Implement dsp lifecycle functions such as core RESET and STALL, SRAM power control and LP clock selection. This also adds functions for handling transport over DW DMA controller.
Signed-off-by: Cezary Rojewski cezary.rojewski@intel.com --- sound/soc/intel/catpt/dsp.c | 491 ++++++++++++++++++++++++++++++++++++ 1 file changed, 491 insertions(+) create mode 100644 sound/soc/intel/catpt/dsp.c
diff --git a/sound/soc/intel/catpt/dsp.c b/sound/soc/intel/catpt/dsp.c new file mode 100644 index 000000000000..137c43ca521d --- /dev/null +++ b/sound/soc/intel/catpt/dsp.c @@ -0,0 +1,491 @@ +// SPDX-License-Identifier: GPL-2.0-only +// +// Copyright(c) 2020 Intel Corporation. All rights reserved. +// +// Author: Cezary Rojewski cezary.rojewski@intel.com +// + +#include <linux/dma-mapping.h> +#include <linux/acpi_dma.h> +#include <linux/firmware.h> +#include "core.h" +#include "messages.h" +#include "registers.h" +#include "trace.h" + +static void catpt_dma_transfer_complete(void *arg) +{ + struct catpt_dev *cdev = arg; + + dev_dbg(cdev->dev, "%s\n", __func__); +} + +static bool catpt_dma_filter(struct dma_chan *chan, void *param) +{ + return chan->device->dev == (struct device *)param; +} + +#define CATPT_DMA_DEVID 1 /* dma engine used */ +#define CATPT_DMA_MAXBURST 0x3 +#define CATPT_DMA_DSP_ADDR_MASK 0xFFF00000 + +struct dma_chan *catpt_dma_request_config_chan(struct catpt_dev *cdev) +{ + struct dma_slave_config config; + struct dma_chan *chan; + dma_cap_mask_t mask; + int ret; + + dma_cap_zero(mask); + dma_cap_set(DMA_SLAVE, mask); + dma_cap_set(DMA_MEMCPY, mask); + + chan = dma_request_channel(mask, catpt_dma_filter, cdev->dev); + if (!chan) { + dev_err(cdev->dev, "request channel failed\n"); + dump_stack(); + return ERR_PTR(-EPROBE_DEFER); + } + + memset(&config, 0, sizeof(config)); + config.direction = DMA_MEM_TO_DEV; + config.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES; + config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES; + config.src_maxburst = CATPT_DMA_MAXBURST; + config.dst_maxburst = CATPT_DMA_MAXBURST; + + ret = dmaengine_slave_config(chan, &config); + if (ret) { + dev_err(cdev->dev, "slave config failed: %d\n", ret); + dma_release_channel(chan); + return ERR_PTR(ret); + } + + return chan; +} + +static int catpt_dma_memcpy(struct catpt_dev *cdev, struct dma_chan *chan, + dma_addr_t dst_addr, dma_addr_t src_addr, + size_t size) +{ + struct dma_async_tx_descriptor *desc; + enum dma_status status; + + desc = dmaengine_prep_dma_memcpy(chan, dst_addr, src_addr, size, + DMA_CTRL_ACK); + if (!desc) { + dev_err(cdev->dev, "prep dma memcpy failed\n"); + return -EIO; + } + + /* enable demand mode for dma channel */ + catpt_updatel_shim(cdev, HMDC, + CATPT_HMDC_HDDA(CATPT_DMA_DEVID, chan->chan_id), + CATPT_HMDC_HDDA(CATPT_DMA_DEVID, chan->chan_id)); + desc->callback = catpt_dma_transfer_complete; + desc->callback_param = cdev; + dmaengine_submit(desc); + + status = dma_wait_for_async_tx(desc); + catpt_updatel_shim(cdev, HMDC, + CATPT_HMDC_HDDA(CATPT_DMA_DEVID, chan->chan_id), 0); + + return (status == DMA_COMPLETE) ? 0 : -EPROTO; +} + +int catpt_dma_memcpy_todsp(struct catpt_dev *cdev, struct dma_chan *chan, + dma_addr_t dst_addr, dma_addr_t src_addr, + size_t size) +{ + return catpt_dma_memcpy(cdev, chan, dst_addr | CATPT_DMA_DSP_ADDR_MASK, + src_addr, size); +} + +int catpt_dma_memcpy_fromdsp(struct catpt_dev *cdev, struct dma_chan *chan, + dma_addr_t dst_addr, dma_addr_t src_addr, + size_t size) +{ + return catpt_dma_memcpy(cdev, chan, dst_addr, + src_addr | CATPT_DMA_DSP_ADDR_MASK, size); +} + +int catpt_dmac_probe(struct catpt_dev *cdev) +{ + struct dw_dma_chip *dmac; + struct resource res; + int ret; + + dmac = devm_kzalloc(cdev->dev, sizeof(*dmac), GFP_KERNEL); + if (!dmac) + return -ENOMEM; + + memset(&res, 0, sizeof(res)); + res.start = cdev->lpe_base + + cdev->pdata->host_dma_offset[CATPT_DMA_DEVID]; + res.end = res.start + (CATPT_DMA_REGS_SIZE - 1); + res.flags = IORESOURCE_MEM; + dmac->dev = cdev->dev; + dmac->irq = cdev->irq; + + dmac->regs = devm_ioremap_resource(cdev->dev, &res); + if (IS_ERR(dmac->regs)) + return PTR_ERR(dmac->regs); + + ret = dma_coerce_mask_and_coherent(cdev->dev, DMA_BIT_MASK(31)); + if (ret < 0) + return ret; + /* + * Caller is responsible for putting device in D0 to allow + * for I/O and memory access before probing DW + */ + ret = dw_dma_probe(dmac); + if (ret < 0) + return ret; + + cdev->dmac = dmac; + return 0; +} + +void catpt_dmac_remove(struct catpt_dev *cdev) +{ + /* + * As do_dma_remove() juggles with pm_runtime_get_xxx() and + * pm_runtime_put_xxx() while both ADSP and DW 'devices' are part of + * the same module, caller makes sure pm_runtime_disable() is invoked + * before removing DW to prevent postmortem resume and suspend + */ + dw_dma_remove(cdev->dmac); +} + +static void catpt_dsp_set_srampge(struct catpt_dev *cdev, + struct catpt_mbank *sram, unsigned long new) +{ + unsigned long old; + u32 off = sram->start; + u32 b = __ffs(sram->mask); + + old = catpt_readl_pci(cdev, VDRTCTL0) & sram->mask; + trace_catpt_updatel("SRAMPGE", sram->mask, old, new); + if (old == new) + return; + + catpt_updatel_pci(cdev, VDRTCTL0, sram->mask, new); + udelay(60); + + /* + * dummy read as the very first access after block enable + * to prevent byte loss in future operations + */ + for_each_clear_bit_from(b, &new, fls(sram->mask)) { + u8 buf[4]; + + /* newly enabled: new bit=0 while old bit=1 */ + if (test_bit(b, &old)) { + dev_dbg(cdev->dev, "sanitize block %ld: off 0x%08x\n", + (b - __ffs(sram->mask)), off); + memcpy_fromio(buf, cdev->lpe_ba + off, sizeof(buf)); + } + off += CATPT_MEMBLOCK_SIZE; + } +} + +void catpt_dsp_update_srampge(struct catpt_dev *cdev, struct catpt_mbank *sram) +{ + struct catpt_mregion *reg; + unsigned long new = 0; + + /* flag all busy blocks */ + list_for_each_entry(reg, &sram->region_list, node) { + u32 h, l; + + if (!reg->busy) + continue; + h = (reg->end - sram->start) / CATPT_MEMBLOCK_SIZE; + l = (reg->start - sram->start) / CATPT_MEMBLOCK_SIZE; + new |= GENMASK(h, l); + } + + /* offset value given mask's start and invert it as ON=b0 */ + new <<= __ffs(sram->mask); + new = ~(new) & sram->mask; + + /* disable core clock gating */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DCLCGE, 0); + + catpt_dsp_set_srampge(cdev, sram, new); + + /* enable core clock gating */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DCLCGE, + CATPT_VDRTCTL2_DCLCGE); +} + +int catpt_dsp_stall(struct catpt_dev *cdev, bool stall) +{ + u32 reg, val; + + val = stall ? CATPT_CS_STALL : 0; + catpt_updatel_shim(cdev, CS, CATPT_CS_STALL, val); + + return catpt_readl_poll_shim(cdev, CS, + reg, (reg & CATPT_CS_STALL) == val, + 500, 10000); +} + +static int catpt_dsp_reset(struct catpt_dev *cdev, bool reset) +{ + u32 reg, val; + + val = reset ? CATPT_CS_RST : 0; + catpt_updatel_shim(cdev, CS, CATPT_CS_RST, val); + + return catpt_readl_poll_shim(cdev, CS, + reg, (reg & CATPT_CS_RST) == val, + 500, 10000); +} + +void lpt_dsp_pll_shutdown(struct catpt_dev *cdev, bool enable) +{ + u32 val; + + val = enable ? LPT_VDRTCTL0_APLLSE : 0; + catpt_updatel_pci(cdev, VDRTCTL0, LPT_VDRTCTL0_APLLSE, val); +} + +void wpt_dsp_pll_shutdown(struct catpt_dev *cdev, bool enable) +{ + u32 val; + + val = enable ? WPT_VDRTCTL2_APLLSE : 0; + catpt_updatel_pci(cdev, VDRTCTL2, WPT_VDRTCTL2_APLLSE, val); +} + +static int catpt_dsp_select_lpclock(struct catpt_dev *cdev, bool lp, bool waiti) +{ + u32 mask, reg, val; + int ret; + + mutex_lock(&cdev->clk_mutex); + + val = lp ? CATPT_CS_LPCS : 0; + reg = catpt_readl_shim(cdev, CS) & CATPT_CS_LPCS; + trace_catpt_updatel("LPCS", CATPT_CS_LPCS, reg, val); + + if (reg == val) { + mutex_unlock(&cdev->clk_mutex); + return 0; + } + + if (waiti) { + /* wait for DSP to signal WAIT state */ + ret = catpt_readl_poll_shim(cdev, ISD, + reg, (reg & CATPT_ISD_DCPWM), + 500, 10000); + if (ret < 0) { + dev_warn(cdev->dev, "await WAITI timeout\n"); + mutex_unlock(&cdev->clk_mutex); + return ret; + } + } + + ret = catpt_readl_poll_shim(cdev, CLKCTL, + reg, !(reg & CATPT_CLKCTL_CFCIP), + 500, 10000); + if (ret < 0) + dev_warn(cdev->dev, "clock change still in progress\n"); + + /* default to DSP core & audio fabric high clock */ + val |= CATPT_CS_DCS_HIGH; + mask = CATPT_CS_LPCS | CATPT_CS_DCS; + catpt_updatel_shim(cdev, CS, mask, val); + + ret = catpt_readl_poll_shim(cdev, CLKCTL, + reg, !(reg & CATPT_CLKCTL_CFCIP), + 500, 10000); + if (ret < 0) + dev_warn(cdev->dev, "clock change still in progress\n"); + + /* update PLL accordingly */ + cdev->pdata->pll_shutdown(cdev, lp); + + mutex_unlock(&cdev->clk_mutex); + return 0; +} + +int catpt_dsp_update_lpclock(struct catpt_dev *cdev) +{ + struct catpt_stream_runtime *stream; + bool lp; + + if (list_empty(&cdev->stream_list)) + return catpt_dsp_select_lpclock(cdev, true, true); + + lp = true; + list_for_each_entry(stream, &cdev->stream_list, node) { + if (stream->prepared) { + lp = false; + break; + } + } + + return catpt_dsp_select_lpclock(cdev, lp, true); +} + +/* bring registers to their defaults as HW won't reset itself */ +static void catpt_dsp_set_regs_defaults(struct catpt_dev *cdev) +{ + int i; + + catpt_writel_shim(cdev, CS, CATPT_CS_DEFAULT); + catpt_writel_shim(cdev, ISC, CATPT_ISC_DEFAULT); + catpt_writel_shim(cdev, ISD, CATPT_ISD_DEFAULT); + catpt_writel_shim(cdev, IMC, CATPT_IMC_DEFAULT); + catpt_writel_shim(cdev, IMD, CATPT_IMD_DEFAULT); + catpt_writel_shim(cdev, IPCC, CATPT_IPCC_DEFAULT); + catpt_writel_shim(cdev, IPCD, CATPT_IPCD_DEFAULT); + catpt_writel_shim(cdev, CLKCTL, CATPT_CLKCTL_DEFAULT); + catpt_writel_shim(cdev, CS2, CATPT_CS2_DEFAULT); + catpt_writel_shim(cdev, LTRC, CATPT_LTRC_DEFAULT); + catpt_writel_shim(cdev, HMDC, CATPT_HMDC_DEFAULT); + + for (i = 0; i < CATPT_SSP_COUNT; i++) { + catpt_writel_ssp(cdev, i, SSC0, CATPT_SSP_SSC0_DEFAULT); + catpt_writel_ssp(cdev, i, SSC1, CATPT_SSP_SSC1_DEFAULT); + catpt_writel_ssp(cdev, i, SSS, CATPT_SSP_SSS_DEFAULT); + catpt_writel_ssp(cdev, i, SSIT, CATPT_SSP_SSIT_DEFAULT); + catpt_writel_ssp(cdev, i, SSD, CATPT_SSP_SSD_DEFAULT); + catpt_writel_ssp(cdev, i, SSTO, CATPT_SSP_SSTO_DEFAULT); + catpt_writel_ssp(cdev, i, SSPSP, CATPT_SSP_SSPSP_DEFAULT); + catpt_writel_ssp(cdev, i, SSTSA, CATPT_SSP_SSTSA_DEFAULT); + catpt_writel_ssp(cdev, i, SSRSA, CATPT_SSP_SSRSA_DEFAULT); + catpt_writel_ssp(cdev, i, SSTSS, CATPT_SSP_SSTSS_DEFAULT); + catpt_writel_ssp(cdev, i, SSC2, CATPT_SSP_SSC2_DEFAULT); + catpt_writel_ssp(cdev, i, SSPSP2, CATPT_SSP_SSPSP2_DEFAULT); + } +} + +int lpt_dsp_power_down(struct catpt_dev *cdev) +{ + catpt_dsp_reset(cdev, true); + + /* set 24Mhz clock for both SSPs */ + catpt_updatel_shim(cdev, CS, CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1), + CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1)); + catpt_dsp_select_lpclock(cdev, true, false); + + /* DRAM power gating all */ + catpt_dsp_set_srampge(cdev, &cdev->dram, cdev->dram.mask); + catpt_dsp_set_srampge(cdev, &cdev->iram, cdev->iram.mask); + + /* set D3 */ + catpt_updatel_pci(cdev, PMCS, CATPT_PMCS_PS, CATPT_PMCS_PS_D3HOT); + udelay(50); + + return 0; +} + +int lpt_dsp_power_up(struct catpt_dev *cdev) +{ + /* SRAM power gating none */ + catpt_dsp_set_srampge(cdev, &cdev->dram, 0); + catpt_dsp_set_srampge(cdev, &cdev->iram, 0); + + /* set D0 */ + catpt_updatel_pci(cdev, PMCS, CATPT_PMCS_PS, 0); + udelay(100); + + catpt_dsp_select_lpclock(cdev, false, false); + catpt_updatel_shim(cdev, CS, + CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1), + CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1)); + udelay(50); + + catpt_dsp_reset(cdev, false); + /* generate int deassert msg to fix inversed int logic */ + catpt_updatel_shim(cdev, IMC, CATPT_IMC_IPCDB | CATPT_IMC_IPCCD, 0); + + return 0; +} + +int wpt_dsp_power_down(struct catpt_dev *cdev) +{ + u32 mask, val; + + /* disable core clock gating */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DCLCGE, 0); + + catpt_dsp_reset(cdev, true); + /* set 24Mhz clock for both SSPs */ + catpt_updatel_shim(cdev, CS, CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1), + CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1)); + catpt_dsp_select_lpclock(cdev, true, false); + /* disable MCLK */ + catpt_updatel_shim(cdev, CLKCTL, CATPT_CLKCTL_SMOS, 0); + + catpt_dsp_set_regs_defaults(cdev); + + /* switch clock gating */ + mask = CATPT_VDRTCTL2_CGEALL & (~CATPT_VDRTCTL2_DCLCGE); + val = mask & (~CATPT_VDRTCTL2_DTCGE); + catpt_updatel_pci(cdev, VDRTCTL2, mask, val); + /* enable DTCGE separatelly */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DTCGE, + CATPT_VDRTCTL2_DTCGE); + + /* SRAM power gating all */ + catpt_dsp_set_srampge(cdev, &cdev->dram, cdev->dram.mask); + catpt_dsp_set_srampge(cdev, &cdev->iram, cdev->iram.mask); + mask = WPT_VDRTCTL0_D3SRAMPGD | WPT_VDRTCTL0_D3PGD; + catpt_updatel_pci(cdev, VDRTCTL0, mask, WPT_VDRTCTL0_D3PGD); + + /* set D3 */ + catpt_updatel_pci(cdev, PMCS, CATPT_PMCS_PS, CATPT_PMCS_PS_D3HOT); + udelay(50); + + /* enable core clock gating */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DCLCGE, + CATPT_VDRTCTL2_DCLCGE); + udelay(50); + + return 0; +} + +int wpt_dsp_power_up(struct catpt_dev *cdev) +{ + u32 mask, val; + + /* disable core clock gating */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DCLCGE, 0); + + /* switch clock gating */ + mask = CATPT_VDRTCTL2_CGEALL & (~CATPT_VDRTCTL2_DCLCGE); + val = mask & (~CATPT_VDRTCTL2_DTCGE); + catpt_updatel_pci(cdev, VDRTCTL2, mask, val); + + /* set D0 */ + catpt_updatel_pci(cdev, PMCS, CATPT_PMCS_PS, 0); + + /* SRAM power gating none */ + mask = WPT_VDRTCTL0_D3SRAMPGD | WPT_VDRTCTL0_D3PGD; + catpt_updatel_pci(cdev, VDRTCTL0, mask, mask); + catpt_dsp_set_srampge(cdev, &cdev->dram, 0); + catpt_dsp_set_srampge(cdev, &cdev->iram, 0); + + catpt_dsp_set_regs_defaults(cdev); + + /* restore MCLK */ + catpt_updatel_shim(cdev, CLKCTL, CATPT_CLKCTL_SMOS, CATPT_CLKCTL_SMOS); + catpt_dsp_select_lpclock(cdev, false, false); + /* set 24Mhz clock for both SSPs */ + catpt_updatel_shim(cdev, CS, CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1), + CATPT_CS_SBCS(0) | CATPT_CS_SBCS(1)); + catpt_dsp_reset(cdev, false); + + /* enable core clock gating */ + catpt_updatel_pci(cdev, VDRTCTL2, CATPT_VDRTCTL2_DCLCGE, + CATPT_VDRTCTL2_DCLCGE); + + /* generate int deassert msg to fix inversed int logic */ + catpt_updatel_shim(cdev, IMC, CATPT_IMC_IPCDB | CATPT_IMC_IPCCD, 0); + + return 0; +}