x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

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) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 456 insertions, 0 deletions

diff --git a/drivers/scsi/aacraid/dpcsup.c b/drivers/scsi/aacraid/dpcsup.c
new file mode 100644
index 000000000..169c41d08
--- /dev/null
+++ b/drivers/scsi/aacraid/dpcsup.c
@@ -0,0 +1,456 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *	Adaptec AAC series RAID controller driver
+ *	(c) Copyright 2001 Red Hat Inc.
+ *
+ * based on the old aacraid driver that is..
+ * Adaptec aacraid device driver for Linux.
+ *
+ * Copyright (c) 2000-2010 Adaptec, Inc.
+ *               2010-2015 PMC-Sierra, Inc. (aacraid@pmc-sierra.com)
+ *		 2016-2017 Microsemi Corp. (aacraid@microsemi.com)
+ *
+ * Module Name:
+ *  dpcsup.c
+ *
+ * Abstract: All DPC processing routines for the cyclone board occur here.
+ */
+
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/types.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/completion.h>
+#include <linux/blkdev.h>
+
+#include "aacraid.h"
+
+/**
+ *	aac_response_normal	-	Handle command replies
+ *	@q: Queue to read from
+ *
+ *	This DPC routine will be run when the adapter interrupts us to let us
+ *	know there is a response on our normal priority queue. We will pull off
+ *	all QE there are and wake up all the waiters before exiting. We will
+ *	take a spinlock out on the queue before operating on it.
+ */
+
+unsigned int aac_response_normal(struct aac_queue * q)
+{
+	struct aac_dev * dev = q->dev;
+	struct aac_entry *entry;
+	struct hw_fib * hwfib;
+	struct fib * fib;
+	int consumed = 0;
+	unsigned long flags, mflags;
+
+	spin_lock_irqsave(q->lock, flags);
+	/*
+	 *	Keep pulling response QEs off the response queue and waking
+	 *	up the waiters until there are no more QEs. We then return
+	 *	back to the system. If no response was requested we just
+	 *	deallocate the Fib here and continue.
+	 */
+	while(aac_consumer_get(dev, q, &entry))
+	{
+		int fast;
+		u32 index = le32_to_cpu(entry->addr);
+		fast = index & 0x01;
+		fib = &dev->fibs[index >> 2];
+		hwfib = fib->hw_fib_va;
+		
+		aac_consumer_free(dev, q, HostNormRespQueue);
+		/*
+		 *	Remove this fib from the Outstanding I/O queue.
+		 *	But only if it has not already been timed out.
+		 *
+		 *	If the fib has been timed out already, then just 
+		 *	continue. The caller has already been notified that
+		 *	the fib timed out.
+		 */
+		atomic_dec(&dev->queues->queue[AdapNormCmdQueue].numpending);
+
+		if (unlikely(fib->flags & FIB_CONTEXT_FLAG_TIMED_OUT)) {
+			spin_unlock_irqrestore(q->lock, flags);
+			aac_fib_complete(fib);
+			aac_fib_free(fib);
+			spin_lock_irqsave(q->lock, flags);
+			continue;
+		}
+		spin_unlock_irqrestore(q->lock, flags);
+
+		if (fast) {
+			/*
+			 *	Doctor the fib
+			 */
+			*(__le32 *)hwfib->data = cpu_to_le32(ST_OK);
+			hwfib->header.XferState |= cpu_to_le32(AdapterProcessed);
+			fib->flags |= FIB_CONTEXT_FLAG_FASTRESP;
+		}
+
+		FIB_COUNTER_INCREMENT(aac_config.FibRecved);
+
+		if (hwfib->header.Command == cpu_to_le16(NuFileSystem))
+		{
+			__le32 *pstatus = (__le32 *)hwfib->data;
+			if (*pstatus & cpu_to_le32(0xffff0000))
+				*pstatus = cpu_to_le32(ST_OK);
+		}
+		if (hwfib->header.XferState & cpu_to_le32(NoResponseExpected | Async)) 
+		{
+			if (hwfib->header.XferState & cpu_to_le32(NoResponseExpected)) {
+				FIB_COUNTER_INCREMENT(aac_config.NoResponseRecved);
+			} else {
+				FIB_COUNTER_INCREMENT(aac_config.AsyncRecved);
+			}
+			/*
+			 *	NOTE:  we cannot touch the fib after this
+			 *	    call, because it may have been deallocated.
+			 */
+			fib->callback(fib->callback_data, fib);
+		} else {
+			unsigned long flagv;
+			spin_lock_irqsave(&fib->event_lock, flagv);
+			if (!fib->done) {
+				fib->done = 1;
+				complete(&fib->event_wait);
+			}
+			spin_unlock_irqrestore(&fib->event_lock, flagv);
+
+			spin_lock_irqsave(&dev->manage_lock, mflags);
+			dev->management_fib_count--;
+			spin_unlock_irqrestore(&dev->manage_lock, mflags);
+
+			FIB_COUNTER_INCREMENT(aac_config.NormalRecved);
+			if (fib->done == 2) {
+				spin_lock_irqsave(&fib->event_lock, flagv);
+				fib->done = 0;
+				spin_unlock_irqrestore(&fib->event_lock, flagv);
+				aac_fib_complete(fib);
+				aac_fib_free(fib);
+			}
+		}
+		consumed++;
+		spin_lock_irqsave(q->lock, flags);
+	}
+
+	if (consumed > aac_config.peak_fibs)
+		aac_config.peak_fibs = consumed;
+	if (consumed == 0) 
+		aac_config.zero_fibs++;
+
+	spin_unlock_irqrestore(q->lock, flags);
+	return 0;
+}
+
+
+/**
+ *	aac_command_normal	-	handle commands
+ *	@q: queue to process
+ *
+ *	This DPC routine will be queued when the adapter interrupts us to 
+ *	let us know there is a command on our normal priority queue. We will 
+ *	pull off all QE there are and wake up all the waiters before exiting.
+ *	We will take a spinlock out on the queue before operating on it.
+ */
+ 
+unsigned int aac_command_normal(struct aac_queue *q)
+{
+	struct aac_dev * dev = q->dev;
+	struct aac_entry *entry;
+	unsigned long flags;
+
+	spin_lock_irqsave(q->lock, flags);
+
+	/*
+	 *	Keep pulling response QEs off the response queue and waking
+	 *	up the waiters until there are no more QEs. We then return
+	 *	back to the system.
+	 */
+	while(aac_consumer_get(dev, q, &entry))
+	{
+		struct fib fibctx;
+		struct hw_fib * hw_fib;
+		u32 index;
+		struct fib *fib = &fibctx;
+		
+		index = le32_to_cpu(entry->addr) / sizeof(struct hw_fib);
+		hw_fib = &dev->aif_base_va[index];
+		
+		/*
+		 *	Allocate a FIB at all costs. For non queued stuff
+		 *	we can just use the stack so we are happy. We need
+		 *	a fib object in order to manage the linked lists
+		 */
+		if (dev->aif_thread)
+			if((fib = kmalloc_obj(struct fib, GFP_ATOMIC)) == NULL)
+				fib = &fibctx;
+		
+		memset(fib, 0, sizeof(struct fib));
+		INIT_LIST_HEAD(&fib->fiblink);
+		fib->type = FSAFS_NTC_FIB_CONTEXT;
+		fib->size = sizeof(struct fib);
+		fib->hw_fib_va = hw_fib;
+		fib->data = hw_fib->data;
+		fib->dev = dev;
+		
+				
+		if (dev->aif_thread && fib != &fibctx) {
+		        list_add_tail(&fib->fiblink, &q->cmdq);
+	 	        aac_consumer_free(dev, q, HostNormCmdQueue);
+		        wake_up_interruptible(&q->cmdready);
+		} else {
+	 	        aac_consumer_free(dev, q, HostNormCmdQueue);
+			spin_unlock_irqrestore(q->lock, flags);
+			/*
+			 *	Set the status of this FIB
+			 */
+			*(__le32 *)hw_fib->data = cpu_to_le32(ST_OK);
+			aac_fib_adapter_complete(fib, sizeof(u32));
+			spin_lock_irqsave(q->lock, flags);
+		}		
+	}
+	spin_unlock_irqrestore(q->lock, flags);
+	return 0;
+}
+
+/*
+ *
+ * aac_aif_callback
+ * @context: the context set in the fib - here it is scsi cmd
+ * @fibptr: pointer to the fib
+ *
+ * Handles the AIFs - new method (SRC)
+ *
+ */
+
+static void aac_aif_callback(void *context, struct fib * fibptr)
+{
+	struct fib *fibctx;
+	struct aac_dev *dev;
+	struct aac_aifcmd *cmd;
+
+	fibctx = (struct fib *)context;
+	BUG_ON(fibptr == NULL);
+	dev = fibptr->dev;
+
+	if ((fibptr->hw_fib_va->header.XferState &
+	    cpu_to_le32(NoMoreAifDataAvailable)) ||
+		dev->sa_firmware) {
+		aac_fib_complete(fibptr);
+		aac_fib_free(fibptr);
+		return;
+	}
+
+	aac_intr_normal(dev, 0, 1, 0, fibptr->hw_fib_va);
+
+	aac_fib_init(fibctx);
+	cmd = (struct aac_aifcmd *) fib_data(fibctx);
+	cmd->command = cpu_to_le32(AifReqEvent);
+
+	aac_fib_send(AifRequest,
+		fibctx,
+		sizeof(struct hw_fib)-sizeof(struct aac_fibhdr),
+		FsaNormal,
+		0, 1,
+		(fib_callback)aac_aif_callback, fibctx);
+}
+
+
+/*
+ *	aac_intr_normal	-	Handle command replies
+ *	@dev: Device
+ *	@index: completion reference
+ *
+ *	This DPC routine will be run when the adapter interrupts us to let us
+ *	know there is a response on our normal priority queue. We will pull off
+ *	all QE there are and wake up all the waiters before exiting.
+ */
+unsigned int aac_intr_normal(struct aac_dev *dev, u32 index, int isAif,
+	int isFastResponse, struct hw_fib *aif_fib)
+{
+	unsigned long mflags;
+	dprintk((KERN_INFO "aac_intr_normal(%p,%x)\n", dev, index));
+	if (isAif == 1) {	/* AIF - common */
+		struct hw_fib * hw_fib;
+		struct fib * fib;
+		struct aac_queue *q = &dev->queues->queue[HostNormCmdQueue];
+		unsigned long flags;
+
+		/*
+		 *	Allocate a FIB. For non queued stuff we can just use
+		 * the stack so we are happy. We need a fib object in order to
+		 * manage the linked lists.
+		 */
+		if ((!dev->aif_thread)
+		 || (!(fib = kzalloc_obj(struct fib, GFP_ATOMIC))))
+			return 1;
+		if (!(hw_fib = kzalloc_obj(struct hw_fib, GFP_ATOMIC))) {
+			kfree (fib);
+			return 1;
+		}
+		if (dev->sa_firmware) {
+			fib->hbacmd_size = index;	/* store event type */
+		} else if (aif_fib != NULL) {
+			memcpy(hw_fib, aif_fib, sizeof(struct hw_fib));
+		} else {
+			memcpy(hw_fib, (struct hw_fib *)
+				(((uintptr_t)(dev->regs.sa)) + index),
+				sizeof(struct hw_fib));
+		}
+		INIT_LIST_HEAD(&fib->fiblink);
+		fib->type = FSAFS_NTC_FIB_CONTEXT;
+		fib->size = sizeof(struct fib);
+		fib->hw_fib_va = hw_fib;
+		fib->data = hw_fib->data;
+		fib->dev = dev;
+	
+		spin_lock_irqsave(q->lock, flags);
+		list_add_tail(&fib->fiblink, &q->cmdq);
+	        wake_up_interruptible(&q->cmdready);
+		spin_unlock_irqrestore(q->lock, flags);
+		return 1;
+	} else if (isAif == 2) {	/* AIF - new (SRC) */
+		struct fib *fibctx;
+		struct aac_aifcmd *cmd;
+
+		fibctx = aac_fib_alloc(dev);
+		if (!fibctx)
+			return 1;
+		aac_fib_init(fibctx);
+
+		cmd = (struct aac_aifcmd *) fib_data(fibctx);
+		cmd->command = cpu_to_le32(AifReqEvent);
+
+		return aac_fib_send(AifRequest,
+			fibctx,
+			sizeof(struct hw_fib)-sizeof(struct aac_fibhdr),
+			FsaNormal,
+			0, 1,
+			(fib_callback)aac_aif_callback, fibctx);
+	} else {
+		struct fib *fib = &dev->fibs[index];
+		int start_callback = 0;
+
+		/*
+		 *	Remove this fib from the Outstanding I/O queue.
+		 *	But only if it has not already been timed out.
+		 *
+		 *	If the fib has been timed out already, then just 
+		 *	continue. The caller has already been notified that
+		 *	the fib timed out.
+		 */
+		atomic_dec(&dev->queues->queue[AdapNormCmdQueue].numpending);
+
+		if (unlikely(fib->flags & FIB_CONTEXT_FLAG_TIMED_OUT)) {
+			aac_fib_complete(fib);
+			aac_fib_free(fib);
+			return 0;
+		}
+
+		FIB_COUNTER_INCREMENT(aac_config.FibRecved);
+
+		if (fib->flags & FIB_CONTEXT_FLAG_NATIVE_HBA) {
+
+			if (isFastResponse)
+				fib->flags |= FIB_CONTEXT_FLAG_FASTRESP;
+
+			if (fib->callback) {
+				start_callback = 1;
+			} else {
+				unsigned long flagv;
+				int completed = 0;
+
+				dprintk((KERN_INFO "event_wait up\n"));
+				spin_lock_irqsave(&fib->event_lock, flagv);
+				if (fib->done == 2) {
+					fib->done = 1;
+					completed = 1;
+				} else {
+					fib->done = 1;
+					complete(&fib->event_wait);
+				}
+				spin_unlock_irqrestore(&fib->event_lock, flagv);
+
+				spin_lock_irqsave(&dev->manage_lock, mflags);
+				dev->management_fib_count--;
+				spin_unlock_irqrestore(&dev->manage_lock,
+					mflags);
+
+				FIB_COUNTER_INCREMENT(aac_config.NativeRecved);
+				if (completed)
+					aac_fib_complete(fib);
+			}
+		} else {
+			struct hw_fib *hwfib = fib->hw_fib_va;
+
+			if (isFastResponse) {
+				/* Doctor the fib */
+				*(__le32 *)hwfib->data = cpu_to_le32(ST_OK);
+				hwfib->header.XferState |=
+					cpu_to_le32(AdapterProcessed);
+				fib->flags |= FIB_CONTEXT_FLAG_FASTRESP;
+			}
+
+			if (hwfib->header.Command ==
+				cpu_to_le16(NuFileSystem)) {
+				__le32 *pstatus = (__le32 *)hwfib->data;
+
+				if (*pstatus & cpu_to_le32(0xffff0000))
+					*pstatus = cpu_to_le32(ST_OK);
+			}
+			if (hwfib->header.XferState &
+				cpu_to_le32(NoResponseExpected | Async)) {
+				if (hwfib->header.XferState & cpu_to_le32(
+					NoResponseExpected)) {
+					FIB_COUNTER_INCREMENT(
+						aac_config.NoResponseRecved);
+				} else {
+					FIB_COUNTER_INCREMENT(
+						aac_config.AsyncRecved);
+				}
+				start_callback = 1;
+			} else {
+				unsigned long flagv;
+				int completed = 0;
+
+				dprintk((KERN_INFO "event_wait up\n"));
+				spin_lock_irqsave(&fib->event_lock, flagv);
+				if (fib->done == 2) {
+					fib->done = 1;
+					completed = 1;
+				} else {
+					fib->done = 1;
+					complete(&fib->event_wait);
+				}
+				spin_unlock_irqrestore(&fib->event_lock, flagv);
+
+				spin_lock_irqsave(&dev->manage_lock, mflags);
+				dev->management_fib_count--;
+				spin_unlock_irqrestore(&dev->manage_lock,
+					mflags);
+
+				FIB_COUNTER_INCREMENT(aac_config.NormalRecved);
+				if (completed)
+					aac_fib_complete(fib);
+			}
+		}
+
+
+		if (start_callback) {
+			/*
+			 * NOTE:  we cannot touch the fib after this
+			 *  call, because it may have been deallocated.
+			 */
+			if (likely(fib->callback && fib->callback_data)) {
+				fib->callback(fib->callback_data, fib);
+			} else {
+				aac_fib_complete(fib);
+				aac_fib_free(fib);
+			}
+
+		}
+		return 0;
+	}
+}