cachepc-linux

adf_transport.c (16718B)

// SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0-only)
/* Copyright(c) 2014 - 2020 Intel Corporation */
#include <linux/delay.h>
#include <linux/nospec.h>
#include "adf_accel_devices.h"
#include "adf_transport_internal.h"
#include "adf_transport_access_macros.h"
#include "adf_cfg.h"
#include "adf_common_drv.h"

#define ADF_MAX_RING_THRESHOLD		80
#define ADF_PERCENT(tot, percent)	(((tot) * (percent)) / 100)

static inline u32 adf_modulo(u32 data, u32 shift)
{
	u32 div = data >> shift;
	u32 mult = div << shift;

	return data - mult;
}

static inline int adf_check_ring_alignment(u64 addr, u64 size)
{
	if (((size - 1) & addr) != 0)
		return -EFAULT;
	return 0;
}

static int adf_verify_ring_size(u32 msg_size, u32 msg_num)
{
	int i = ADF_MIN_RING_SIZE;

	for (; i <= ADF_MAX_RING_SIZE; i++)
		if ((msg_size * msg_num) == ADF_SIZE_TO_RING_SIZE_IN_BYTES(i))
			return i;

	return ADF_DEFAULT_RING_SIZE;
}

static int adf_reserve_ring(struct adf_etr_bank_data *bank, u32 ring)
{
	spin_lock(&bank->lock);
	if (bank->ring_mask & (1 << ring)) {
		spin_unlock(&bank->lock);
		return -EFAULT;
	}
	bank->ring_mask |= (1 << ring);
	spin_unlock(&bank->lock);
	return 0;
}

static void adf_unreserve_ring(struct adf_etr_bank_data *bank, u32 ring)
{
	spin_lock(&bank->lock);
	bank->ring_mask &= ~(1 << ring);
	spin_unlock(&bank->lock);
}

static void adf_enable_ring_irq(struct adf_etr_bank_data *bank, u32 ring)
{
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(bank->accel_dev);

	spin_lock_bh(&bank->lock);
	bank->irq_mask |= (1 << ring);
	spin_unlock_bh(&bank->lock);
	csr_ops->write_csr_int_col_en(bank->csr_addr, bank->bank_number,
				      bank->irq_mask);
	csr_ops->write_csr_int_col_ctl(bank->csr_addr, bank->bank_number,
				       bank->irq_coalesc_timer);
}

static void adf_disable_ring_irq(struct adf_etr_bank_data *bank, u32 ring)
{
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(bank->accel_dev);

	spin_lock_bh(&bank->lock);
	bank->irq_mask &= ~(1 << ring);
	spin_unlock_bh(&bank->lock);
	csr_ops->write_csr_int_col_en(bank->csr_addr, bank->bank_number,
				      bank->irq_mask);
}

bool adf_ring_nearly_full(struct adf_etr_ring_data *ring)
{
	return atomic_read(ring->inflights) > ring->threshold;
}

int adf_send_message(struct adf_etr_ring_data *ring, u32 *msg)
{
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(ring->bank->accel_dev);

	if (atomic_add_return(1, ring->inflights) >
	    ADF_MAX_INFLIGHTS(ring->ring_size, ring->msg_size)) {
		atomic_dec(ring->inflights);
		return -EAGAIN;
	}
	spin_lock_bh(&ring->lock);
	memcpy((void *)((uintptr_t)ring->base_addr + ring->tail), msg,
	       ADF_MSG_SIZE_TO_BYTES(ring->msg_size));

	ring->tail = adf_modulo(ring->tail +
				ADF_MSG_SIZE_TO_BYTES(ring->msg_size),
				ADF_RING_SIZE_MODULO(ring->ring_size));
	csr_ops->write_csr_ring_tail(ring->bank->csr_addr,
				     ring->bank->bank_number, ring->ring_number,
				     ring->tail);
	spin_unlock_bh(&ring->lock);

	return 0;
}

static int adf_handle_response(struct adf_etr_ring_data *ring)
{
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(ring->bank->accel_dev);
	u32 msg_counter = 0;
	u32 *msg = (u32 *)((uintptr_t)ring->base_addr + ring->head);

	while (*msg != ADF_RING_EMPTY_SIG) {
		ring->callback((u32 *)msg);
		atomic_dec(ring->inflights);
		*msg = ADF_RING_EMPTY_SIG;
		ring->head = adf_modulo(ring->head +
					ADF_MSG_SIZE_TO_BYTES(ring->msg_size),
					ADF_RING_SIZE_MODULO(ring->ring_size));
		msg_counter++;
		msg = (u32 *)((uintptr_t)ring->base_addr + ring->head);
	}
	if (msg_counter > 0) {
		csr_ops->write_csr_ring_head(ring->bank->csr_addr,
					     ring->bank->bank_number,
					     ring->ring_number, ring->head);
	}
	return 0;
}

static void adf_configure_tx_ring(struct adf_etr_ring_data *ring)
{
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(ring->bank->accel_dev);
	u32 ring_config = BUILD_RING_CONFIG(ring->ring_size);

	csr_ops->write_csr_ring_config(ring->bank->csr_addr,
				       ring->bank->bank_number,
				       ring->ring_number, ring_config);

}

static void adf_configure_rx_ring(struct adf_etr_ring_data *ring)
{
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(ring->bank->accel_dev);
	u32 ring_config =
			BUILD_RESP_RING_CONFIG(ring->ring_size,
					       ADF_RING_NEAR_WATERMARK_512,
					       ADF_RING_NEAR_WATERMARK_0);

	csr_ops->write_csr_ring_config(ring->bank->csr_addr,
				       ring->bank->bank_number,
				       ring->ring_number, ring_config);
}

static int adf_init_ring(struct adf_etr_ring_data *ring)
{
	struct adf_etr_bank_data *bank = ring->bank;
	struct adf_accel_dev *accel_dev = bank->accel_dev;
	struct adf_hw_device_data *hw_data = accel_dev->hw_device;
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(accel_dev);
	u64 ring_base;
	u32 ring_size_bytes =
			ADF_SIZE_TO_RING_SIZE_IN_BYTES(ring->ring_size);

	ring_size_bytes = ADF_RING_SIZE_BYTES_MIN(ring_size_bytes);
	ring->base_addr = dma_alloc_coherent(&GET_DEV(accel_dev),
					     ring_size_bytes, &ring->dma_addr,
					     GFP_KERNEL);
	if (!ring->base_addr)
		return -ENOMEM;

	memset(ring->base_addr, 0x7F, ring_size_bytes);
	/* The base_addr has to be aligned to the size of the buffer */
	if (adf_check_ring_alignment(ring->dma_addr, ring_size_bytes)) {
		dev_err(&GET_DEV(accel_dev), "Ring address not aligned\n");
		dma_free_coherent(&GET_DEV(accel_dev), ring_size_bytes,
				  ring->base_addr, ring->dma_addr);
		ring->base_addr = NULL;
		return -EFAULT;
	}

	if (hw_data->tx_rings_mask & (1 << ring->ring_number))
		adf_configure_tx_ring(ring);

	else
		adf_configure_rx_ring(ring);

	ring_base = csr_ops->build_csr_ring_base_addr(ring->dma_addr,
						      ring->ring_size);

	csr_ops->write_csr_ring_base(ring->bank->csr_addr,
				     ring->bank->bank_number, ring->ring_number,
				     ring_base);
	spin_lock_init(&ring->lock);
	return 0;
}

static void adf_cleanup_ring(struct adf_etr_ring_data *ring)
{
	u32 ring_size_bytes =
			ADF_SIZE_TO_RING_SIZE_IN_BYTES(ring->ring_size);
	ring_size_bytes = ADF_RING_SIZE_BYTES_MIN(ring_size_bytes);

	if (ring->base_addr) {
		memset(ring->base_addr, 0x7F, ring_size_bytes);
		dma_free_coherent(&GET_DEV(ring->bank->accel_dev),
				  ring_size_bytes, ring->base_addr,
				  ring->dma_addr);
	}
}

int adf_create_ring(struct adf_accel_dev *accel_dev, const char *section,
		    u32 bank_num, u32 num_msgs,
		    u32 msg_size, const char *ring_name,
		    adf_callback_fn callback, int poll_mode,
		    struct adf_etr_ring_data **ring_ptr)
{
	struct adf_etr_data *transport_data = accel_dev->transport;
	u8 num_rings_per_bank = GET_NUM_RINGS_PER_BANK(accel_dev);
	struct adf_etr_bank_data *bank;
	struct adf_etr_ring_data *ring;
	char val[ADF_CFG_MAX_VAL_LEN_IN_BYTES];
	int max_inflights;
	u32 ring_num;
	int ret;

	if (bank_num >= GET_MAX_BANKS(accel_dev)) {
		dev_err(&GET_DEV(accel_dev), "Invalid bank number\n");
		return -EFAULT;
	}
	if (msg_size > ADF_MSG_SIZE_TO_BYTES(ADF_MAX_MSG_SIZE)) {
		dev_err(&GET_DEV(accel_dev), "Invalid msg size\n");
		return -EFAULT;
	}
	if (ADF_MAX_INFLIGHTS(adf_verify_ring_size(msg_size, num_msgs),
			      ADF_BYTES_TO_MSG_SIZE(msg_size)) < 2) {
		dev_err(&GET_DEV(accel_dev),
			"Invalid ring size for given msg size\n");
		return -EFAULT;
	}
	if (adf_cfg_get_param_value(accel_dev, section, ring_name, val)) {
		dev_err(&GET_DEV(accel_dev), "Section %s, no such entry : %s\n",
			section, ring_name);
		return -EFAULT;
	}
	if (kstrtouint(val, 10, &ring_num)) {
		dev_err(&GET_DEV(accel_dev), "Can't get ring number\n");
		return -EFAULT;
	}
	if (ring_num >= num_rings_per_bank) {
		dev_err(&GET_DEV(accel_dev), "Invalid ring number\n");
		return -EFAULT;
	}

	ring_num = array_index_nospec(ring_num, num_rings_per_bank);
	bank = &transport_data->banks[bank_num];
	if (adf_reserve_ring(bank, ring_num)) {
		dev_err(&GET_DEV(accel_dev), "Ring %d, %s already exists.\n",
			ring_num, ring_name);
		return -EFAULT;
	}
	ring = &bank->rings[ring_num];
	ring->ring_number = ring_num;
	ring->bank = bank;
	ring->callback = callback;
	ring->msg_size = ADF_BYTES_TO_MSG_SIZE(msg_size);
	ring->ring_size = adf_verify_ring_size(msg_size, num_msgs);
	ring->head = 0;
	ring->tail = 0;
	max_inflights = ADF_MAX_INFLIGHTS(ring->ring_size, ring->msg_size);
	ring->threshold = ADF_PERCENT(max_inflights, ADF_MAX_RING_THRESHOLD);
	atomic_set(ring->inflights, 0);
	ret = adf_init_ring(ring);
	if (ret)
		goto err;

	/* Enable HW arbitration for the given ring */
	adf_update_ring_arb(ring);

	if (adf_ring_debugfs_add(ring, ring_name)) {
		dev_err(&GET_DEV(accel_dev),
			"Couldn't add ring debugfs entry\n");
		ret = -EFAULT;
		goto err;
	}

	/* Enable interrupts if needed */
	if (callback && (!poll_mode))
		adf_enable_ring_irq(bank, ring->ring_number);
	*ring_ptr = ring;
	return 0;
err:
	adf_cleanup_ring(ring);
	adf_unreserve_ring(bank, ring_num);
	adf_update_ring_arb(ring);
	return ret;
}

void adf_remove_ring(struct adf_etr_ring_data *ring)
{
	struct adf_etr_bank_data *bank = ring->bank;
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(bank->accel_dev);

	/* Disable interrupts for the given ring */
	adf_disable_ring_irq(bank, ring->ring_number);

	/* Clear PCI config space */

	csr_ops->write_csr_ring_config(bank->csr_addr, bank->bank_number,
				       ring->ring_number, 0);
	csr_ops->write_csr_ring_base(bank->csr_addr, bank->bank_number,
				     ring->ring_number, 0);
	adf_ring_debugfs_rm(ring);
	adf_unreserve_ring(bank, ring->ring_number);
	/* Disable HW arbitration for the given ring */
	adf_update_ring_arb(ring);
	adf_cleanup_ring(ring);
}

static void adf_ring_response_handler(struct adf_etr_bank_data *bank)
{
	struct adf_accel_dev *accel_dev = bank->accel_dev;
	u8 num_rings_per_bank = GET_NUM_RINGS_PER_BANK(accel_dev);
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(accel_dev);
	unsigned long empty_rings;
	int i;

	empty_rings = csr_ops->read_csr_e_stat(bank->csr_addr,
					       bank->bank_number);
	empty_rings = ~empty_rings & bank->irq_mask;

	for_each_set_bit(i, &empty_rings, num_rings_per_bank)
		adf_handle_response(&bank->rings[i]);
}

void adf_response_handler(uintptr_t bank_addr)
{
	struct adf_etr_bank_data *bank = (void *)bank_addr;
	struct adf_hw_csr_ops *csr_ops = GET_CSR_OPS(bank->accel_dev);

	/* Handle all the responses and reenable IRQs */
	adf_ring_response_handler(bank);

	csr_ops->write_csr_int_flag_and_col(bank->csr_addr, bank->bank_number,
					    bank->irq_mask);
}

static inline int adf_get_cfg_int(struct adf_accel_dev *accel_dev,
				  const char *section, const char *format,
				  u32 key, u32 *value)
{
	char key_buf[ADF_CFG_MAX_KEY_LEN_IN_BYTES];
	char val_buf[ADF_CFG_MAX_VAL_LEN_IN_BYTES];

	snprintf(key_buf, ADF_CFG_MAX_KEY_LEN_IN_BYTES, format, key);

	if (adf_cfg_get_param_value(accel_dev, section, key_buf, val_buf))
		return -EFAULT;

	if (kstrtouint(val_buf, 10, value))
		return -EFAULT;
	return 0;
}

static void adf_get_coalesc_timer(struct adf_etr_bank_data *bank,
				  const char *section,
				  u32 bank_num_in_accel)
{
	if (adf_get_cfg_int(bank->accel_dev, section,
			    ADF_ETRMGR_COALESCE_TIMER_FORMAT,
			    bank_num_in_accel, &bank->irq_coalesc_timer))
		bank->irq_coalesc_timer = ADF_COALESCING_DEF_TIME;

	if (ADF_COALESCING_MAX_TIME < bank->irq_coalesc_timer ||
	    ADF_COALESCING_MIN_TIME > bank->irq_coalesc_timer)
		bank->irq_coalesc_timer = ADF_COALESCING_DEF_TIME;
}

static int adf_init_bank(struct adf_accel_dev *accel_dev,
			 struct adf_etr_bank_data *bank,
			 u32 bank_num, void __iomem *csr_addr)
{
	struct adf_hw_device_data *hw_data = accel_dev->hw_device;
	u8 num_rings_per_bank = hw_data->num_rings_per_bank;
	struct adf_hw_csr_ops *csr_ops = &hw_data->csr_ops;
	u32 irq_mask = BIT(num_rings_per_bank) - 1;
	struct adf_etr_ring_data *ring;
	struct adf_etr_ring_data *tx_ring;
	u32 i, coalesc_enabled = 0;
	unsigned long ring_mask;
	int size;

	memset(bank, 0, sizeof(*bank));
	bank->bank_number = bank_num;
	bank->csr_addr = csr_addr;
	bank->accel_dev = accel_dev;
	spin_lock_init(&bank->lock);

	/* Allocate the rings in the bank */
	size = num_rings_per_bank * sizeof(struct adf_etr_ring_data);
	bank->rings = kzalloc_node(size, GFP_KERNEL,
				   dev_to_node(&GET_DEV(accel_dev)));
	if (!bank->rings)
		return -ENOMEM;

	/* Enable IRQ coalescing always. This will allow to use
	 * the optimised flag and coalesc register.
	 * If it is disabled in the config file just use min time value */
	if ((adf_get_cfg_int(accel_dev, "Accelerator0",
			     ADF_ETRMGR_COALESCING_ENABLED_FORMAT, bank_num,
			     &coalesc_enabled) == 0) && coalesc_enabled)
		adf_get_coalesc_timer(bank, "Accelerator0", bank_num);
	else
		bank->irq_coalesc_timer = ADF_COALESCING_MIN_TIME;

	for (i = 0; i < num_rings_per_bank; i++) {
		csr_ops->write_csr_ring_config(csr_addr, bank_num, i, 0);
		csr_ops->write_csr_ring_base(csr_addr, bank_num, i, 0);

		ring = &bank->rings[i];
		if (hw_data->tx_rings_mask & (1 << i)) {
			ring->inflights =
				kzalloc_node(sizeof(atomic_t),
					     GFP_KERNEL,
					     dev_to_node(&GET_DEV(accel_dev)));
			if (!ring->inflights)
				goto err;
		} else {
			if (i < hw_data->tx_rx_gap) {
				dev_err(&GET_DEV(accel_dev),
					"Invalid tx rings mask config\n");
				goto err;
			}
			tx_ring = &bank->rings[i - hw_data->tx_rx_gap];
			ring->inflights = tx_ring->inflights;
		}
	}
	if (adf_bank_debugfs_add(bank)) {
		dev_err(&GET_DEV(accel_dev),
			"Failed to add bank debugfs entry\n");
		goto err;
	}

	csr_ops->write_csr_int_flag(csr_addr, bank_num, irq_mask);
	csr_ops->write_csr_int_srcsel(csr_addr, bank_num);

	return 0;
err:
	ring_mask = hw_data->tx_rings_mask;
	for_each_set_bit(i, &ring_mask, num_rings_per_bank) {
		ring = &bank->rings[i];
		kfree(ring->inflights);
		ring->inflights = NULL;
	}
	kfree(bank->rings);
	return -ENOMEM;
}

/**
 * adf_init_etr_data() - Initialize transport rings for acceleration device
 * @accel_dev:  Pointer to acceleration device.
 *
 * Function is the initializes the communications channels (rings) to the
 * acceleration device accel_dev.
 * To be used by QAT device specific drivers.
 *
 * Return: 0 on success, error code otherwise.
 */
int adf_init_etr_data(struct adf_accel_dev *accel_dev)
{
	struct adf_etr_data *etr_data;
	struct adf_hw_device_data *hw_data = accel_dev->hw_device;
	void __iomem *csr_addr;
	u32 size;
	u32 num_banks = 0;
	int i, ret;

	etr_data = kzalloc_node(sizeof(*etr_data), GFP_KERNEL,
				dev_to_node(&GET_DEV(accel_dev)));
	if (!etr_data)
		return -ENOMEM;

	num_banks = GET_MAX_BANKS(accel_dev);
	size = num_banks * sizeof(struct adf_etr_bank_data);
	etr_data->banks = kzalloc_node(size, GFP_KERNEL,
				       dev_to_node(&GET_DEV(accel_dev)));
	if (!etr_data->banks) {
		ret = -ENOMEM;
		goto err_bank;
	}

	accel_dev->transport = etr_data;
	i = hw_data->get_etr_bar_id(hw_data);
	csr_addr = accel_dev->accel_pci_dev.pci_bars[i].virt_addr;

	/* accel_dev->debugfs_dir should always be non-NULL here */
	etr_data->debug = debugfs_create_dir("transport",
					     accel_dev->debugfs_dir);

	for (i = 0; i < num_banks; i++) {
		ret = adf_init_bank(accel_dev, &etr_data->banks[i], i,
				    csr_addr);
		if (ret)
			goto err_bank_all;
	}

	return 0;

err_bank_all:
	debugfs_remove(etr_data->debug);
	kfree(etr_data->banks);
err_bank:
	kfree(etr_data);
	accel_dev->transport = NULL;
	return ret;
}
EXPORT_SYMBOL_GPL(adf_init_etr_data);

static void cleanup_bank(struct adf_etr_bank_data *bank)
{
	struct adf_accel_dev *accel_dev = bank->accel_dev;
	struct adf_hw_device_data *hw_data = accel_dev->hw_device;
	u8 num_rings_per_bank = hw_data->num_rings_per_bank;
	u32 i;

	for (i = 0; i < num_rings_per_bank; i++) {
		struct adf_etr_ring_data *ring = &bank->rings[i];

		if (bank->ring_mask & (1 << i))
			adf_cleanup_ring(ring);

		if (hw_data->tx_rings_mask & (1 << i))
			kfree(ring->inflights);
	}
	kfree(bank->rings);
	adf_bank_debugfs_rm(bank);
	memset(bank, 0, sizeof(*bank));
}

static void adf_cleanup_etr_handles(struct adf_accel_dev *accel_dev)
{
	struct adf_etr_data *etr_data = accel_dev->transport;
	u32 i, num_banks = GET_MAX_BANKS(accel_dev);

	for (i = 0; i < num_banks; i++)
		cleanup_bank(&etr_data->banks[i]);
}

/**
 * adf_cleanup_etr_data() - Clear transport rings for acceleration device
 * @accel_dev:  Pointer to acceleration device.
 *
 * Function is the clears the communications channels (rings) of the
 * acceleration device accel_dev.
 * To be used by QAT device specific drivers.
 *
 * Return: void
 */
void adf_cleanup_etr_data(struct adf_accel_dev *accel_dev)
{
	struct adf_etr_data *etr_data = accel_dev->transport;

	if (etr_data) {
		adf_cleanup_etr_handles(accel_dev);
		debugfs_remove(etr_data->debug);
		kfree(etr_data->banks->rings);
		kfree(etr_data->banks);
		kfree(etr_data);
		accel_dev->transport = NULL;
	}
}
EXPORT_SYMBOL_GPL(adf_cleanup_etr_data);