x86/boot/compressed: Register dummy NMI handler in EFI boot loader, to avoid kdump crashes

hulk inclusion
category: bugfix
issue: https://gitee.com/openeuler/kernel/issues/I69VF6

--------------------------------

If kdump is enabled, when using mce_inject to inject errors, EFI
boot loader would decompress & load second kernel for saving the
vmcore file.

For normal errors that is fine. However, in the MCE case, the panic
CPU that firstly enters into mce_panic() is running within NMI
interrupt context, and the processor blocks delivery of subsequent
NMIs until the next execution of the IRET instruction.

When the panic CPU takes long time in the panic processing route,
and causes the watchdog timeout, at this moment, the processor
already receives NMI interrupt in the background.

In the reproducer sequence below, panic CPU would run into EFI loader
and raise page fault exception (like visiting `vidmem` variable
when attempting to call debug_putstr()), the CPU would execute IRET
instruction when it exits from the page fault handler.

But the loader never registers handler for NMI vector in IDT,
lack of vector handler would cause reboot, which interrupts
kdump procedure and fails to save the vmcore file.

Here is steps to reproduce the above issue (it's sporadic):

  1. # cat uncorrected
     CPU 1 BANK 4
     STATUS uncorrected 0xc0
     MCGSTATUS  EIPV MCIP
     ADDR 0x1234
     RIP 0xdeadbabe
     RAISINGCPU 0
     MCGCAP SER CMCI TES 0x6
  2. # modprobe mce_inject
  3. # mce-inject uncorrected

For increasing the probability of reproduction of this issue, there are
two ways to increase the probability of the bug:

  1. modify the threshold value of watchdog (increase NMI frequency);
  2. and/or add delays before panic() in mce_panic() and modify
     PANIC_TIMEOUT macro;

Fixes: ca0e22d4 ("x86/boot/compressed/64: Always switch to own page table")
Signed-off-by: NZeng Heng <zengheng4@huawei.com>
[ Tidy up changelog, add comments. ]
Signed-off-by: NIngo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20230110102745.2514694-1-zengheng4@huawei.com

arch/x86/boot/compressed/ident_map_64.c

@@ -359,3 +359,15 @@ void do_boot_page_fault(struct pt_regs *regs, unsigned long error_code)
 	 */
 	add_identity_map(address, end);
 }
+
+void do_boot_nmi_fault(struct pt_regs *regs, unsigned long error_code)
+{
+	/*
+	 * Default boot loader placeholder fault handler - there's no real
+	 * kernel running yet, so there's not much we can do - but NMIs
+	 * can arrive in a kdump scenario, for example by the NMI watchdog.
+	 *
+	 * Not having any handler would cause the CPU to silently reboot,
+	 * so we do the second-worst thing here and ignore the NMI.
+	 */
+}

arch/x86/boot/compressed/idt_64.c

@@ -44,6 +44,7 @@ void load_stage2_idt(void)
 {
 	boot_idt_desc.address = (unsigned long)boot_idt;
 
+	set_idt_entry(X86_TRAP_NMI, boot_nmi_fault);
 	set_idt_entry(X86_TRAP_PF, boot_page_fault);
 
 #ifdef CONFIG_AMD_MEM_ENCRYPT

arch/x86/boot/compressed/idt_handlers_64.S

@@ -69,6 +69,7 @@ SYM_FUNC_END(\name)
 	.text
 	.code64
 
+EXCEPTION_HANDLER	boot_nmi_fault do_boot_nmi_fault error_code=0
 EXCEPTION_HANDLER	boot_page_fault do_boot_page_fault error_code=1
 
 #ifdef CONFIG_AMD_MEM_ENCRYPT

arch/x86/boot/compressed/misc.h

@@ -155,6 +155,7 @@ extern gate_desc boot_idt[BOOT_IDT_ENTRIES];
 extern struct desc_ptr boot_idt_desc;
 
 /* IDT Entry Points */
+void boot_nmi_fault(void);
 void boot_page_fault(void);
 void boot_stage1_vc(void);
 void boot_stage2_vc(void);