tegrakernel/kernel/nvidia/drivers/virt/tegra/vm_err.c

558 lines
16 KiB
C
Raw Normal View History

2022-02-16 09:13:02 -06:00
/*
* Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*/
#define pr_fmt(fmt) "vm-err: " fmt
#include <linux/interrupt.h>
#include <linux/of_irq.h>
#include <linux/platform_device.h>
#include <linux/vm_err.h>
#include <asm/traps.h>
#include <asm-generic/irq_regs.h>
#include <asm/system_misc.h>
#include <soc/tegra/virt/syscalls.h>
#include <soc/tegra/chip-id.h>
struct tegra_hv_err_ctrl {
struct device *dev;
struct err_info_t *err_info;
unsigned int async_err_arr_items;
int hv_peer_err_irq_id;
unsigned int vcpu_cnt;
struct serr_hook hook;
struct tegra_hv_vm_err_handlers handlers;
};
static struct tegra_hv_config config;
static unsigned int tegra_hv_virq_intr_info[3]; /* intr_property_size = 3 */
static struct property tegra_hv_virq_intr_prop = {
.name = "interrupts",
};
static bool check_sync_err(const unsigned int vcpu_id,
const struct tegra_hv_err_ctrl *const ctrl,
bool *send_sync_err_ack)
{
uint64_t rd_idx;
const struct err_data_t *err_data;
if (vcpu_id >= ctrl->vcpu_cnt) {
dev_crit(ctrl->dev, "%s: Invalid vcpu id %u\n", __func__,
vcpu_id);
*send_sync_err_ack = false;
/* Unexpected vcpu id. Enter bad mode. */
return true;
}
/* Shared memory layout is:
* |--async-err-metadata--|--async-errors-array-|--sync-errors-array-|
* Size of async errors array = Max errors + 1(to avoid same empty
* and full conditions of the buffer)
* Size of sync errors array = 1 error per VCPU * number of VCPUs in VM
*/
rd_idx = ctrl->async_err_arr_items + vcpu_id;
/* It's already validated at init time that sufficient memory is
* allocated to hold async_err_arr_items + sync error per vcpu. Hence,
* after validating the vcpu_id above, no need to validate rd_idx here.
*/
err_data = &(ctrl->err_info->err_data[rd_idx]);
if (err_data->err_reason != REASON_SYNC) {
dev_crit(ctrl->dev, "%s: unexpected reason id %u\n", __func__,
err_data->err_reason);
*send_sync_err_ack = true;
/* Invalid reason. Enter bad mode. */
return true;
}
if (err_data->err_type != SYNC) {
dev_crit(ctrl->dev, "%s: unexpected error Type %d\n",
__func__, err_data->err_type);
*send_sync_err_ack = true;
/* Unexpected error id. Enter bad mode. */
return true;
}
if (err_data->offending_guest_id != config.guest_id_self) {
dev_crit(ctrl->dev, "%s: invalid offender id %u\n", __func__,
err_data->offending_guest_id);
*send_sync_err_ack = true;
/* Invalid id of offending guest. Enter bad mode. */
return true;
}
dev_err(ctrl->dev, "Synchronous error on vcpu %u\n", vcpu_id);
if (ctrl->handlers.fn_self_sync) {
*send_sync_err_ack = true;
/* Enter bad_mode (or otherwise) as custom handler dictates */
return ctrl->handlers.fn_self_sync(err_data);
}
/* should never reach here */
*send_sync_err_ack = true;
/* Reaching here is unexpected. Enter bad mode. */
return true;
}
static irqreturn_t async_err_handler(int irq, void *context)
{
unsigned int num_async_errs_read = 0;
bool enter_bad_mode = false;
const struct tegra_hv_err_ctrl *const ctrl = context;
const unsigned int vcpu_id = hyp_read_vcpu_id();
uint64_t local_rd_idx, next_rd_idx;
const struct err_data_t *err_data;
bool (*fn_self_async)(const struct err_data_t *const err_data);
bool (*fn_peer)(const struct err_data_t *const err_data);
bool (*handler)(const struct err_data_t *const err_data);
struct pt_regs *regs;
if (vcpu_id != 0) {
dev_err(ctrl->dev, "Asynchronous error on vcpu %u\n", vcpu_id);
/* Only VCPU0 is expected to receive async error vIRQ */
return IRQ_HANDLED;
}
fn_self_async = ctrl->handlers.fn_self_async;
fn_peer = ctrl->handlers.fn_peer;
if ((fn_self_async == NULL) && (fn_peer == NULL)) {
dev_err(ctrl->dev, "Asynchronous error handlers absent\n");
return IRQ_HANDLED;
}
local_rd_idx = ctrl->err_info->async_metadata.rd_idx;
dev_dbg(ctrl->dev, "Local Rd Idx = %llu, shared Wr Idx = %llu\n",
local_rd_idx, ctrl->err_info->async_metadata.wr_idx);
/* Check async error. Read until error queue gets empty */
while (local_rd_idx != ctrl->err_info->async_metadata.wr_idx) {
next_rd_idx = (local_rd_idx + 1) % ctrl->async_err_arr_items;
err_data = &(ctrl->err_info->err_data[next_rd_idx]);
if (err_data->offending_guest_id == config.guest_id_self)
handler = fn_self_async;
else
handler = fn_peer;
if (handler) {
enter_bad_mode = handler(err_data);
}
local_rd_idx = next_rd_idx;
num_async_errs_read++;
dev_dbg(ctrl->dev, "Local Rd Idx = %llu\n", local_rd_idx);
}
if (num_async_errs_read) {
dev_err(ctrl->dev, "%u asynchronous error(s) read\n",
num_async_errs_read);
/* Send ack for async error(s) to HV */
if (hyp_send_async_err_ack(local_rd_idx) != 0) {
dev_crit(ctrl->dev,
"%s: Sending ack failed. Setting bad mode\n",
__func__);
/* Unexpected */
enter_bad_mode = true;
}
}
if (enter_bad_mode) {
regs = get_irq_regs();
die("Oops - bad mode", regs, 0);
panic("bad mode");
}
return IRQ_HANDLED;
}
static int sync_err_handler(struct pt_regs *regs, int reason,
uint32_t esr, void *context)
{
bool enter_bad_mode = false;
bool send_sync_err_ack = false;
const struct tegra_hv_err_ctrl *const ctrl = context;
const unsigned int vcpu_id = hyp_read_vcpu_id();
/* Check sync error */
enter_bad_mode = check_sync_err(vcpu_id, ctrl, &send_sync_err_ack);
/* Send ack for error to HV. */
if (send_sync_err_ack) {
if (hyp_send_sync_err_ack() != 0) {
dev_crit(ctrl->dev,
"%s: Sending ack failed. Setting bad mode\n",
__func__);
/* Unexpected */
enter_bad_mode = true;
}
}
/* bad_mode() will call die(). Force the latter to panic. */
if (enter_bad_mode) {
panic_on_oops = 1;
wmb();
}
/* Caller expects 0 to enter bad mode */
return (!enter_bad_mode);
}
void tegra_hv_get_config(struct tegra_hv_config *cfg)
{
cfg->guest_id_self = config.guest_id_self;
cfg->num_guests = config.num_guests;
}
EXPORT_SYMBOL(tegra_hv_get_config);
static int virq_handler_init(struct platform_device *pdev)
{
int ret;
struct irq_data *peer_err_irq_data;
int lin_peer_err_irq_id;
struct tegra_hv_err_ctrl *ctrl = platform_get_drvdata(pdev);
struct device *dev = &pdev->dev;
dev_info(ctrl->dev, "Error notification HV IRQ id: %d\n",
ctrl->hv_peer_err_irq_id);
/* Ensure HV returned valid irq */
if (ctrl->hv_peer_err_irq_id == -1)
return 0;
/* Set indicate irq type 0 to indicate Shared Peripheral Irq */
tegra_hv_virq_intr_info[0] = cpu_to_be32(0);
/* Id in SPI namespace - subtract number of PPIs
* (Private Peripheral Irqs) which is = 32
*/
tegra_hv_virq_intr_info[1] = cpu_to_be32(ctrl->hv_peer_err_irq_id - 32);
/* Trigger irq on low-to-high edge (0x1) */
tegra_hv_virq_intr_info[2] = cpu_to_be32(IRQF_TRIGGER_RISING);
tegra_hv_virq_intr_prop.length = sizeof(tegra_hv_virq_intr_info);
dev_info(ctrl->dev, "tegra_hv_virq_intr_prop.length %u\n",
tegra_hv_virq_intr_prop.length);
tegra_hv_virq_intr_prop.value = tegra_hv_virq_intr_info;
if (of_add_property(dev->of_node, &tegra_hv_virq_intr_prop)) {
dev_err(ctrl->dev, "%s: failed to add interrupts property\n",
__func__);
return -EACCES;
}
lin_peer_err_irq_id = of_irq_get(dev->of_node, 0);
if (lin_peer_err_irq_id < 0) {
dev_err(ctrl->dev, "%s: Unable to get Linux irq for id %d\n",
__func__, ctrl->hv_peer_err_irq_id);
return lin_peer_err_irq_id;
}
peer_err_irq_data = irq_get_irq_data(lin_peer_err_irq_id);
if (peer_err_irq_data == NULL) {
dev_err(ctrl->dev, "%s: Failed to get data for Linux irq %d\n",
__func__, lin_peer_err_irq_id);
return -ENODEV;
}
ret = devm_request_irq(dev, lin_peer_err_irq_id, async_err_handler,
IRQ_NOTHREAD, dev_name(dev), ctrl);
if (ret < 0) {
dev_err(ctrl->dev,
"%s: failed to register IRQ %d, Err %d, %s\n",
__func__, lin_peer_err_irq_id, ret, dev_name(dev));
return ret;
}
dev_info(ctrl->dev, "Registered Linux IRQ %d for peer notification\n",
lin_peer_err_irq_id);
return 0;
}
static int serr_handler_init(struct platform_device *pdev)
{
struct tegra_hv_err_ctrl *ctrl = platform_get_drvdata(pdev);
ctrl->hook.fn = sync_err_handler;
ctrl->hook.priv = platform_get_drvdata(pdev);
register_serr_hook(&ctrl->hook);
return 0;
}
static int shared_mem_map(struct platform_device *pdev)
{
uint64_t ipa, buff_size, required_size;
int ret;
struct tegra_hv_err_ctrl *ctrl = platform_get_drvdata(pdev);
/* Get error info details */
ret = hyp_read_err_info_get(&ipa, &buff_size,
&ctrl->async_err_arr_items, &ctrl->hv_peer_err_irq_id,
&ctrl->vcpu_cnt);
if (ret != 0) {
/* It could come here if DTS and defconfig enable execution
* of this code, but HV hasn't implemented the hypercall.
* Flag error.
*/
dev_err(ctrl->dev,
"%s: failed to get err memory address. Err %d\n",
__func__, ret);
return -ENODEV;
}
if ((ipa == 0) || (buff_size == 0) ||
(ctrl->async_err_arr_items == 0)) {
/* It could come here if DTS and defconfig enable execution
* of this code, but PCT hasn't enabled error injection.
* A warning should suffice.
*/
dev_warn(ctrl->dev, "%s: invalid shared memory parameters.\n",
__func__);
dev_warn(ctrl->dev,
"%s: make sure error injection is enabled in PCT\n",
__func__);
return -ENOMEM;
}
/* Shared memory layout is:
* |--async-err-metadata--|--async-errors-array-|--sync-errors-array-|
* Size of async errors array = Max errors + 1 (to avoid same empty and
* full conditions of the buffer)
* Size of sync errors array = 1 error per VCPU * number of VCPUs on
* a VM
*/
required_size = sizeof(struct async_metadata_t) +
(sizeof(struct err_data_t) *
(ctrl->async_err_arr_items + ctrl->vcpu_cnt));
if (buff_size < required_size) {
dev_err(ctrl->dev,
"%s:invalid params. size %llu. required size %llu\n",
__func__, buff_size, required_size);
dev_err(ctrl->dev, "%s: async arr size %u. vcpus %u\n",
__func__, ctrl->async_err_arr_items, ctrl->vcpu_cnt);
return -ENOMEM;
}
dev_info(ctrl->dev, "%s: Err info IPA for guest %u: 0x%llx\n",
__func__, config.guest_id_self, ipa);
dev_info(ctrl->dev, "Err info buf size 0x%llX\n", buff_size);
dev_info(ctrl->dev, "Async err arr size %u. Number of VCPUs %u\n",
ctrl->async_err_arr_items, ctrl->vcpu_cnt);
/* Map shared memory */
ctrl->err_info = (struct err_info_t *) ioremap_cache(ipa, buff_size);
if (ctrl->err_info == NULL)
return -ENOMEM;
return 0;
}
static int hyp_config_init(struct device *dev)
{
int ret = hyp_read_gid(&config.guest_id_self);
if (ret != 0) {
dev_err(dev, "%s: failed to read guest id. Err %d\n",
__func__, ret);
return ret;
}
ret = hyp_read_nguests(&config.num_guests);
if (ret != 0) {
/* Only privileged guest can query number of guests */
dev_warn(dev, "%s: can't read number of guests. Err %d\n",
__func__, ret);
}
dev_info(dev, "%s: guest id %u num guests %u\n", __func__,
config.guest_id_self, config.num_guests);
return 0;
}
static void shared_structs_check(struct device *dev)
{
/* Ensure coherency with common header */
BUILD_BUG_ON(REASON_ENUM_SIZE !=
(ARRAY_SIZE(tegra_hv_err_reason_desc)));
/* Dump shared structures sizes. This is needed to manually compare
* with HV console dump, if a build bug fires below.
*/
dev_info(dev, "async_metadata size 0x%lx\n",
sizeof(struct async_metadata_t));
dev_info(dev, "async_bridge_err size 0x%lx\n",
sizeof(struct async_bridge_err_t));
dev_info(dev, "async_cbb_err size 0x%lx\n",
sizeof(struct async_cbb_err_t));
dev_info(dev, "async_smmu_err size 0x%lx\n",
sizeof(struct async_smmu_err_t));
dev_info(dev, "async_mc_err size 0x%lx\n",
sizeof(struct async_mc_err_t));
dev_info(dev, "async_mc_err_t19x size 0x%lx\n",
sizeof(struct async_mc_err_t19x_t));
dev_info(dev, "sync size 0x%lx\n",
sizeof(struct sync_t));
dev_info(dev, "err_data size 0x%lx\n", sizeof(struct err_data_t));
/* Ensure common structures shared by HV and Linux are in sync */
BUILD_BUG_ON(sizeof(struct async_metadata_t) != 0x10);
BUILD_BUG_ON(sizeof(struct async_bridge_err_t) != 0x74);
BUILD_BUG_ON(sizeof(struct async_cbb_err_t) != 0x259);
BUILD_BUG_ON(sizeof(struct async_smmu_err_t) != 0x1C);
BUILD_BUG_ON(sizeof(struct async_mc_err_t) != 0x24);
BUILD_BUG_ON(sizeof(struct async_mc_err_t19x_t) != 0xBB);
BUILD_BUG_ON(sizeof(struct sync_t) != 0x149);
BUILD_BUG_ON(sizeof(struct err_data_t) != 0x265);
}
static int vm_err_handler_init(struct platform_device *pdev)
{
int ret;
struct tegra_hv_err_ctrl *ctrl;
struct device *dev = &pdev->dev;
if (!is_tegra_hypervisor_mode()) {
dev_err(dev, "%s: hypervisor is not present\n", __func__);
return -ENODEV;
}
shared_structs_check(dev);
ctrl = devm_kzalloc(dev, sizeof(*ctrl), GFP_KERNEL);
if (!ctrl)
return -ENOMEM;
ctrl->dev = dev;
platform_set_drvdata(pdev, ctrl);
ret = hyp_config_init(dev);
if (ret)
return ret;
ret = shared_mem_map(pdev);
if (ret)
return -ENOMEM;
ret = serr_handler_init(pdev);
if (ret)
return ret;
ret = virq_handler_init(pdev);
if (ret)
return ret;
return 0;
}
static int vm_err_handler_remove(struct platform_device *pdev)
{
struct tegra_hv_err_ctrl *ctrl = platform_get_drvdata(pdev);
struct device_node *node = pdev->dev.of_node;
if (of_remove_property(node,
of_find_property(node, "interrupts", NULL))) {
dev_err(ctrl->dev, "%s: failed to add interrupts property\n",
__func__);
return -EACCES;
}
unregister_serr_hook(&ctrl->hook);
iounmap(ctrl->err_info);
dev_info(ctrl->dev, "%s: cleaned up and unregistered handler\n",
__func__);
return 0;
}
static const struct of_device_id tegra_hv_err_match[] = {
{ .compatible = "nvidia,tegra-hv-err", .data = NULL},
{},
};
static struct platform_driver tegra_hv_err_pdriver = {
.driver = {
.name = "tegra-hv-err-handler",
.owner = THIS_MODULE,
.of_match_table = of_match_ptr(tegra_hv_err_match),
},
.probe = vm_err_handler_init,
.remove = vm_err_handler_remove,
};
static int tegra_hv_register_hooks_for_device(struct device *dev,
void *handlers)
{
struct tegra_hv_err_ctrl *ctrl;
const struct platform_device *pd = container_of(dev,
struct platform_device, dev);
const struct tegra_hv_vm_err_handlers *_handlers =
(struct tegra_hv_vm_err_handlers *) handlers;
ctrl = platform_get_drvdata(pd);
if (!ctrl) {
dev_err(dev, "%s: no platform data", __func__);
return 0;
}
if (ctrl->handlers.fn_self_async == NULL)
ctrl->handlers.fn_self_async = _handlers->fn_self_async;
if (ctrl->handlers.fn_self_sync == NULL)
ctrl->handlers.fn_self_sync = _handlers->fn_self_sync;
if (ctrl->handlers.fn_peer == NULL)
ctrl->handlers.fn_peer = _handlers->fn_peer;
return 0;
}
int tegra_hv_register_vm_err_hooks(struct tegra_hv_vm_err_handlers *handlers)
{
int ret;
if (!handlers) {
pr_err("%s: invalid error handlers\n", __func__);
return 1;
}
if (!handlers->fn_self_async && !handlers->fn_self_sync
&& !handlers->fn_peer) {
platform_driver_unregister(&tegra_hv_err_pdriver);
return 0;
}
if (!tegra_hv_err_pdriver.driver.p) {
/* Not registered/bound yet */
ret = platform_driver_register(&tegra_hv_err_pdriver);
if (ret) {
pr_err("%s: failed to register driver. Err %d\n",
__func__, ret);
return ret;
}
}
ret = driver_for_each_device(&tegra_hv_err_pdriver.driver, NULL,
handlers, tegra_hv_register_hooks_for_device);
if (ret) {
pr_err("%s: failed to attach driver. Err %d\n", __func__, ret);
return ret;
}
return 0;
}
EXPORT_SYMBOL(tegra_hv_register_vm_err_hooks);