KVM: SEV: Do cache maintenance on the source VM during intra-host migrationgrafted

Manually perform cache maintenance on the source VM during intra-host
migration to ensure no stale data is left in CPU caches after the VM is
destroyed.  Because the source VM is "converted" to a non-SEV VM, KVM's
memory reclaim flows won't trigger cache maintenance, e.g. when all guest
memory is reclaimed in response to detaching from the mmu_notifier.

Note, relying on the destination VM to do cache maintenance isn't an option
as KVM doesn't require identical guest memory configurations, i.e. the
source VM may have access to memory that the destination VM does not.
Enforcing equivalent memory configurations is infeasible, as it would
require a *deep* comparison of memslots, e.g. to verify that not only are
the memslot identical, but what the memslots point at is also identical.

Fixes: b56639318bb2 ("KVM: SEV: Add support for SEV intra host migration")
Cc: stable@vger.kernel.org
Reported-by: Stefan Teodorescu <fane@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20260923163721.1584779-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

1 files changed, 244 insertions, 0 deletions

diff --git a/arch/hexagon/mm/init.c b/arch/hexagon/mm/init.c
new file mode 100644
index 000000000..07086dbd3
--- /dev/null
+++ b/arch/hexagon/mm/init.c
@@ -0,0 +1,244 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Memory subsystem initialization for Hexagon
+ *
+ * Copyright (c) 2010-2013, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/memblock.h>
+#include <asm/atomic.h>
+#include <linux/highmem.h>
+#include <asm/tlb.h>
+#include <asm/sections.h>
+#include <asm/setup.h>
+#include <asm/vm_mmu.h>
+
+/*
+ * Define a startpg just past the end of the kernel image and a lastpg
+ * that corresponds to the end of real or simulated platform memory.
+ */
+#define bootmem_startpg (PFN_UP(((unsigned long) _end) - PAGE_OFFSET + PHYS_OFFSET))
+
+unsigned long bootmem_lastpg;	/*  Should be set by platform code  */
+unsigned long __phys_offset;	/*  physical kernel offset >> 12  */
+
+/*  Set as variable to limit PMD copies  */
+int max_kernel_seg = 0x303;
+
+/*  indicate pfn's of high memory  */
+unsigned long highstart_pfn, highend_pfn;
+
+/* Default cache attribute for newly created page tables */
+unsigned long _dflt_cache_att = CACHEDEF;
+
+/*
+ * The current "generation" of kernel map, which should not roll
+ * over until Hell freezes over.  Actual bound in years needs to be
+ * calculated to confirm.
+ */
+DEFINE_SPINLOCK(kmap_gen_lock);
+
+/*  checkpatch says don't init this to 0.  */
+unsigned long long kmap_generation;
+
+void sync_icache_dcache(pte_t pte)
+{
+	unsigned long addr;
+	struct page *page;
+
+	page = pte_page(pte);
+	addr = (unsigned long) page_address(page);
+
+	__vmcache_idsync(addr, PAGE_SIZE);
+}
+
+void __init arch_zone_limits_init(unsigned long *max_zone_pfns)
+{
+	/*
+	 *  This is not particularly well documented anywhere, but
+	 *  give ZONE_NORMAL all the memory, including the big holes
+	 *  left by the kernel+bootmem_map which are already left as reserved
+	 *  in the bootmem_map; free_area_init should see those bits and
+	 *  adjust accordingly.
+	 */
+	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
+}
+
+static void __init paging_init(void)
+{
+	/*
+	 * Set the init_mm descriptors "context" value to point to the
+	 * initial kernel segment table's physical address.
+	 */
+	init_mm.context.ptbase = __pa(init_mm.pgd);
+}
+
+#ifndef DMA_RESERVE
+#define DMA_RESERVE		(4)
+#endif
+
+#define DMA_CHUNKSIZE		(1<<22)
+#define DMA_RESERVED_BYTES	(DMA_RESERVE * DMA_CHUNKSIZE)
+
+/*
+ * Pick out the memory size.  We look for mem=size,
+ * where size is "size[KkMm]"
+ */
+static int __init early_mem(char *p)
+{
+	unsigned long size;
+	char *endp;
+
+	size = memparse(p, &endp);
+
+	bootmem_lastpg = PFN_DOWN(size);
+
+	return 0;
+}
+early_param("mem", early_mem);
+
+size_t hexagon_coherent_pool_size = (size_t) (DMA_RESERVE << 22);
+
+void __init setup_arch_memory(void)
+{
+	/*  XXX Todo: this probably should be cleaned up  */
+	u32 *segtable = (u32 *) &swapper_pg_dir[0];
+	u32 *segtable_end;
+
+	/*
+	 * Set up boot memory allocator
+	 *
+	 * The Gorman book also talks about these functions.
+	 * This needs to change for highmem setups.
+	 */
+
+	/*  Prior to this, bootmem_lastpg is actually mem size  */
+	bootmem_lastpg += ARCH_PFN_OFFSET;
+
+	/* Memory size needs to be a multiple of 16M */
+	bootmem_lastpg = PFN_DOWN((bootmem_lastpg << PAGE_SHIFT) &
+		~((BIG_KERNEL_PAGE_SIZE) - 1));
+
+	memblock_add(PHYS_OFFSET,
+		     (bootmem_lastpg - ARCH_PFN_OFFSET) << PAGE_SHIFT);
+
+	/* Reserve kernel text/data/bss */
+	memblock_reserve(PHYS_OFFSET,
+			 (bootmem_startpg - ARCH_PFN_OFFSET) << PAGE_SHIFT);
+	/*
+	 * Reserve the top DMA_RESERVE bytes of RAM for DMA (uncached)
+	 * memory allocation
+	 */
+	max_low_pfn = bootmem_lastpg - PFN_DOWN(DMA_RESERVED_BYTES);
+	min_low_pfn = ARCH_PFN_OFFSET;
+	memblock_reserve(PFN_PHYS(max_low_pfn), DMA_RESERVED_BYTES);
+
+	printk(KERN_INFO "bootmem_startpg:  0x%08lx\n", bootmem_startpg);
+	printk(KERN_INFO "bootmem_lastpg:  0x%08lx\n", bootmem_lastpg);
+	printk(KERN_INFO "min_low_pfn:  0x%08lx\n", min_low_pfn);
+	printk(KERN_INFO "max_low_pfn:  0x%08lx\n", max_low_pfn);
+
+	/*
+	 * The default VM page tables (will be) populated with
+	 * VA=PA+PAGE_OFFSET mapping.  We go in and invalidate entries
+	 * higher than what we have memory for.
+	 */
+
+	/*  this is pointer arithmetic; each entry covers 4MB  */
+	segtable = segtable + (PAGE_OFFSET >> 22);
+
+	/*  this actually only goes to the end of the first gig  */
+	segtable_end = segtable + (1<<(30-22));
+
+	/*
+	 * Move forward to the start of empty pages; take into account
+	 * phys_offset shift.
+	 */
+
+	segtable += (bootmem_lastpg-ARCH_PFN_OFFSET)>>(22-PAGE_SHIFT);
+	{
+		int i;
+
+		for (i = 1 ; i <= DMA_RESERVE ; i++)
+			segtable[-i] = ((segtable[-i] & __HVM_PTE_PGMASK_4MB)
+				| __HVM_PTE_R | __HVM_PTE_W | __HVM_PTE_X
+				| __HEXAGON_C_UNC << 6
+				| __HVM_PDE_S_4MB);
+	}
+
+	printk(KERN_INFO "clearing segtable from %p to %p\n", segtable,
+		segtable_end);
+	while (segtable < (segtable_end-8))
+		*(segtable++) = __HVM_PDE_S_INVALID;
+	/* stop the pointer at the device I/O 4MB page  */
+
+	printk(KERN_INFO "segtable = %p (should be equal to _K_io_map)\n",
+		segtable);
+
+#if 0
+	/*  Other half of the early device table from vm_init_segtable. */
+	printk(KERN_INFO "&_K_init_devicetable = 0x%08x\n",
+		(unsigned long) _K_init_devicetable-PAGE_OFFSET);
+	*segtable = ((u32) (unsigned long) _K_init_devicetable-PAGE_OFFSET) |
+		__HVM_PDE_S_4KB;
+	printk(KERN_INFO "*segtable = 0x%08x\n", *segtable);
+#endif
+
+	/*
+	 *  The bootmem allocator seemingly just lives to feed memory
+	 *  to the paging system
+	 */
+	printk(KERN_INFO "PAGE_SIZE=%lu\n", PAGE_SIZE);
+	paging_init();  /*  See Gorman Book, 2.3  */
+
+	/*
+	 *  At this point, the page allocator is kind of initialized, but
+	 *  apparently no pages are available (just like with the bootmem
+	 *  allocator), and need to be freed themselves via mem_init(),
+	 *  which is called by start_kernel() later on in the process
+	 */
+}
+
+static const pgprot_t protection_map[16] = {
+	[VM_NONE]					= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   CACHEDEF),
+	[VM_READ]					= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_READ | CACHEDEF),
+	[VM_WRITE]					= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   CACHEDEF),
+	[VM_WRITE | VM_READ]				= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_READ | CACHEDEF),
+	[VM_EXEC]					= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | CACHEDEF),
+	[VM_EXEC | VM_READ]				= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | _PAGE_READ |
+								   CACHEDEF),
+	[VM_EXEC | VM_WRITE]				= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | CACHEDEF),
+	[VM_EXEC | VM_WRITE | VM_READ]			= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | _PAGE_READ |
+								   CACHEDEF),
+	[VM_SHARED]                                     = __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   CACHEDEF),
+	[VM_SHARED | VM_READ]				= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_READ | CACHEDEF),
+	[VM_SHARED | VM_WRITE]				= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_WRITE | CACHEDEF),
+	[VM_SHARED | VM_WRITE | VM_READ]		= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_READ | _PAGE_WRITE |
+								   CACHEDEF),
+	[VM_SHARED | VM_EXEC]				= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | CACHEDEF),
+	[VM_SHARED | VM_EXEC | VM_READ]			= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | _PAGE_READ |
+								   CACHEDEF),
+	[VM_SHARED | VM_EXEC | VM_WRITE]		= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_EXECUTE | _PAGE_WRITE |
+								   CACHEDEF),
+	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= __pgprot(_PAGE_PRESENT | _PAGE_USER |
+								   _PAGE_READ | _PAGE_EXECUTE |
+								   _PAGE_WRITE | CACHEDEF)
+};
+DECLARE_VM_GET_PAGE_PROT