/*
* Guts of ath10k_ce_completed_send_next.
* The caller takes responsibility for any necessary locking.
*/
int ath10k_ce_completed_send_next_nolock(struct ath10k_ce_pipe *ce_state,
void **per_transfer_contextp)
{
struct ath10k_ce_ring *src_ring = ce_state->src_ring;
u32 ctrl_addr = ce_state->ctrl_addr;
struct ath10k *ar = ce_state->ar;
unsigned int nentries_mask = src_ring->nentries_mask;
unsigned int sw_index = src_ring->sw_index;
unsigned int read_index;
if (src_ring->hw_index == sw_index) {
/*
* The SW completion index has caught up with the cached
* version of the HW completion index.
* Update the cached HW completion index to see whether
* the SW has really caught up to the HW, or if the cached
* value of the HW index has become stale.
*/
read_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
if (read_index == 0xffffffff)
return -ENODEV;
read_index &= nentries_mask;
src_ring->hw_index = read_index;
}
read_index = src_ring->hw_index;
if (read_index == sw_index)
return -EIO;
if (per_transfer_contextp)
*per_transfer_contextp =
src_ring->per_transfer_context[sw_index];
/* sanity */
src_ring->per_transfer_context[sw_index] = NULL;
/* Update sw_index */
sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
src_ring->sw_index = sw_index;
return 0;
}
void ath10k_ce_dump_registers(struct ath10k *ar,
struct ath10k_fw_crash_data *crash_data)
{
struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
struct ath10k_ce_crash_data ce;
u32 addr, id;
lockdep_assert_held(&ar->data_lock);
ath10k_err(ar, "Copy Engine register dump:\n");
spin_lock_bh(&ar_pci->ce_lock);
for (id = 0; id < CE_COUNT; id++) {
addr = ath10k_ce_base_address(ar, id);
ce.base_addr = cpu_to_le32(addr);
ce.src_wr_idx =
cpu_to_le32(ath10k_ce_src_ring_write_index_get(ar, addr));
ce.src_r_idx =
cpu_to_le32(ath10k_ce_src_ring_read_index_get(ar, addr));
ce.dst_wr_idx =
cpu_to_le32(ath10k_ce_dest_ring_write_index_get(ar, addr));
ce.dst_r_idx =
cpu_to_le32(ath10k_ce_dest_ring_read_index_get(ar, addr));
if (crash_data)
crash_data->ce_crash_data[id] = ce;
ath10k_err(ar, "[%02d]: 0x%08x %3u %3u %3u %3u", id,
le32_to_cpu(ce.base_addr),
le32_to_cpu(ce.src_wr_idx),
le32_to_cpu(ce.src_r_idx),
le32_to_cpu(ce.dst_wr_idx),
le32_to_cpu(ce.dst_r_idx));
}
spin_unlock_bh(&ar_pci->ce_lock);
}
static int ath10k_ce_init_src_ring(struct ath10k *ar,
unsigned int ce_id,
const struct ce_attr *attr)
{
struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
struct ath10k_ce_ring *src_ring = ce_state->src_ring;
u32 nentries, ctrl_addr = ath10k_ce_base_address(ar, ce_id);
nentries = roundup_pow_of_two(attr->src_nentries);
memset(src_ring->base_addr_owner_space, 0,
nentries * sizeof(struct ce_desc));
src_ring->sw_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
src_ring->sw_index &= src_ring->nentries_mask;
src_ring->hw_index = src_ring->sw_index;
src_ring->write_index =
ath10k_ce_src_ring_write_index_get(ar, ctrl_addr);
src_ring->write_index &= src_ring->nentries_mask;
ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
src_ring->base_addr_ce_space);
ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
ath10k_dbg(ar, ATH10K_DBG_BOOT,
"boot init ce src ring id %d entries %d base_addr %pK\n",
ce_id, nentries, src_ring->base_addr_owner_space);
return 0;
}
static int ath10k_ce_init_src_ring(struct ath10k *ar,
unsigned int ce_id,
struct ath10k_ce_pipe *ce_state,
const struct ce_attr *attr)
{
struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
struct ath10k_ce_ring *src_ring;
unsigned int nentries = attr->src_nentries;
unsigned int ce_nbytes;
u32 ctrl_addr = ath10k_ce_base_address(ce_id);
dma_addr_t base_addr;
char *ptr;
nentries = roundup_pow_of_two(nentries);
if (ce_state->src_ring) {
WARN_ON(ce_state->src_ring->nentries != nentries);
return 0;
}
ce_nbytes = sizeof(struct ath10k_ce_ring) + (nentries * sizeof(void *));
ptr = kzalloc(ce_nbytes, GFP_KERNEL);
if (ptr == NULL)
return -ENOMEM;
ce_state->src_ring = (struct ath10k_ce_ring *)ptr;
src_ring = ce_state->src_ring;
ptr += sizeof(struct ath10k_ce_ring);
src_ring->nentries = nentries;
src_ring->nentries_mask = nentries - 1;
src_ring->sw_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
src_ring->sw_index &= src_ring->nentries_mask;
src_ring->hw_index = src_ring->sw_index;
src_ring->write_index =
ath10k_ce_src_ring_write_index_get(ar, ctrl_addr);
src_ring->write_index &= src_ring->nentries_mask;
src_ring->per_transfer_context = (void **)ptr;
/*
* Legacy platforms that do not support cache
* coherent DMA are unsupported
*/
src_ring->base_addr_owner_space_unaligned =
pci_alloc_consistent(ar_pci->pdev,
(nentries * sizeof(struct ce_desc) +
CE_DESC_RING_ALIGN),
&base_addr);
if (!src_ring->base_addr_owner_space_unaligned) {
kfree(ce_state->src_ring);
ce_state->src_ring = NULL;
return -ENOMEM;
}
src_ring->base_addr_ce_space_unaligned = base_addr;
src_ring->base_addr_owner_space = PTR_ALIGN(
src_ring->base_addr_owner_space_unaligned,
CE_DESC_RING_ALIGN);
src_ring->base_addr_ce_space = ALIGN(
src_ring->base_addr_ce_space_unaligned,
CE_DESC_RING_ALIGN);
/*
* Also allocate a shadow src ring in regular
* mem to use for faster access.
*/
src_ring->shadow_base_unaligned =
kmalloc((nentries * sizeof(struct ce_desc) +
CE_DESC_RING_ALIGN), GFP_KERNEL);
if (!src_ring->shadow_base_unaligned) {
pci_free_consistent(ar_pci->pdev,
(nentries * sizeof(struct ce_desc) +
CE_DESC_RING_ALIGN),
src_ring->base_addr_owner_space,
src_ring->base_addr_ce_space);
kfree(ce_state->src_ring);
ce_state->src_ring = NULL;
return -ENOMEM;
}
src_ring->shadow_base = PTR_ALIGN(
src_ring->shadow_base_unaligned,
CE_DESC_RING_ALIGN);
ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
src_ring->base_addr_ce_space);
ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
ath10k_dbg(ATH10K_DBG_BOOT,
"boot ce src ring id %d entries %d base_addr %p\n",
ce_id, nentries, src_ring->base_addr_owner_space);
return 0;
}
/*
* Guts of ath10k_ce_completed_send_next.
* The caller takes responsibility for any necessary locking.
*/
static int ath10k_ce_completed_send_next_nolock(struct ath10k_ce_pipe *ce_state,
void **per_transfer_contextp,
u32 *bufferp,
unsigned int *nbytesp,
unsigned int *transfer_idp)
{
struct ath10k_ce_ring *src_ring = ce_state->src_ring;
u32 ctrl_addr = ce_state->ctrl_addr;
struct ath10k *ar = ce_state->ar;
unsigned int nentries_mask = src_ring->nentries_mask;
unsigned int sw_index = src_ring->sw_index;
struct ce_desc *sdesc, *sbase;
unsigned int read_index;
int ret;
if (src_ring->hw_index == sw_index) {
/*
* The SW completion index has caught up with the cached
* version of the HW completion index.
* Update the cached HW completion index to see whether
* the SW has really caught up to the HW, or if the cached
* value of the HW index has become stale.
*/
ret = ath10k_pci_wake(ar);
if (ret)
return ret;
src_ring->hw_index =
ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
src_ring->hw_index &= nentries_mask;
ath10k_pci_sleep(ar);
}
read_index = src_ring->hw_index;
if ((read_index == sw_index) || (read_index == 0xffffffff))
return -EIO;
sbase = src_ring->shadow_base;
sdesc = CE_SRC_RING_TO_DESC(sbase, sw_index);
/* Return data from completed source descriptor */
*bufferp = __le32_to_cpu(sdesc->addr);
*nbytesp = __le16_to_cpu(sdesc->nbytes);
*transfer_idp = MS(__le16_to_cpu(sdesc->flags),
CE_DESC_FLAGS_META_DATA);
if (per_transfer_contextp)
*per_transfer_contextp =
src_ring->per_transfer_context[sw_index];
/* sanity */
src_ring->per_transfer_context[sw_index] = NULL;
/* Update sw_index */
sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
src_ring->sw_index = sw_index;
return 0;
}
