/*=== CSR (control and status registers) ===*/
#define CSR_BASE (0x000)
-#define CSR_SW_VER (CSR_BASE+0x000)
#define CSR_HW_IF_CONFIG_REG (CSR_BASE+0x000) /* hardware interface config */
#define CSR_INT_COALESCING (CSR_BASE+0x004) /* accum ints, 32-usec units */
#define CSR_INT (CSR_BASE+0x008) /* host interrupt status/ack */
*/
#define CSR_HW_REV_WA_REG (CSR_BASE+0x22C)
+/* Hardware interface configuration bits */
+#define CSR_HW_IF_CONFIG_REG_BIT_KEDRON_R (0x00000010)
+#define CSR_HW_IF_CONFIG_REG_MSK_BOARD_VER (0x00000C00)
+#define CSR_HW_IF_CONFIG_REG_BIT_MAC_SI (0x00000100)
+#define CSR_HW_IF_CONFIG_REG_BIT_RADIO_SI (0x00000200)
+#define CSR_HW_IF_CONFIG_REG_BIT_EEPROM_OWN_SEM (0x00200000)
+
/* interrupt flags in INTA, set by uCode or hardware (e.g. dma),
* acknowledged (reset) by host writing "1" to flagged bits. */
-#define CSR_INT_BIT_FH_RX (1<<31) /* Rx DMA, cmd responses, FH_INT[17:16] */
-#define CSR_INT_BIT_HW_ERR (1<<29) /* DMA hardware error FH_INT[31] */
-#define CSR_INT_BIT_DNLD (1<<28) /* uCode Download */
-#define CSR_INT_BIT_FH_TX (1<<27) /* Tx DMA FH_INT[1:0] */
-#define CSR_INT_BIT_MAC_CLK_ACTV (1<<26) /* NIC controller's clock toggled on/off */
-#define CSR_INT_BIT_SW_ERR (1<<25) /* uCode error */
-#define CSR_INT_BIT_RF_KILL (1<<7) /* HW RFKILL switch GP_CNTRL[27] toggled */
-#define CSR_INT_BIT_CT_KILL (1<<6) /* Critical temp (chip too hot) rfkill */
-#define CSR_INT_BIT_SW_RX (1<<3) /* Rx, command responses, 3945 */
-#define CSR_INT_BIT_WAKEUP (1<<1) /* NIC controller waking up (pwr mgmt) */
-#define CSR_INT_BIT_ALIVE (1<<0) /* uCode interrupts once it initializes */
+#define CSR_INT_BIT_FH_RX (1 << 31) /* Rx DMA, cmd responses, FH_INT[17:16] */
+#define CSR_INT_BIT_HW_ERR (1 << 29) /* DMA hardware error FH_INT[31] */
+#define CSR_INT_BIT_DNLD (1 << 28) /* uCode Download */
+#define CSR_INT_BIT_FH_TX (1 << 27) /* Tx DMA FH_INT[1:0] */
+#define CSR_INT_BIT_SCD (1 << 26) /* TXQ pointer advanced */
+#define CSR_INT_BIT_SW_ERR (1 << 25) /* uCode error */
+#define CSR_INT_BIT_RF_KILL (1 << 7) /* HW RFKILL switch GP_CNTRL[27] toggled */
+#define CSR_INT_BIT_CT_KILL (1 << 6) /* Critical temp (chip too hot) rfkill */
+#define CSR_INT_BIT_SW_RX (1 << 3) /* Rx, command responses, 3945 */
+#define CSR_INT_BIT_WAKEUP (1 << 1) /* NIC controller waking up (pwr mgmt) */
+#define CSR_INT_BIT_ALIVE (1 << 0) /* uCode interrupts once it initializes */
#define CSR_INI_SET_MASK (CSR_INT_BIT_FH_RX | \
CSR_INT_BIT_HW_ERR | \
CSR_INT_BIT_ALIVE)
/* interrupt flags in FH (flow handler) (PCI busmaster DMA) */
-#define CSR_FH_INT_BIT_ERR (1<<31) /* Error */
-#define CSR_FH_INT_BIT_HI_PRIOR (1<<30) /* High priority Rx, bypass coalescing */
-#define CSR_FH_INT_BIT_RX_CHNL1 (1<<17) /* Rx channel 1 */
-#define CSR_FH_INT_BIT_RX_CHNL0 (1<<16) /* Rx channel 0 */
-#define CSR_FH_INT_BIT_TX_CHNL1 (1<<1) /* Tx channel 1 */
-#define CSR_FH_INT_BIT_TX_CHNL0 (1<<0) /* Tx channel 0 */
+#define CSR_FH_INT_BIT_ERR (1 << 31) /* Error */
+#define CSR_FH_INT_BIT_HI_PRIOR (1 << 30) /* High priority Rx, bypass coalescing */
+#define CSR_FH_INT_BIT_RX_CHNL1 (1 << 17) /* Rx channel 1 */
+#define CSR_FH_INT_BIT_RX_CHNL0 (1 << 16) /* Rx channel 0 */
+#define CSR_FH_INT_BIT_TX_CHNL1 (1 << 1) /* Tx channel 1 */
+#define CSR_FH_INT_BIT_TX_CHNL0 (1 << 0) /* Tx channel 0 */
#define CSR_FH_INT_RX_MASK (CSR_FH_INT_BIT_HI_PRIOR | \
CSR_FH_INT_BIT_RX_CHNL1 | \
/*
* Per-Tx-queue write pointer (index, really!) (3945 and 4965).
- * Indicates index to next TFD that driver will fill (1 past latest filled).
+ * Driver sets this to indicate index to next TFD that driver will fill
+ * (1 past latest filled).
* Bit usage:
* 0-7: queue write index (0-255)
* 11-8: queue selector (0-15)
#define HBUS_TARG_MBX_C (HBUS_BASE+0x030)
-/*=== FH (data Flow Handler) ===*/
-#define FH_BASE (0x800)
-
-#define FH_RSCSR_CHNL0_WPTR (FH_RSCSR_CHNL0_RBDCB_WPTR_REG)
-
-/* RSSR */
-#define FH_RSSR_CTRL (FH_RSSR_TABLE+0x000)
-#define FH_RSSR_STATUS (FH_RSSR_TABLE+0x004)
-/* TCSR */
-#define FH_TCSR(_channel) (FH_TCSR_TABLE+(_channel)*0x20)
-#define FH_TCSR_CONFIG(_channel) (FH_TCSR(_channel)+0x00)
-#define FH_TCSR_CREDIT(_channel) (FH_TCSR(_channel)+0x04)
-#define FH_TCSR_BUFF_STTS(_channel) (FH_TCSR(_channel)+0x08)
-/* TSSR */
-#define FH_TSSR_CBB_BASE (FH_TSSR_TABLE+0x000)
-#define FH_TSSR_MSG_CONFIG (FH_TSSR_TABLE+0x008)
-#define FH_TSSR_TX_STATUS (FH_TSSR_TABLE+0x010)
-
-
#define HBUS_TARG_MBX_C_REG_BIT_CMD_BLOCKED (0x00000004)
#define TFD_QUEUE_SIZE_MAX (256)
#define RX_LOW_WATERMARK 8
/* Size of one Rx buffer in host DRAM */
-#define IWL_RX_BUF_SIZE (4 * 1024)
+#define IWL_RX_BUF_SIZE_4K (4 * 1024)
+#define IWL_RX_BUF_SIZE_8K (8 * 1024)
/* Sizes and addresses for instruction and data memory (SRAM) in
* 4965's embedded processor. Driver access is via HBUS_TARG_MEM_* regs. */
/* Flow Handler Definitions */
/****************************/
-/*
+/**
* This I/O area is directly read/writable by driver (e.g. Linux uses writel())
* Addresses are offsets from device's PCI hardware base address.
*/
#define IWL_FH_KW_MEM_ADDR_REG (FH_MEM_LOWER_BOUND + 0x97C)
-/* CBBC Area - Circular buffers base address cache pointers table */
+/**
+ * TFD Circular Buffers Base (CBBC) addresses
+ *
+ * 4965 has 16 base pointer registers, one for each of 16 host-DRAM-resident
+ * circular buffers (CBs/queues) containing Transmit Frame Descriptors (TFDs)
+ * (see struct iwl_tfd_frame). These 16 pointer registers are offset by 0x04
+ * bytes from one another. Each TFD circular buffer in DRAM must be 256-byte
+ * aligned (address bits 0-7 must be 0).
+ *
+ * Bit fields in each pointer register:
+ * 27-0: TFD CB physical base address [35:8], must be 256-byte aligned
+ */
#define FH_MEM_CBBC_LOWER_BOUND (FH_MEM_LOWER_BOUND + 0x9D0)
#define FH_MEM_CBBC_UPPER_BOUND (FH_MEM_LOWER_BOUND + 0xA10)
-/* queues 0 - 15 */
+
+/* Find TFD CB base pointer for given queue (range 0-15). */
#define FH_MEM_CBBC_QUEUE(x) (FH_MEM_CBBC_LOWER_BOUND + (x) * 0x4)
-/* RSCSR Area */
+
+/**
+ * Rx SRAM Control and Status Registers (RSCSR)
+ *
+ * These registers provide handshake between driver and 4965 for the Rx queue
+ * (this queue handles *all* command responses, notifications, Rx data, etc.
+ * sent from 4965 uCode to host driver). Unlike Tx, there is only one Rx
+ * queue, and only one Rx DMA/FIFO channel. Also unlike Tx, which can
+ * concatenate up to 20 DRAM buffers to form a Tx frame, each Receive Buffer
+ * Descriptor (RBD) points to only one Rx Buffer (RB); there is a 1:1
+ * mapping between RBDs and RBs.
+ *
+ * Driver must allocate host DRAM memory for the following, and set the
+ * physical address of each into 4965 registers:
+ *
+ * 1) Receive Buffer Descriptor (RBD) circular buffer (CB), typically with 256
+ * entries (although any power of 2, up to 4096, is selectable by driver).
+ * Each entry (1 dword) points to a receive buffer (RB) of consistent size
+ * (typically 4K, although 8K or 16K are also selectable by driver).
+ * Driver sets up RB size and number of RBDs in the CB via Rx config
+ * register FH_MEM_RCSR_CHNL0_CONFIG_REG.
+ *
+ * Bit fields within one RBD:
+ * 27-0: Receive Buffer physical address bits [35:8], 256-byte aligned
+ *
+ * Driver sets physical address [35:8] of base of RBD circular buffer
+ * into FH_RSCSR_CHNL0_RBDCB_BASE_REG [27:0].
+ *
+ * 2) Rx status buffer, 8 bytes, in which 4965 indicates which Rx Buffers
+ * (RBs) have been filled, via a "write pointer", actually the index of
+ * the RB's corresponding RBD within the circular buffer. Driver sets
+ * physical address [35:4] into FH_RSCSR_CHNL0_STTS_WPTR_REG [31:0].
+ *
+ * Bit fields in lower dword of Rx status buffer (upper dword not used
+ * by driver; see struct iwl4965_shared, val0):
+ * 31-12: Not used by driver
+ * 11- 0: Index of last filled Rx buffer descriptor
+ * (4965 writes, driver reads this value)
+ *
+ * As the driver prepares Receive Buffers (RBs) for 4965 to fill, driver must
+ * enter pointers to these RBs into contiguous RBD circular buffer entries,
+ * and update the 4965's "write" index register, FH_RSCSR_CHNL0_RBDCB_WPTR_REG.
+ *
+ * This "write" index corresponds to the *next* RBD that the driver will make
+ * available, i.e. one RBD past the tail of the ready-to-fill RBDs within
+ * the circular buffer. This value should initially be 0 (before preparing any
+ * RBs), should be 8 after preparing the first 8 RBs (for example), and must
+ * wrap back to 0 at the end of the circular buffer (but don't wrap before
+ * "read" index has advanced past 1! See below).
+ * NOTE: 4965 EXPECTS THE WRITE INDEX TO BE INCREMENTED IN MULTIPLES OF 8.
+ *
+ * As the 4965 fills RBs (referenced from contiguous RBDs within the circular
+ * buffer), it updates the Rx status buffer in host DRAM, 2) described above,
+ * to tell the driver the index of the latest filled RBD. The driver must
+ * read this "read" index from DRAM after receiving an Rx interrupt from 4965.
+ *
+ * The driver must also internally keep track of a third index, which is the
+ * next RBD to process. When receiving an Rx interrupt, driver should process
+ * all filled but unprocessed RBs up to, but not including, the RB
+ * corresponding to the "read" index. For example, if "read" index becomes "1",
+ * driver may process the RB pointed to by RBD 0. Depending on volume of
+ * traffic, there may be many RBs to process.
+ *
+ * If read index == write index, 4965 thinks there is no room to put new data.
+ * Due to this, the maximum number of filled RBs is 255, instead of 256. To
+ * be safe, make sure that there is a gap of at least 2 RBDs between "write"
+ * and "read" indexes; that is, make sure that there are no more than 254
+ * buffers waiting to be filled.
+ */
#define FH_MEM_RSCSR_LOWER_BOUND (FH_MEM_LOWER_BOUND + 0xBC0)
#define FH_MEM_RSCSR_UPPER_BOUND (FH_MEM_LOWER_BOUND + 0xC00)
#define FH_MEM_RSCSR_CHNL0 (FH_MEM_RSCSR_LOWER_BOUND)
+/**
+ * Physical base address of 8-byte Rx Status buffer.
+ * Bit fields:
+ * 31-0: Rx status buffer physical base address [35:4], must 16-byte aligned.
+ */
#define FH_RSCSR_CHNL0_STTS_WPTR_REG (FH_MEM_RSCSR_CHNL0)
+
+/**
+ * Physical base address of Rx Buffer Descriptor Circular Buffer.
+ * Bit fields:
+ * 27-0: RBD CD physical base address [35:8], must be 256-byte aligned.
+ */
#define FH_RSCSR_CHNL0_RBDCB_BASE_REG (FH_MEM_RSCSR_CHNL0 + 0x004)
+
+/**
+ * Rx write pointer (index, really!).
+ * Bit fields:
+ * 11-0: Index of driver's most recent prepared-to-be-filled RBD, + 1.
+ * NOTE: For 256-entry circular buffer, use only bits [7:0].
+ */
#define FH_RSCSR_CHNL0_RBDCB_WPTR_REG (FH_MEM_RSCSR_CHNL0 + 0x008)
+#define FH_RSCSR_CHNL0_WPTR (FH_RSCSR_CHNL0_RBDCB_WPTR_REG)
-/* RCSR Area - Registers address map */
+
+/**
+ * Rx Config/Status Registers (RCSR)
+ * Rx Config Reg for channel 0 (only channel used)
+ *
+ * Driver must initialize FH_MEM_RCSR_CHNL0_CONFIG_REG as follows for
+ * normal operation (see bit fields).
+ *
+ * Clearing FH_MEM_RCSR_CHNL0_CONFIG_REG to 0 turns off Rx DMA.
+ * Driver should poll FH_MEM_RSSR_RX_STATUS_REG for
+ * FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE (bit 24) before continuing.
+ *
+ * Bit fields:
+ * 31-30: Rx DMA channel enable: '00' off/pause, '01' pause at end of frame,
+ * '10' operate normally
+ * 29-24: reserved
+ * 23-20: # RBDs in circular buffer = 2^value; use "8" for 256 RBDs (normal),
+ * min "5" for 32 RBDs, max "12" for 4096 RBDs.
+ * 19-18: reserved
+ * 17-16: size of each receive buffer; '00' 4K (normal), '01' 8K,
+ * '10' 12K, '11' 16K.
+ * 15-14: reserved
+ * 13-12: IRQ destination; '00' none, '01' host driver (normal operation)
+ * 11- 4: timeout for closing Rx buffer and interrupting host (units 32 usec)
+ * typical value 0x10 (about 1/2 msec)
+ * 3- 0: reserved
+ */
#define FH_MEM_RCSR_LOWER_BOUND (FH_MEM_LOWER_BOUND + 0xC00)
#define FH_MEM_RCSR_UPPER_BOUND (FH_MEM_LOWER_BOUND + 0xCC0)
#define FH_MEM_RCSR_CHNL0 (FH_MEM_RCSR_LOWER_BOUND)
#define FH_MEM_RCSR_CHNL0_CONFIG_REG (FH_MEM_RCSR_CHNL0)
-/* RSSR Area - Rx shared ctrl & status registers */
+#define FH_RCSR_CHNL0_RX_CONFIG_RB_TIMEOUT_MASK (0x00000FF0) /* bit 4-11 */
+#define FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_MASK (0x00001000) /* bit 12 */
+#define FH_RCSR_CHNL0_RX_CONFIG_SINGLE_FRAME_MASK (0x00008000) /* bit 15 */
+#define FH_RCSR_CHNL0_RX_CONFIG_RB_SIZE_MASK (0x00030000) /* bits 16-17 */
+#define FH_RCSR_CHNL0_RX_CONFIG_RBDBC_SIZE_MASK (0x00F00000) /* bits 20-23 */
+#define FH_RCSR_CHNL0_RX_CONFIG_DMA_CHNL_EN_MASK (0xC0000000) /* bits 30-31 */
+
+#define FH_RCSR_RX_CONFIG_RBDCB_SIZE_BITSHIFT (20)
+#define FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_BITSHIFT (4)
+#define RX_RB_TIMEOUT (0x10)
+
+#define FH_RCSR_RX_CONFIG_CHNL_EN_PAUSE_VAL (0x00000000)
+#define FH_RCSR_RX_CONFIG_CHNL_EN_PAUSE_EOF_VAL (0x40000000)
+#define FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL (0x80000000)
+
+#define FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K (0x00000000)
+#define FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K (0x00010000)
+#define FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K (0x00020000)
+#define FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_16K (0x00030000)
+
+#define FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_NO_INT_VAL (0x00000000)
+#define FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL (0x00001000)
+
+
+/**
+ * Rx Shared Status Registers (RSSR)
+ *
+ * After stopping Rx DMA channel (writing 0 to FH_MEM_RCSR_CHNL0_CONFIG_REG),
+ * driver must poll FH_MEM_RSSR_RX_STATUS_REG until Rx channel is idle.
+ *
+ * Bit fields:
+ * 24: 1 = Channel 0 is idle
+ *
+ * FH_MEM_RSSR_SHARED_CTRL_REG and FH_MEM_RSSR_RX_ENABLE_ERR_IRQ2DRV contain
+ * default values that should not be altered by the driver.
+ */
#define FH_MEM_RSSR_LOWER_BOUND (FH_MEM_LOWER_BOUND + 0xC40)
#define FH_MEM_RSSR_UPPER_BOUND (FH_MEM_LOWER_BOUND + 0xD00)
+
#define FH_MEM_RSSR_SHARED_CTRL_REG (FH_MEM_RSSR_LOWER_BOUND)
#define FH_MEM_RSSR_RX_STATUS_REG (FH_MEM_RSSR_LOWER_BOUND + 0x004)
#define FH_MEM_RSSR_RX_ENABLE_ERR_IRQ2DRV (FH_MEM_RSSR_LOWER_BOUND + 0x008)
-/* TCSR */
+#define FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE (0x01000000)
+
+
+/**
+ * Transmit DMA Channel Control/Status Registers (TCSR)
+ *
+ * 4965 has one configuration register for each of 8 Tx DMA/FIFO channels
+ * supported in hardware (don't confuse these with the 16 Tx queues in DRAM,
+ * which feed the DMA/FIFO channels); config regs are separated by 0x20 bytes.
+ *
+ * To use a Tx DMA channel, driver must initialize its
+ * IWL_FH_TCSR_CHNL_TX_CONFIG_REG(chnl) with:
+ *
+ * IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
+ * IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE_VAL
+ *
+ * All other bits should be 0.
+ *
+ * Bit fields:
+ * 31-30: Tx DMA channel enable: '00' off/pause, '01' pause at end of frame,
+ * '10' operate normally
+ * 29- 4: Reserved, set to "0"
+ * 3: Enable internal DMA requests (1, normal operation), disable (0)
+ * 2- 0: Reserved, set to "0"
+ */
#define IWL_FH_TCSR_LOWER_BOUND (FH_MEM_LOWER_BOUND + 0xD00)
#define IWL_FH_TCSR_UPPER_BOUND (FH_MEM_LOWER_BOUND + 0xE60)
+/* Find Control/Status reg for given Tx DMA/FIFO channel */
#define IWL_FH_TCSR_CHNL_TX_CONFIG_REG(_chnl) \
(IWL_FH_TCSR_LOWER_BOUND + 0x20 * _chnl)
-/* TSSR Area - Tx shared status registers */
-/* TSSR */
+#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_DISABLE_VAL (0x00000000)
+#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE_VAL (0x00000008)
+
+#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_PAUSE (0x00000000)
+#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_PAUSE_EOF (0x40000000)
+#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE (0x80000000)
+
+/**
+ * Tx Shared Status Registers (TSSR)
+ *
+ * After stopping Tx DMA channel (writing 0 to
+ * IWL_FH_TCSR_CHNL_TX_CONFIG_REG(chnl)), driver must poll
+ * IWL_FH_TSSR_TX_STATUS_REG until selected Tx channel is idle
+ * (channel's buffers empty | no pending requests).
+ *
+ * Bit fields:
+ * 31-24: 1 = Channel buffers empty (channel 7:0)
+ * 23-16: 1 = No pending requests (channel 7:0)
+ */
#define IWL_FH_TSSR_LOWER_BOUND (FH_MEM_LOWER_BOUND + 0xEA0)
#define IWL_FH_TSSR_UPPER_BOUND (FH_MEM_LOWER_BOUND + 0xEC0)
(IWL_FH_TSSR_TX_STATUS_REG_BIT_BUFS_EMPTY(_chnl) | \
IWL_FH_TSSR_TX_STATUS_REG_BIT_NO_PEND_REQ(_chnl))
-#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE_VAL (0x00000008)
-#define IWL_FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE (0x80000000)
+/********************* START TX SCHEDULER *************************************/
-/* RCSR: channel 0 rx_config register defines */
+/**
+ * 4965 Tx Scheduler
+ *
+ * The Tx Scheduler selects the next frame to be transmitted, chosing TFDs
+ * (Transmit Frame Descriptors) from up to 16 circular Tx queues resident in
+ * host DRAM. It steers each frame's Tx command (which contains the frame
+ * data) into one of up to 7 prioritized Tx DMA FIFO channels within the
+ * device. A queue maps to only one (selectable by driver) Tx DMA channel,
+ * but one DMA channel may take input from several queues.
+ *
+ * Tx DMA channels have dedicated purposes. For 4965, they are used as follows:
+ *
+ * 0 -- EDCA BK (background) frames, lowest priority
+ * 1 -- EDCA BE (best effort) frames, normal priority
+ * 2 -- EDCA VI (video) frames, higher priority
+ * 3 -- EDCA VO (voice) and management frames, highest priority
+ * 4 -- Commands (e.g. RXON, etc.)
+ * 5 -- HCCA short frames
+ * 6 -- HCCA long frames
+ * 7 -- not used by driver (device-internal only)
+ *
+ * Driver should normally map queues 0-6 to Tx DMA/FIFO channels 0-6.
+ * In addition, driver can map queues 7-15 to Tx DMA/FIFO channels 0-3 to
+ * support 11n aggregation via EDCA DMA channels.
+ *
+ * The driver sets up each queue to work in one of two modes:
+ *
+ * 1) Scheduler-Ack, in which the scheduler automatically supports a
+ * block-ack (BA) window of up to 64 TFDs. In this mode, each queue
+ * contains TFDs for a unique combination of Recipient Address (RA)
+ * and Traffic Identifier (TID), that is, traffic of a given
+ * Quality-Of-Service (QOS) priority, destined for a single station.
+ *
+ * In scheduler-ack mode, the scheduler keeps track of the Tx status of
+ * each frame within the BA window, including whether it's been transmitted,
+ * and whether it's been acknowledged by the receiving station. The device
+ * automatically processes block-acks received from the receiving STA,
+ * and reschedules un-acked frames to be retransmitted (successful
+ * Tx completion may end up being out-of-order).
+ *
+ * The driver must maintain the queue's Byte Count table in host DRAM
+ * (struct iwl4965_sched_queue_byte_cnt_tbl) for this mode.
+ * This mode does not support fragmentation.
+ *
+ * 2) FIFO (a.k.a. non-Scheduler-ACK), in which each TFD is processed in order.
+ * The device may automatically retry Tx, but will retry only one frame
+ * at a time, until receiving ACK from receiving station, or reaching
+ * retry limit and giving up.
+ *
+ * The command queue (#4) must use this mode!
+ * This mode does not require use of the Byte Count table in host DRAM.
+ *
+ * Driver controls scheduler operation via 3 means:
+ * 1) Scheduler registers
+ * 2) Shared scheduler data base in internal 4956 SRAM
+ * 3) Shared data in host DRAM
+ *
+ * Initialization:
+ *
+ * When loading, driver should allocate memory for:
+ * 1) 16 TFD circular buffers, each with space for (typically) 256 TFDs.
+ * 2) 16 Byte Count circular buffers in 16 KBytes contiguous memory
+ * (1024 bytes for each queue).
+ *
+ * After receiving "Alive" response from uCode, driver must initialize
+ * the scheduler (especially for queue #4, the command queue, otherwise
+ * the driver can't issue commands!):
+ */
-#define FH_RCSR_RX_CONFIG_RBDCB_SIZE_BITSHIFT (20)
+/**
+ * Max Tx window size is the max number of contiguous TFDs that the scheduler
+ * can keep track of at one time when creating block-ack chains of frames.
+ * Note that "64" matches the number of ack bits in a block-ack packet.
+ * Driver should use SCD_WIN_SIZE and SCD_FRAME_LIMIT values to initialize
+ * SCD_CONTEXT_QUEUE_OFFSET(x) values.
+ */
+#define SCD_WIN_SIZE 64
+#define SCD_FRAME_LIMIT 64
-#define FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL (0x80000000)
+/* SCD registers are internal, must be accessed via HBUS_TARG_PRPH regs */
+#define SCD_START_OFFSET 0xa02c00
-#define IWL_FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K (0x00000000)
+/*
+ * 4965 tells driver SRAM address for internal scheduler structs via this reg.
+ * Value is valid only after "Alive" response from uCode.
+ */
+#define SCD_SRAM_BASE_ADDR (SCD_START_OFFSET + 0x0)
-/* RCSR channel 0 config register values */
-#define FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL (0x00001000)
+/*
+ * Driver may need to update queue-empty bits after changing queue's
+ * write and read pointers (indexes) during (re-)initialization (i.e. when
+ * scheduler is not tracking what's happening).
+ * Bit fields:
+ * 31-16: Write mask -- 1: update empty bit, 0: don't change empty bit
+ * 15-00: Empty state, one for each queue -- 1: empty, 0: non-empty
+ * NOTE: This register is not used by Linux driver.
+ */
+#define SCD_EMPTY_BITS (SCD_START_OFFSET + 0x4)
-#define SCD_WIN_SIZE 64
-#define SCD_FRAME_LIMIT 64
+/*
+ * Physical base address of array of byte count (BC) circular buffers (CBs).
+ * Each Tx queue has a BC CB in host DRAM to support Scheduler-ACK mode.
+ * This register points to BC CB for queue 0, must be on 1024-byte boundary.
+ * Others are spaced by 1024 bytes.
+ * Each BC CB is 2 bytes * (256 + 64) = 740 bytes, followed by 384 bytes pad.
+ * (Index into a queue's BC CB) = (index into queue's TFD CB) = (SSN & 0xff).
+ * Bit fields:
+ * 25-00: Byte Count CB physical address [35:10], must be 1024-byte aligned.
+ */
+#define SCD_DRAM_BASE_ADDR (SCD_START_OFFSET + 0x10)
-/* SRAM structures */
-#define SCD_CONTEXT_DATA_OFFSET 0x380
-#define SCD_TX_STTS_BITMAP_OFFSET 0x400
-#define SCD_TRANSLATE_TBL_OFFSET 0x500
-#define SCD_CONTEXT_QUEUE_OFFSET(x) (SCD_CONTEXT_DATA_OFFSET + ((x) * 8))
-#define SCD_TRANSLATE_TBL_OFFSET_QUEUE(x) \
- ((SCD_TRANSLATE_TBL_OFFSET + ((x) * 2)) & 0xfffffffc)
+/*
+ * Enables any/all Tx DMA/FIFO channels.
+ * Scheduler generates requests for only the active channels.
+ * Set this to 0xff to enable all 8 channels (normal usage).
+ * Bit fields:
+ * 7- 0: Enable (1), disable (0), one bit for each channel 0-7
+ */
+#define SCD_TXFACT (SCD_START_OFFSET + 0x1c)
+/* Mask to enable contiguous Tx DMA/FIFO channels between "lo" and "hi". */
#define SCD_TXFACT_REG_TXFIFO_MASK(lo, hi) \
- ((1<<(hi))|((1<<(hi))-(1<<(lo))))
+ ((1 << (hi)) | ((1 << (hi)) - (1 << (lo))))
+/*
+ * Queue (x) Write Pointers (indexes, really!), one for each Tx queue.
+ * Initialized and updated by driver as new TFDs are added to queue.
+ * NOTE: If using Block Ack, index must correspond to frame's
+ * Start Sequence Number; index = (SSN & 0xff)
+ * NOTE: Alternative to HBUS_TARG_WRPTR, which is what Linux driver uses?
+ */
+#define SCD_QUEUE_WRPTR(x) (SCD_START_OFFSET + 0x24 + (x) * 4)
+
+/*
+ * Queue (x) Read Pointers (indexes, really!), one for each Tx queue.
+ * For FIFO mode, index indicates next frame to transmit.
+ * For Scheduler-ACK mode, index indicates first frame in Tx window.
+ * Initialized by driver, updated by scheduler.
+ */
+#define SCD_QUEUE_RDPTR(x) (SCD_START_OFFSET + 0x64 + (x) * 4)
+
+/*
+ * Select which queues work in chain mode (1) vs. not (0).
+ * Use chain mode to build chains of aggregated frames.
+ * Bit fields:
+ * 31-16: Reserved
+ * 15-00: Mode, one bit for each queue -- 1: Chain mode, 0: one-at-a-time
+ * NOTE: If driver sets up queue for chain mode, it should be also set up
+ * Scheduler-ACK mode as well, via SCD_QUEUE_STATUS_BITS(x).
+ */
+#define SCD_QUEUECHAIN_SEL (SCD_START_OFFSET + 0xd0)
+
+/*
+ * Select which queues interrupt driver when scheduler increments
+ * a queue's read pointer (index).
+ * Bit fields:
+ * 31-16: Reserved
+ * 15-00: Interrupt enable, one bit for each queue -- 1: enabled, 0: disabled
+ * NOTE: This functionality is apparently a no-op; driver relies on interrupts
+ * from Rx queue to read Tx command responses and update Tx queues.
+ */
+#define SCD_INTERRUPT_MASK (SCD_START_OFFSET + 0xe4)
+
+/*
+ * Queue search status registers. One for each queue.
+ * Sets up queue mode and assigns queue to Tx DMA channel.
+ * Bit fields:
+ * 19-10: Write mask/enable bits for bits 0-9
+ * 9: Driver should init to "0"
+ * 8: Scheduler-ACK mode (1), non-Scheduler-ACK (i.e. FIFO) mode (0).
+ * Driver should init to "1" for aggregation mode, or "0" otherwise.
+ * 7-6: Driver should init to "0"
+ * 5: Window Size Left; indicates whether scheduler can request
+ * another TFD, based on window size, etc. Driver should init
+ * this bit to "1" for aggregation mode, or "0" for non-agg.
+ * 4-1: Tx FIFO to use (range 0-7).
+ * 0: Queue is active (1), not active (0).
+ * Other bits should be written as "0"
+ *
+ * NOTE: If enabling Scheduler-ACK mode, chain mode should also be enabled
+ * via SCD_QUEUECHAIN_SEL.
+ */
+#define SCD_QUEUE_STATUS_BITS(x) (SCD_START_OFFSET + 0x104 + (x) * 4)
+
+/* Bit field positions */
#define SCD_QUEUE_STTS_REG_POS_ACTIVE (0)
#define SCD_QUEUE_STTS_REG_POS_TXF (1)
#define SCD_QUEUE_STTS_REG_POS_WSL (5)
#define SCD_QUEUE_STTS_REG_POS_SCD_ACK (8)
+
+/* Write masks */
#define SCD_QUEUE_STTS_REG_POS_SCD_ACT_EN (10)
#define SCD_QUEUE_STTS_REG_MSK (0x0007FC00)
+/**
+ * 4965 internal SRAM structures for scheduler, shared with driver ...
+ *
+ * Driver should clear and initialize the following areas after receiving
+ * "Alive" response from 4965 uCode, i.e. after initial
+ * uCode load, or after a uCode load done for error recovery:
+ *
+ * SCD_CONTEXT_DATA_OFFSET (size 128 bytes)
+ * SCD_TX_STTS_BITMAP_OFFSET (size 256 bytes)
+ * SCD_TRANSLATE_TBL_OFFSET (size 32 bytes)
+ *
+ * Driver accesses SRAM via HBUS_TARG_MEM_* registers.
+ * Driver reads base address of this scheduler area from SCD_SRAM_BASE_ADDR.
+ * All OFFSET values must be added to this base address.
+ */
+
+/*
+ * Queue context. One 8-byte entry for each of 16 queues.
+ *
+ * Driver should clear this entire area (size 0x80) to 0 after receiving
+ * "Alive" notification from uCode. Additionally, driver should init
+ * each queue's entry as follows:
+ *
+ * LS Dword bit fields:
+ * 0-06: Max Tx window size for Scheduler-ACK. Driver should init to 64.
+ *
+ * MS Dword bit fields:
+ * 16-22: Frame limit. Driver should init to 10 (0xa).
+ *
+ * Driver should init all other bits to 0.
+ *
+ * Init must be done after driver receives "Alive" response from 4965 uCode,
+ * and when setting up queue for aggregation.
+ */
+#define SCD_CONTEXT_DATA_OFFSET 0x380
+#define SCD_CONTEXT_QUEUE_OFFSET(x) (SCD_CONTEXT_DATA_OFFSET + ((x) * 8))
+
#define SCD_QUEUE_CTX_REG1_WIN_SIZE_POS (0)
#define SCD_QUEUE_CTX_REG1_WIN_SIZE_MSK (0x0000007F)
-
#define SCD_QUEUE_CTX_REG2_FRAME_LIMIT_POS (16)
#define SCD_QUEUE_CTX_REG2_FRAME_LIMIT_MSK (0x007F0000)
-#define CSR_HW_IF_CONFIG_REG_BIT_KEDRON_R (0x00000010)
-#define CSR_HW_IF_CONFIG_REG_MSK_BOARD_VER (0x00000C00)
-#define CSR_HW_IF_CONFIG_REG_BIT_MAC_SI (0x00000100)
-#define CSR_HW_IF_CONFIG_REG_BIT_RADIO_SI (0x00000200)
-#define CSR_HW_IF_CONFIG_REG_BIT_EEPROM_OWN_SEM (0x00200000)
+/*
+ * Tx Status Bitmap
+ *
+ * Driver should clear this entire area (size 0x100) to 0 after receiving
+ * "Alive" notification from uCode. Area is used only by device itself;
+ * no other support (besides clearing) is required from driver.
+ */
+#define SCD_TX_STTS_BITMAP_OFFSET 0x400
+
+/*
+ * RAxTID to queue translation mapping.
+ *
+ * When queue is in Scheduler-ACK mode, frames placed in a that queue must be
+ * for only one combination of receiver address (RA) and traffic ID (TID), i.e.
+ * one QOS priority level destined for one station (for this wireless link,
+ * not final destination). The SCD_TRANSLATE_TABLE area provides 16 16-bit
+ * mappings, one for each of the 16 queues. If queue is not in Scheduler-ACK
+ * mode, the device ignores the mapping value.
+ *
+ * Bit fields, for each 16-bit map:
+ * 15-9: Reserved, set to 0
+ * 8-4: Index into device's station table for recipient station
+ * 3-0: Traffic ID (tid), range 0-15
+ *
+ * Driver should clear this entire area (size 32 bytes) to 0 after receiving
+ * "Alive" notification from uCode. To update a 16-bit map value, driver
+ * must read a dword-aligned value from device SRAM, replace the 16-bit map
+ * value of interest, and write the dword value back into device SRAM.
+ */
+#define SCD_TRANSLATE_TBL_OFFSET 0x500
+
+/* Find translation table dword to read/write for given queue */
+#define SCD_TRANSLATE_TBL_OFFSET_QUEUE(x) \
+ ((SCD_TRANSLATE_TBL_OFFSET + ((x) * 2)) & 0xfffffffc)
+
+#define SCD_TXFIFO_POS_TID (0)
+#define SCD_TXFIFO_POS_RA (4)
+#define SCD_QUEUE_RA_TID_MAP_RATID_MSK (0x01FF)
+
+/*********************** END TX SCHEDULER *************************************/
static inline u8 iwl4965_hw_get_rate(__le32 rate_n_flags)
{
return cpu_to_le32(flags|(u16)rate);
}
+
+/**
+ * Tx/Rx Queues
+ *
+ * Most communication between driver and 4965 is via queues of data buffers.
+ * For example, all commands that the driver issues to device's embedded
+ * controller (uCode) are via the command queue (one of the Tx queues). All
+ * uCode command responses/replies/notifications, including Rx frames, are
+ * conveyed from uCode to driver via the Rx queue.
+ *
+ * Most support for these queues, including handshake support, resides in
+ * structures in host DRAM, shared between the driver and the device. When
+ * allocating this memory, the driver must make sure that data written by
+ * the host CPU updates DRAM immediately (and does not get "stuck" in CPU's
+ * cache memory), so DRAM and cache are consistent, and the device can
+ * immediately see changes made by the driver.
+ *
+ * 4965 supports up to 16 DRAM-based Tx queues, and services these queues via
+ * up to 7 DMA channels (FIFOs). Each Tx queue is supported by a circular array
+ * in DRAM containing 256 Transmit Frame Descriptors (TFDs).
+ */
+#define IWL4965_MAX_WIN_SIZE 64
+#define IWL4965_QUEUE_SIZE 256
+#define IWL4965_NUM_FIFOS 7
+#define IWL_MAX_NUM_QUEUES 16
+
+
+/**
+ * struct iwl4965_tfd_frame_data
+ *
+ * Describes up to 2 buffers containing (contiguous) portions of a Tx frame.
+ * Each buffer must be on dword boundary.
+ * Up to 10 iwl_tfd_frame_data structures, describing up to 20 buffers,
+ * may be filled within a TFD (iwl_tfd_frame).
+ *
+ * Bit fields in tb1_addr:
+ * 31- 0: Tx buffer 1 address bits [31:0]
+ *
+ * Bit fields in val1:
+ * 31-16: Tx buffer 2 address bits [15:0]
+ * 15- 4: Tx buffer 1 length (bytes)
+ * 3- 0: Tx buffer 1 address bits [32:32]
+ *
+ * Bit fields in val2:
+ * 31-20: Tx buffer 2 length (bytes)
+ * 19- 0: Tx buffer 2 address bits [35:16]
+ */
struct iwl4965_tfd_frame_data {
__le32 tb1_addr;
#define IWL_tb2_len_SYM val2
} __attribute__ ((packed));
+
+/**
+ * struct iwl4965_tfd_frame
+ *
+ * Transmit Frame Descriptor (TFD)
+ *
+ * 4965 supports up to 16 Tx queues resident in host DRAM.
+ * Each Tx queue uses a circular buffer of 256 TFDs stored in host DRAM.
+ * Both driver and device share these circular buffers, each of which must be
+ * contiguous 256 TFDs x 128 bytes-per-TFD = 32 KBytes for 4965.
+ *
+ * Driver must indicate the physical address of the base of each
+ * circular buffer via the 4965's FH_MEM_CBBC_QUEUE registers.
+ *
+ * Each TFD contains pointer/size information for up to 20 data buffers
+ * in host DRAM. These buffers collectively contain the (one) frame described
+ * by the TFD. Each buffer must be a single contiguous block of memory within
+ * itself, but buffers may be scattered in host DRAM. Each buffer has max size
+ * of (4K - 4). The 4965 concatenates all of a TFD's buffers into a single
+ * Tx frame, up to 8 KBytes in size.
+ *
+ * Bit fields in the control dword (val0):
+ * 31-30: # dwords (0-3) of padding required at end of frame for 16-byte bound
+ * 29: reserved
+ * 28-24: # Transmit Buffer Descriptors in TFD
+ * 23- 0: reserved
+ *
+ * A maximum of 255 (not 256!) TFDs may be on a queue waiting for Tx.
+ */
struct iwl4965_tfd_frame {
__le32 val0;
/* __le32 rsvd1:24; */
__le32 reserved;
} __attribute__ ((packed));
-#define IWL4965_MAX_WIN_SIZE 64
-#define IWL4965_QUEUE_SIZE 256
-#define IWL4965_NUM_FIFOS 7
-#define IWL_MAX_NUM_QUEUES 16
+/**
+ * struct iwl4965_queue_byte_cnt_entry
+ *
+ * Byte Count Table Entry
+ *
+ * Bit fields:
+ * 15-12: reserved
+ * 11- 0: total to-be-transmitted byte count of frame (does not include command)
+ */
struct iwl4965_queue_byte_cnt_entry {
__le16 val;
/* __le16 byte_cnt:12; */
/* __le16 rsvd:4; */
} __attribute__ ((packed));
+
+/**
+ * struct iwl4965_sched_queue_byte_cnt_tbl
+ *
+ * Byte Count table
+ *
+ * Each Tx queue uses a byte-count table containing 320 entries:
+ * one 16-bit entry for each of 256 TFDs, plus an additional 64 entries that
+ * duplicate the first 64 entries (to avoid wrap-around within a Tx window;
+ * max Tx window is 64 TFDs).
+ *
+ * When driver sets up a new TFD, it must also enter the total byte count
+ * of the frame to be transmitted into the corresponding entry in the byte
+ * count table for the chosen Tx queue. If the TFD index is 0-63, the driver
+ * must duplicate the byte count entry in corresponding index 256-319.
+ *
+ * "dont_care" padding puts each byte count table on a 1024-byte boundary;
+ * 4965 assumes tables are separated by 1024 bytes.
+ */
struct iwl4965_sched_queue_byte_cnt_tbl {
struct iwl4965_queue_byte_cnt_entry tfd_offset[IWL4965_QUEUE_SIZE +
IWL4965_MAX_WIN_SIZE];
sizeof(__le16)];
} __attribute__ ((packed));
-/* Base physical address of iwl4965_shared is provided to KDR_SCD_DRAM_BASE_ADDR
- * and &iwl4965_shared.val0 is provided to FH_RSCSR_CHNL0_STTS_WPTR_REG */
+
+/**
+ * struct iwl4965_shared - handshake area for Tx and Rx
+ *
+ * For convenience in allocating memory, this structure combines 2 areas of
+ * DRAM which must be shared between driver and 4965. These do not need to
+ * be combined, if better allocation would result from keeping them separate:
+ *
+ * 1) The Tx byte count tables occupy 1024 bytes each (16 KBytes total for
+ * 16 queues). Driver uses SCD_DRAM_BASE_ADDR to tell 4965 where to find
+ * the first of these tables. 4965 assumes tables are 1024 bytes apart.
+ *
+ * 2) The Rx status (val0 and val1) occupies only 8 bytes. Driver uses
+ * FH_RSCSR_CHNL0_STTS_WPTR_REG to tell 4965 where to find this area.
+ * Driver reads val0 to determine the latest Receive Buffer Descriptor (RBD)
+ * that has been filled by the 4965.
+ *
+ * Bit fields val0:
+ * 31-12: Not used
+ * 11- 0: Index of last filled Rx buffer descriptor (4965 writes, driver reads)
+ *
+ * Bit fields val1:
+ * 31- 0: Not used
+ */
struct iwl4965_shared {
struct iwl4965_sched_queue_byte_cnt_tbl
queues_byte_cnt_tbls[IWL_MAX_NUM_QUEUES];