Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [High] x86/hw_breakpoints: arch_modify_local_hw_breakpoint_addr()
corrupts DRs
--
commit 490e341c5679ea8c1cf5b3a39b5b39ecc93de7d0
Author: Jinchao Wang [off-list ref]
x86/hw_breakpoints: Add arch_modify_local_hw_breakpoint_addr() API
This commit adds arch_modify_local_hw_breakpoint_addr() to update the
address of an installed per-CPU hardware breakpoint on the local CPU
without releasing and re-allocating its debug register slot.
Link: https://lore.kernel.org/all/59637b96946653393a7ad3c7de094094796b39c2.1785067572.git.wangjinchao600@gmail.com/ (local)
quoted hunk ↗ jump to hunk
diff --git a/arch/x86/kernel/hw_breakpoint.c b/arch/x86/kernel/hw_breakpoint.c
index f846c15f21ca1..3f63f7899f95e 100644
--- a/arch/x86/kernel/hw_breakpoint.c
+++ b/arch/x86/kernel/hw_breakpoint.c
@@ -133,6 +133,57 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
return 0;
}
+/*
+ * Modify the address of an installed perf counter breakpoint on local CPU.
+ *
+ * This updates the debug address register and per-CPU shadow state without
+ * changing slot allocation or DR7. The core layer (modify_local_hw_breakpoint_addr())
+ * manages updating bp->attr.bp_addr and counter_arch_bp(bp)->address.
+ *
+ * Atomic: called with IRQs disabled and handles variables and registers
+ * local to this CPU.
+ */
+int arch_modify_local_hw_breakpoint_addr(struct perf_event *bp,
+ unsigned long addr)
+{
+ struct arch_hw_breakpoint *info = counter_arch_bp(bp);
+ struct arch_hw_breakpoint hw;
+ struct perf_event_attr attr = bp->attr;
+ unsigned long dr7;
+ int i, ret;
+
+ lockdep_assert_irqs_disabled();
+
+ attr.bp_addr = addr;
+ ret = hw_breakpoint_arch_parse(bp, &attr, &hw);
+ if (ret)
+ return ret;
+
+ for (i = 0; i < HBP_NUM; i++) {
+ if (this_cpu_read(bp_per_reg[i]) == bp)
+ break;
+ }
+
+ if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
+ return -ENOENT;
+
+ /*
+ * Temporarily clear the breakpoint enable bit in DR7 before updating the
+ * debug address register, and re-enable it afterward, to comply with the
+ * x86 hardware specification and avoid spurious #DB exceptions.
+ */
+ dr7 = this_cpu_read(cpu_dr7);
+ set_debugreg(dr7 & ~__encode_dr7(i, info->len, info->type), 7);
+
+ this_cpu_write(cpu_debugreg[i], addr);
+ barrier();
+ set_debugreg(addr, i);
+
+ set_debugreg(dr7, 7);
[Severity: High]
Can arch_modify_local_hw_breakpoint_addr() corrupt hardware debug
registers or prematurely re-enable hardware DR7 when host breakpoints are
temporarily disabled or switched to KVM guest state?
Several paths disable hardware DR7 (writing DR7_FIXED_1) or replace
hardware DR0-DR3 without clearing the per-CPU shadow cpu_dr7:
arch/x86/kvm/x86.c:vcpu_enter_guest() {
...
if (unlikely(vcpu->arch.switch_db_regs &&
!(vcpu->arch.switch_db_regs & KVM_DEBUGREG_AUTO_SWITCH))) {
set_debugreg(DR7_FIXED_1, 7);
set_debugreg(vcpu->arch.eff_db[0], 0);
set_debugreg(vcpu->arch.eff_db[1], 1);
set_debugreg(vcpu->arch.eff_db[2], 2);
set_debugreg(vcpu->arch.eff_db[3], 3);
}
...
}
vcpu_enter_guest() runs in normal context with IRQs disabled and invokes
instrumentable functions and tracepoints (such as trace_kvm_entry and
trace_kvm_exit) before sync_dirty_debug_regs() and
hw_breakpoint_restore(). If a wprobe trigger calls
arch_modify_local_hw_breakpoint_addr() in this window, set_debugreg(addr,
i) overwrites the guest's hardware DR[i] with the host kernel address
(which sync_dirty_debug_regs() then saves into vcpu->arch.db[i], leaking
the host address to the guest), and set_debugreg(dr7, 7) activates host
cpu_dr7 while the remaining hardware DR[j] registers still hold guest
addresses.
Similarly, if called while use_temporary_mm() -> hw_breakpoint_disable()
or local_db_save() has suppressed hardware breakpoints, writing cpu_dr7
back to hardware DR7 re-enables the breakpoints prematurely while temp_mm
is loaded (with 0UL in the other DR registers) or while inside a #DB
exception context.
--
Sashiko AI review · https://sashiko.dev/#/patchset/179480675130.38065.3086970856119381693.stgit@devnote2?part=1