qed_status = qed_dmae_operation_wait(p_hwfn);
if (qed_status) {
DP_NOTICE(p_hwfn,
"qed_dmae_host2grc: Wait Failed. source_addr 0x%llx, grc_addr 0x%llx, size_in_dwords 0x%x\n",
src_addr, dst_addr, length_dw);
return qed_status;
}
if (dst_type == QED_DMAE_ADDRESS_HOST_VIRT)
memcpy((void *)(uintptr_t)(dst_addr),
&p_hwfn->dmae_info.p_intermediate_buffer[0],
length_dw * sizeof(u32));
return 0;
}
static int qed_dmae_execute_command(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
u64 src_addr, u64 dst_addr,
u8 src_type, u8 dst_type,
u32 size_in_dwords,
struct qed_dmae_params *p_params)
{
dma_addr_t phys = p_hwfn->dmae_info.completion_word_phys_addr;
u16 length_cur = 0, i = 0, cnt_split = 0, length_mod = 0;
struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
u64 src_addr_split = 0, dst_addr_split = 0;
u16 length_limit = DMAE_MAX_RW_SIZE;
int qed_status = 0;
u32 offset = 0;
if (p_hwfn->cdev->recov_in_prog) {
DP_VERBOSE(p_hwfn,
NETIF_MSG_HW,
"Recovery is in progress. Avoid DMAE transaction [{src: addr 0x%llx, type %d}, {dst: addr 0x%llx, type %d}, size %d].\n",
src_addr, src_type, dst_addr, dst_type,
size_in_dwords);
/* Let the flow complete w/o any error handling */
return 0;
}
qed_dmae_opcode(p_hwfn,
(src_type == QED_DMAE_ADDRESS_GRC),
(dst_type == QED_DMAE_ADDRESS_GRC),
p_params);
cmd->comp_addr_lo = cpu_to_le32(lower_32_bits(phys));
cmd->comp_addr_hi = cpu_to_le32(upper_32_bits(phys));
cmd->comp_val = cpu_to_le32(DMAE_COMPLETION_VAL);
/* Check if the grc_addr is valid like < MAX_GRC_OFFSET */
cnt_split = size_in_dwords / length_limit;
length_mod = size_in_dwords % length_limit;
src_addr_split = src_addr;
dst_addr_split = dst_addr;
for (i = 0; i <= cnt_split; i++) {
offset = length_limit * i;
if (!QED_DMAE_FLAGS_IS_SET(p_params, RW_REPL_SRC)) {
if (src_type == QED_DMAE_ADDRESS_GRC)
src_addr_split = src_addr + offset;
else
src_addr_split = src_addr + (offset * 4);
}
if (dst_type == QED_DMAE_ADDRESS_GRC)
dst_addr_split = dst_addr + offset;
else
dst_addr_split = dst_addr + (offset * 4);
length_cur = (cnt_split == i) ? length_mod : length_limit;
/* might be zero on last iteration */
if (!length_cur)
continue;
qed_status = qed_dmae_execute_sub_operation(p_hwfn,
p_ptt,
src_addr_split,
dst_addr_split,
src_type,
dst_type,
length_cur);
if (qed_status) {
qed_hw_err_notify(p_hwfn, p_ptt, QED_HW_ERR_DMAE_FAIL,
"qed_dmae_execute_sub_operation Failed with error 0x%x. source_addr 0x%llx, destination addr 0x%llx, size_in_dwords 0x%x\n",
qed_status, src_addr,
dst_addr, length_cur);
break;
}
}
return qed_status;
}
int qed_dmae_host2grc(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
u64 source_addr, u32 grc_addr, u32 size_in_dwords,
struct qed_dmae_params *p_params)
{
u32 grc_addr_in_dw = grc_addr / sizeof(u32);
int rc;
mutex_lock(&p_hwfn->dmae_info.mutex);
rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
grc_addr_in_dw,
QED_DMAE_ADDRESS_HOST_VIRT,
QED_DMAE_ADDRESS_GRC,
size_in_dwords, p_params);
mutex_unlock(&p_hwfn->dmae_info.mutex);
return rc;
}
int qed_dmae_grc2host(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
u32 grc_addr,
dma_addr_t dest_addr, u32 size_in_dwords,
struct qed_dmae_params *p_params)
{
u32 grc_addr_in_dw = grc_addr / sizeof(u32);
int rc;
mutex_lock(&p_hwfn->dmae_info.mutex);
rc = qed_dmae_execute_command(p_hwfn, p_ptt, grc_addr_in_dw,
dest_addr, QED_DMAE_ADDRESS_GRC,
QED_DMAE_ADDRESS_HOST_VIRT,
size_in_dwords, p_params);
mutex_unlock(&p_hwfn->dmae_info.mutex);
return rc;
}
int qed_dmae_host2host(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
dma_addr_t source_addr,
dma_addr_t dest_addr,
u32 size_in_dwords, struct qed_dmae_params *p_params)
{
int rc;
mutex_lock(&(p_hwfn->dmae_info.mutex));
rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
dest_addr,
QED_DMAE_ADDRESS_HOST_PHYS,
QED_DMAE_ADDRESS_HOST_PHYS,
size_in_dwords, p_params);
mutex_unlock(&(p_hwfn->dmae_info.mutex));
return rc;
}
void qed_hw_err_notify(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
enum qed_hw_err_type err_type, const char *fmt, ...)
{
char buf[QED_HW_ERR_MAX_STR_SIZE];
va_list vl;
int len;
if (fmt) {
va_start(vl, fmt);
len = vsnprintf(buf, QED_HW_ERR_MAX_STR_SIZE, fmt, vl);
va_end(vl);
if (len > QED_HW_ERR_MAX_STR_SIZE - 1)
len = QED_HW_ERR_MAX_STR_SIZE - 1;
DP_NOTICE(p_hwfn, "%s", buf);
}
/* Fan failure cannot be masked by handling of another HW error */
if (p_hwfn->cdev->recov_in_prog &&
err_type != QED_HW_ERR_FAN_FAIL) {
DP_VERBOSE(p_hwfn,
NETIF_MSG_DRV,
"Recovery is in progress. Avoid notifying about HW error %d.\n",
err_type);
return;
}
qed_hw_error_occurred(p_hwfn, err_type);
if (fmt)
qed_mcp_send_raw_debug_data(p_hwfn, p_ptt, buf, len);
}
int qed_dmae_sanity(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, const char *phase)
{
u32 size = PAGE_SIZE / 2, val;
int rc = 0;
dma_addr_t p_phys;
void *p_virt;
u32 *p_tmp;
p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2 * size, &p_phys, GFP_KERNEL);
if (!p_virt) {
DP_NOTICE(p_hwfn,
"DMAE sanity [%s]: failed to allocate memory\n",
phase);
return -ENOMEM;
}
/* Fill the bottom half of the allocated memory with a known pattern */
for (p_tmp = (u32 *)p_virt;
p_tmp < (u32 *)((u8 *)p_virt + size); p_tmp++) {
/* Save the address itself as the value */
val = (u32)(uintptr_t)p_tmp;
*p_tmp = val;
}
/* Zero the top half of the allocated memory */
memset((u8 *)p_virt + size, 0, size);
DP_VERBOSE(p_hwfn,
QED_MSG_SP,
"DMAE sanity [%s]: src_addr={phys 0x%llx, virt %p}, dst_addr={phys 0x%llx, virt %p}, size 0x%x\n",
phase,
(u64)p_phys,
p_virt, (u64)(p_phys + size), (u8 *)p_virt + size, size);
rc = qed_dmae_host2host(p_hwfn, p_ptt, p_phys, p_phys + size,
size / 4, NULL);
if (rc) {
DP_NOTICE(p_hwfn,
"DMAE sanity [%s]: qed_dmae_host2host() failed. rc = %d.\n",
phase, rc);
goto out;
}
/* Verify that the top half of the allocated memory has the pattern */
for (p_tmp = (u32 *)((u8 *)p_virt + size);
p_tmp < (u32 *)((u8 *)p_virt + (2 * size)); p_tmp++) {
/* The corresponding address in the bottom half */
val = (u32)(uintptr_t)p_tmp - size;
if (*p_tmp != val) {
DP_NOTICE(p_hwfn,
"DMAE sanity [%s]: addr={phys 0x%llx, virt %p}, read_val 0x%08x, expected_val 0x%08x\n",
phase,
(u64)p_phys + ((u8 *)p_tmp - (u8 *)p_virt),
p_tmp, *p_tmp, val);
rc = -EINVAL;
goto out;
}
}
out:
dma_free_coherent(&p_hwfn->cdev->pdev->dev, 2 * size, p_virt, p_phys);
return rc;
}