/* * Copyright (c) 2016-2019 The Linux Foundation. All rights reserved. * * Permission to use, copy, modify, and/or distribute this software for * any purpose with or without fee is hereby granted, provided that the * above copyright notice and this permission notice appear in all * copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL * WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE * AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR * PERFORMANCE OF THIS SOFTWARE. */ #ifndef _DP_RX_H #define _DP_RX_H #include "hal_rx.h" #include "dp_tx.h" #include "dp_peer.h" #include "dp_internal.h" #ifdef RXDMA_OPTIMIZATION #ifdef NO_RX_PKT_HDR_TLV #define RX_BUFFER_ALIGNMENT 0 #else #define RX_BUFFER_ALIGNMENT 128 #endif /* NO_RX_PKT_HDR_TLV */ #else /* RXDMA_OPTIMIZATION */ #define RX_BUFFER_ALIGNMENT 4 #endif /* RXDMA_OPTIMIZATION */ #ifdef QCA_HOST2FW_RXBUF_RING #define DP_WBM2SW_RBM HAL_RX_BUF_RBM_SW1_BM /** * For MCL cases, allocate as many RX descriptors as buffers in the SW2RXDMA * ring. This value may need to be tuned later. */ #define DP_RX_DESC_ALLOC_MULTIPLIER 1 #else #define DP_WBM2SW_RBM HAL_RX_BUF_RBM_SW3_BM /** * AP use cases need to allocate more RX Descriptors than the number of * entries avaialable in the SW2RXDMA buffer replenish ring. This is to account * for frames sitting in REO queues, HW-HW DMA rings etc. Hence using a * multiplication factor of 3, to allocate three times as many RX descriptors * as RX buffers. */ #define DP_RX_DESC_ALLOC_MULTIPLIER 3 #endif /* QCA_HOST2FW_RXBUF_RING */ #define RX_BUFFER_RESERVATION 0 #define DP_PEER_METADATA_PEER_ID_MASK 0x0000ffff #define DP_PEER_METADATA_PEER_ID_SHIFT 0 #define DP_PEER_METADATA_VDEV_ID_MASK 0x00070000 #define DP_PEER_METADATA_VDEV_ID_SHIFT 16 #define DP_PEER_METADATA_PEER_ID_GET(_peer_metadata) \ (((_peer_metadata) & DP_PEER_METADATA_PEER_ID_MASK) \ >> DP_PEER_METADATA_PEER_ID_SHIFT) #define DP_PEER_METADATA_ID_GET(_peer_metadata) \ (((_peer_metadata) & DP_PEER_METADATA_VDEV_ID_MASK) \ >> DP_PEER_METADATA_VDEV_ID_SHIFT) #define DP_RX_DESC_MAGIC 0xdec0de /** * struct dp_rx_desc * * @nbuf : VA of the "skb" posted * @rx_buf_start : VA of the original Rx buffer, before * movement of any skb->data pointer * @cookie : index into the sw array which holds * the sw Rx descriptors * Cookie space is 21 bits: * lower 18 bits -- index * upper 3 bits -- pool_id * @pool_id : pool Id for which this allocated. * Can only be used if there is no flow * steering * @in_use rx_desc is in use * @unmapped used to mark rx_desc an unmapped if the corresponding * nbuf is already unmapped */ struct dp_rx_desc { qdf_nbuf_t nbuf; uint8_t *rx_buf_start; uint32_t cookie; uint8_t pool_id; #ifdef RX_DESC_DEBUG_CHECK uint32_t magic; #endif uint8_t in_use:1, unmapped:1; }; /* RX Descriptor Multi Page memory alloc related */ #define DP_RX_DESC_OFFSET_NUM_BITS 8 #define DP_RX_DESC_PAGE_ID_NUM_BITS 8 #define DP_RX_DESC_POOL_ID_NUM_BITS 4 #define DP_RX_DESC_PAGE_ID_SHIFT DP_RX_DESC_OFFSET_NUM_BITS #define DP_RX_DESC_POOL_ID_SHIFT \ (DP_RX_DESC_OFFSET_NUM_BITS + DP_RX_DESC_PAGE_ID_NUM_BITS) #define RX_DESC_MULTI_PAGE_COOKIE_POOL_ID_MASK \ (((1 << DP_RX_DESC_POOL_ID_NUM_BITS) - 1) << DP_RX_DESC_POOL_ID_SHIFT) #define RX_DESC_MULTI_PAGE_COOKIE_PAGE_ID_MASK \ (((1 << DP_RX_DESC_PAGE_ID_NUM_BITS) - 1) << \ DP_RX_DESC_PAGE_ID_SHIFT) #define RX_DESC_MULTI_PAGE_COOKIE_OFFSET_MASK \ ((1 << DP_RX_DESC_OFFSET_NUM_BITS) - 1) #define DP_RX_DESC_MULTI_PAGE_COOKIE_GET_POOL_ID(_cookie) \ (((_cookie) & RX_DESC_MULTI_PAGE_COOKIE_POOL_ID_MASK) >> \ DP_RX_DESC_POOL_ID_SHIFT) #define DP_RX_DESC_MULTI_PAGE_COOKIE_GET_PAGE_ID(_cookie) \ (((_cookie) & RX_DESC_MULTI_PAGE_COOKIE_PAGE_ID_MASK) >> \ DP_RX_DESC_PAGE_ID_SHIFT) #define DP_RX_DESC_MULTI_PAGE_COOKIE_GET_OFFSET(_cookie) \ ((_cookie) & RX_DESC_MULTI_PAGE_COOKIE_OFFSET_MASK) #define RX_DESC_COOKIE_INDEX_SHIFT 0 #define RX_DESC_COOKIE_INDEX_MASK 0x3ffff /* 18 bits */ #define RX_DESC_COOKIE_POOL_ID_SHIFT 18 #define RX_DESC_COOKIE_POOL_ID_MASK 0x1c0000 #define DP_RX_DESC_COOKIE_MAX \ (RX_DESC_COOKIE_INDEX_MASK | RX_DESC_COOKIE_POOL_ID_MASK) #define DP_RX_DESC_COOKIE_POOL_ID_GET(_cookie) \ (((_cookie) & RX_DESC_COOKIE_POOL_ID_MASK) >> \ RX_DESC_COOKIE_POOL_ID_SHIFT) #define DP_RX_DESC_COOKIE_INDEX_GET(_cookie) \ (((_cookie) & RX_DESC_COOKIE_INDEX_MASK) >> \ RX_DESC_COOKIE_INDEX_SHIFT) /* DOC: Offset to obtain LLC hdr * * In the case of Wifi parse error * to reach LLC header from beginning * of VLAN tag we need to skip 8 bytes. * Vlan_tag(4)+length(2)+length added * by HW(2) = 8 bytes. */ #define DP_SKIP_VLAN 8 /** * struct dp_rx_cached_buf - rx cached buffer * @list: linked list node * @buf: skb buffer */ struct dp_rx_cached_buf { qdf_list_node_t node; qdf_nbuf_t buf; }; /* *dp_rx_xor_block() - xor block of data *@b: destination data block *@a: source data block *@len: length of the data to process * *Returns: None */ static inline void dp_rx_xor_block(uint8_t *b, const uint8_t *a, qdf_size_t len) { qdf_size_t i; for (i = 0; i < len; i++) b[i] ^= a[i]; } /* *dp_rx_rotl() - rotate the bits left *@val: unsigned integer input value *@bits: number of bits * *Returns: Integer with left rotated by number of 'bits' */ static inline uint32_t dp_rx_rotl(uint32_t val, int bits) { return (val << bits) | (val >> (32 - bits)); } /* *dp_rx_rotr() - rotate the bits right *@val: unsigned integer input value *@bits: number of bits * *Returns: Integer with right rotated by number of 'bits' */ static inline uint32_t dp_rx_rotr(uint32_t val, int bits) { return (val >> bits) | (val << (32 - bits)); } /* * dp_set_rx_queue() - set queue_mapping in skb * @nbuf: skb * @queue_id: rx queue_id * * Return: void */ #ifdef QCA_OL_RX_MULTIQ_SUPPORT static inline void dp_set_rx_queue(qdf_nbuf_t nbuf, uint8_t queue_id) { qdf_nbuf_record_rx_queue(nbuf, queue_id); return; } #else static inline void dp_set_rx_queue(qdf_nbuf_t nbuf, uint8_t queue_id) { } #endif /* *dp_rx_xswap() - swap the bits left *@val: unsigned integer input value * *Returns: Integer with bits swapped */ static inline uint32_t dp_rx_xswap(uint32_t val) { return ((val & 0x00ff00ff) << 8) | ((val & 0xff00ff00) >> 8); } /* *dp_rx_get_le32_split() - get little endian 32 bits split *@b0: byte 0 *@b1: byte 1 *@b2: byte 2 *@b3: byte 3 * *Returns: Integer with split little endian 32 bits */ static inline uint32_t dp_rx_get_le32_split(uint8_t b0, uint8_t b1, uint8_t b2, uint8_t b3) { return b0 | (b1 << 8) | (b2 << 16) | (b3 << 24); } /* *dp_rx_get_le32() - get little endian 32 bits *@b0: byte 0 *@b1: byte 1 *@b2: byte 2 *@b3: byte 3 * *Returns: Integer with little endian 32 bits */ static inline uint32_t dp_rx_get_le32(const uint8_t *p) { return dp_rx_get_le32_split(p[0], p[1], p[2], p[3]); } /* * dp_rx_put_le32() - put little endian 32 bits * @p: destination char array * @v: source 32-bit integer * * Returns: None */ static inline void dp_rx_put_le32(uint8_t *p, uint32_t v) { p[0] = (v) & 0xff; p[1] = (v >> 8) & 0xff; p[2] = (v >> 16) & 0xff; p[3] = (v >> 24) & 0xff; } /* Extract michal mic block of data */ #define dp_rx_michael_block(l, r) \ do { \ r ^= dp_rx_rotl(l, 17); \ l += r; \ r ^= dp_rx_xswap(l); \ l += r; \ r ^= dp_rx_rotl(l, 3); \ l += r; \ r ^= dp_rx_rotr(l, 2); \ l += r; \ } while (0) /** * struct dp_rx_desc_list_elem_t * * @next : Next pointer to form free list * @rx_desc : DP Rx descriptor */ union dp_rx_desc_list_elem_t { union dp_rx_desc_list_elem_t *next; struct dp_rx_desc rx_desc; }; #ifdef RX_DESC_MULTI_PAGE_ALLOC /** * dp_rx_desc_find() - find dp rx descriptor from page ID and offset * @page_id: Page ID * @offset: Offset of the descriptor element * * Return: RX descriptor element */ union dp_rx_desc_list_elem_t *dp_rx_desc_find(uint16_t page_id, uint16_t offset, struct rx_desc_pool *rx_pool); static inline struct dp_rx_desc *dp_get_rx_desc_from_cookie(struct dp_soc *soc, struct rx_desc_pool *pool, uint32_t cookie) { uint8_t pool_id = DP_RX_DESC_MULTI_PAGE_COOKIE_GET_POOL_ID(cookie); uint16_t page_id = DP_RX_DESC_MULTI_PAGE_COOKIE_GET_PAGE_ID(cookie); uint8_t offset = DP_RX_DESC_MULTI_PAGE_COOKIE_GET_OFFSET(cookie); struct rx_desc_pool *rx_desc_pool; union dp_rx_desc_list_elem_t *rx_desc_elem; if (qdf_unlikely(pool_id >= MAX_RXDESC_POOLS)) return NULL; rx_desc_pool = &pool[pool_id]; rx_desc_elem = (union dp_rx_desc_list_elem_t *) (rx_desc_pool->desc_pages.cacheable_pages[page_id] + rx_desc_pool->elem_size * offset); return &rx_desc_elem->rx_desc; } /** * dp_rx_cookie_2_va_rxdma_buf() - Converts cookie to a virtual address of * the Rx descriptor on Rx DMA source ring buffer * @soc: core txrx main context * @cookie: cookie used to lookup virtual address * * Return: Pointer to the Rx descriptor */ static inline struct dp_rx_desc *dp_rx_cookie_2_va_rxdma_buf(struct dp_soc *soc, uint32_t cookie) { return dp_get_rx_desc_from_cookie(soc, &soc->rx_desc_buf[0], cookie); } /** * dp_rx_cookie_2_va_mon_buf() - Converts cookie to a virtual address of * the Rx descriptor on monitor ring buffer * @soc: core txrx main context * @cookie: cookie used to lookup virtual address * * Return: Pointer to the Rx descriptor */ static inline struct dp_rx_desc *dp_rx_cookie_2_va_mon_buf(struct dp_soc *soc, uint32_t cookie) { return dp_get_rx_desc_from_cookie(soc, &soc->rx_desc_mon[0], cookie); } /** * dp_rx_cookie_2_va_mon_status() - Converts cookie to a virtual address of * the Rx descriptor on monitor status ring buffer * @soc: core txrx main context * @cookie: cookie used to lookup virtual address * * Return: Pointer to the Rx descriptor */ static inline struct dp_rx_desc *dp_rx_cookie_2_va_mon_status(struct dp_soc *soc, uint32_t cookie) { return dp_get_rx_desc_from_cookie(soc, &soc->rx_desc_status[0], cookie); } #else /** * dp_rx_cookie_2_va_rxdma_buf() - Converts cookie to a virtual address of * the Rx descriptor on Rx DMA source ring buffer * @soc: core txrx main context * @cookie: cookie used to lookup virtual address * * Return: void *: Virtual Address of the Rx descriptor */ static inline void *dp_rx_cookie_2_va_rxdma_buf(struct dp_soc *soc, uint32_t cookie) { uint8_t pool_id = DP_RX_DESC_COOKIE_POOL_ID_GET(cookie); uint16_t index = DP_RX_DESC_COOKIE_INDEX_GET(cookie); struct rx_desc_pool *rx_desc_pool; if (qdf_unlikely(pool_id >= MAX_RXDESC_POOLS)) return NULL; rx_desc_pool = &soc->rx_desc_buf[pool_id]; if (qdf_unlikely(index >= rx_desc_pool->pool_size)) return NULL; return &(soc->rx_desc_buf[pool_id].array[index].rx_desc); } /** * dp_rx_cookie_2_va_mon_buf() - Converts cookie to a virtual address of * the Rx descriptor on monitor ring buffer * @soc: core txrx main context * @cookie: cookie used to lookup virtual address * * Return: void *: Virtual Address of the Rx descriptor */ static inline void *dp_rx_cookie_2_va_mon_buf(struct dp_soc *soc, uint32_t cookie) { uint8_t pool_id = DP_RX_DESC_COOKIE_POOL_ID_GET(cookie); uint16_t index = DP_RX_DESC_COOKIE_INDEX_GET(cookie); /* TODO */ /* Add sanity for pool_id & index */ return &(soc->rx_desc_mon[pool_id].array[index].rx_desc); } /** * dp_rx_cookie_2_va_mon_status() - Converts cookie to a virtual address of * the Rx descriptor on monitor status ring buffer * @soc: core txrx main context * @cookie: cookie used to lookup virtual address * * Return: void *: Virtual Address of the Rx descriptor */ static inline void *dp_rx_cookie_2_va_mon_status(struct dp_soc *soc, uint32_t cookie) { uint8_t pool_id = DP_RX_DESC_COOKIE_POOL_ID_GET(cookie); uint16_t index = DP_RX_DESC_COOKIE_INDEX_GET(cookie); /* TODO */ /* Add sanity for pool_id & index */ return &(soc->rx_desc_status[pool_id].array[index].rx_desc); } #endif /* RX_DESC_MULTI_PAGE_ALLOC */ void dp_rx_add_desc_list_to_free_list(struct dp_soc *soc, union dp_rx_desc_list_elem_t **local_desc_list, union dp_rx_desc_list_elem_t **tail, uint16_t pool_id, struct rx_desc_pool *rx_desc_pool); uint16_t dp_rx_get_free_desc_list(struct dp_soc *soc, uint32_t pool_id, struct rx_desc_pool *rx_desc_pool, uint16_t num_descs, union dp_rx_desc_list_elem_t **desc_list, union dp_rx_desc_list_elem_t **tail); QDF_STATUS dp_rx_pdev_attach(struct dp_pdev *pdev); void dp_rx_pdev_detach(struct dp_pdev *pdev); uint32_t dp_rx_process(struct dp_intr *int_ctx, void *hal_ring, uint8_t reo_ring_num, uint32_t quota); uint32_t dp_rx_err_process(struct dp_soc *soc, void *hal_ring, uint32_t quota); uint32_t dp_rx_wbm_err_process(struct dp_soc *soc, void *hal_ring, uint32_t quota); /** * dp_rx_sg_create() - create a frag_list for MSDUs which are spread across * multiple nbufs. * @nbuf: pointer to the first msdu of an amsdu. * @rx_tlv_hdr: pointer to the start of RX TLV headers. * * This function implements the creation of RX frag_list for cases * where an MSDU is spread across multiple nbufs. * * Return: returns the head nbuf which contains complete frag_list. */ qdf_nbuf_t dp_rx_sg_create(qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr); /* * dp_rx_desc_pool_alloc() - create a pool of software rx_descs * at the time of dp rx initialization * * @soc: core txrx main context * @pool_id: pool_id which is one of 3 mac_ids * @pool_size: number of Rx descriptor in the pool * @rx_desc_pool: rx descriptor pool pointer * * Return: QDF status */ QDF_STATUS dp_rx_desc_pool_alloc(struct dp_soc *soc, uint32_t pool_id, uint32_t pool_size, struct rx_desc_pool *pool); /* * dp_rx_desc_nbuf_and_pool_free() - free the sw rx desc pool called during * de-initialization of wifi module. * * @soc: core txrx main context * @pool_id: pool_id which is one of 3 mac_ids * @rx_desc_pool: rx descriptor pool pointer * * Return: None */ void dp_rx_desc_nbuf_and_pool_free(struct dp_soc *soc, uint32_t pool_id, struct rx_desc_pool *rx_desc_pool); /* * dp_rx_desc_nbuf_free() - free the sw rx desc nbufs called during * de-initialization of wifi module. * * @soc: core txrx main context * @pool_id: pool_id which is one of 3 mac_ids * @rx_desc_pool: rx descriptor pool pointer * * Return: None */ void dp_rx_desc_nbuf_free(struct dp_soc *soc, struct rx_desc_pool *rx_desc_pool); /* * dp_rx_desc_pool_free() - free the sw rx desc array called during * de-initialization of wifi module. * * @soc: core txrx main context * @rx_desc_pool: rx descriptor pool pointer * * Return: None */ void dp_rx_desc_pool_free(struct dp_soc *soc, struct rx_desc_pool *rx_desc_pool); void dp_rx_deliver_raw(struct dp_vdev *vdev, qdf_nbuf_t nbuf_list, struct dp_peer *peer); /** * dp_rx_add_to_free_desc_list() - Adds to a local free descriptor list * * @head: pointer to the head of local free list * @tail: pointer to the tail of local free list * @new: new descriptor that is added to the free list * * Return: void: */ static inline void dp_rx_add_to_free_desc_list(union dp_rx_desc_list_elem_t **head, union dp_rx_desc_list_elem_t **tail, struct dp_rx_desc *new) { qdf_assert(head && new); new->nbuf = NULL; new->in_use = 0; ((union dp_rx_desc_list_elem_t *)new)->next = *head; *head = (union dp_rx_desc_list_elem_t *)new; if (!*tail) *tail = *head; } uint8_t dp_rx_process_invalid_peer(struct dp_soc *soc, qdf_nbuf_t nbuf); void dp_rx_process_invalid_peer_wrapper(struct dp_soc *soc, qdf_nbuf_t mpdu, bool mpdu_done); void dp_rx_process_mic_error(struct dp_soc *soc, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr, struct dp_peer *peer); void dp_2k_jump_handle(struct dp_soc *soc, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr, uint16_t peer_id, uint8_t tid); #define DP_RX_LIST_APPEND(head, tail, elem) \ do { \ if (!(head)) { \ (head) = (elem); \ QDF_NBUF_CB_RX_NUM_ELEMENTS_IN_LIST(head) = 1;\ } else { \ qdf_nbuf_set_next((tail), (elem)); \ QDF_NBUF_CB_RX_NUM_ELEMENTS_IN_LIST(head)++; \ } \ (tail) = (elem); \ qdf_nbuf_set_next((tail), NULL); \ } while (0) #ifndef BUILD_X86 static inline int check_x86_paddr(struct dp_soc *dp_soc, qdf_nbuf_t *rx_netbuf, qdf_dma_addr_t *paddr, struct dp_pdev *pdev) { return QDF_STATUS_SUCCESS; } #else #define MAX_RETRY 100 static inline int check_x86_paddr(struct dp_soc *dp_soc, qdf_nbuf_t *rx_netbuf, qdf_dma_addr_t *paddr, struct dp_pdev *pdev) { uint32_t nbuf_retry = 0; int32_t ret; const uint32_t x86_phy_addr = 0x50000000; /* * in M2M emulation platforms (x86) the memory below 0x50000000 * is reserved for target use, so any memory allocated in this * region should not be used by host */ do { if (qdf_likely(*paddr > x86_phy_addr)) return QDF_STATUS_SUCCESS; else { QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_INFO, "phy addr %pK exceeded 0x50000000 trying again", paddr); nbuf_retry++; if ((*rx_netbuf)) { qdf_nbuf_unmap_single(dp_soc->osdev, *rx_netbuf, QDF_DMA_FROM_DEVICE); /* Not freeing buffer intentionally. * Observed that same buffer is getting * re-allocated resulting in longer load time * WMI init timeout. * This buffer is anyway not useful so skip it. **/ } *rx_netbuf = qdf_nbuf_alloc(dp_soc->osdev, RX_BUFFER_SIZE, RX_BUFFER_RESERVATION, RX_BUFFER_ALIGNMENT, FALSE); if (qdf_unlikely(!(*rx_netbuf))) return QDF_STATUS_E_FAILURE; ret = qdf_nbuf_map_single(dp_soc->osdev, *rx_netbuf, QDF_DMA_FROM_DEVICE); if (qdf_unlikely(ret == QDF_STATUS_E_FAILURE)) { qdf_nbuf_free(*rx_netbuf); *rx_netbuf = NULL; continue; } *paddr = qdf_nbuf_get_frag_paddr(*rx_netbuf, 0); } } while (nbuf_retry < MAX_RETRY); if ((*rx_netbuf)) { qdf_nbuf_unmap_single(dp_soc->osdev, *rx_netbuf, QDF_DMA_FROM_DEVICE); qdf_nbuf_free(*rx_netbuf); } return QDF_STATUS_E_FAILURE; } #endif /** * dp_rx_cookie_2_link_desc_va() - Converts cookie to a virtual address of * the MSDU Link Descriptor * @soc: core txrx main context * @buf_info: buf_info include cookie that used to lookup virtual address of * link descriptor Normally this is just an index into a per SOC array. * * This is the VA of the link descriptor, that HAL layer later uses to * retrieve the list of MSDU's for a given MPDU. * * Return: void *: Virtual Address of the Rx descriptor */ static inline void *dp_rx_cookie_2_link_desc_va(struct dp_soc *soc, struct hal_buf_info *buf_info) { void *link_desc_va; uint32_t bank_id = LINK_DESC_COOKIE_BANK_ID(buf_info->sw_cookie); /* TODO */ /* Add sanity for cookie */ link_desc_va = soc->link_desc_banks[bank_id].base_vaddr + (buf_info->paddr - soc->link_desc_banks[bank_id].base_paddr); return link_desc_va; } /** * dp_rx_cookie_2_mon_link_desc_va() - Converts cookie to a virtual address of * the MSDU Link Descriptor * @pdev: core txrx pdev context * @buf_info: buf_info includes cookie that used to lookup virtual address of * link descriptor. Normally this is just an index into a per pdev array. * * This is the VA of the link descriptor in monitor mode destination ring, * that HAL layer later uses to retrieve the list of MSDU's for a given MPDU. * * Return: void *: Virtual Address of the Rx descriptor */ static inline void *dp_rx_cookie_2_mon_link_desc_va(struct dp_pdev *pdev, struct hal_buf_info *buf_info, int mac_id) { void *link_desc_va; int mac_for_pdev = dp_get_mac_id_for_mac(pdev->soc, mac_id); /* TODO */ /* Add sanity for cookie */ link_desc_va = pdev->link_desc_banks[mac_for_pdev][buf_info->sw_cookie].base_vaddr + (buf_info->paddr - pdev->link_desc_banks[mac_for_pdev][buf_info->sw_cookie].base_paddr); return link_desc_va; } /** * dp_rx_defrag_concat() - Concatenate the fragments * * @dst: destination pointer to the buffer * @src: source pointer from where the fragment payload is to be copied * * Return: QDF_STATUS */ static inline QDF_STATUS dp_rx_defrag_concat(qdf_nbuf_t dst, qdf_nbuf_t src) { /* * Inside qdf_nbuf_cat, if it is necessary to reallocate dst * to provide space for src, the headroom portion is copied from * the original dst buffer to the larger new dst buffer. * (This is needed, because the headroom of the dst buffer * contains the rx desc.) */ if (!qdf_nbuf_cat(dst, src)) { /* * qdf_nbuf_cat does not free the src memory. * Free src nbuf before returning * For failure case the caller takes of freeing the nbuf */ qdf_nbuf_free(src); return QDF_STATUS_SUCCESS; } return QDF_STATUS_E_DEFRAG_ERROR; } #ifndef FEATURE_WDS static inline QDF_STATUS dp_rx_ast_set_active(struct dp_soc *soc, uint16_t sa_idx, bool is_active) { return QDF_STATUS_SUCCESS; } static inline void dp_rx_wds_srcport_learn(struct dp_soc *soc, uint8_t *rx_tlv_hdr, struct dp_peer *ta_peer, qdf_nbuf_t nbuf) { } #endif /* * dp_rx_desc_dump() - dump the sw rx descriptor * * @rx_desc: sw rx descriptor */ static inline void dp_rx_desc_dump(struct dp_rx_desc *rx_desc) { QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_FATAL, "rx_desc->nbuf: %pK, rx_desc->cookie: %d, rx_desc->pool_id: %d, rx_desc->in_use: %d, rx_desc->unmapped: %d", rx_desc->nbuf, rx_desc->cookie, rx_desc->pool_id, rx_desc->in_use, rx_desc->unmapped); } /* * check_qwrap_multicast_loopback() - Check if rx packet is a loopback packet. * In qwrap mode, packets originated from * any vdev should not loopback and * should be dropped. * @vdev: vdev on which rx packet is received * @nbuf: rx pkt * */ #if ATH_SUPPORT_WRAP static inline bool check_qwrap_multicast_loopback(struct dp_vdev *vdev, qdf_nbuf_t nbuf) { struct dp_vdev *psta_vdev; struct dp_pdev *pdev = vdev->pdev; uint8_t *data = qdf_nbuf_data(nbuf); if (qdf_unlikely(vdev->proxysta_vdev)) { /* In qwrap isolation mode, allow loopback packets as all * packets go to RootAP and Loopback on the mpsta. */ if (vdev->isolation_vdev) return false; TAILQ_FOREACH(psta_vdev, &pdev->vdev_list, vdev_list_elem) { if (qdf_unlikely(psta_vdev->proxysta_vdev && !qdf_mem_cmp(psta_vdev->mac_addr.raw, &data[QDF_MAC_ADDR_SIZE], QDF_MAC_ADDR_SIZE))) { /* Drop packet if source address is equal to * any of the vdev addresses. */ return true; } } } return false; } #else static inline bool check_qwrap_multicast_loopback(struct dp_vdev *vdev, qdf_nbuf_t nbuf) { return false; } #endif #if defined(WLAN_SUPPORT_RX_TAG_STATISTICS) && \ defined(WLAN_SUPPORT_RX_PROTOCOL_TYPE_TAG) /** * dp_rx_update_rx_protocol_tag_stats() - Increments the protocol tag stats * for the given protocol type * @soc: core txrx main context * @pdev: TXRX pdev context for which stats should be incremented * @protocol_index: Protocol index for which the stats should be incremented * @ring_index: REO ring number from which this tag was received. * * Since HKv2 is a SMP, two or more cores may simultaneously receive packets * of same type, and hence attempt to increment counters for the same protocol * type at the same time. This creates the possibility of missing stats. * * For example, when two or more CPUs have each read the old tag value, V, * for protocol type, P and each increment the value to V+1. Instead, the * operations should have been sequenced to achieve a final value of V+2. * * In order to avoid this scenario, we can either use locks or store stats * on a per-CPU basis. Since tagging happens in the core data path, locks * are not preferred. Instead, we use a per-ring counter, since each CPU * operates on a REO ring. * * Return: void */ static inline void dp_rx_update_rx_protocol_tag_stats(struct dp_pdev *pdev, uint16_t protocol_index, uint16_t ring_index) { if (ring_index >= MAX_REO_DEST_RINGS) return; pdev->reo_proto_tag_stats[ring_index][protocol_index].tag_ctr++; } #else static inline void dp_rx_update_rx_protocol_tag_stats(struct dp_pdev *pdev, uint16_t protocol_index, uint16_t ring_index) { } #endif /* WLAN_SUPPORT_RX_TAG_STATISTICS */ #if defined(WLAN_SUPPORT_RX_TAG_STATISTICS) && \ defined(WLAN_SUPPORT_RX_PROTOCOL_TYPE_TAG) /** * dp_rx_update_rx_err_protocol_tag_stats() - Increments the protocol tag stats * for the given protocol type * received from exception ring * @soc: core txrx main context * @pdev: TXRX pdev context for which stats should be incremented * @protocol_index: Protocol index for which the stats should be incremented * * In HKv2, all exception packets are received on Ring-0 (along with normal * Rx). Hence tags are maintained separately for exception ring as well. * * Return: void */ static inline void dp_rx_update_rx_err_protocol_tag_stats(struct dp_pdev *pdev, uint16_t protocol_index) { pdev->rx_err_proto_tag_stats[protocol_index].tag_ctr++; } #else static inline void dp_rx_update_rx_err_protocol_tag_stats(struct dp_pdev *pdev, uint16_t protocol_index) { } #endif /* WLAN_SUPPORT_RX_TAG_STATISTICS */ /** * dp_rx_update_protocol_tag() - Reads CCE metadata from the RX MSDU end TLV * and set the corresponding tag in QDF packet * @soc: core txrx main context * @vdev: vdev on which the packet is received * @nbuf: QDF pkt buffer on which the protocol tag should be set * @rx_tlv_hdr: rBbase address where the RX TLVs starts * @ring_index: REO ring number, not used for error & monitor ring * @is_reo_exception: flag to indicate if rx from REO ring or exception ring * @is_update_stats: flag to indicate whether to update stats or not * Return: void */ #ifdef WLAN_SUPPORT_RX_PROTOCOL_TYPE_TAG static inline void dp_rx_update_protocol_tag(struct dp_soc *soc, struct dp_vdev *vdev, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr, uint16_t ring_index, bool is_reo_exception, bool is_update_stats) { uint16_t cce_metadata = RX_PROTOCOL_TAG_START_OFFSET; bool cce_match = false; struct dp_pdev *pdev; uint16_t protocol_tag = 0; if (qdf_unlikely(!vdev)) return; pdev = vdev->pdev; if (qdf_likely(!pdev->is_rx_protocol_tagging_enabled)) return; /* * In case of raw frames, rx_attention and rx_msdu_end tlv * may be stale or invalid. Do not tag such frames. * Default decap_type is set to ethernet for monitor vdev, * therefore, cannot check decap_type for monitor mode. * We will call this only for eth frames from dp_rx_mon_dest.c. */ if (qdf_likely(!(pdev->monitor_vdev && pdev->monitor_vdev == vdev) && (vdev->rx_decap_type != htt_cmn_pkt_type_ethernet))) return; /* * Check whether HW has filled in the CCE metadata in * this packet, if not filled, just return */ if (qdf_likely(!hal_rx_msdu_cce_match_get(rx_tlv_hdr))) return; cce_match = true; /* Get the cce_metadata from RX MSDU TLV */ cce_metadata = (hal_rx_msdu_cce_metadata_get(rx_tlv_hdr) & RX_MSDU_END_16_CCE_METADATA_MASK); /* * Received CCE metadata should be within the * valid limits */ qdf_assert_always((cce_metadata >= RX_PROTOCOL_TAG_START_OFFSET) && (cce_metadata < (RX_PROTOCOL_TAG_START_OFFSET + RX_PROTOCOL_TAG_MAX))); /* * The CCE metadata received is just the * packet_type + RX_PROTOCOL_TAG_START_OFFSET */ cce_metadata -= RX_PROTOCOL_TAG_START_OFFSET; /* * Update the QDF packet with the user-specified * tag/metadata by looking up tag value for * received protocol type. */ protocol_tag = pdev->rx_proto_tag_map[cce_metadata].tag; qdf_nbuf_set_rx_protocol_tag(nbuf, protocol_tag); QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_INFO_LOW, "Seq:%u decap:%u CCE Match:%d ProtoID:%u Tag:%u US:%d", hal_rx_get_rx_sequence(rx_tlv_hdr), vdev->rx_decap_type, cce_match, cce_metadata, protocol_tag, is_update_stats); if (qdf_likely(!is_update_stats)) return; if (qdf_unlikely(is_reo_exception)) { dp_rx_update_rx_err_protocol_tag_stats(pdev, cce_metadata); } else { dp_rx_update_rx_protocol_tag_stats(pdev, cce_metadata, ring_index); } } #else static inline void dp_rx_update_protocol_tag(struct dp_soc *soc, struct dp_vdev *vdev, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr, uint16_t ring_index, bool is_reo_exception, bool is_update_stats) { /* Stub API */ } #endif /* WLAN_SUPPORT_RX_PROTOCOL_TYPE_TAG */ /** * dp_rx_mon_update_protocol_tag() - Performs necessary checks for monitor mode * and then tags appropriate packets * @soc: core txrx main context * @vdev: pdev on which packet is received * @msdu: QDF packet buffer on which the protocol tag should be set * @rx_desc: base address where the RX TLVs start * Return: void */ #ifdef WLAN_SUPPORT_RX_PROTOCOL_TYPE_TAG static inline void dp_rx_mon_update_protocol_tag(struct dp_soc *soc, struct dp_pdev *dp_pdev, qdf_nbuf_t msdu, void *rx_desc) { uint32_t msdu_ppdu_id = 0; struct mon_rx_status *mon_recv_status; if (qdf_likely(!dp_pdev->is_rx_protocol_tagging_enabled)) return; if (qdf_likely(!dp_pdev->monitor_vdev)) return; if (qdf_likely(1 != dp_pdev->ppdu_info.rx_status.rxpcu_filter_pass)) return; msdu_ppdu_id = HAL_RX_HW_DESC_GET_PPDUID_GET(rx_desc); if (msdu_ppdu_id != dp_pdev->ppdu_info.com_info.ppdu_id) { QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR, "msdu_ppdu_id=%x,com_info.ppdu_id=%x", msdu_ppdu_id, dp_pdev->ppdu_info.com_info.ppdu_id); return; } /* * Update the protocol tag in SKB for packets received on BSS. * Do not update tag stats since it would double actual received count */ mon_recv_status = &dp_pdev->ppdu_info.rx_status; if (mon_recv_status->frame_control_info_valid && ((mon_recv_status->frame_control & IEEE80211_FC0_TYPE_MASK) == IEEE80211_FC0_TYPE_DATA)) { dp_rx_update_protocol_tag(soc, dp_pdev->monitor_vdev, msdu, rx_desc, MAX_REO_DEST_RINGS, false, false); } } #else static inline void dp_rx_mon_update_protocol_tag(struct dp_soc *soc, struct dp_pdev *dp_pdev, qdf_nbuf_t msdu, void *rx_desc) { /* Stub API */ } #endif /* WLAN_SUPPORT_RX_PROTOCOL_TYPE_TAG */ /* * dp_rx_buffers_replenish() - replenish rxdma ring with rx nbufs * called during dp rx initialization * and at the end of dp_rx_process. * * @soc: core txrx main context * @mac_id: mac_id which is one of 3 mac_ids * @dp_rxdma_srng: dp rxdma circular ring * @rx_desc_pool: Pointer to free Rx descriptor pool * @num_req_buffers: number of buffer to be replenished * @desc_list: list of descs if called from dp_rx_process * or NULL during dp rx initialization or out of buffer * interrupt. * @tail: tail of descs list * Return: return success or failure */ QDF_STATUS dp_rx_buffers_replenish(struct dp_soc *dp_soc, uint32_t mac_id, struct dp_srng *dp_rxdma_srng, struct rx_desc_pool *rx_desc_pool, uint32_t num_req_buffers, union dp_rx_desc_list_elem_t **desc_list, union dp_rx_desc_list_elem_t **tail); /** * dp_rx_link_desc_return() - Return a MPDU link descriptor to HW * (WBM), following error handling * * @soc: core DP main context * @buf_addr_info: opaque pointer to the REO error ring descriptor * @buf_addr_info: void pointer to the buffer_addr_info * @bm_action: put to idle_list or release to msdu_list * Return: QDF_STATUS */ QDF_STATUS dp_rx_link_desc_return(struct dp_soc *soc, void *ring_desc, uint8_t bm_action); QDF_STATUS dp_rx_link_desc_buf_return(struct dp_soc *soc, struct dp_srng *dp_rxdma_srng, void *buf_addr_info, uint8_t bm_action); /** * dp_rx_link_desc_return_by_addr - Return a MPDU link descriptor to * (WBM) by address * * @soc: core DP main context * @link_desc_addr: link descriptor addr * * Return: QDF_STATUS */ QDF_STATUS dp_rx_link_desc_return_by_addr(struct dp_soc *soc, void *link_desc_addr, uint8_t bm_action); uint32_t dp_rxdma_err_process(struct dp_soc *soc, uint32_t mac_id, uint32_t quota); void dp_rx_fill_mesh_stats(struct dp_vdev *vdev, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr, struct dp_peer *peer); QDF_STATUS dp_rx_filter_mesh_packets(struct dp_vdev *vdev, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr); int dp_wds_rx_policy_check(uint8_t *rx_tlv_hdr, struct dp_vdev *vdev, struct dp_peer *peer); qdf_nbuf_t dp_rx_nbuf_prepare(struct dp_soc *soc, struct dp_pdev *pdev); void dp_rx_dump_info_and_assert(struct dp_soc *soc, void *hal_ring, void *ring_desc, struct dp_rx_desc *rx_desc); void dp_rx_compute_delay(struct dp_vdev *vdev, qdf_nbuf_t nbuf); #ifdef RX_DESC_DEBUG_CHECK /** * dp_rx_desc_check_magic() - check the magic value in dp_rx_desc * @rx_desc: rx descriptor pointer * * Return: true, if magic is correct, else false. */ static inline bool dp_rx_desc_check_magic(struct dp_rx_desc *rx_desc) { if (qdf_unlikely(rx_desc->magic != DP_RX_DESC_MAGIC)) return false; rx_desc->magic = 0; return true; } /** * dp_rx_desc_prep() - prepare rx desc * @rx_desc: rx descriptor pointer to be prepared * @nbuf: nbuf to be associated with rx_desc * * Note: assumption is that we are associating a nbuf which is mapped * * Return: none */ static inline void dp_rx_desc_prep(struct dp_rx_desc *rx_desc, qdf_nbuf_t nbuf) { rx_desc->magic = DP_RX_DESC_MAGIC; rx_desc->nbuf = nbuf; rx_desc->unmapped = 0; } #else static inline bool dp_rx_desc_check_magic(struct dp_rx_desc *rx_desc) { return true; } static inline void dp_rx_desc_prep(struct dp_rx_desc *rx_desc, qdf_nbuf_t nbuf) { rx_desc->nbuf = nbuf; rx_desc->unmapped = 0; } #endif /* RX_DESC_DEBUG_CHECK */ void dp_rx_process_rxdma_err(struct dp_soc *soc, qdf_nbuf_t nbuf, uint8_t *rx_tlv_hdr, struct dp_peer *peer, uint8_t err_code); #ifdef PEER_CACHE_RX_PKTS /** * dp_rx_flush_rx_cached() - flush cached rx frames * @peer: peer * @drop: set flag to drop frames * * Return: None */ void dp_rx_flush_rx_cached(struct dp_peer *peer, bool drop); #else static inline void dp_rx_flush_rx_cached(struct dp_peer *peer, bool drop) { } #endif #endif /* _DP_RX_H */