Re: [PATCH 06/11] x86/xen: Add i386 kexec/kdump implementation

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Thu, Sep 27, 2012 at 08:06:33PM +0200, Daniel Kiper wrote:
> Add i386 kexec/kdump implementation.
> 
> Signed-off-by: Daniel Kiper <daniel.kiper@oracle.com>
> ---
>  arch/x86/xen/machine_kexec_32.c   |  245 ++++++++++++++++++++++++++++
>  arch/x86/xen/relocate_kernel_32.S |  323 +++++++++++++++++++++++++++++++++++++
>  2 files changed, 568 insertions(+), 0 deletions(-)
>  create mode 100644 arch/x86/xen/machine_kexec_32.c
>  create mode 100644 arch/x86/xen/relocate_kernel_32.S
> 
> diff --git a/arch/x86/xen/machine_kexec_32.c b/arch/x86/xen/machine_kexec_32.c
> new file mode 100644
> index 0000000..6b5141e
> --- /dev/null
> +++ b/arch/x86/xen/machine_kexec_32.c
> @@ -0,0 +1,245 @@
> +/*
> + * Copyright (c) 2011 Daniel Kiper
> + * Copyright (c) 2012 Daniel Kiper, Oracle Corporation
> + *
> + * kexec/kdump implementation for Xen was written by Daniel Kiper.
> + * Initial work on it was sponsored by Google under Google Summer
> + * of Code 2011 program and Citrix. Konrad Rzeszutek Wilk from Oracle
> + * was the mentor for this project.
> + *
> + * Some ideas are taken from:
> + *   - native kexec/kdump implementation,
> + *   - kexec/kdump implementation for Xen Linux Kernel Ver. 2.6.18,
> + *   - PV-GRUB.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 of the License, or
> + * (at your option) any later version.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License along
> + * with this program.  If not, see <http://www.gnu.org/licenses/>.
> + */
> +
> +#include <linux/errno.h>
> +#include <linux/init.h>
> +#include <linux/kernel.h>
> +#include <linux/kexec.h>
> +#include <linux/mm.h>
> +#include <linux/string.h>
> +
> +#include <xen/xen.h>
> +#include <xen/xen-ops.h>
> +
> +#include <asm/xen/hypercall.h>
> +#include <asm/xen/kexec.h>
> +#include <asm/xen/page.h>
> +
> +#define __ma(vaddr)	(virt_to_machine(vaddr).maddr)
> +
> +static struct page *kimage_alloc_pages(gfp_t gfp_mask,
> +					unsigned int order,
> +					unsigned long limit)
> +{
> +	struct page *pages;
> +	unsigned int address_bits, i;
> +
> +	pages = alloc_pages(gfp_mask, order);
> +
> +	if (!pages)
> +		return NULL;
> +
> +	address_bits = (limit == ULONG_MAX) ? BITS_PER_LONG : ilog2(limit);
> +
> +	/* Relocate set of pages below given limit. */
> +	if (xen_create_contiguous_region((unsigned long)page_address(pages),
> +							order, address_bits)) {
> +		__free_pages(pages, order);
> +		return NULL;
> +	}
> +
> +	pages->mapping = NULL;

It shouldn't matter (as you did the alloc_page) but could you
add:
	BUG_ON(PagePrivate(pages))
in case somebody did do something weird beforehand.

> +	set_page_private(pages, order);
> +
> +	for (i = 0; i < (1 << order); ++i)
> +		SetPageReserved(pages + i);
> +
> +	return pages;
> +}
> +
> +static void kimage_free_pages(struct page *page)
> +{
> +	unsigned int i, order;
> +
> +	order = page_private(page);
> +
> +	for (i = 0; i < (1 << order); ++i)
> +		ClearPageReserved(page + i);
> +
> +	xen_destroy_contiguous_region((unsigned long)page_address(page), order);
> +	__free_pages(page, order);
> +}
> +
> +static unsigned long xen_page_to_mfn(struct page *page)
> +{
> +	return pfn_to_mfn(page_to_pfn(page));
> +}
> +
> +static struct page *xen_mfn_to_page(unsigned long mfn)
> +{
> +	return pfn_to_page(mfn_to_pfn(mfn));
> +}
> +
> +static unsigned long xen_virt_to_machine(volatile void *address)
> +{
> +	return virt_to_machine(address).maddr;
> +}
> +
> +static void *xen_machine_to_virt(unsigned long address)
> +{
> +	return phys_to_virt(machine_to_phys(XMADDR(address)).paddr);
> +}
> +
> +static void free_transition_pgtable(struct kimage *image)
> +{
> +	free_page((unsigned long)image->arch.pgd);
> +	free_page((unsigned long)image->arch.pmd0);
> +	free_page((unsigned long)image->arch.pmd1);
> +	free_page((unsigned long)image->arch.pte0);
> +	free_page((unsigned long)image->arch.pte1);
> +}
> +
> +static int alloc_transition_pgtable(struct kimage *image)
> +{
> +	image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
> +
> +	if (!image->arch.pgd)
> +		goto err;
> +
> +	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
> +
> +	if (!image->arch.pmd0)
> +		goto err;
> +
> +	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
> +
> +	if (!image->arch.pmd1)
> +		goto err;
> +
> +	image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
> +
> +	if (!image->arch.pte0)
> +		goto err;
> +
> +	image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
> +
> +	if (!image->arch.pte1)
> +		goto err;
> +
> +	return 0;
> +
> +err:
> +	free_transition_pgtable(image);
> +
> +	return -ENOMEM;
> +}
> +
> +static int machine_xen_kexec_prepare(struct kimage *image)
> +{
> +#ifdef CONFIG_KEXEC_JUMP
> +	if (image->preserve_context) {
> +		pr_info_once("kexec: Context preservation is not "
> +				"supported in Xen domains.\n");
> +		return -ENOSYS;
> +	}
> +#endif
> +
> +	return alloc_transition_pgtable(image);
> +}
> +
> +static int machine_xen_kexec_load(struct kimage *image)
> +{
> +	void *control_page;
> +	struct xen_kexec_load xkl = {};
> +
> +	if (!image)
> +		return 0;

Not -EINVAL?

> +
> +	control_page = page_address(image->control_code_page);
> +	memcpy(control_page, xen_relocate_kernel, xen_kexec_control_code_size);
> +
> +	xkl.type = image->type;
> +	xkl.image.page_list[XK_MA_CONTROL_PAGE] = __ma(control_page);
> +	xkl.image.page_list[XK_MA_TABLE_PAGE] = 0; /* Unused. */
> +	xkl.image.page_list[XK_MA_PGD_PAGE] = __ma(image->arch.pgd);
> +	xkl.image.page_list[XK_MA_PUD0_PAGE] = 0; /* Unused. */
> +	xkl.image.page_list[XK_MA_PUD1_PAGE] = 0; /* Unused. */
> +	xkl.image.page_list[XK_MA_PMD0_PAGE] = __ma(image->arch.pmd0);
> +	xkl.image.page_list[XK_MA_PMD1_PAGE] = __ma(image->arch.pmd1);
> +	xkl.image.page_list[XK_MA_PTE0_PAGE] = __ma(image->arch.pte0);
> +	xkl.image.page_list[XK_MA_PTE1_PAGE] = __ma(image->arch.pte1);
> +	xkl.image.indirection_page = image->head;
> +	xkl.image.start_address = image->start;
> +
> +	return HYPERVISOR_kexec_op(KEXEC_CMD_kexec_load, &xkl);
> +}
> +
> +static void machine_xen_kexec_cleanup(struct kimage *image)
> +{
> +	free_transition_pgtable(image);
> +}
> +
> +static void machine_xen_kexec_unload(struct kimage *image)
> +{
> +	int rc;
> +	struct xen_kexec_load xkl = {};
> +
> +	if (!image)
> +		return;
> +
> +	xkl.type = image->type;
> +	rc = HYPERVISOR_kexec_op(KEXEC_CMD_kexec_unload, &xkl);
> +
> +	WARN(rc, "kexec: %s: HYPERVISOR_kexec_op(): %i\n", __func__, rc);
> +}
> +
> +static void machine_xen_kexec_shutdown(void)
> +{
> +}
> +
> +static void machine_xen_kexec(struct kimage *image)
> +{
> +	int rc;
> +	struct xen_kexec_exec xke = {};
> +
> +	xke.type = image->type;
> +	rc = HYPERVISOR_kexec_op(KEXEC_CMD_kexec, &xke);
> +
> +	pr_emerg("kexec: %s: HYPERVISOR_kexec_op(): %i\n", __func__, rc);
> +	BUG();
> +}
> +
> +void __init xen_init_kexec_ops(void)
> +{
> +	if (!xen_initial_domain())
> +		return;
> +
> +	kexec_ops.always_use_normal_alloc = true;
> +	kexec_ops.kimage_alloc_pages = kimage_alloc_pages;
> +	kexec_ops.kimage_free_pages = kimage_free_pages;
> +	kexec_ops.page_to_pfn = xen_page_to_mfn;
> +	kexec_ops.pfn_to_page = xen_mfn_to_page;
> +	kexec_ops.virt_to_phys = xen_virt_to_machine;
> +	kexec_ops.phys_to_virt = xen_machine_to_virt;
> +	kexec_ops.machine_kexec_prepare = machine_xen_kexec_prepare;
> +	kexec_ops.machine_kexec_load = machine_xen_kexec_load;
> +	kexec_ops.machine_kexec_cleanup = machine_xen_kexec_cleanup;
> +	kexec_ops.machine_kexec_unload = machine_xen_kexec_unload;
> +	kexec_ops.machine_kexec_shutdown = machine_xen_kexec_shutdown;
> +	kexec_ops.machine_kexec = machine_xen_kexec;
> +}
> diff --git a/arch/x86/xen/relocate_kernel_32.S b/arch/x86/xen/relocate_kernel_32.S
> new file mode 100644
> index 0000000..0e81830
> --- /dev/null
> +++ b/arch/x86/xen/relocate_kernel_32.S
> @@ -0,0 +1,323 @@
> +/*
> + * Copyright (c) 2002-2005 Eric Biederman <ebiederm@xmission.com>
> + * Copyright (c) 2011 Daniel Kiper
> + * Copyright (c) 2012 Daniel Kiper, Oracle Corporation
> + *
> + * kexec/kdump implementation for Xen was written by Daniel Kiper.
> + * Initial work on it was sponsored by Google under Google Summer
> + * of Code 2011 program and Citrix. Konrad Rzeszutek Wilk from Oracle
> + * was the mentor for this project.
> + *
> + * Some ideas are taken from:
> + *   - native kexec/kdump implementation,
> + *   - kexec/kdump implementation for Xen Linux Kernel Ver. 2.6.18,
> + *   - PV-GRUB.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either veesion 2 of the License, or
> + * (at your option) any later veesion.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License along
> + * with this program.  If not, see <http://www.gnu.org/licenses/>.
> + */
> +
> +#include <asm/cache.h>
> +#include <asm/page_types.h>
> +#include <asm/pgtable_types.h>
> +#include <asm/processor-flags.h>
> +
> +#include <asm/xen/kexec.h>
> +
> +#define ARG_INDIRECTION_PAGE	0x4
> +#define ARG_PAGE_LIST		0x8
> +#define ARG_START_ADDRESS	0xc
> +
> +#define PTR(x)	(x << 2)
> +
> +	.text
> +	.align	PAGE_SIZE
> +	.globl	xen_kexec_control_code_size, xen_relocate_kernel
> +
> +xen_relocate_kernel:
> +	/*
> +	 * Must be relocatable PIC code callable as a C function.
> +	 *
> +	 * This function is called by Xen but here hypervisor is dead.
> +	 * We are playing on bare metal.
> +	 *
> +	 * Every machine address passed to this function through
> +	 * page_list (e.g. XK_MA_CONTROL_PAGE) is established
> +	 * by dom0 during kexec load phase.
> +	 *
> +	 * Every virtual address passed to this function through page_list
> +	 * (e.g. XK_VA_CONTROL_PAGE) is established by hypervisor during
> +	 * HYPERVISOR_kexec_op(KEXEC_CMD_kexec_load) hypercall.
> +	 *
> +	 * 0x4(%esp) - indirection_page,
> +	 * 0x8(%esp) - page_list,
> +	 * 0xc(%esp) - start_address,
> +	 * 0x10(%esp) - cpu_has_pae (ignored),
> +	 * 0x14(%esp) - preserve_context (ignored).
> +	 */
> +
> +	/* Zero out flags, and disable interrupts. */
> +	pushl	$0
> +	popfl
> +
> +	/* Get page_list address. */
> +	movl	ARG_PAGE_LIST(%esp), %esi
> +
> +	/*
> +	 * Map the control page at its virtual address
> +	 * in transition page table.
> +	 */
> +	movl	PTR(XK_VA_CONTROL_PAGE)(%esi), %eax
> +
> +	/* Get PGD address and PGD entry index. */
> +	movl	PTR(XK_VA_PGD_PAGE)(%esi), %ebx
> +	movl	%eax, %ecx
> +	shrl	$PGDIR_SHIFT, %ecx
> +	andl	$(PTRS_PER_PGD - 1), %ecx
> +
> +	/* Fill PGD entry with PMD0 reference. */
> +	movl	PTR(XK_MA_PMD0_PAGE)(%esi), %edx
> +	orl	$_PAGE_PRESENT, %edx
> +	movl	%edx, (%ebx, %ecx, 8)
> +
> +	/* Get PMD0 address and PMD0 entry index. */
> +	movl	PTR(XK_VA_PMD0_PAGE)(%esi), %ebx
> +	movl	%eax, %ecx
> +	shrl	$PMD_SHIFT, %ecx
> +	andl	$(PTRS_PER_PMD - 1), %ecx
> +
> +	/* Fill PMD0 entry with PTE0 reference. */
> +	movl	PTR(XK_MA_PTE0_PAGE)(%esi), %edx
> +	orl	$_KERNPG_TABLE, %edx
> +	movl	%edx, (%ebx, %ecx, 8)
> +
> +	/* Get PTE0 address and PTE0 entry index. */
> +	movl	PTR(XK_VA_PTE0_PAGE)(%esi), %ebx
> +	movl	%eax, %ecx
> +	shrl	$PAGE_SHIFT, %ecx
> +	andl	$(PTRS_PER_PTE - 1), %ecx
> +
> +	/* Fill PTE0 entry with control page reference. */
> +	movl	PTR(XK_MA_CONTROL_PAGE)(%esi), %edx
> +	orl	$__PAGE_KERNEL_EXEC, %edx
> +	movl	%edx, (%ebx, %ecx, 8)
> +
> +	/*
> +	 * Identity map the control page at its machine address
> +	 * in transition page table.
> +	 */
> +	movl	PTR(XK_MA_CONTROL_PAGE)(%esi), %eax
> +
> +	/* Get PGD address and PGD entry index. */
> +	movl	PTR(XK_VA_PGD_PAGE)(%esi), %ebx
> +	movl	%eax, %ecx
> +	shrl	$PGDIR_SHIFT, %ecx
> +	andl	$(PTRS_PER_PGD - 1), %ecx
> +
> +	/* Fill PGD entry with PMD1 reference. */
> +	movl	PTR(XK_MA_PMD1_PAGE)(%esi), %edx
> +	orl	$_PAGE_PRESENT, %edx
> +	movl	%edx, (%ebx, %ecx, 8)
> +
> +	/* Get PMD1 address and PMD1 entry index. */
> +	movl	PTR(XK_VA_PMD1_PAGE)(%esi), %ebx
> +	movl	%eax, %ecx
> +	shrl	$PMD_SHIFT, %ecx
> +	andl	$(PTRS_PER_PMD - 1), %ecx
> +
> +	/* Fill PMD1 entry with PTE1 reference. */
> +	movl	PTR(XK_MA_PTE1_PAGE)(%esi), %edx
> +	orl	$_KERNPG_TABLE, %edx
> +	movl	%edx, (%ebx, %ecx, 8)
> +
> +	/* Get PTE1 address and PTE1 entry index. */
> +	movl	PTR(XK_VA_PTE1_PAGE)(%esi), %ebx
> +	movl	%eax, %ecx
> +	shrl	$PAGE_SHIFT, %ecx
> +	andl	$(PTRS_PER_PTE - 1), %ecx
> +
> +	/* Fill PTE1 entry with control page reference. */
> +	movl	PTR(XK_MA_CONTROL_PAGE)(%esi), %edx
> +	orl	$__PAGE_KERNEL_EXEC, %edx
> +	movl	%edx, (%ebx, %ecx, 8)
> +
> +	/*
> +	 * Get machine address of control page now.
> +	 * This is impossible after page table switch.
> +	 */
> +	movl	PTR(XK_MA_CONTROL_PAGE)(%esi), %ebx
> +
> +	/* Get machine address of transition page table now too. */
> +	movl	PTR(XK_MA_PGD_PAGE)(%esi), %ecx
> +
> +	/* Get start_address too. */
> +	movl	ARG_START_ADDRESS(%esp), %edx
> +
> +	/* Get indirection_page address too. */
> +	movl	ARG_INDIRECTION_PAGE(%esp), %edi
> +
> +	/* Switch to transition page table. */
> +	movl	%ecx, %cr3
> +
> +	/* Load IDT. */
> +	lidtl	(idt_48 - xen_relocate_kernel)(%ebx)
> +
> +	/* Load GDT. */
> +	leal	(gdt - xen_relocate_kernel)(%ebx), %eax
> +	movl	%eax, (gdt_48 - xen_relocate_kernel + 2)(%ebx)
> +	lgdtl	(gdt_48 - xen_relocate_kernel)(%ebx)
> +
> +	/* Load data segment registers. */
> +	movl	$(gdt_ds - gdt), %eax
> +	movl	%eax, %ds
> +	movl	%eax, %es
> +	movl	%eax, %fs
> +	movl	%eax, %gs
> +	movl	%eax, %ss
> +
> +	/* Setup a new stack at the end of machine address of control page. */
> +	leal	PAGE_SIZE(%ebx), %esp
> +
> +	/* Store start_address on the stack. */
> +	pushl   %edx
> +
> +	/* Jump to identity mapped page. */
> +	pushl	$0
> +	pushl	$(gdt_cs - gdt)
> +	addl	$(identity_mapped - xen_relocate_kernel), %ebx
> +	pushl	%ebx
> +	iretl
> +
> +identity_mapped:
> +	/*
> +	 * Set %cr0 to a known state:
> +	 *   - disable alignment check,
> +	 *   - disable floating point emulation,
> +	 *   - disable paging,
> +	 *   - no task switch,
> +	 *   - disable write protect,
> +	 *   - enable protected mode.
> +	 */
> +	movl	%cr0, %eax
> +	andl	$~(X86_CR0_AM | X86_CR0_EM | X86_CR0_PG | X86_CR0_TS | X86_CR0_WP), %eax
> +	orl	$(X86_CR0_PE), %eax
> +	movl	%eax, %cr0
> +
> +	/* Set %cr4 to a known state. */
> +	xorl	%eax, %eax
> +	movl	%eax, %cr4
> +
> +	jmp	1f
> +
> +1:
> +	/* Flush the TLB (needed?). */
> +	movl	%eax, %cr3
> +
> +	/* Do the copies. */
> +	movl	%edi, %ecx	/* Put the indirection_page in %ecx. */
> +	xorl	%edi, %edi
> +	xorl	%esi, %esi
> +	jmp	1f
> +
> +0:
> +	/*
> +	 * Top, read another doubleword from the indirection page.
> +	 * Indirection page is an array which contains source
> +	 * and destination address pairs. If all pairs could
> +	 * not fit in one page then at the end of given
> +	 * indirection page is pointer to next one.
> +	 * Copy is stopped when done indicator
> +	 * is found in indirection page.
> +	 */
> +	movl	(%ebx), %ecx
> +	addl	$4, %ebx
> +
> +1:
> +	testl	$0x1, %ecx	/* Is it a destination page? */
> +	jz	2f
> +
> +	movl	%ecx, %edi
> +	andl	$PAGE_MASK, %edi
> +	jmp	0b
> +
> +2:
> +	testl	$0x2, %ecx	/* Is it an indirection page? */
> +	jz	2f
> +
> +	movl	%ecx, %ebx
> +	andl	$PAGE_MASK, %ebx
> +	jmp	0b
> +
> +2:
> +	testl	$0x4, %ecx	/* Is it the done indicator? */
> +	jz	2f
> +	jmp	3f
> +
> +2:
> +	testl	$0x8, %ecx	/* Is it the source indicator? */
> +	jz	0b		/* Ignore it otherwise. */
> +
> +	movl	%ecx, %esi
> +	andl	$PAGE_MASK, %esi
> +	movl	$1024, %ecx
> +
> +	/* Copy page. */
> +	rep	movsl
> +	jmp	0b
> +
> +3:
> +	/*
> +	 * To be certain of avoiding problems with self-modifying code
> +	 * I need to execute a serializing instruction here.
> +	 * So I flush the TLB by reloading %cr3 here, it's handy,
> +	 * and not processor dependent.
> +	 */
> +	xorl	%eax, %eax
> +	movl	%eax, %cr3
> +
> +	/*
> +	 * Set all of the registers to known values.
> +	 * Leave %esp alone.
> +	 */
> +	xorl	%ebx, %ebx
> +	xorl    %ecx, %ecx
> +	xorl    %edx, %edx
> +	xorl    %esi, %esi
> +	xorl    %edi, %edi
> +	xorl    %ebp, %ebp
> +
> +	/* Jump to start_address. */
> +	retl
> +
> +	.align	L1_CACHE_BYTES
> +
> +gdt:
> +	.quad	0x0000000000000000	/* NULL descriptor. */
> +
> +gdt_cs:
> +	.quad	0x00cf9a000000ffff	/* 4 GiB code segment at 0x00000000. */
> +
> +gdt_ds:
> +	.quad	0x00cf92000000ffff	/* 4 GiB data segment at 0x00000000. */
> +gdt_end:
> +
> +gdt_48:
> +	.word	gdt_end - gdt - 1	/* GDT limit. */
> +	.long	0			/* GDT base - filled in by code above. */
> +
> +idt_48:
> +	.word	0			/* IDT limit. */
> +	.long	0			/* IDT base. */
> +
> +xen_kexec_control_code_size:
> +	.long	. - xen_relocate_kernel
> -- 
> 1.5.6.5