From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- drivers/crypto/ccp/ccp-dev.c | 670 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 670 insertions(+) create mode 100644 drivers/crypto/ccp/ccp-dev.c (limited to 'drivers/crypto/ccp/ccp-dev.c') diff --git a/drivers/crypto/ccp/ccp-dev.c b/drivers/crypto/ccp/ccp-dev.c new file mode 100644 index 000000000..246801912 --- /dev/null +++ b/drivers/crypto/ccp/ccp-dev.c @@ -0,0 +1,670 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * AMD Cryptographic Coprocessor (CCP) driver + * + * Copyright (C) 2013,2019 Advanced Micro Devices, Inc. + * + * Author: Tom Lendacky + * Author: Gary R Hook + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#ifdef CONFIG_X86 +#include +#endif +#include + +#include "ccp-dev.h" + +#define MAX_CCPS 32 + +/* Limit CCP use to a specifed number of queues per device */ +static unsigned int nqueues; +module_param(nqueues, uint, 0444); +MODULE_PARM_DESC(nqueues, "Number of queues per CCP (minimum 1; default: all available)"); + +/* Limit the maximum number of configured CCPs */ +static atomic_t dev_count = ATOMIC_INIT(0); +static unsigned int max_devs = MAX_CCPS; +module_param(max_devs, uint, 0444); +MODULE_PARM_DESC(max_devs, "Maximum number of CCPs to enable (default: all; 0 disables all CCPs)"); + +struct ccp_tasklet_data { + struct completion completion; + struct ccp_cmd *cmd; +}; + +/* Human-readable error strings */ +#define CCP_MAX_ERROR_CODE 64 +static char *ccp_error_codes[] = { + "", + "ILLEGAL_ENGINE", + "ILLEGAL_KEY_ID", + "ILLEGAL_FUNCTION_TYPE", + "ILLEGAL_FUNCTION_MODE", + "ILLEGAL_FUNCTION_ENCRYPT", + "ILLEGAL_FUNCTION_SIZE", + "Zlib_MISSING_INIT_EOM", + "ILLEGAL_FUNCTION_RSVD", + "ILLEGAL_BUFFER_LENGTH", + "VLSB_FAULT", + "ILLEGAL_MEM_ADDR", + "ILLEGAL_MEM_SEL", + "ILLEGAL_CONTEXT_ID", + "ILLEGAL_KEY_ADDR", + "0xF Reserved", + "Zlib_ILLEGAL_MULTI_QUEUE", + "Zlib_ILLEGAL_JOBID_CHANGE", + "CMD_TIMEOUT", + "IDMA0_AXI_SLVERR", + "IDMA0_AXI_DECERR", + "0x15 Reserved", + "IDMA1_AXI_SLAVE_FAULT", + "IDMA1_AIXI_DECERR", + "0x18 Reserved", + "ZLIBVHB_AXI_SLVERR", + "ZLIBVHB_AXI_DECERR", + "0x1B Reserved", + "ZLIB_UNEXPECTED_EOM", + "ZLIB_EXTRA_DATA", + "ZLIB_BTYPE", + "ZLIB_UNDEFINED_SYMBOL", + "ZLIB_UNDEFINED_DISTANCE_S", + "ZLIB_CODE_LENGTH_SYMBOL", + "ZLIB _VHB_ILLEGAL_FETCH", + "ZLIB_UNCOMPRESSED_LEN", + "ZLIB_LIMIT_REACHED", + "ZLIB_CHECKSUM_MISMATCH0", + "ODMA0_AXI_SLVERR", + "ODMA0_AXI_DECERR", + "0x28 Reserved", + "ODMA1_AXI_SLVERR", + "ODMA1_AXI_DECERR", +}; + +void ccp_log_error(struct ccp_device *d, unsigned int e) +{ + if (WARN_ON(e >= CCP_MAX_ERROR_CODE)) + return; + + if (e < ARRAY_SIZE(ccp_error_codes)) + dev_err(d->dev, "CCP error %d: %s\n", e, ccp_error_codes[e]); + else + dev_err(d->dev, "CCP error %d: Unknown Error\n", e); +} + +/* List of CCPs, CCP count, read-write access lock, and access functions + * + * Lock structure: get ccp_unit_lock for reading whenever we need to + * examine the CCP list. While holding it for reading we can acquire + * the RR lock to update the round-robin next-CCP pointer. The unit lock + * must be acquired before the RR lock. + * + * If the unit-lock is acquired for writing, we have total control over + * the list, so there's no value in getting the RR lock. + */ +static DEFINE_RWLOCK(ccp_unit_lock); +static LIST_HEAD(ccp_units); + +/* Round-robin counter */ +static DEFINE_SPINLOCK(ccp_rr_lock); +static struct ccp_device *ccp_rr; + +/** + * ccp_add_device - add a CCP device to the list + * + * @ccp: ccp_device struct pointer + * + * Put this CCP on the unit list, which makes it available + * for use. + * + * Returns zero if a CCP device is present, -ENODEV otherwise. + */ +void ccp_add_device(struct ccp_device *ccp) +{ + unsigned long flags; + + write_lock_irqsave(&ccp_unit_lock, flags); + list_add_tail(&ccp->entry, &ccp_units); + if (!ccp_rr) + /* We already have the list lock (we're first) so this + * pointer can't change on us. Set its initial value. + */ + ccp_rr = ccp; + write_unlock_irqrestore(&ccp_unit_lock, flags); +} + +/** + * ccp_del_device - remove a CCP device from the list + * + * @ccp: ccp_device struct pointer + * + * Remove this unit from the list of devices. If the next device + * up for use is this one, adjust the pointer. If this is the last + * device, NULL the pointer. + */ +void ccp_del_device(struct ccp_device *ccp) +{ + unsigned long flags; + + write_lock_irqsave(&ccp_unit_lock, flags); + if (ccp_rr == ccp) { + /* ccp_unit_lock is read/write; any read access + * will be suspended while we make changes to the + * list and RR pointer. + */ + if (list_is_last(&ccp_rr->entry, &ccp_units)) + ccp_rr = list_first_entry(&ccp_units, struct ccp_device, + entry); + else + ccp_rr = list_next_entry(ccp_rr, entry); + } + list_del(&ccp->entry); + if (list_empty(&ccp_units)) + ccp_rr = NULL; + write_unlock_irqrestore(&ccp_unit_lock, flags); +} + + + +int ccp_register_rng(struct ccp_device *ccp) +{ + int ret = 0; + + dev_dbg(ccp->dev, "Registering RNG...\n"); + /* Register an RNG */ + ccp->hwrng.name = ccp->rngname; + ccp->hwrng.read = ccp_trng_read; + ret = hwrng_register(&ccp->hwrng); + if (ret) + dev_err(ccp->dev, "error registering hwrng (%d)\n", ret); + + return ret; +} + +void ccp_unregister_rng(struct ccp_device *ccp) +{ + if (ccp->hwrng.name) + hwrng_unregister(&ccp->hwrng); +} + +static struct ccp_device *ccp_get_device(void) +{ + unsigned long flags; + struct ccp_device *dp = NULL; + + /* We round-robin through the unit list. + * The (ccp_rr) pointer refers to the next unit to use. + */ + read_lock_irqsave(&ccp_unit_lock, flags); + if (!list_empty(&ccp_units)) { + spin_lock(&ccp_rr_lock); + dp = ccp_rr; + if (list_is_last(&ccp_rr->entry, &ccp_units)) + ccp_rr = list_first_entry(&ccp_units, struct ccp_device, + entry); + else + ccp_rr = list_next_entry(ccp_rr, entry); + spin_unlock(&ccp_rr_lock); + } + read_unlock_irqrestore(&ccp_unit_lock, flags); + + return dp; +} + +/** + * ccp_present - check if a CCP device is present + * + * Returns zero if a CCP device is present, -ENODEV otherwise. + */ +int ccp_present(void) +{ + unsigned long flags; + int ret; + + read_lock_irqsave(&ccp_unit_lock, flags); + ret = list_empty(&ccp_units); + read_unlock_irqrestore(&ccp_unit_lock, flags); + + return ret ? -ENODEV : 0; +} +EXPORT_SYMBOL_GPL(ccp_present); + +/** + * ccp_version - get the version of the CCP device + * + * Returns the version from the first unit on the list; + * otherwise a zero if no CCP device is present + */ +unsigned int ccp_version(void) +{ + struct ccp_device *dp; + unsigned long flags; + int ret = 0; + + read_lock_irqsave(&ccp_unit_lock, flags); + if (!list_empty(&ccp_units)) { + dp = list_first_entry(&ccp_units, struct ccp_device, entry); + ret = dp->vdata->version; + } + read_unlock_irqrestore(&ccp_unit_lock, flags); + + return ret; +} +EXPORT_SYMBOL_GPL(ccp_version); + +/** + * ccp_enqueue_cmd - queue an operation for processing by the CCP + * + * @cmd: ccp_cmd struct to be processed + * + * Queue a cmd to be processed by the CCP. If queueing the cmd + * would exceed the defined length of the cmd queue the cmd will + * only be queued if the CCP_CMD_MAY_BACKLOG flag is set and will + * result in a return code of -EBUSY. + * + * The callback routine specified in the ccp_cmd struct will be + * called to notify the caller of completion (if the cmd was not + * backlogged) or advancement out of the backlog. If the cmd has + * advanced out of the backlog the "err" value of the callback + * will be -EINPROGRESS. Any other "err" value during callback is + * the result of the operation. + * + * The cmd has been successfully queued if: + * the return code is -EINPROGRESS or + * the return code is -EBUSY and CCP_CMD_MAY_BACKLOG flag is set + */ +int ccp_enqueue_cmd(struct ccp_cmd *cmd) +{ + struct ccp_device *ccp; + unsigned long flags; + unsigned int i; + int ret; + + /* Some commands might need to be sent to a specific device */ + ccp = cmd->ccp ? cmd->ccp : ccp_get_device(); + + if (!ccp) + return -ENODEV; + + /* Caller must supply a callback routine */ + if (!cmd->callback) + return -EINVAL; + + cmd->ccp = ccp; + + spin_lock_irqsave(&ccp->cmd_lock, flags); + + i = ccp->cmd_q_count; + + if (ccp->cmd_count >= MAX_CMD_QLEN) { + if (cmd->flags & CCP_CMD_MAY_BACKLOG) { + ret = -EBUSY; + list_add_tail(&cmd->entry, &ccp->backlog); + } else { + ret = -ENOSPC; + } + } else { + ret = -EINPROGRESS; + ccp->cmd_count++; + list_add_tail(&cmd->entry, &ccp->cmd); + + /* Find an idle queue */ + if (!ccp->suspending) { + for (i = 0; i < ccp->cmd_q_count; i++) { + if (ccp->cmd_q[i].active) + continue; + + break; + } + } + } + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); + + /* If we found an idle queue, wake it up */ + if (i < ccp->cmd_q_count) + wake_up_process(ccp->cmd_q[i].kthread); + + return ret; +} +EXPORT_SYMBOL_GPL(ccp_enqueue_cmd); + +static void ccp_do_cmd_backlog(struct work_struct *work) +{ + struct ccp_cmd *cmd = container_of(work, struct ccp_cmd, work); + struct ccp_device *ccp = cmd->ccp; + unsigned long flags; + unsigned int i; + + cmd->callback(cmd->data, -EINPROGRESS); + + spin_lock_irqsave(&ccp->cmd_lock, flags); + + ccp->cmd_count++; + list_add_tail(&cmd->entry, &ccp->cmd); + + /* Find an idle queue */ + for (i = 0; i < ccp->cmd_q_count; i++) { + if (ccp->cmd_q[i].active) + continue; + + break; + } + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); + + /* If we found an idle queue, wake it up */ + if (i < ccp->cmd_q_count) + wake_up_process(ccp->cmd_q[i].kthread); +} + +static struct ccp_cmd *ccp_dequeue_cmd(struct ccp_cmd_queue *cmd_q) +{ + struct ccp_device *ccp = cmd_q->ccp; + struct ccp_cmd *cmd = NULL; + struct ccp_cmd *backlog = NULL; + unsigned long flags; + + spin_lock_irqsave(&ccp->cmd_lock, flags); + + cmd_q->active = 0; + + if (ccp->suspending) { + cmd_q->suspended = 1; + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); + wake_up_interruptible(&ccp->suspend_queue); + + return NULL; + } + + if (ccp->cmd_count) { + cmd_q->active = 1; + + cmd = list_first_entry(&ccp->cmd, struct ccp_cmd, entry); + list_del(&cmd->entry); + + ccp->cmd_count--; + } + + if (!list_empty(&ccp->backlog)) { + backlog = list_first_entry(&ccp->backlog, struct ccp_cmd, + entry); + list_del(&backlog->entry); + } + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); + + if (backlog) { + INIT_WORK(&backlog->work, ccp_do_cmd_backlog); + schedule_work(&backlog->work); + } + + return cmd; +} + +static void ccp_do_cmd_complete(unsigned long data) +{ + struct ccp_tasklet_data *tdata = (struct ccp_tasklet_data *)data; + struct ccp_cmd *cmd = tdata->cmd; + + cmd->callback(cmd->data, cmd->ret); + + complete(&tdata->completion); +} + +/** + * ccp_cmd_queue_thread - create a kernel thread to manage a CCP queue + * + * @data: thread-specific data + */ +int ccp_cmd_queue_thread(void *data) +{ + struct ccp_cmd_queue *cmd_q = (struct ccp_cmd_queue *)data; + struct ccp_cmd *cmd; + struct ccp_tasklet_data tdata; + struct tasklet_struct tasklet; + + tasklet_init(&tasklet, ccp_do_cmd_complete, (unsigned long)&tdata); + + set_current_state(TASK_INTERRUPTIBLE); + while (!kthread_should_stop()) { + schedule(); + + set_current_state(TASK_INTERRUPTIBLE); + + cmd = ccp_dequeue_cmd(cmd_q); + if (!cmd) + continue; + + __set_current_state(TASK_RUNNING); + + /* Execute the command */ + cmd->ret = ccp_run_cmd(cmd_q, cmd); + + /* Schedule the completion callback */ + tdata.cmd = cmd; + init_completion(&tdata.completion); + tasklet_schedule(&tasklet); + wait_for_completion(&tdata.completion); + } + + __set_current_state(TASK_RUNNING); + + return 0; +} + +/** + * ccp_alloc_struct - allocate and initialize the ccp_device struct + * + * @sp: sp_device struct of the CCP + */ +struct ccp_device *ccp_alloc_struct(struct sp_device *sp) +{ + struct device *dev = sp->dev; + struct ccp_device *ccp; + + ccp = devm_kzalloc(dev, sizeof(*ccp), GFP_KERNEL); + if (!ccp) + return NULL; + ccp->dev = dev; + ccp->sp = sp; + ccp->axcache = sp->axcache; + + INIT_LIST_HEAD(&ccp->cmd); + INIT_LIST_HEAD(&ccp->backlog); + + spin_lock_init(&ccp->cmd_lock); + mutex_init(&ccp->req_mutex); + mutex_init(&ccp->sb_mutex); + ccp->sb_count = KSB_COUNT; + ccp->sb_start = 0; + + /* Initialize the wait queues */ + init_waitqueue_head(&ccp->sb_queue); + init_waitqueue_head(&ccp->suspend_queue); + + snprintf(ccp->name, MAX_CCP_NAME_LEN, "ccp-%u", sp->ord); + snprintf(ccp->rngname, MAX_CCP_NAME_LEN, "ccp-%u-rng", sp->ord); + + return ccp; +} + +int ccp_trng_read(struct hwrng *rng, void *data, size_t max, bool wait) +{ + struct ccp_device *ccp = container_of(rng, struct ccp_device, hwrng); + u32 trng_value; + int len = min(sizeof(trng_value), max); + + /* Locking is provided by the caller so we can update device + * hwrng-related fields safely + */ + trng_value = ioread32(ccp->io_regs + TRNG_OUT_REG); + if (!trng_value) { + /* Zero is returned if not data is available or if a + * bad-entropy error is present. Assume an error if + * we exceed TRNG_RETRIES reads of zero. + */ + if (ccp->hwrng_retries++ > TRNG_RETRIES) + return -EIO; + + return 0; + } + + /* Reset the counter and save the rng value */ + ccp->hwrng_retries = 0; + memcpy(data, &trng_value, len); + + return len; +} + +bool ccp_queues_suspended(struct ccp_device *ccp) +{ + unsigned int suspended = 0; + unsigned long flags; + unsigned int i; + + spin_lock_irqsave(&ccp->cmd_lock, flags); + + for (i = 0; i < ccp->cmd_q_count; i++) + if (ccp->cmd_q[i].suspended) + suspended++; + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); + + return ccp->cmd_q_count == suspended; +} + +void ccp_dev_suspend(struct sp_device *sp) +{ + struct ccp_device *ccp = sp->ccp_data; + unsigned long flags; + unsigned int i; + + /* If there's no device there's nothing to do */ + if (!ccp) + return; + + spin_lock_irqsave(&ccp->cmd_lock, flags); + + ccp->suspending = 1; + + /* Wake all the queue kthreads to prepare for suspend */ + for (i = 0; i < ccp->cmd_q_count; i++) + wake_up_process(ccp->cmd_q[i].kthread); + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); + + /* Wait for all queue kthreads to say they're done */ + while (!ccp_queues_suspended(ccp)) + wait_event_interruptible(ccp->suspend_queue, + ccp_queues_suspended(ccp)); +} + +void ccp_dev_resume(struct sp_device *sp) +{ + struct ccp_device *ccp = sp->ccp_data; + unsigned long flags; + unsigned int i; + + /* If there's no device there's nothing to do */ + if (!ccp) + return; + + spin_lock_irqsave(&ccp->cmd_lock, flags); + + ccp->suspending = 0; + + /* Wake up all the kthreads */ + for (i = 0; i < ccp->cmd_q_count; i++) { + ccp->cmd_q[i].suspended = 0; + wake_up_process(ccp->cmd_q[i].kthread); + } + + spin_unlock_irqrestore(&ccp->cmd_lock, flags); +} + +int ccp_dev_init(struct sp_device *sp) +{ + struct device *dev = sp->dev; + struct ccp_device *ccp; + int ret; + + /* + * Check how many we have so far, and stop after reaching + * that number + */ + if (atomic_inc_return(&dev_count) > max_devs) + return 0; /* don't fail the load */ + + ret = -ENOMEM; + ccp = ccp_alloc_struct(sp); + if (!ccp) + goto e_err; + sp->ccp_data = ccp; + + if (!nqueues || (nqueues > MAX_HW_QUEUES)) + ccp->max_q_count = MAX_HW_QUEUES; + else + ccp->max_q_count = nqueues; + + ccp->vdata = (struct ccp_vdata *)sp->dev_vdata->ccp_vdata; + if (!ccp->vdata || !ccp->vdata->version) { + ret = -ENODEV; + dev_err(dev, "missing driver data\n"); + goto e_err; + } + + ccp->use_tasklet = sp->use_tasklet; + + ccp->io_regs = sp->io_map + ccp->vdata->offset; + if (ccp->vdata->setup) + ccp->vdata->setup(ccp); + + ret = ccp->vdata->perform->init(ccp); + if (ret) { + /* A positive number means that the device cannot be initialized, + * but no additional message is required. + */ + if (ret > 0) + goto e_quiet; + + /* An unexpected problem occurred, and should be reported in the log */ + goto e_err; + } + + dev_notice(dev, "ccp enabled\n"); + + return 0; + +e_err: + dev_notice(dev, "ccp initialization failed\n"); + +e_quiet: + sp->ccp_data = NULL; + + return ret; +} + +void ccp_dev_destroy(struct sp_device *sp) +{ + struct ccp_device *ccp = sp->ccp_data; + + if (!ccp) + return; + + ccp->vdata->perform->destroy(ccp); +} -- cgit v1.3.1