On 4/7/24 06:21, Vinod Koul wrote:
On 05-03-24, 13:49, Lizhi Hou wrote:From: Nishad Saraf <nishads@xxxxxxx> Adds driver to enable PCIe board which uses AMD QDMA (the Queue-based Direct Memory Access) subsystem. For example, Xilinx Alveo V70 AI Accelerator devices. https://www.xilinx.com/applications/data-center/v70.html The QDMA subsystem is used in conjunction with the PCI Express IP block to provide high performance data transfer between host memory and the card's DMA subsystem. +-------+ +-------+ +-----------+ PCIe | | | | | | Tx/Rx | | | | AXI | | <=======> | PCIE | <===> | QDMA | <====>| User Logic| | | | | | | +-------+ +-------+ +-----------+ The primary mechanism to transfer data using the QDMA is for the QDMA engine to operate on instructions (descriptors) provided by the host operating system. Using the descriptors, the QDMA can move data in both the Host to Card (H2C) direction, or the Card to Host (C2H) direction. The QDMA provides a per-queue basis option whether DMA traffic goes to an AXI4 memory map (MM) interface or to an AXI4-Stream interface. The hardware detail is provided by https://docs.xilinx.com/r/en-US/pg302-qdma Implements dmaengine APIs to support MM DMA transfers. - probe the available DMA channels - use dma_slave_map for channel lookup - use virtual channel to manage dmaengine tx descriptors - implement device_prep_slave_sg callback to handle host scatter gather list - implement descriptor metadata operations to set device address for DMA transfer Signed-off-by: Nishad Saraf <nishads@xxxxxxx> Signed-off-by: Lizhi Hou <lizhi.hou@xxxxxxx> --- MAINTAINERS | 8 + drivers/dma/Kconfig | 2 + drivers/dma/Makefile | 1 + drivers/dma/amd/Kconfig | 14 + drivers/dma/amd/Makefile | 6 + drivers/dma/amd/qdma/Makefile | 8 + drivers/dma/amd/qdma/qdma-comm-regs.c | 64 ++ drivers/dma/amd/qdma/qdma.c | 1162 ++++++++++++++++++++++++ drivers/dma/amd/qdma/qdma.h | 265 ++++++ include/linux/platform_data/amd_qdma.h | 36 + 10 files changed, 1566 insertions(+) create mode 100644 drivers/dma/amd/Kconfig create mode 100644 drivers/dma/amd/Makefile create mode 100644 drivers/dma/amd/qdma/Makefile create mode 100644 drivers/dma/amd/qdma/qdma-comm-regs.c create mode 100644 drivers/dma/amd/qdma/qdma.c create mode 100644 drivers/dma/amd/qdma/qdma.h create mode 100644 include/linux/platform_data/amd_qdma.h diff --git a/MAINTAINERS b/MAINTAINERS index 9ed4d3868539..d438214685d9 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1063,6 +1063,14 @@ L: dmaengine@xxxxxxxxxxxxxxx S: Maintained F: drivers/dma/ptdma/+AMD QDMA DRIVER+M: Nishad Saraf <nishads@xxxxxxx> +M: Lizhi Hou <lizhi.hou@xxxxxxx> +L: dmaengine@xxxxxxxxxxxxxxx +S: Supported +F: drivers/dma/amd/qdma/ +F: include/linux/platform_data/amd_qdma.h + AMD SEATTLE DEVICE TREE SUPPORT M: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx> M: Tom Lendacky <thomas.lendacky@xxxxxxx> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig index e928f2ca0f1e..62d09090b94d 100644 --- a/drivers/dma/Kconfig +++ b/drivers/dma/Kconfig @@ -748,6 +748,8 @@ config XILINX_ZYNQMP_DPDMA display driver.# driver files+source "drivers/dma/amd/Kconfig" + source "drivers/dma/bestcomm/Kconfig"source "drivers/dma/mediatek/Kconfig"diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile index dfd40d14e408..f2c614fcf2a4 100644 --- a/drivers/dma/Makefile +++ b/drivers/dma/Makefile @@ -84,6 +84,7 @@ obj-$(CONFIG_ST_FDMA) += st_fdma.o obj-$(CONFIG_FSL_DPAA2_QDMA) += fsl-dpaa2-qdma/ obj-$(CONFIG_INTEL_LDMA) += lgm/+obj-y += amd/obj-y += mediatek/ obj-y += qcom/ obj-y += ti/ diff --git a/drivers/dma/amd/Kconfig b/drivers/dma/amd/Kconfig new file mode 100644 index 000000000000..7d1f51d69675 --- /dev/null +++ b/drivers/dma/amd/Kconfig @@ -0,0 +1,14 @@ +# SPDX-License-Identifier: GPL-2.0-only + +config AMD_QDMA + tristate "AMD Queue-based DMA" + depends on HAS_IOMEM + select DMA_ENGINE + select DMA_VIRTUAL_CHANNELS + select REGMAP_MMIO + help + Enable support for the AMD Queue-based DMA subsystem. The primary + mechanism to transfer data using the QDMA is for the QDMA engine to + operate on instructions (descriptors) provided by the host operating + system. Using the descriptors, the QDMA can move data in either the + Host to Card (H2C) direction or the Card to Host (C2H) direction. diff --git a/drivers/dma/amd/Makefile b/drivers/dma/amd/Makefile new file mode 100644 index 000000000000..99c77de9a9dd --- /dev/null +++ b/drivers/dma/amd/Makefile @@ -0,0 +1,6 @@ +# SPDX-License-Identifier: GPL-2.0 +# +# Copyright (C) 2023, Advanced Micro Devices, Inc.Copyright for Makefile?
Ok. I will remove the Copyright
+# + +obj-$(CONFIG_AMD_QDMA) += qdma/ diff --git a/drivers/dma/amd/qdma/Makefile b/drivers/dma/amd/qdma/Makefile new file mode 100644 index 000000000000..ba53971d2714 --- /dev/null +++ b/drivers/dma/amd/qdma/Makefile @@ -0,0 +1,8 @@ +# SPDX-License-Identifier: GPL-2.0 +# +# Copyright (C) 2023, Advanced Micro Devices, Inc.2024
I will fix this.
+# + +obj-$(CONFIG_AMD_QDMA) += amd-qdma.o + +amd-qdma-$(CONFIG_AMD_QDMA) := qdma.o qdma-comm-regs.o diff --git a/drivers/dma/amd/qdma/qdma-comm-regs.c b/drivers/dma/amd/qdma/qdma-comm-regs.c new file mode 100644 index 000000000000..14c7a73268b4 --- /dev/null +++ b/drivers/dma/amd/qdma/qdma-comm-regs.c @@ -0,0 +1,64 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * Copyright (C) 2023, Advanced Micro Devices, Inc. + */ + +#ifndef __QDMA_REGS_DEF_H +#define __QDMA_REGS_DEF_H + +#include "qdma.h" + +const struct qdma_reg qdma_regos_default[QDMA_REGO_MAX] = { + [QDMA_REGO_CTXT_DATA] = QDMA_REGO(0x804, 8), + [QDMA_REGO_CTXT_CMD] = QDMA_REGO(0x844, 1), + [QDMA_REGO_CTXT_MASK] = QDMA_REGO(0x824, 8), + [QDMA_REGO_MM_H2C_CTRL] = QDMA_REGO(0x1004, 1), + [QDMA_REGO_MM_C2H_CTRL] = QDMA_REGO(0x1204, 1), + [QDMA_REGO_QUEUE_COUNT] = QDMA_REGO(0x120, 1), + [QDMA_REGO_RING_SIZE] = QDMA_REGO(0x204, 1), + [QDMA_REGO_H2C_PIDX] = QDMA_REGO(0x18004, 1), + [QDMA_REGO_C2H_PIDX] = QDMA_REGO(0x18008, 1), + [QDMA_REGO_INTR_CIDX] = QDMA_REGO(0x18000, 1), + [QDMA_REGO_FUNC_ID] = QDMA_REGO(0x12c, 1), + [QDMA_REGO_ERR_INT] = QDMA_REGO(0xb04, 1), + [QDMA_REGO_ERR_STAT] = QDMA_REGO(0x248, 1), +}; + +const struct qdma_reg_field qdma_regfs_default[QDMA_REGF_MAX] = { + /* QDMA_REGO_CTXT_DATA fields */ + [QDMA_REGF_IRQ_ENABLE] = QDMA_REGF(53, 53), + [QDMA_REGF_WBK_ENABLE] = QDMA_REGF(52, 52), + [QDMA_REGF_WBI_CHECK] = QDMA_REGF(34, 34), + [QDMA_REGF_IRQ_ARM] = QDMA_REGF(16, 16), + [QDMA_REGF_IRQ_VEC] = QDMA_REGF(138, 128), + [QDMA_REGF_IRQ_AGG] = QDMA_REGF(139, 139), + [QDMA_REGF_WBI_INTVL_ENABLE] = QDMA_REGF(35, 35), + [QDMA_REGF_MRKR_DISABLE] = QDMA_REGF(62, 62), + [QDMA_REGF_QUEUE_ENABLE] = QDMA_REGF(32, 32), + [QDMA_REGF_QUEUE_MODE] = QDMA_REGF(63, 63), + [QDMA_REGF_DESC_BASE] = QDMA_REGF(127, 64), + [QDMA_REGF_DESC_SIZE] = QDMA_REGF(49, 48), + [QDMA_REGF_RING_ID] = QDMA_REGF(47, 44), + [QDMA_REGF_QUEUE_BASE] = QDMA_REGF(11, 0), + [QDMA_REGF_QUEUE_MAX] = QDMA_REGF(44, 32), + [QDMA_REGF_FUNCTION_ID] = QDMA_REGF(24, 17), + [QDMA_REGF_INTR_AGG_BASE] = QDMA_REGF(66, 15), + [QDMA_REGF_INTR_VECTOR] = QDMA_REGF(11, 1), + [QDMA_REGF_INTR_SIZE] = QDMA_REGF(69, 67), + [QDMA_REGF_INTR_VALID] = QDMA_REGF(0, 0), + [QDMA_REGF_INTR_COLOR] = QDMA_REGF(14, 14), + [QDMA_REGF_INTR_FUNCTION_ID] = QDMA_REGF(125, 114), + /* QDMA_REGO_CTXT_CMD fields */ + [QDMA_REGF_CMD_INDX] = QDMA_REGF(19, 7), + [QDMA_REGF_CMD_CMD] = QDMA_REGF(6, 5), + [QDMA_REGF_CMD_TYPE] = QDMA_REGF(4, 1), + [QDMA_REGF_CMD_BUSY] = QDMA_REGF(0, 0), + /* QDMA_REGO_QUEUE_COUNT fields */ + [QDMA_REGF_QUEUE_COUNT] = QDMA_REGF(11, 0), + /* QDMA_REGO_ERR_INT fields */ + [QDMA_REGF_ERR_INT_FUNC] = QDMA_REGF(11, 0), + [QDMA_REGF_ERR_INT_VEC] = QDMA_REGF(22, 12), + [QDMA_REGF_ERR_INT_ARM] = QDMA_REGF(24, 24), +}; + +#endif /* __QDMA_REGS_DEF_H */ diff --git a/drivers/dma/amd/qdma/qdma.c b/drivers/dma/amd/qdma/qdma.c new file mode 100644 index 000000000000..a1427a72c002 --- /dev/null +++ b/drivers/dma/amd/qdma/qdma.c @@ -0,0 +1,1162 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * DMA driver for AMD Queue-based DMA Subsystem + * + * Copyright (C) 2023, Advanced Micro Devices, Inc. + */ +#include <linux/bitfield.h> +#include <linux/bitops.h> +#include <linux/dmaengine.h> +#include <linux/module.h> +#include <linux/mod_devicetable.h> +#include <linux/dma-map-ops.h> +#include <linux/platform_device.h> +#include <linux/platform_data/amd_qdma.h> +#include <linux/regmap.h> + +#include "qdma.h" + +#define CHAN_STR(q) (((q)->dir == DMA_MEM_TO_DEV) ? "H2C" : "C2H") +#define QDMA_REG_OFF(d, r) ((d)->roffs[r].off) + +/* MMIO regmap config for all QDMA registers */ +static const struct regmap_config qdma_regmap_config = { + .reg_bits = 32, + .val_bits = 32, + .reg_stride = 4, +}; + +static inline struct qdma_queue *to_qdma_queue(struct dma_chan *chan) +{ + return container_of(chan, struct qdma_queue, vchan.chan); +} + +static inline struct qdma_mm_vdesc *to_qdma_vdesc(struct virt_dma_desc *vdesc) +{ + return container_of(vdesc, struct qdma_mm_vdesc, vdesc); +} + +static inline u32 qdma_get_intr_ring_idx(struct qdma_device *qdev) +{ + u32 idx; + + idx = qdev->qintr_rings[qdev->qintr_ring_idx++].ridx; + qdev->qintr_ring_idx %= qdev->qintr_ring_num; + + return idx; +} + +static u64 qdma_get_field(const struct qdma_device *qdev, const u32 *data, + enum qdma_reg_fields field) +{ + const struct qdma_reg_field *f = &qdev->rfields[field]; + u16 low_pos, hi_pos, low_bit, hi_bit; + u64 value = 0, mask; + + low_pos = f->lsb / BITS_PER_TYPE(*data); + hi_pos = f->msb / BITS_PER_TYPE(*data); + + if (low_pos == hi_pos) { + low_bit = f->lsb % BITS_PER_TYPE(*data); + hi_bit = f->msb % BITS_PER_TYPE(*data);why not use {upper|lower}_xx_bits
f->lsb and f->msb is not multiple of 16 or 32. The code here is to get the value of data(f->msb : f->lsb).
+ mask = GENMASK(hi_bit, low_bit); + value = (data[low_pos] & mask) >> low_bit; + } else if (hi_pos == low_pos + 1) { + low_bit = f->lsb % BITS_PER_TYPE(*data); + hi_bit = low_bit + (f->msb - f->lsb); + value = ((u64)data[hi_pos] << BITS_PER_TYPE(*data)) | + data[low_pos]; + mask = GENMASK_ULL(hi_bit, low_bit); + value = (value & mask) >> low_bit; + } else { + hi_bit = f->msb % BITS_PER_TYPE(*data); + mask = GENMASK(hi_bit, 0); + value = data[hi_pos] & mask; + low_bit = f->msb - f->lsb - hi_bit; + value <<= low_bit; + low_bit -= 32; + value |= (u64)data[hi_pos - 1] << low_bit; + mask = GENMASK(31, 32 - low_bit); + value |= (data[hi_pos - 2] & mask) >> low_bit; + } + + return value; +} + +static void qdma_set_field(const struct qdma_device *qdev, u32 *data, + enum qdma_reg_fields field, u64 value) +{ + const struct qdma_reg_field *f = &qdev->rfields[field]; + u16 low_pos, hi_pos, low_bit; + + low_pos = f->lsb / BITS_PER_TYPE(*data); + hi_pos = f->msb / BITS_PER_TYPE(*data); + low_bit = f->lsb % BITS_PER_TYPE(*data); + + data[low_pos++] |= value << low_bit; + if (low_pos <= hi_pos) + data[low_pos++] |= (u32)(value >> (32 - low_bit)); + if (low_pos <= hi_pos) + data[low_pos] |= (u32)(value >> (64 - low_bit)); +} + +static inline int qdma_reg_write(const struct qdma_device *qdev, + const u32 *data, enum qdma_regs reg) +{ + const struct qdma_reg *r = &qdev->roffs[reg]; + int ret; + + if (r->count > 1) + ret = regmap_bulk_write(qdev->regmap, r->off, data, r->count); + else + ret = regmap_write(qdev->regmap, r->off, *data); + + return ret; +} + +static inline int qdma_reg_read(const struct qdma_device *qdev, u32 *data, + enum qdma_regs reg) +{ + const struct qdma_reg *r = &qdev->roffs[reg]; + int ret; + + if (r->count > 1) + ret = regmap_bulk_read(qdev->regmap, r->off, data, r->count); + else + ret = regmap_read(qdev->regmap, r->off, data); + + return ret; +} + +static int qdma_context_cmd_execute(const struct qdma_device *qdev, + enum qdma_ctxt_type type, + enum qdma_ctxt_cmd cmd, u16 index) +{ + u32 value = 0; + int ret; + + qdma_set_field(qdev, &value, QDMA_REGF_CMD_INDX, index); + qdma_set_field(qdev, &value, QDMA_REGF_CMD_CMD, cmd); + qdma_set_field(qdev, &value, QDMA_REGF_CMD_TYPE, type); + + ret = qdma_reg_write(qdev, &value, QDMA_REGO_CTXT_CMD); + if (ret) + return ret; + + ret = regmap_read_poll_timeout(qdev->regmap, + QDMA_REG_OFF(qdev, QDMA_REGO_CTXT_CMD), + value, + !qdma_get_field(qdev, &value, + QDMA_REGF_CMD_BUSY), + QDMA_POLL_INTRVL_US, + QDMA_POLL_TIMEOUT_US); + if (ret) { + qdma_err(qdev, "Context command execution timed out"); + return ret; + } + + return 0; +} + +static int qdma_context_write_data(const struct qdma_device *qdev, + const u32 *data) +{ + u32 mask[QDMA_CTXT_REGMAP_LEN]; + int ret; + + memset(mask, ~0, sizeof(mask)); + + ret = qdma_reg_write(qdev, mask, QDMA_REGO_CTXT_MASK); + if (ret) + return ret; + + ret = qdma_reg_write(qdev, data, QDMA_REGO_CTXT_DATA); + if (ret) + return ret; + + return 0; +} + +static void qdma_prep_sw_desc_context(const struct qdma_device *qdev, + const struct qdma_ctxt_sw_desc *ctxt, + u32 *data) +{ + memset(data, 0, QDMA_CTXT_REGMAP_LEN * sizeof(*data)); + qdma_set_field(qdev, data, QDMA_REGF_DESC_BASE, ctxt->desc_base); + qdma_set_field(qdev, data, QDMA_REGF_IRQ_VEC, ctxt->vec); + qdma_set_field(qdev, data, QDMA_REGF_FUNCTION_ID, qdev->fid); + + qdma_set_field(qdev, data, QDMA_REGF_DESC_SIZE, QDMA_DESC_SIZE_32B); + qdma_set_field(qdev, data, QDMA_REGF_RING_ID, QDMA_DEFAULT_RING_ID); + qdma_set_field(qdev, data, QDMA_REGF_QUEUE_MODE, QDMA_QUEUE_OP_MM); + qdma_set_field(qdev, data, QDMA_REGF_IRQ_ENABLE, 1); + qdma_set_field(qdev, data, QDMA_REGF_WBK_ENABLE, 1); + qdma_set_field(qdev, data, QDMA_REGF_WBI_CHECK, 1); + qdma_set_field(qdev, data, QDMA_REGF_IRQ_ARM, 1); + qdma_set_field(qdev, data, QDMA_REGF_IRQ_AGG, 1); + qdma_set_field(qdev, data, QDMA_REGF_WBI_INTVL_ENABLE, 1); + qdma_set_field(qdev, data, QDMA_REGF_QUEUE_ENABLE, 1); + qdma_set_field(qdev, data, QDMA_REGF_MRKR_DISABLE, 1); +} + +static void qdma_prep_intr_context(const struct qdma_device *qdev, + const struct qdma_ctxt_intr *ctxt, + u32 *data) +{ + memset(data, 0, QDMA_CTXT_REGMAP_LEN * sizeof(*data)); + qdma_set_field(qdev, data, QDMA_REGF_INTR_AGG_BASE, ctxt->agg_base); + qdma_set_field(qdev, data, QDMA_REGF_INTR_VECTOR, ctxt->vec); + qdma_set_field(qdev, data, QDMA_REGF_INTR_SIZE, ctxt->size); + qdma_set_field(qdev, data, QDMA_REGF_INTR_VALID, ctxt->valid); + qdma_set_field(qdev, data, QDMA_REGF_INTR_COLOR, ctxt->color); + qdma_set_field(qdev, data, QDMA_REGF_INTR_FUNCTION_ID, qdev->fid); +} + +static void qdma_prep_fmap_context(const struct qdma_device *qdev, + const struct qdma_ctxt_fmap *ctxt, + u32 *data) +{ + memset(data, 0, QDMA_CTXT_REGMAP_LEN * sizeof(*data)); + qdma_set_field(qdev, data, QDMA_REGF_QUEUE_BASE, ctxt->qbase); + qdma_set_field(qdev, data, QDMA_REGF_QUEUE_MAX, ctxt->qmax); +} + +/* + * Program the indirect context register space + * + * Once the queue is enabled, context is dynamically updated by hardware. Any + * modification of the context through this API when the queue is enabled can + * result in unexpected behavior. Reading the context when the queue is enabled + * is not recommended as it can result in reduced performance. + */ +static int qdma_prog_context(struct qdma_device *qdev, enum qdma_ctxt_type type, + enum qdma_ctxt_cmd cmd, u16 index, u32 *ctxt) +{ + int ret; + + mutex_lock(&qdev->ctxt_lock); + if (cmd == QDMA_CTXT_WRITE) { + ret = qdma_context_write_data(qdev, ctxt); + if (ret) + goto failed; + } + + ret = qdma_context_cmd_execute(qdev, type, cmd, index); + if (ret) + goto failed; + + if (cmd == QDMA_CTXT_READ) { + ret = qdma_reg_read(qdev, ctxt, QDMA_REGO_CTXT_DATA); + if (ret) + goto failed; + } + +failed: + mutex_unlock(&qdev->ctxt_lock); + + return ret; +} + +static int qdma_check_queue_status(struct qdma_device *qdev, + enum dma_transfer_direction dir, u16 qid) +{ + u32 status, data[QDMA_CTXT_REGMAP_LEN] = {0}; + enum qdma_ctxt_type type; + int ret; + + if (dir == DMA_MEM_TO_DEV) + type = QDMA_CTXT_DESC_SW_H2C; + else + type = QDMA_CTXT_DESC_SW_C2H; + + ret = qdma_prog_context(qdev, type, QDMA_CTXT_READ, qid, data); + if (ret) + return ret; + + status = qdma_get_field(qdev, data, QDMA_REGF_QUEUE_ENABLE); + if (status) { + qdma_err(qdev, "queue %d already in use", qid); + return -EBUSY; + } + + return 0; +} + +static int qdma_clear_queue_context(const struct qdma_queue *queue) +{ + enum qdma_ctxt_type h2c_types[] = { QDMA_CTXT_DESC_SW_H2C, + QDMA_CTXT_DESC_HW_H2C, + QDMA_CTXT_DESC_CR_H2C, + QDMA_CTXT_PFTCH, }; + enum qdma_ctxt_type c2h_types[] = { QDMA_CTXT_DESC_SW_C2H, + QDMA_CTXT_DESC_HW_C2H, + QDMA_CTXT_DESC_CR_C2H, + QDMA_CTXT_PFTCH, }; + struct qdma_device *qdev = queue->qdev; + enum qdma_ctxt_type *type; + int ret, num, i; + + if (queue->dir == DMA_MEM_TO_DEV) { + type = h2c_types; + num = ARRAY_SIZE(h2c_types); + } else { + type = c2h_types; + num = ARRAY_SIZE(c2h_types); + } + for (i = 0; i < num; i++) { + ret = qdma_prog_context(qdev, type[i], QDMA_CTXT_CLEAR, + queue->qid, NULL); + if (ret) { + qdma_err(qdev, "Failed to clear ctxt %d", type[i]); + return ret; + } + } + + return 0; +} + +static int qdma_setup_fmap_context(struct qdma_device *qdev) +{ + u32 ctxt[QDMA_CTXT_REGMAP_LEN]; + struct qdma_ctxt_fmap fmap; + int ret; + + ret = qdma_prog_context(qdev, QDMA_CTXT_FMAP, QDMA_CTXT_CLEAR, + qdev->fid, NULL); + if (ret) { + qdma_err(qdev, "Failed clearing context"); + return ret; + } + + fmap.qbase = 0; + fmap.qmax = qdev->chan_num * 2; + qdma_prep_fmap_context(qdev, &fmap, ctxt); + ret = qdma_prog_context(qdev, QDMA_CTXT_FMAP, QDMA_CTXT_WRITE, + qdev->fid, ctxt); + if (ret) + qdma_err(qdev, "Failed setup fmap, ret %d", ret); + + return ret; +} + +static int qdma_setup_queue_context(struct qdma_device *qdev, + const struct qdma_ctxt_sw_desc *sw_desc, + enum dma_transfer_direction dir, u16 qid) +{ + u32 ctxt[QDMA_CTXT_REGMAP_LEN]; + enum qdma_ctxt_type type; + int ret; + + if (dir == DMA_MEM_TO_DEV) + type = QDMA_CTXT_DESC_SW_H2C; + else + type = QDMA_CTXT_DESC_SW_C2H; + + qdma_prep_sw_desc_context(qdev, sw_desc, ctxt); + /* Setup SW descriptor context */ + ret = qdma_prog_context(qdev, type, QDMA_CTXT_WRITE, qid, ctxt); + if (ret) + qdma_err(qdev, "Failed setup SW desc ctxt for queue: %d", qid); + + return ret; +} + +/* + * Enable or disable memory-mapped DMA engines + * 1: enable, 0: disable + */ +static int qdma_sgdma_control(struct qdma_device *qdev, u32 ctrl) +{ + int ret; + + ret = qdma_reg_write(qdev, &ctrl, QDMA_REGO_MM_H2C_CTRL); + ret |= qdma_reg_write(qdev, &ctrl, QDMA_REGO_MM_C2H_CTRL); + + return ret; +} + +static int qdma_get_hw_info(struct qdma_device *qdev) +{ + struct qdma_platdata *pdata = dev_get_platdata(&qdev->pdev->dev); + u32 value = 0; + int ret; + + ret = qdma_reg_read(qdev, &value, QDMA_REGO_QUEUE_COUNT); + if (ret) + return ret; + + value = qdma_get_field(qdev, &value, QDMA_REGF_QUEUE_COUNT) + 1; + if (pdata->max_mm_channels * 2 > value) { + qdma_err(qdev, "not enough hw queues %d", value); + return -EINVAL; + } + qdev->chan_num = pdata->max_mm_channels; + + ret = qdma_reg_read(qdev, &qdev->fid, QDMA_REGO_FUNC_ID); + if (ret) + return ret; + + qdma_info(qdev, "max channel %d, function id %d", + qdev->chan_num, qdev->fid); + + return 0; +} + +static inline int qdma_update_pidx(const struct qdma_queue *queue, u16 pidx) +{ + struct qdma_device *qdev = queue->qdev; + + return regmap_write(qdev->regmap, queue->pidx_reg, + pidx | QDMA_QUEUE_ARM_BIT); +} + +static inline int qdma_update_cidx(const struct qdma_queue *queue, + u16 ridx, u16 cidx) +{ + struct qdma_device *qdev = queue->qdev; + + return regmap_write(qdev->regmap, queue->cidx_reg, + ((u32)ridx << 16) | cidx); +} + +/** + * qdma_free_vdesc - Free descriptor + * @vdesc: Virtual DMA descriptor + */ +static void qdma_free_vdesc(struct virt_dma_desc *vdesc) +{ + struct qdma_mm_vdesc *vd = to_qdma_vdesc(vdesc); + + kfree(vd); +} + +static int qdma_alloc_queues(struct qdma_device *qdev, + enum dma_transfer_direction dir) +{ + struct qdma_queue *q, **queues; + u32 i, pidx_base; + int ret; + + if (dir == DMA_MEM_TO_DEV) { + queues = &qdev->h2c_queues; + pidx_base = QDMA_REG_OFF(qdev, QDMA_REGO_H2C_PIDX); + } else { + queues = &qdev->c2h_queues; + pidx_base = QDMA_REG_OFF(qdev, QDMA_REGO_C2H_PIDX); + } + + *queues = devm_kcalloc(&qdev->pdev->dev, qdev->chan_num, sizeof(*q), + GFP_KERNEL); + if (!*queues) + return -ENOMEM; + + for (i = 0; i < qdev->chan_num; i++) { + ret = qdma_check_queue_status(qdev, dir, i); + if (ret) + return ret; + + q = &(*queues)[i]; + q->ring_size = QDMA_DEFAULT_RING_SIZE; + q->idx_mask = q->ring_size - 2; + q->qdev = qdev; + q->dir = dir; + q->qid = i; + q->pidx_reg = pidx_base + i * QDMA_DMAP_REG_STRIDE; + q->cidx_reg = QDMA_REG_OFF(qdev, QDMA_REGO_INTR_CIDX) + + i * QDMA_DMAP_REG_STRIDE; + q->vchan.desc_free = qdma_free_vdesc; + vchan_init(&q->vchan, &qdev->dma_dev); + } + + return 0; +} + +static int qdma_device_verify(struct qdma_device *qdev) +{ + u32 value; + int ret; + + ret = regmap_read(qdev->regmap, QDMA_IDENTIFIER_REGOFF, &value); + if (ret) + return ret; + + value = FIELD_GET(QDMA_IDENTIFIER_MASK, value); + if (value != QDMA_IDENTIFIER) { + qdma_err(qdev, "Invalid identifier"); + return -ENODEV; + } + qdev->rfields = qdma_regfs_default; + qdev->roffs = qdma_regos_default; + + return 0; +} + +static int qdma_device_setup(struct qdma_device *qdev) +{ + struct device *dev = &qdev->pdev->dev; + u32 ring_sz = QDMA_DEFAULT_RING_SIZE; + int ret = 0; + + while (dev && get_dma_ops(dev)) + dev = dev->parent; + if (!dev) { + qdma_err(qdev, "dma device not found"); + return -EINVAL; + } + set_dma_ops(&qdev->pdev->dev, get_dma_ops(dev)); + + ret = qdma_setup_fmap_context(qdev); + if (ret) { + qdma_err(qdev, "Failed setup fmap context"); + return ret; + } + + /* Setup global ring buffer size at QDMA_DEFAULT_RING_ID index */ + ret = qdma_reg_write(qdev, &ring_sz, QDMA_REGO_RING_SIZE); + if (ret) { + qdma_err(qdev, "Failed to setup ring %d of size %ld", + QDMA_DEFAULT_RING_ID, QDMA_DEFAULT_RING_SIZE); + return ret; + } + + /* Enable memory-mapped DMA engine in both directions */ + ret = qdma_sgdma_control(qdev, 1); + if (ret) { + qdma_err(qdev, "Failed to SGDMA with error %d", ret); + return ret; + } + + ret = qdma_alloc_queues(qdev, DMA_MEM_TO_DEV); + if (ret) { + qdma_err(qdev, "Failed to alloc H2C queues, ret %d", ret); + return ret; + } + + ret = qdma_alloc_queues(qdev, DMA_DEV_TO_MEM); + if (ret) { + qdma_err(qdev, "Failed to alloc C2H queues, ret %d", ret); + return ret; + } + + return 0; +} + +/** + * qdma_free_queue_resources() - Free queue resources + * @chan: DMA channel + */ +static void qdma_free_queue_resources(struct dma_chan *chan) +{ + struct qdma_queue *queue = to_qdma_queue(chan); + struct qdma_device *qdev = queue->qdev; + struct device *dev = qdev->dma_dev.dev; + + qdma_clear_queue_context(queue); + vchan_free_chan_resources(&queue->vchan); + dma_free_coherent(dev, queue->ring_size * QDMA_MM_DESC_SIZE, + queue->desc_base, queue->dma_desc_base); +} + +/** + * qdma_alloc_queue_resources() - Allocate queue resources + * @chan: DMA channel + */ +static int qdma_alloc_queue_resources(struct dma_chan *chan) +{ + struct qdma_queue *queue = to_qdma_queue(chan); + struct qdma_device *qdev = queue->qdev; + struct qdma_ctxt_sw_desc desc; + size_t size; + int ret; + + ret = qdma_clear_queue_context(queue); + if (ret) + return ret; + + size = queue->ring_size * QDMA_MM_DESC_SIZE; + queue->desc_base = dma_alloc_coherent(qdev->dma_dev.dev, size, + &queue->dma_desc_base, + GFP_KERNEL); + if (!queue->desc_base) { + qdma_err(qdev, "Failed to allocate descriptor ring"); + return -ENOMEM; + } + + /* Setup SW descriptor queue context for DMA memory map */ + desc.vec = qdma_get_intr_ring_idx(qdev); + desc.desc_base = queue->dma_desc_base; + ret = qdma_setup_queue_context(qdev, &desc, queue->dir, queue->qid); + if (ret) { + qdma_err(qdev, "Failed to setup SW desc ctxt for %s", + chan->name); + dma_free_coherent(qdev->dma_dev.dev, size, queue->desc_base, + queue->dma_desc_base); + return ret; + } + + queue->pidx = 0; + queue->cidx = 0; + + return 0; +} + +static bool qdma_filter_fn(struct dma_chan *chan, void *param) +{ + struct qdma_queue *queue = to_qdma_queue(chan); + struct qdma_queue_info *info = param; + + return info->dir == queue->dir; +} + +static int qdma_xfer_start(struct qdma_queue *queue) +{ + struct qdma_device *qdev = queue->qdev; + int ret; + + if (!vchan_next_desc(&queue->vchan)) + return 0; + + qdma_dbg(qdev, "Tnx kickoff with P: %d for %s%d", + queue->issued_vdesc->pidx, CHAN_STR(queue), queue->qid); + + ret = qdma_update_pidx(queue, queue->issued_vdesc->pidx); + if (ret) { + qdma_err(qdev, "Failed to update PIDX to %d for %s queue: %d", + queue->pidx, CHAN_STR(queue), queue->qid); + } + + return ret; +} + +static void qdma_issue_pending(struct dma_chan *chan) +{ + struct qdma_queue *queue = to_qdma_queue(chan); + unsigned long flags; + + spin_lock_irqsave(&queue->vchan.lock, flags); + if (vchan_issue_pending(&queue->vchan)) { + if (queue->submitted_vdesc) { + queue->issued_vdesc = queue->submitted_vdesc; + queue->submitted_vdesc = NULL; + } + qdma_xfer_start(queue); + } + + spin_unlock_irqrestore(&queue->vchan.lock, flags); +} + +static struct qdma_mm_desc *qdma_get_desc(struct qdma_queue *q) +{ + struct qdma_mm_desc *desc; + + if (((q->pidx + 1) & q->idx_mask) == q->cidx) + return NULL; + + desc = q->desc_base + q->pidx; + q->pidx = (q->pidx + 1) & q->idx_mask; + + return desc; +} + +static int qdma_hw_enqueue(struct qdma_queue *q, struct qdma_mm_vdesc *vdesc) +{ + struct qdma_mm_desc *desc; + struct scatterlist *sg; + u64 addr, *src, *dst; + u32 rest, len; + int ret = 0; + u32 i; + + if (!vdesc->sg_len) + return 0; + + if (q->dir == DMA_MEM_TO_DEV) { + dst = &vdesc->dev_addr; + src = &addr; + } else { + dst = &addr; + src = &vdesc->dev_addr; + } + + for_each_sg(vdesc->sgl, sg, vdesc->sg_len, i) { + addr = sg_dma_address(sg) + vdesc->sg_off; + rest = sg_dma_len(sg) - vdesc->sg_off; + while (rest) { + len = min_t(u32, rest, QDMA_MM_DESC_MAX_LEN); + desc = qdma_get_desc(q); + if (!desc) { + ret = -EBUSY; + goto out; + } + + desc->src_addr = cpu_to_le64(*src); + desc->dst_addr = cpu_to_le64(*dst); + desc->len = cpu_to_le32(len); + + vdesc->dev_addr += len; + vdesc->sg_off += len; + vdesc->pending_descs++; + addr += len; + rest -= len; + } + vdesc->sg_off = 0; + } +out: + vdesc->sg_len -= i; + vdesc->pidx = q->pidx; + return ret; +} + +static void qdma_fill_pending_vdesc(struct qdma_queue *q) +{ + struct virt_dma_chan *vc = &q->vchan; + struct qdma_mm_vdesc *vdesc = NULL; + struct virt_dma_desc *vd; + int ret; + + if (!list_empty(&vc->desc_issued)) { + vd = &q->issued_vdesc->vdesc; + list_for_each_entry_from(vd, &vc->desc_issued, node) { + vdesc = to_qdma_vdesc(vd); + ret = qdma_hw_enqueue(q, vdesc); + if (ret) { + q->issued_vdesc = vdesc; + return; + } + } + q->issued_vdesc = vdesc; + } + + if (list_empty(&vc->desc_submitted)) + return; + + if (q->submitted_vdesc) + vd = &q->submitted_vdesc->vdesc; + else + vd = list_first_entry(&vc->desc_submitted, typeof(*vd), node); + + list_for_each_entry_from(vd, &vc->desc_submitted, node) { + vdesc = to_qdma_vdesc(vd); + ret = qdma_hw_enqueue(q, vdesc); + if (ret) + break; + } + q->submitted_vdesc = vdesc; +} + +static dma_cookie_t qdma_tx_submit(struct dma_async_tx_descriptor *tx) +{ + struct virt_dma_chan *vc = to_virt_chan(tx->chan); + struct qdma_queue *q = to_qdma_queue(&vc->chan); + struct virt_dma_desc *vd; + unsigned long flags; + dma_cookie_t cookie; + + vd = container_of(tx, struct virt_dma_desc, tx); + spin_lock_irqsave(&vc->lock, flags); + cookie = dma_cookie_assign(tx); + + list_move_tail(&vd->node, &vc->desc_submitted); + qdma_fill_pending_vdesc(q); + spin_unlock_irqrestore(&vc->lock, flags); + + return cookie; +} + +static void *qdma_get_metadata_ptr(struct dma_async_tx_descriptor *tx, + size_t *payload_len, size_t *max_len) +{ + struct qdma_mm_vdesc *vdesc; + + vdesc = container_of(tx, typeof(*vdesc), vdesc.tx); + if (payload_len) + *payload_len = sizeof(vdesc->dev_addr); + if (max_len) + *max_len = sizeof(vdesc->dev_addr); + + return &vdesc->dev_addr;Can you describe what metadata is being used here for?
The metadata is the device address the dma request will transfer data to / from. Please see the example usage here: https://github.com/houlz0507/XRT-1/blob/qdma_v1_usage/src/runtime_src/core/pcie/driver/linux/xocl/subdev/qdma.c#L311 Before dmaengine_submit(), it specifies the device address.
+} + +static int qdma_set_metadata_len(struct dma_async_tx_descriptor *tx, + size_t payload_len) +{ + struct qdma_mm_vdesc *vdesc; + + vdesc = container_of(tx, typeof(*vdesc), vdesc.tx); + if (payload_len != sizeof(vdesc->dev_addr)) + return -EINVAL; + + return 0; +} + +static struct dma_descriptor_metadata_ops metadata_ops = { + .get_ptr = qdma_get_metadata_ptr, + .set_len = qdma_set_metadata_len, +}; + +static struct dma_async_tx_descriptor * +qdma_prep_device_sg(struct dma_chan *chan, struct scatterlist *sgl, + unsigned int sg_len, enum dma_transfer_direction dir, + unsigned long flags, void *context) +{ + struct qdma_queue *q = to_qdma_queue(chan); + struct dma_async_tx_descriptor *tx; + struct qdma_mm_vdesc *vdesc; + + vdesc = kzalloc(sizeof(*vdesc), GFP_NOWAIT); + if (!vdesc) + return NULL; + vdesc->sgl = sgl; + vdesc->sg_len = sg_len; + + tx = vchan_tx_prep(&q->vchan, &vdesc->vdesc, flags); + tx->tx_submit = qdma_tx_submit; + tx->metadata_ops = &metadata_ops; + + return tx; +} + +static int qdma_arm_err_intr(const struct qdma_device *qdev) +{ + u32 value = 0; + + qdma_set_field(qdev, &value, QDMA_REGF_ERR_INT_FUNC, qdev->fid); + qdma_set_field(qdev, &value, QDMA_REGF_ERR_INT_VEC, qdev->err_irq_idx); + qdma_set_field(qdev, &value, QDMA_REGF_ERR_INT_ARM, 1); + + return qdma_reg_write(qdev, &value, QDMA_REGO_ERR_INT); +} + +static irqreturn_t qdma_error_isr(int irq, void *data) +{ + struct qdma_device *qdev = data; + u32 err_stat = 0; + int ret; + + ret = qdma_reg_read(qdev, &err_stat, QDMA_REGO_ERR_STAT); + if (ret) { + qdma_err(qdev, "read error state failed, ret %d", ret); + goto out; + } + + qdma_err(qdev, "global error %d", err_stat); + ret = qdma_reg_write(qdev, &err_stat, QDMA_REGO_ERR_STAT); + if (ret) + qdma_err(qdev, "clear error state failed, ret %d", ret); + +out: + qdma_arm_err_intr(qdev); + return IRQ_HANDLED; +} + +static irqreturn_t qdma_queue_isr(int irq, void *data) +{ + struct qdma_intr_ring *intr = data; + struct qdma_queue *q = NULL; + struct qdma_device *qdev; + u32 index, comp_desc; + u64 intr_ent; + u8 color; + int ret; + u16 qid; + + qdev = intr->qdev; + index = intr->cidx; + while (1) { + struct virt_dma_desc *vd; + struct qdma_mm_vdesc *vdesc; + unsigned long flags; + u32 cidx; + + intr_ent = le64_to_cpu(intr->base[index]); + color = FIELD_GET(QDMA_INTR_MASK_COLOR, intr_ent); + if (color != intr->color) + break; + + qid = FIELD_GET(QDMA_INTR_MASK_QID, intr_ent); + if (FIELD_GET(QDMA_INTR_MASK_TYPE, intr_ent)) + q = qdev->c2h_queues; + else + q = qdev->h2c_queues; + q += qid; + + cidx = FIELD_GET(QDMA_INTR_MASK_CIDX, intr_ent); + + spin_lock_irqsave(&q->vchan.lock, flags); + comp_desc = (cidx - q->cidx) & q->idx_mask; + + vd = vchan_next_desc(&q->vchan); + if (!vd) + goto skip; + + vdesc = to_qdma_vdesc(vd); + while (comp_desc > vdesc->pending_descs) { + list_del(&vd->node); + vchan_cookie_complete(vd); + comp_desc -= vdesc->pending_descs; + vd = vchan_next_desc(&q->vchan); + vdesc = to_qdma_vdesc(vd); + } + vdesc->pending_descs -= comp_desc; + if (!vdesc->pending_descs && QDMA_VDESC_QUEUED(vdesc)) { + list_del(&vd->node); + vchan_cookie_complete(vd); + } + q->cidx = cidx; + + qdma_fill_pending_vdesc(q); + qdma_xfer_start(q); + +skip: + spin_unlock_irqrestore(&q->vchan.lock, flags); + + /* + * Wrap the index value and flip the expected color value if + * interrupt aggregation PIDX has wrapped around. + */ + index++; + index &= QDMA_INTR_RING_IDX_MASK; + if (!index) + intr->color = !intr->color; + } + + /* + * Update the software interrupt aggregation ring CIDX if a valid entry + * was found. + */ + if (q) { + qdma_dbg(qdev, "update intr ring%d %d", intr->ridx, index); + + /* + * Record the last read index of status descriptor from the + * interrupt aggregation ring. + */ + intr->cidx = index; + + ret = qdma_update_cidx(q, intr->ridx, index); + if (ret) { + qdma_err(qdev, "Failed to update IRQ CIDX"); + return IRQ_NONE; + } + } + + return IRQ_HANDLED; +} + +static int qdma_init_error_irq(struct qdma_device *qdev) +{ + struct device *dev = &qdev->pdev->dev; + int ret; + u32 vec; + + vec = qdev->queue_irq_start - 1; + + ret = devm_request_threaded_irq(dev, vec, NULL, qdma_error_isr, + IRQF_ONESHOT, "amd-qdma-error", qdev); + if (ret) { + qdma_err(qdev, "Failed to request error IRQ vector: %d", vec); + return ret; + } + + ret = qdma_arm_err_intr(qdev); + if (ret) + qdma_err(qdev, "Failed to arm err interrupt, ret %d", ret); + + return ret; +} + +static int qdmam_alloc_qintr_rings(struct qdma_device *qdev) +{ + u32 ctxt[QDMA_CTXT_REGMAP_LEN]; + struct device *dev = &qdev->pdev->dev; + struct qdma_intr_ring *ring; + struct qdma_ctxt_intr intr_ctxt; + u32 vector; + int ret, i; + + qdev->qintr_ring_num = qdev->queue_irq_num; + qdev->qintr_rings = devm_kcalloc(dev, qdev->qintr_ring_num, + sizeof(*qdev->qintr_rings), + GFP_KERNEL); + if (!qdev->qintr_rings) + return -ENOMEM; + + vector = qdev->queue_irq_start; + for (i = 0; i < qdev->qintr_ring_num; i++, vector++) { + ring = &qdev->qintr_rings[i]; + ring->qdev = qdev; + ring->msix_id = qdev->err_irq_idx + i + 1; + ring->ridx = i; + ring->color = 1; + ring->base = dmam_alloc_coherent(dev, QDMA_INTR_RING_SIZE, + &ring->dev_base, GFP_KERNEL); + if (!ring->base) { + qdma_err(qdev, "Failed to alloc intr ring %d", i); + return -ENOMEM; + } + intr_ctxt.agg_base = QDMA_INTR_RING_BASE(ring->dev_base); + intr_ctxt.size = (QDMA_INTR_RING_SIZE - 1) / 4096; + intr_ctxt.vec = ring->msix_id; + intr_ctxt.valid = true; + intr_ctxt.color = true; + ret = qdma_prog_context(qdev, QDMA_CTXT_INTR_COAL, + QDMA_CTXT_CLEAR, ring->ridx, NULL); + if (ret) { + qdma_err(qdev, "Failed clear intr ctx, ret %d", ret); + return ret; + } + + qdma_prep_intr_context(qdev, &intr_ctxt, ctxt); + ret = qdma_prog_context(qdev, QDMA_CTXT_INTR_COAL, + QDMA_CTXT_WRITE, ring->ridx, ctxt); + if (ret) { + qdma_err(qdev, "Failed setup intr ctx, ret %d", ret); + return ret; + } + + ret = devm_request_threaded_irq(dev, vector, NULL, + qdma_queue_isr, IRQF_ONESHOT, + "amd-qdma-queue", ring); + if (ret) { + qdma_err(qdev, "Failed to request irq %d", vector); + return ret; + } + } + + return 0; +} + +static int qdma_intr_init(struct qdma_device *qdev) +{ + int ret; + + ret = qdma_init_error_irq(qdev); + if (ret) { + qdma_err(qdev, "Failed to init error IRQs, ret %d", ret); + return ret; + } + + ret = qdmam_alloc_qintr_rings(qdev); + if (ret) { + qdma_err(qdev, "Failed to init queue IRQs, ret %d", ret); + return ret; + } + + return 0; +} + +static void amd_qdma_remove(struct platform_device *pdev) +{ + struct qdma_device *qdev = platform_get_drvdata(pdev); + + qdma_sgdma_control(qdev, 0); + dma_async_device_unregister(&qdev->dma_dev); + + mutex_destroy(&qdev->ctxt_lock); +} + +static int amd_qdma_probe(struct platform_device *pdev) +{ + struct qdma_platdata *pdata = dev_get_platdata(&pdev->dev); + struct qdma_device *qdev; + struct resource *res; + void __iomem *regs; + int ret; + + qdev = devm_kzalloc(&pdev->dev, sizeof(*qdev), GFP_KERNEL); + if (!qdev) + return -ENOMEM; + + platform_set_drvdata(pdev, qdev); + qdev->pdev = pdev; + mutex_init(&qdev->ctxt_lock); + + res = platform_get_resource(pdev, IORESOURCE_IRQ, 0); + if (!res) { + qdma_err(qdev, "Failed to get IRQ resource"); + ret = -ENODEV; + goto failed; + } + qdev->err_irq_idx = pdata->irq_index; + qdev->queue_irq_start = res->start + 1; + qdev->queue_irq_num = resource_size(res) - 1; + + regs = devm_platform_get_and_ioremap_resource(pdev, 0, NULL); + if (IS_ERR(regs)) { + ret = PTR_ERR(regs); + qdma_err(qdev, "Failed to map IO resource, err %d", ret); + goto failed; + } + + qdev->regmap = devm_regmap_init_mmio(&pdev->dev, regs, + &qdma_regmap_config); + if (IS_ERR(qdev->regmap)) { + ret = PTR_ERR(qdev->regmap); + qdma_err(qdev, "Regmap init failed, err %d", ret); + goto failed; + } + + ret = qdma_device_verify(qdev); + if (ret) + goto failed; + + ret = qdma_get_hw_info(qdev); + if (ret) + goto failed; + + INIT_LIST_HEAD(&qdev->dma_dev.channels); + + ret = qdma_device_setup(qdev); + if (ret) + goto failed; + + ret = qdma_intr_init(qdev); + if (ret) { + qdma_err(qdev, "Failed to initialize IRQs %d", ret); + goto failed_disable_engine; + } + + dma_cap_set(DMA_SLAVE, qdev->dma_dev.cap_mask); + dma_cap_set(DMA_PRIVATE, qdev->dma_dev.cap_mask); + + qdev->dma_dev.dev = &pdev->dev; + qdev->dma_dev.filter.map = pdata->device_map; + qdev->dma_dev.filter.mapcnt = qdev->chan_num * 2; + qdev->dma_dev.filter.fn = qdma_filter_fn; + qdev->dma_dev.desc_metadata_modes = DESC_METADATA_ENGINE; + qdev->dma_dev.device_alloc_chan_resources = qdma_alloc_queue_resources; + qdev->dma_dev.device_free_chan_resources = qdma_free_queue_resources; + qdev->dma_dev.device_prep_slave_sg = qdma_prep_device_sg; + qdev->dma_dev.device_issue_pending = qdma_issue_pending; + qdev->dma_dev.device_tx_status = dma_cookie_status; + qdev->dma_dev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV); + + ret = dma_async_device_register(&qdev->dma_dev); + if (ret) { + qdma_err(qdev, "Failed to register AMD QDMA: %d", ret); + goto failed_disable_engine; + } + + return 0; + +failed_disable_engine: + qdma_sgdma_control(qdev, 0); +failed: + mutex_destroy(&qdev->ctxt_lock); + qdma_err(qdev, "Failed to probe AMD QDMA driver"); + return ret; +} + +static struct platform_driver amd_qdma_driver = { + .driver = { + .name = "amd-qdma", + }, + .probe = amd_qdma_probe, + .remove_new = amd_qdma_remove, +}; + +module_platform_driver(amd_qdma_driver); + +MODULE_DESCRIPTION("AMD QDMA driver"); +MODULE_AUTHOR("XRT Team <runtimeca39d@xxxxxxx>"); +MODULE_LICENSE("GPL"); diff --git a/drivers/dma/amd/qdma/qdma.h b/drivers/dma/amd/qdma/qdma.h new file mode 100644 index 000000000000..268fbad9f0a2 --- /dev/null +++ b/drivers/dma/amd/qdma/qdma.h @@ -0,0 +1,265 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +/* + * DMA header for AMD Queue-based DMA Subsystem + * + * Copyright (C) 2023, Advanced Micro Devices, Inc. + */ + +#ifndef __QDMA_H +#define __QDMA_H + +#include <linux/bitfield.h> +#include <linux/dmaengine.h> +#include <linux/kernel.h> +#include <linux/platform_device.h> +#include <linux/regmap.h> + +#include "../../virt-dma.h" + +#define DISABLE 0 +#define ENABLE 1 + +#define QDMA_MIN_IRQ 3 +#define QDMA_INTR_NAME_MAX_LEN 30 +#define QDMA_INTR_PREFIX "amd-qdma" + +#define QDMA_IDENTIFIER 0x1FD3 +#define QDMA_DEFAULT_RING_SIZE (BIT(10) + 1) +#define QDMA_DEFAULT_RING_ID 0 +#define QDMA_POLL_INTRVL_US 10 /* 10us */ +#define QDMA_POLL_TIMEOUT_US (500 * 1000) /* 500ms */ +#define QDMA_DMAP_REG_STRIDE 16 +#define QDMA_CTXT_REGMAP_LEN 8 /* 8 regs */ +#define QDMA_MM_DESC_SIZE 32 /* Bytes */ +#define QDMA_MM_DESC_LEN_BITS 28 +#define QDMA_MM_DESC_MAX_LEN (BIT(QDMA_MM_DESC_LEN_BITS) - 1) +#define QDMA_MIN_DMA_ALLOC_SIZE 4096 +#define QDMA_INTR_RING_SIZE BIT(13) +#define QDMA_INTR_RING_IDX_MASK GENMASK(9, 0) +#define QDMA_INTR_RING_BASE(_addr) ((_addr) >> 12) + +#define QDMA_IDENTIFIER_REGOFF 0x0 +#define QDMA_IDENTIFIER_MASK GENMASK(31, 16) +#define QDMA_QUEUE_ARM_BIT BIT(16) + +#define qdma_err(qdev, fmt, args...) \ + dev_err(&(qdev)->pdev->dev, fmt, ##args) + +#define qdma_dbg(qdev, fmt, args...) \ + dev_dbg(&(qdev)->pdev->dev, fmt, ##args) + +#define qdma_info(qdev, fmt, args...) \ + dev_info(&(qdev)->pdev->dev, fmt, ##args) + +enum qdma_reg_fields { + QDMA_REGF_IRQ_ENABLE, + QDMA_REGF_WBK_ENABLE, + QDMA_REGF_WBI_CHECK, + QDMA_REGF_IRQ_ARM, + QDMA_REGF_IRQ_VEC, + QDMA_REGF_IRQ_AGG, + QDMA_REGF_WBI_INTVL_ENABLE, + QDMA_REGF_MRKR_DISABLE, + QDMA_REGF_QUEUE_ENABLE, + QDMA_REGF_QUEUE_MODE, + QDMA_REGF_DESC_BASE, + QDMA_REGF_DESC_SIZE, + QDMA_REGF_RING_ID, + QDMA_REGF_CMD_INDX, + QDMA_REGF_CMD_CMD, + QDMA_REGF_CMD_TYPE, + QDMA_REGF_CMD_BUSY, + QDMA_REGF_QUEUE_COUNT, + QDMA_REGF_QUEUE_MAX, + QDMA_REGF_QUEUE_BASE, + QDMA_REGF_FUNCTION_ID, + QDMA_REGF_INTR_AGG_BASE, + QDMA_REGF_INTR_VECTOR, + QDMA_REGF_INTR_SIZE, + QDMA_REGF_INTR_VALID, + QDMA_REGF_INTR_COLOR, + QDMA_REGF_INTR_FUNCTION_ID, + QDMA_REGF_ERR_INT_FUNC, + QDMA_REGF_ERR_INT_VEC, + QDMA_REGF_ERR_INT_ARM, + QDMA_REGF_MAX +}; + +enum qdma_regs { + QDMA_REGO_CTXT_DATA, + QDMA_REGO_CTXT_CMD, + QDMA_REGO_CTXT_MASK, + QDMA_REGO_MM_H2C_CTRL, + QDMA_REGO_MM_C2H_CTRL, + QDMA_REGO_QUEUE_COUNT, + QDMA_REGO_RING_SIZE, + QDMA_REGO_H2C_PIDX, + QDMA_REGO_C2H_PIDX, + QDMA_REGO_INTR_CIDX, + QDMA_REGO_FUNC_ID, + QDMA_REGO_ERR_INT, + QDMA_REGO_ERR_STAT, + QDMA_REGO_MAX +}; + +struct qdma_reg_field { + u16 lsb; /* Least significant bit of field */ + u16 msb; /* Most significant bit of field */ +}; + +struct qdma_reg { + u32 off; + u32 count; +}; + +#define QDMA_REGF(_msb, _lsb) { \ + .lsb = (_lsb), \ + .msb = (_msb), \ +} + +#define QDMA_REGO(_off, _count) { \ + .off = (_off), \ + .count = (_count), \ +} + +enum qdma_desc_size { + QDMA_DESC_SIZE_8B, + QDMA_DESC_SIZE_16B, + QDMA_DESC_SIZE_32B, + QDMA_DESC_SIZE_64B, +}; + +enum qdma_queue_op_mode { + QDMA_QUEUE_OP_STREAM, + QDMA_QUEUE_OP_MM, +}; + +enum qdma_ctxt_type { + QDMA_CTXT_DESC_SW_C2H, + QDMA_CTXT_DESC_SW_H2C, + QDMA_CTXT_DESC_HW_C2H, + QDMA_CTXT_DESC_HW_H2C, + QDMA_CTXT_DESC_CR_C2H, + QDMA_CTXT_DESC_CR_H2C, + QDMA_CTXT_WRB, + QDMA_CTXT_PFTCH, + QDMA_CTXT_INTR_COAL, + QDMA_CTXT_RSVD, + QDMA_CTXT_HOST_PROFILE, + QDMA_CTXT_TIMER, + QDMA_CTXT_FMAP, + QDMA_CTXT_FNC_STS, +}; + +enum qdma_ctxt_cmd { + QDMA_CTXT_CLEAR, + QDMA_CTXT_WRITE, + QDMA_CTXT_READ, + QDMA_CTXT_INVALIDATE, + QDMA_CTXT_MAX +}; + +struct qdma_ctxt_sw_desc { + u64 desc_base; + u16 vec; +}; + +struct qdma_ctxt_intr { + u64 agg_base; + u16 vec; + u32 size; + bool valid; + bool color; +}; + +struct qdma_ctxt_fmap { + u16 qbase; + u16 qmax; +}; + +struct qdma_device; + +struct qdma_mm_desc { + __le64 src_addr; + __le32 len; + __le32 reserved1; + __le64 dst_addr; + __le64 reserved2; +} __packed; + +struct qdma_mm_vdesc { + struct virt_dma_desc vdesc; + struct qdma_queue *queue; + struct scatterlist *sgl; + u64 sg_off; + u32 sg_len; + u64 dev_addr; + u32 pidx; + u32 pending_descs; +}; + +#define QDMA_VDESC_QUEUED(vdesc) (!(vdesc)->sg_len) + +struct qdma_queue { + struct qdma_device *qdev; + struct virt_dma_chan vchan; + enum dma_transfer_direction dir; + struct dma_slave_config cfg; + struct qdma_mm_desc *desc_base; + struct qdma_mm_vdesc *submitted_vdesc; + struct qdma_mm_vdesc *issued_vdesc; + dma_addr_t dma_desc_base; + u32 pidx_reg; + u32 cidx_reg; + u32 ring_size; + u32 idx_mask; + u16 qid; + u32 pidx; + u32 cidx; +}; + +struct qdma_intr_ring { + struct qdma_device *qdev; + __le64 *base; + dma_addr_t dev_base; + char msix_name[QDMA_INTR_NAME_MAX_LEN]; + u32 msix_vector; + u16 msix_id; + u32 ring_size; + u16 ridx; + u16 cidx; + u8 color; +}; + +#define QDMA_INTR_MASK_PIDX GENMASK_ULL(15, 0) +#define QDMA_INTR_MASK_CIDX GENMASK_ULL(31, 16) +#define QDMA_INTR_MASK_DESC_COLOR GENMASK_ULL(32, 32) +#define QDMA_INTR_MASK_STATE GENMASK_ULL(34, 33) +#define QDMA_INTR_MASK_ERROR GENMASK_ULL(36, 35) +#define QDMA_INTR_MASK_TYPE GENMASK_ULL(38, 38) +#define QDMA_INTR_MASK_QID GENMASK_ULL(62, 39) +#define QDMA_INTR_MASK_COLOR GENMASK_ULL(63, 63) + +struct qdma_device { + struct platform_device *pdev; + struct dma_device dma_dev; + struct regmap *regmap; + struct mutex ctxt_lock; /* protect ctxt registers */ + const struct qdma_reg_field *rfields; + const struct qdma_reg *roffs; + struct qdma_queue *h2c_queues; + struct qdma_queue *c2h_queues; + struct qdma_intr_ring *qintr_rings; + u32 qintr_ring_num; + u32 qintr_ring_idx; + u32 chan_num; + u32 queue_irq_start; + u32 queue_irq_num; + u32 err_irq_idx; + u32 fid; +}; + +extern const struct qdma_reg qdma_regos_default[QDMA_REGO_MAX]; +extern const struct qdma_reg_field qdma_regfs_default[QDMA_REGF_MAX]; + +#endif /* __QDMA_H */ diff --git a/include/linux/platform_data/amd_qdma.h b/include/linux/platform_data/amd_qdma.h new file mode 100644 index 000000000000..59fa0c174f70 --- /dev/null +++ b/include/linux/platform_data/amd_qdma.h @@ -0,0 +1,36 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +/* + * Copyright (C) 2023, Advanced Micro Devices, Inc. + */ + +#ifndef _PLATDATA_AMD_QDMA_H +#define _PLATDATA_AMD_QDMA_H + +#include <linux/dmaengine.h> + +/** + * struct qdma_queue_info - DMA queue information. This information is used to + * match queue when DMA channel is requested + * @dir: Channel transfer direction + */ +struct qdma_queue_info { + enum dma_transfer_direction dir; +};why is dma_transfer_direction here, why not use prep_ calls for passing that around?
This is used for dma_dev.filter.fn callback. It is an extra check for dma_request_chan().
It happens before prep_ calls.
+ +#define QDMA_FILTER_PARAM(qinfo) ((void *)(qinfo)) + +struct dma_slave_map; + +/** + * struct qdma_platdata - Platform specific data for QDMA engine + * @max_mm_channels: Maximum number of MM DMA channels in each direction + * @device_map: DMA slave map + * @irq_index: The index of first IRQ + */ +struct qdma_platdata { + u32 max_mm_channels; + u32 irq_index; + struct dma_slave_map *device_map; +};why should this be plat data?
The QDMA IP has the same usage cases with Xilinx XDMA IP. (platform_data/amd_xdma.h)
The device driver for the product (such as Alveo) which uses XDMA IP will instantiate the
platform driver and populate the platdata. Thanks, Lizhi
