[RFC PATCH 2/2] x86/kexec: Add data section to relocate_kernel

Fri Nov 8 04:29:33 PST 2024

On 8 November 2024 03:26:58 GMT-08:00, "H. Peter Anvin" <hpa at zytor.com> wrote:
>On November 8, 2024 6:22:41 AM GMT+01:00, David Woodhouse <dwmw2 at infradead.org> wrote:
>>From: David Woodhouse <dwmw at amazon.co.uk>
>>
>>Now that it's handled sanely by a linker script we can have actual data,
>>and just use %rip-relative addressing to access it.
>>
>>If we could call the *copy* instead of the original relocate_kernel in
>>the kernel text, then we could use %rip-relative addressing everywhere.
>>
>>Signed-off-by: David Woodhouse <dwmw at amazon.co.uk>
>>---
>> arch/x86/kernel/relocate_kernel_64.S | 58 ++++++++++++++++------------
>> arch/x86/kernel/vmlinux.lds.S        |  2 +-
>> 2 files changed, 35 insertions(+), 25 deletions(-)
>>
>>diff --git a/arch/x86/kernel/relocate_kernel_64.S b/arch/x86/kernel/relocate_kernel_64.S
>>index 1efcbd340528..577aa1672349 100644
>>--- a/arch/x86/kernel/relocate_kernel_64.S
>>+++ b/arch/x86/kernel/relocate_kernel_64.S
>>@@ -27,18 +27,28 @@
>>  * ~ control_page + PAGE_SIZE are used as data storage and stack for
>>  * jumping back
>>  */
>>-#define DATA(offset)		(KEXEC_CONTROL_CODE_MAX_SIZE+(offset))
>> 
>>+	.section .data.relocate_kernel,"a";
>> /* Minimal CPU state */
>>-#define RSP			DATA(0x0)
>>-#define CR0			DATA(0x8)
>>-#define CR3			DATA(0x10)
>>-#define CR4			DATA(0x18)
>>-
>>+SYM_DATA_LOCAL(saved_rsp, .quad 0)
>>+SYM_DATA_LOCAL(saved_cr0, .quad 0)
>>+SYM_DATA_LOCAL(saved_cr3, .quad 0)
>>+SYM_DATA_LOCAL(saved_cr4, .quad 0)
>> /* other data */
>>-#define CP_PA_TABLE_PAGE	DATA(0x20)
>>-#define CP_PA_SWAP_PAGE		DATA(0x28)
>>-#define CP_PA_BACKUP_PAGES_MAP	DATA(0x30)
>>+SYM_DATA_LOCAL(pa_table_page, .quad 0)
>>+SYM_DATA_LOCAL(pa_swap_page, .quad 0)
>>+SYM_DATA_LOCAL(pa_backup_pages_map, .quad 0)
>>+
>>+/*
>>+ * There are two physical copies of relocate_kernel(), one in the original
>>+ * Kernel text and the other copied to the control page. There is a virtual
>>+ * mapping of each, in the original kernel. It is the *original* which is
>>+ * called from machine_kexec(), largely becaose the copy isn't mapped as an
>>+ * executable page. Thus, this code cannot just use %rip-relative addressing
>>+ * until after the %cr3 change and the jump to identity_mapped(). Until
>>+ * then, some pointer arithmetic is required.
>>+ */
>>+#define DATA(x) (x - relocate_kernel)
>> 
>> 	.section .text.relocate_kernel,"ax";
>> 	.code64
>>@@ -63,13 +73,13 @@ SYM_CODE_START_NOALIGN(relocate_kernel)
>> 	pushf
>> 
>> 	movq	PTR(VA_CONTROL_PAGE)(%rsi), %r11
>>-	movq	%rsp, RSP(%r11)
>>+	movq	%rsp, DATA(saved_rsp)(%r11)
>> 	movq	%cr0, %rax
>>-	movq	%rax, CR0(%r11)
>>+	movq	%rax, DATA(saved_cr0)(%r11)
>> 	movq	%cr3, %rax
>>-	movq	%rax, CR3(%r11)
>>+	movq	%rax, DATA(saved_cr3)(%r11)
>> 	movq	%cr4, %rax
>>-	movq	%rax, CR4(%r11)
>>+	movq	%rax, DATA(saved_cr4)(%r11)
>> 
>> 	/* Save CR4. Required to enable the right paging mode later. */
>> 	movq	%rax, %r13
>>@@ -94,9 +104,9 @@ SYM_CODE_START_NOALIGN(relocate_kernel)
>> 	movq	PTR(PA_SWAP_PAGE)(%rsi), %r10
>> 
>> 	/* save some information for jumping back */
>>-	movq	%r9, CP_PA_TABLE_PAGE(%r11)
>>-	movq	%r10, CP_PA_SWAP_PAGE(%r11)
>>-	movq	%rdi, CP_PA_BACKUP_PAGES_MAP(%r11)
>>+	movq	%r9, DATA(pa_table_page)(%r11)
>>+	movq	%r10, DATA(pa_swap_page)(%r11)
>>+	movq	%rdi, DATA(pa_backup_pages_map)(%r11)
>> 
>> 	/* Save the preserve_context to %r11 as swap_pages clobbers %rcx. */
>> 	movq	%rcx, %r11
>>@@ -128,7 +138,7 @@ SYM_CODE_START_LOCAL_NOALIGN(identity_mapped)
>> 	/* set return address to 0 if not preserving context */
>> 	pushq	$0
>> 	/* store the start address on the stack */
>>-	pushq   %rdx
>>+	pushq   start_address(%rip)
>> 
>> 	/*
>> 	 * Clear X86_CR4_CET (if it was set) such that we can clear CR0_WP
>>@@ -227,9 +237,9 @@ SYM_CODE_START_LOCAL_NOALIGN(identity_mapped)
>> 	/* get the re-entry point of the peer system */
>> 	movq	0(%rsp), %rbp
>> 	leaq	relocate_kernel(%rip), %r8
>>-	movq	CP_PA_SWAP_PAGE(%r8), %r10
>>-	movq	CP_PA_BACKUP_PAGES_MAP(%r8), %rdi
>>-	movq	CP_PA_TABLE_PAGE(%r8), %rax
>>+	movq	pa_swap_page(%rip), %r10
>>+	movq	pa_backup_pages_map(%rip), %rdi
>>+	movq	pa_table_page(%rip), %rax
>> 	movq	%rax, %cr3
>> 	lea	PAGE_SIZE(%r8), %rsp
>> 	call	swap_pages
>>@@ -243,11 +253,11 @@ SYM_CODE_END(identity_mapped)
>> SYM_CODE_START_LOCAL_NOALIGN(virtual_mapped)
>> 	UNWIND_HINT_END_OF_STACK
>> 	ANNOTATE_NOENDBR // RET target, above
>>-	movq	RSP(%r8), %rsp
>>-	movq	CR4(%r8), %rax
>>+	movq	saved_rsp(%rip), %rsp
>>+	movq	saved_cr4(%rip), %rax
>> 	movq	%rax, %cr4
>>-	movq	CR3(%r8), %rax
>>-	movq	CR0(%r8), %r8
>>+	movq	saved_cr3(%rip), %rax
>>+	movq	saved_cr0(%r8), %r8
>> 	movq	%rax, %cr3
>> 	movq	%r8, %cr0
>> 	movq	%rbp, %rax
>>diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
>>index ad451371e179..65f879b31a82 100644
>>--- a/arch/x86/kernel/vmlinux.lds.S
>>+++ b/arch/x86/kernel/vmlinux.lds.S
>>@@ -100,7 +100,7 @@ const_pcpu_hot = pcpu_hot;
>> 	. = ALIGN(PAGE_SIZE);					\
>> 	__relocate_kernel_start = .;				\
>> 	*(.text.relocate_kernel);				\
>>-	*(.rodata.relocate_kernel);				\
>>+	*(.data.relocate_kernel);				\
>> 	__relocate_kernel_end = .;
>> #else
>> #define KEXEC_RELOCATE_KERNEL_TEXT
>
>Looks good at first glance. I'm currently traveling so I haven't fully reviewed it though.

Ta. That's good enough for me to go ahead and port the rest over. 

Is there a selftest for the preserve-context mode somewhere, with a payload that just does a "ret"?