hid: Add support for Intel Touch Host Controller

Based on quo/ithc-linux@55803a2

Signed-off-by: Dorian Stoll <[email protected]>
Patchset: ithc

drivers/hid/Kconfig

@@ -1292,4 +1292,6 @@ source "drivers/hid/surface-hid/Kconfig"
 
 source "drivers/hid/ipts/Kconfig"
 
+source "drivers/hid/ithc/Kconfig"
+
 endmenu

drivers/hid/Makefile

@@ -166,3 +166,4 @@ obj-$(CONFIG_AMD_SFH_HID)       += amd-sfh-hid/
 obj-$(CONFIG_SURFACE_HID_CORE)  += surface-hid/
 
 obj-$(CONFIG_HID_IPTS)          += ipts/
+obj-$(CONFIG_HID_ITHC)          += ithc/

drivers/hid/ithc/Kbuild

@@ -0,0 +1,6 @@
+obj-$(CONFIG_HID_ITHC) := ithc.o
+
+ithc-objs := ithc-main.o ithc-regs.o ithc-dma.o ithc-debug.o
+
+ccflags-y := -std=gnu11 -Wno-declaration-after-statement
+

drivers/hid/ithc/Kconfig

@@ -0,0 +1,12 @@
+config HID_ITHC
+	tristate "Intel Touch Host Controller"
+	depends on PCI
+	depends on HID
+	help
+	  Say Y here if your system has a touchscreen using Intels
+	  Touch Host Controller (ITHC / IPTS) technology.
+
+	  If unsure say N.
+
+	  To compile this driver as a module, choose M here: the
+	  module will be called ithc.

drivers/hid/ithc/ithc-debug.c

@@ -0,0 +1,96 @@
+#include "ithc.h"
+
+void ithc_log_regs(struct ithc *ithc) {
+	if (!ithc->prev_regs) return;
+	u32 __iomem *cur = (__iomem void*)ithc->regs;
+	u32 *prev = (void*)ithc->prev_regs;
+	for (int i = 1024; i < sizeof *ithc->regs / 4; i++) {
+		u32 x = readl(cur + i);
+		if (x != prev[i]) {
+			pci_info(ithc->pci, "reg %04x: %08x -> %08x\n", i * 4, prev[i], x);
+			prev[i] = x;
+		}
+	}
+}
+
+static ssize_t ithc_debugfs_cmd_write(struct file *f, const char __user *buf, size_t len, loff_t *offset) {
+	struct ithc *ithc = file_inode(f)->i_private;
+	char cmd[256];
+	if (!ithc || !ithc->pci) return -ENODEV;
+	if (!len) return -EINVAL;
+	if (len >= sizeof cmd) return -EINVAL;
+	if (copy_from_user(cmd, buf, len)) return -EFAULT;
+	cmd[len] = 0;
+	if (cmd[len-1] == '\n') cmd[len-1] = 0;
+	pci_info(ithc->pci, "debug command: %s\n", cmd);
+	u32 n = 0;
+	const char *s = cmd + 1;
+	u32 a[32];
+	while (*s && *s != '\n') {
+		if (n >= ARRAY_SIZE(a)) return -EINVAL;
+		if (*s++ != ' ') return -EINVAL;
+		char *e;
+		a[n++] = simple_strtoul(s, &e, 0);
+		if (e == s) return -EINVAL;
+		s = e;
+	}
+	ithc_log_regs(ithc);
+	switch(cmd[0]) {
+	case 'x': // reset
+		ithc_reset(ithc);
+		break;
+	case 'w': // write register: offset mask value
+		if (n != 3 || (a[0] & 3)) return -EINVAL;
+		pci_info(ithc->pci, "debug write 0x%04x = 0x%08x (mask 0x%08x)\n", a[0], a[2], a[1]);
+		bitsl(((__iomem u32 *)ithc->regs) + a[0] / 4, a[1], a[2]);
+		break;
+	case 'r': // read register: offset
+		if (n != 1 || (a[0] & 3)) return -EINVAL;
+		pci_info(ithc->pci, "debug read 0x%04x = 0x%08x\n", a[0], readl(((__iomem u32 *)ithc->regs) + a[0] / 4));
+		break;
+	case 's': // spi command: cmd offset len data...
+		// read config: s 4 0 64 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+		// set touch cfg: s 6 12 4 XX
+		if (n < 3 || a[2] > (n - 3) * 4) return -EINVAL;
+		pci_info(ithc->pci, "debug spi command %u with %u bytes of data\n", a[0], a[2]);
+		if (!CHECK(ithc_spi_command, ithc, a[0], a[1], a[2], a + 3))
+			for (u32 i = 0; i < (a[2] + 3) / 4; i++) pci_info(ithc->pci, "resp %u = 0x%08x\n", i, a[3+i]);
+		break;
+	case 'd': // dma command: cmd len data...
+		// get report descriptor: d 7 8 0 0
+		// enable multitouch: d 3 2 0x0105
+		if (n < 2 || a[1] > (n - 2) * 4) return -EINVAL;
+		pci_info(ithc->pci, "debug dma command %u with %u bytes of data\n", a[0], a[1]);
+		if (ithc_dma_tx(ithc, a[0], a[1], a + 2)) pci_err(ithc->pci, "dma tx failed\n");
+		break;
+	default:
+		return -EINVAL;
+	}
+	ithc_log_regs(ithc);
+	return len;
+}
+
+static const struct file_operations ithc_debugfops_cmd = {
+	.owner = THIS_MODULE,
+	.write = ithc_debugfs_cmd_write,
+};
+
+static void ithc_debugfs_devres_release(struct device *dev, void *res) {
+	struct dentry **dbgm = res;
+	if (*dbgm) debugfs_remove_recursive(*dbgm);
+}
+
+int ithc_debug_init(struct ithc *ithc) {
+	struct dentry **dbgm = devres_alloc(ithc_debugfs_devres_release, sizeof *dbgm, GFP_KERNEL);
+	if (!dbgm) return -ENOMEM;
+	devres_add(&ithc->pci->dev, dbgm);
+	struct dentry *dbg = debugfs_create_dir(DEVNAME, NULL);
+	if (IS_ERR(dbg)) return PTR_ERR(dbg);
+	*dbgm = dbg;
+
+	struct dentry *cmd = debugfs_create_file("cmd", 0220, dbg, ithc, &ithc_debugfops_cmd);
+	if (IS_ERR(cmd)) return PTR_ERR(cmd);
+
+	return 0;
+}
+

drivers/hid/ithc/ithc-dma.c

@@ -0,0 +1,258 @@
+#include "ithc.h"
+
+static int ithc_dma_prd_alloc(struct ithc *ithc, struct ithc_dma_prd_buffer *p, unsigned num_buffers, unsigned num_pages, enum dma_data_direction dir) {
+	p->num_pages = num_pages;
+	p->dir = dir;
+	p->size = round_up(num_buffers * num_pages * sizeof(struct ithc_phys_region_desc), PAGE_SIZE);
+	p->addr = dmam_alloc_coherent(&ithc->pci->dev, p->size, &p->dma_addr, GFP_KERNEL);
+	if (!p->addr) return -ENOMEM;
+	if (p->dma_addr & (PAGE_SIZE - 1)) return -EFAULT;
+	return 0;
+}
+
+struct ithc_sg_table {
+	void *addr;
+	struct sg_table sgt;
+	enum dma_data_direction dir;
+};
+static void ithc_dma_sgtable_free(struct sg_table *sgt) {
+	struct scatterlist *sg;
+	int i;
+	for_each_sgtable_sg(sgt, sg, i) {
+		struct page *p = sg_page(sg);
+		if (p) __free_page(p);
+	}
+	sg_free_table(sgt);
+}
+static void ithc_dma_data_devres_release(struct device *dev, void *res) {
+	struct ithc_sg_table *sgt = res;
+	if (sgt->addr) vunmap(sgt->addr);
+	dma_unmap_sgtable(dev, &sgt->sgt, sgt->dir, 0);
+	ithc_dma_sgtable_free(&sgt->sgt);
+}
+
+static int ithc_dma_data_alloc(struct ithc* ithc, struct ithc_dma_prd_buffer *prds, struct ithc_dma_data_buffer *b) {
+	// We don't use dma_alloc_coherent for data buffers, because they don't have to be contiguous (we can use one PRD per page) or coherent (they are unidirectional).
+	// Instead we use an sg_table of individually allocated pages (5.13 has dma_alloc_noncontiguous for this, but we'd like to support 5.10 for now).
+	struct page *pages[16];
+	if (prds->num_pages == 0 || prds->num_pages > ARRAY_SIZE(pages)) return -EINVAL;
+	b->active_idx = -1;
+	struct ithc_sg_table *sgt = devres_alloc(ithc_dma_data_devres_release, sizeof *sgt, GFP_KERNEL);
+	if (!sgt) return -ENOMEM;
+	sgt->dir = prds->dir;
+	if (!sg_alloc_table(&sgt->sgt, prds->num_pages, GFP_KERNEL)) {
+		struct scatterlist *sg;
+		int i;
+		bool ok = true;
+		for_each_sgtable_sg(&sgt->sgt, sg, i) {
+			struct page *p = pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO); // don't need __GFP_DMA for PCI DMA
+			if (!p) { ok = false; break; }
+			sg_set_page(sg, p, PAGE_SIZE, 0);
+		}
+		if (ok && !dma_map_sgtable(&ithc->pci->dev, &sgt->sgt, prds->dir, 0)) {
+			devres_add(&ithc->pci->dev, sgt);
+			b->sgt = &sgt->sgt;
+			b->addr = sgt->addr = vmap(pages, prds->num_pages, 0, PAGE_KERNEL);
+			if (!b->addr) return -ENOMEM;
+			return 0;
+		}
+		ithc_dma_sgtable_free(&sgt->sgt);
+	}
+	devres_free(sgt);
+	return -ENOMEM;
+}
+
+static int ithc_dma_data_buffer_put(struct ithc *ithc, struct ithc_dma_prd_buffer *prds, struct ithc_dma_data_buffer *b, unsigned idx) {
+	struct ithc_phys_region_desc *prd = prds->addr;
+	prd += idx * prds->num_pages;
+	if (b->active_idx >= 0) { pci_err(ithc->pci, "buffer already active\n"); return -EINVAL; }
+	b->active_idx = idx;
+	if (prds->dir == DMA_TO_DEVICE) {
+		if (b->data_size > PAGE_SIZE) return -EINVAL;
+		prd->addr = sg_dma_address(b->sgt->sgl) >> 10;
+		prd->size = b->data_size | PRD_FLAG_END;
+		flush_kernel_vmap_range(b->addr, b->data_size);
+	} else if (prds->dir == DMA_FROM_DEVICE) {
+		struct scatterlist *sg;
+		int i;
+		for_each_sgtable_dma_sg(b->sgt, sg, i) {
+			prd->addr = sg_dma_address(sg) >> 10;
+			prd->size = sg_dma_len(sg);
+			prd++;
+		}
+		prd[-1].size |= PRD_FLAG_END;
+	}
+	dma_wmb(); // for the prds
+	dma_sync_sgtable_for_device(&ithc->pci->dev, b->sgt, prds->dir);
+	return 0;
+}
+
+static int ithc_dma_data_buffer_get(struct ithc *ithc, struct ithc_dma_prd_buffer *prds, struct ithc_dma_data_buffer *b, unsigned idx) {
+	struct ithc_phys_region_desc *prd = prds->addr;
+	prd += idx * prds->num_pages;
+	if (b->active_idx != idx) { pci_err(ithc->pci, "wrong buffer index\n"); return -EINVAL; }
+	b->active_idx = -1;
+	if (prds->dir == DMA_FROM_DEVICE) {
+		dma_rmb(); // for the prds
+		b->data_size = 0;
+		struct scatterlist *sg;
+		int i;
+		for_each_sgtable_dma_sg(b->sgt, sg, i) {
+			unsigned size = prd->size;
+			b->data_size += size & PRD_SIZE_MASK;
+			if (size & PRD_FLAG_END) break;
+			if ((size & PRD_SIZE_MASK) != sg_dma_len(sg)) { pci_err(ithc->pci, "truncated prd\n"); break; }
+			prd++;
+		}
+		invalidate_kernel_vmap_range(b->addr, b->data_size);
+	}
+	dma_sync_sgtable_for_cpu(&ithc->pci->dev, b->sgt, prds->dir);
+	return 0;
+}
+
+int ithc_dma_rx_init(struct ithc *ithc, u8 channel, const char *devname) {
+	struct ithc_dma_rx *rx = &ithc->dma_rx[channel];
+	mutex_init(&rx->mutex);
+	u32 buf_size = DEVCFG_DMA_RX_SIZE(ithc->config.dma_buf_sizes);
+	unsigned num_pages = (buf_size + PAGE_SIZE - 1) / PAGE_SIZE;
+	pci_dbg(ithc->pci, "allocating rx buffers: num = %u, size = %u, pages = %u\n", NUM_RX_BUF, buf_size, num_pages);
+	CHECK_RET(ithc_dma_prd_alloc, ithc, &rx->prds, NUM_RX_BUF, num_pages, DMA_FROM_DEVICE);
+	for (unsigned i = 0; i < NUM_RX_BUF; i++)
+		CHECK_RET(ithc_dma_data_alloc, ithc, &rx->prds, &rx->bufs[i]);
+	writeb(DMA_RX_CONTROL2_RESET, &ithc->regs->dma_rx[channel].control2);
+	lo_hi_writeq(rx->prds.dma_addr, &ithc->regs->dma_rx[channel].addr);
+	writeb(NUM_RX_BUF - 1, &ithc->regs->dma_rx[channel].num_bufs);
+	writeb(num_pages - 1, &ithc->regs->dma_rx[channel].num_prds);
+	u8 head = readb(&ithc->regs->dma_rx[channel].head);
+	if (head) { pci_err(ithc->pci, "head is nonzero (%u)\n", head); return -EIO; }
+	for (unsigned i = 0; i < NUM_RX_BUF; i++)
+		CHECK_RET(ithc_dma_data_buffer_put, ithc, &rx->prds, &rx->bufs[i], i);
+	writeb(head ^ DMA_RX_WRAP_FLAG, &ithc->regs->dma_rx[channel].tail);
+	return 0;
+}
+void ithc_dma_rx_enable(struct ithc *ithc, u8 channel) {
+	bitsb_set(&ithc->regs->dma_rx[channel].control, DMA_RX_CONTROL_ENABLE | DMA_RX_CONTROL_IRQ_ERROR | DMA_RX_CONTROL_IRQ_DATA);
+	CHECK(waitl, ithc, &ithc->regs->dma_rx[1].status, DMA_RX_STATUS_ENABLED, DMA_RX_STATUS_ENABLED);
+}
+
+int ithc_dma_tx_init(struct ithc *ithc) {
+	struct ithc_dma_tx *tx = &ithc->dma_tx;
+	mutex_init(&tx->mutex);
+	tx->max_size = DEVCFG_DMA_TX_SIZE(ithc->config.dma_buf_sizes);
+	unsigned num_pages = (tx->max_size + PAGE_SIZE - 1) / PAGE_SIZE;
+	pci_dbg(ithc->pci, "allocating tx buffers: size = %u, pages = %u\n", tx->max_size, num_pages);
+	CHECK_RET(ithc_dma_prd_alloc, ithc, &tx->prds, 1, num_pages, DMA_TO_DEVICE);
+	CHECK_RET(ithc_dma_data_alloc, ithc, &tx->prds, &tx->buf);
+	lo_hi_writeq(tx->prds.dma_addr, &ithc->regs->dma_tx.addr);
+	writeb(num_pages - 1, &ithc->regs->dma_tx.num_prds);
+	CHECK_RET(ithc_dma_data_buffer_put, ithc, &ithc->dma_tx.prds, &ithc->dma_tx.buf, 0);
+	return 0;
+}
+
+static int ithc_dma_rx_process_buf(struct ithc *ithc, struct ithc_dma_data_buffer *data, u8 channel, u8 buf) {
+	if (buf >= NUM_RX_BUF) {
+		pci_err(ithc->pci, "invalid dma ringbuffer index\n");
+		return -EINVAL;
+	}
+	ithc_set_active(ithc);
+	u32 len = data->data_size;
+	struct ithc_dma_rx_header *hdr = data->addr;
+	u8 *hiddata = (void *)(hdr + 1);
+	if (len >= sizeof *hdr && hdr->code == DMA_RX_CODE_RESET) {
+		CHECK(ithc_reset, ithc);
+	} else if (len < sizeof *hdr || len != sizeof *hdr + hdr->data_size) {
+		if (hdr->code == DMA_RX_CODE_INPUT_REPORT) {
+			// When the CPU enters a low power state during DMA, we can get truncated messages.
+			// Typically this will be a single touch HID report that is only 1 byte, or a multitouch report that is 257 bytes.
+			// See also ithc_set_active().
+		} else {
+			pci_err(ithc->pci, "invalid dma rx data! channel %u, buffer %u, size %u, code %u, data size %u\n", channel, buf, len, hdr->code, hdr->data_size);
+			print_hex_dump_debug(DEVNAME " data: ", DUMP_PREFIX_OFFSET, 32, 1, hdr, min(len, 0x400u), 0);
+		}
+	} else if (hdr->code == DMA_RX_CODE_REPORT_DESCRIPTOR && hdr->data_size > 8) {
+		CHECK(hid_parse_report, ithc->hid, hiddata + 8, hdr->data_size - 8);
+		WRITE_ONCE(ithc->hid_parse_done, true);
+		wake_up(&ithc->wait_hid_parse);
+	} else if (hdr->code == DMA_RX_CODE_INPUT_REPORT) {
+		CHECK(hid_input_report, ithc->hid, HID_INPUT_REPORT, hiddata, hdr->data_size, 1);
+	} else if (hdr->code == DMA_RX_CODE_FEATURE_REPORT) {
+		bool done = false;
+		mutex_lock(&ithc->hid_get_feature_mutex);
+		if (ithc->hid_get_feature_buf) {
+			if (hdr->data_size < ithc->hid_get_feature_size) ithc->hid_get_feature_size = hdr->data_size;
+			memcpy(ithc->hid_get_feature_buf, hiddata, ithc->hid_get_feature_size);
+			ithc->hid_get_feature_buf = NULL;
+			done = true;
+		}
+		mutex_unlock(&ithc->hid_get_feature_mutex);
+		if (done) wake_up(&ithc->wait_hid_get_feature);
+		else CHECK(hid_input_report, ithc->hid, HID_FEATURE_REPORT, hiddata, hdr->data_size, 1);
+	} else {
+		pci_dbg(ithc->pci, "unhandled dma rx data! channel %u, buffer %u, size %u, code %u\n", channel, buf, len, hdr->code);
+		print_hex_dump_debug(DEVNAME " data: ", DUMP_PREFIX_OFFSET, 32, 1, hdr, min(len, 0x400u), 0);
+	}
+	return 0;
+}
+
+static int ithc_dma_rx_unlocked(struct ithc *ithc, u8 channel) {
+	struct ithc_dma_rx *rx = &ithc->dma_rx[channel];
+	unsigned n = rx->num_received;
+	u8 head_wrap = readb(&ithc->regs->dma_rx[channel].head);
+	while (1) {
+		u8 tail = n % NUM_RX_BUF;
+		u8 tail_wrap = tail | ((n / NUM_RX_BUF) & 1 ? 0 : DMA_RX_WRAP_FLAG);
+		writeb(tail_wrap, &ithc->regs->dma_rx[channel].tail);
+		// ringbuffer is full if tail_wrap == head_wrap
+		// ringbuffer is empty if tail_wrap == head_wrap ^ WRAP_FLAG
+		if (tail_wrap == (head_wrap ^ DMA_RX_WRAP_FLAG)) return 0;
+
+		// take the buffer that the device just filled
+		struct ithc_dma_data_buffer *b = &rx->bufs[n % NUM_RX_BUF];
+		CHECK_RET(ithc_dma_data_buffer_get, ithc, &rx->prds, b, tail);
+		rx->num_received = ++n;
+
+		// process data
+		CHECK(ithc_dma_rx_process_buf, ithc, b, channel, tail);
+
+		// give the buffer back to the device
+		CHECK_RET(ithc_dma_data_buffer_put, ithc, &rx->prds, b, tail);
+	}
+}
+int ithc_dma_rx(struct ithc *ithc, u8 channel) {
+	struct ithc_dma_rx *rx = &ithc->dma_rx[channel];
+	mutex_lock(&rx->mutex);
+	int ret = ithc_dma_rx_unlocked(ithc, channel);
+	mutex_unlock(&rx->mutex);
+	return ret;
+}
+
+static int ithc_dma_tx_unlocked(struct ithc *ithc, u32 cmdcode, u32 datasize, void *data) {
+	pci_dbg(ithc->pci, "dma tx command %u, size %u\n", cmdcode, datasize);
+	struct ithc_dma_tx_header *hdr;
+	u8 padding = datasize & 3 ? 4 - (datasize & 3) : 0;
+	unsigned fullsize = sizeof *hdr + datasize + padding;
+	if (fullsize > ithc->dma_tx.max_size || fullsize > PAGE_SIZE) return -EINVAL;
+	CHECK_RET(ithc_dma_data_buffer_get, ithc, &ithc->dma_tx.prds, &ithc->dma_tx.buf, 0);
+
+	ithc->dma_tx.buf.data_size = fullsize;
+	hdr = ithc->dma_tx.buf.addr;
+	hdr->code = cmdcode;
+	hdr->data_size = datasize;
+	u8 *dest = (void *)(hdr + 1);
+	memcpy(dest, data, datasize);
+	dest += datasize;
+	for (u8 p = 0; p < padding; p++) *dest++ = 0;
+	CHECK_RET(ithc_dma_data_buffer_put, ithc, &ithc->dma_tx.prds, &ithc->dma_tx.buf, 0);
+
+	bitsb_set(&ithc->regs->dma_tx.control, DMA_TX_CONTROL_SEND);
+	CHECK_RET(waitb, ithc, &ithc->regs->dma_tx.control, DMA_TX_CONTROL_SEND, 0);
+	writel(DMA_TX_STATUS_DONE, &ithc->regs->dma_tx.status);
+	return 0;
+}
+int ithc_dma_tx(struct ithc *ithc, u32 cmdcode, u32 datasize, void *data) {
+	mutex_lock(&ithc->dma_tx.mutex);
+	int ret = ithc_dma_tx_unlocked(ithc, cmdcode, datasize, data);
+	mutex_unlock(&ithc->dma_tx.mutex);
+	return ret;
+}
+