x86/microcode: Rework microcode loading

Yeah, I know, I know, this is a huuge patch and reviewing it is hard.

Sorry but this is the only way I could think of in which I can rewrite
the microcode patches loading procedure without breaking (knowingly) the
driver.

So maybe this patch is easier to review if one looks at the files after
the patch has been applied instead at the diff. Because then it becomes
pretty obvious:

* The BSP-loading path - load_ucode_bsp() is working independently from
  the AP path now and it doesn't save any pointers or patches anymore -
  it solely parses the builtin or initrd microcode and applies the patch.
  That's it.

This fixes the CONFIG_RANDOMIZE_MEMORY offset fun more solidly.

* The AP-loading path - load_ucode_ap() then goes and scans
  builtin/initrd *again* for the microcode patches but it caches them this
  time so that we don't have to do that scan on each AP but only once.

This simplifies the code considerably.

Then, when we save the microcode from the initrd/builtin, we go and
add the relevant patches to our own cache. The AMD side did do that
and now the Intel side does it too. So no more pointer copying and
blabla, we save the microcode patches ourselves and are independent from
initrd/builtin.

This whole conversion gives us other benefits like unifying the
initrd parsing into a single function: find_microcode_in_initrd() is
used by both.

The diffstat speaks for itself: 456 insertions(+), 695 deletions(-)

Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20161025095522.11964-12-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

4 files changed, 446 insertions, 697 deletions

diff --git a/arch/x86/include/asm/microcode.h b/arch/x86/include/asm/microcode.h
index 91595ecb3689..38711df3bcb5 100644
--- a/arch/x86/include/asm/microcode.h
+++ b/arch/x86/include/asm/microcode.h
@@ -64,6 +64,7 @@ struct ucode_cpu_info {
 	void			*mc;
 };
 extern struct ucode_cpu_info ucode_cpu_info[];
+struct cpio_data find_microcode_in_initrd(const char *path, bool use_pa);
 
 #ifdef CONFIG_MICROCODE_INTEL
 extern struct microcode_ops * __init init_intel_microcode(void);
diff --git a/arch/x86/kernel/cpu/microcode/amd.c b/arch/x86/kernel/cpu/microcode/amd.c
index 03c19a68d61f..0dd8db1213d3 100644
--- a/arch/x86/kernel/cpu/microcode/amd.c
+++ b/arch/x86/kernel/cpu/microcode/amd.c
@@ -41,53 +41,23 @@ static struct equiv_cpu_entry *equiv_cpu_table;
 
 /*
  * This points to the current valid container of microcode patches which we will
- * save from the initrd before jettisoning its contents.
+ * save from the initrd/builtin before jettisoning its contents.
  */
-static u8 *container;
-static size_t container_size;
-static bool ucode_builtin;
+struct container {
+	u8 *data;
+	size_t size;
+} cont;
 
 static u32 ucode_new_rev;
 static u8 amd_ucode_patch[PATCH_MAX_SIZE];
 static u16 this_equiv_id;
 
-static struct cpio_data ucode_cpio;
-
-static struct cpio_data __init find_ucode_in_initrd(void)
-{
-#ifdef CONFIG_BLK_DEV_INITRD
-	char *path;
-	void *start;
-	size_t size;
-
-	/*
-	 * Microcode patch container file is prepended to the initrd in cpio
-	 * format. See Documentation/x86/early-microcode.txt
-	 */
-	static __initdata char ucode_path[] = "kernel/x86/microcode/AuthenticAMD.bin";
-
-#ifdef CONFIG_X86_32
-	struct boot_params *p;
-
-	/*
-	 * On 32-bit, early load occurs before paging is turned on so we need
-	 * to use physical addresses.
-	 */
-	p       = (struct boot_params *)__pa_nodebug(&boot_params);
-	path    = (char *)__pa_nodebug(ucode_path);
-	start   = (void *)p->hdr.ramdisk_image;
-	size    = p->hdr.ramdisk_size;
-#else
-	path    = ucode_path;
-	start   = (void *)(boot_params.hdr.ramdisk_image + PAGE_OFFSET);
-	size    = boot_params.hdr.ramdisk_size;
-#endif /* !CONFIG_X86_32 */
-
-	return find_cpio_data(path, start, size, NULL);
-#else
-	return (struct cpio_data){ NULL, 0, "" };
-#endif
-}
+/*
+ * Microcode patch container file is prepended to the initrd in cpio
+ * format. See Documentation/x86/early-microcode.txt
+ */
+static const char
+ucode_path[] __maybe_unused = "kernel/x86/microcode/AuthenticAMD.bin";
 
 static size_t compute_container_size(u8 *data, u32 total_size)
 {
@@ -143,62 +113,31 @@ static inline u16 find_equiv_id(struct equiv_cpu_entry *equiv_cpu_table,
 	return 0;
 }
 
-static int __apply_microcode_amd(struct microcode_amd *mc_amd)
-{
-	u32 rev, dummy;
-
-	native_wrmsrl(MSR_AMD64_PATCH_LOADER, (u64)(long)&mc_amd->hdr.data_code);
-
-	/* verify patch application was successful */
-	native_rdmsr(MSR_AMD64_PATCH_LEVEL, rev, dummy);
-	if (rev != mc_amd->hdr.patch_id)
-		return -1;
-
-	return 0;
-}
-
 /*
- * Early load occurs before we can vmalloc(). So we look for the microcode
- * patch container file in initrd, traverse equivalent cpu table, look for a
- * matching microcode patch, and update, all in initrd memory in place.
- * When vmalloc() is available for use later -- on 64-bit during first AP load,
- * and on 32-bit during save_microcode_in_initrd_amd() -- we can call
- * load_microcode_amd() to save equivalent cpu table and microcode patches in
- * kernel heap memory.
+ * This scans the ucode blob for the proper container as we can have multiple
+ * containers glued together.
  */
-static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
+static struct container
+find_proper_container(u8 *ucode, size_t size, u16 *ret_id)
 {
+	struct container ret = { NULL, 0 };
+	u32 eax, ebx, ecx, edx;
 	struct equiv_cpu_entry *eq;
-	size_t *cont_sz;
-	u32 *header;
-	u8  *data, **cont;
-	u8 (*patch)[PATCH_MAX_SIZE];
-	u16 eq_id = 0;
 	int offset, left;
-	u32 rev, eax, ebx, ecx, edx;
-	u32 *new_rev;
-
-#ifdef CONFIG_X86_32
-	new_rev = (u32 *)__pa_nodebug(&ucode_new_rev);
-	cont_sz = (size_t *)__pa_nodebug(&container_size);
-	cont	= (u8 **)__pa_nodebug(&container);
-	patch	= (u8 (*)[PATCH_MAX_SIZE])__pa_nodebug(&amd_ucode_patch);
-#else
-	new_rev = &ucode_new_rev;
-	cont_sz = &container_size;
-	cont	= &container;
-	patch	= &amd_ucode_patch;
-#endif
+	u16 eq_id = 0;
+	u32 *header;
+	u8 *data;
 
 	data   = ucode;
 	left   = size;
 	header = (u32 *)data;
 
+
 	/* find equiv cpu table */
 	if (header[0] != UCODE_MAGIC ||
 	    header[1] != UCODE_EQUIV_CPU_TABLE_TYPE || /* type */
 	    header[2] == 0)                            /* size */
-		return;
+		return ret;
 
 	eax = 0x00000001;
 	ecx = 0;
@@ -207,7 +146,7 @@ static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
 	while (left > 0) {
 		eq = (struct equiv_cpu_entry *)(data + CONTAINER_HDR_SZ);
 
-		*cont = data;
+		ret.data = data;
 
 		/* Advance past the container header */
 		offset = header[2] + CONTAINER_HDR_SZ;
@@ -216,15 +155,15 @@ static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
 
 		eq_id = find_equiv_id(eq, eax);
 		if (eq_id) {
-			this_equiv_id = eq_id;
-			*cont_sz = compute_container_size(*cont, left + offset);
+			ret.size = compute_container_size(ret.data, left + offset);
 
 			/*
 			 * truncate how much we need to iterate over in the
 			 * ucode update loop below
 			 */
-			left = *cont_sz - offset;
-			break;
+			left = ret.size - offset;
+			*ret_id = eq_id;
+			return ret;
 		}
 
 		/*
@@ -234,6 +173,7 @@ static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
 		 */
 		while (left > 0) {
 			header = (u32 *)data;
+
 			if (header[0] == UCODE_MAGIC &&
 			    header[1] == UCODE_EQUIV_CPU_TABLE_TYPE)
 				break;
@@ -248,14 +188,64 @@ static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
 		ucode     = data;
 	}
 
-	if (!eq_id) {
-		*cont = NULL;
-		*cont_sz = 0;
-		return;
-	}
+	return ret;
+}
+
+static int __apply_microcode_amd(struct microcode_amd *mc_amd)
+{
+	u32 rev, dummy;
+
+	native_wrmsrl(MSR_AMD64_PATCH_LOADER, (u64)(long)&mc_amd->hdr.data_code);
+
+	/* verify patch application was successful */
+	native_rdmsr(MSR_AMD64_PATCH_LEVEL, rev, dummy);
+	if (rev != mc_amd->hdr.patch_id)
+		return -1;
+
+	return 0;
+}
+
+/*
+ * Early load occurs before we can vmalloc(). So we look for the microcode
+ * patch container file in initrd, traverse equivalent cpu table, look for a
+ * matching microcode patch, and update, all in initrd memory in place.
+ * When vmalloc() is available for use later -- on 64-bit during first AP load,
+ * and on 32-bit during save_microcode_in_initrd_amd() -- we can call
+ * load_microcode_amd() to save equivalent cpu table and microcode patches in
+ * kernel heap memory.
+ */
+static struct container
+apply_microcode_early_amd(void *ucode, size_t size, bool save_patch)
+{
+	struct container ret = { NULL, 0 };
+	u8 (*patch)[PATCH_MAX_SIZE];
+	int offset, left;
+	u32 rev, *header;
+	u8  *data;
+	u16 eq_id = 0;
+	u32 *new_rev;
+
+#ifdef CONFIG_X86_32
+	new_rev = (u32 *)__pa_nodebug(&ucode_new_rev);
+	patch	= (u8 (*)[PATCH_MAX_SIZE])__pa_nodebug(&amd_ucode_patch);
+#else
+	new_rev = &ucode_new_rev;
+	patch	= &amd_ucode_patch;
+#endif
 
 	if (check_current_patch_level(&rev, true))
-		return;
+		return (struct container){ NULL, 0 };
+
+	ret = find_proper_container(ucode, size, &eq_id);
+	if (!eq_id)
+		return (struct container){ NULL, 0 };
+
+	this_equiv_id = eq_id;
+	header = (u32 *)ret.data;
+
+	/* We're pointing to an equiv table, skip over it. */
+	data = ret.data +  header[2] + CONTAINER_HDR_SZ;
+	left = ret.size - (header[2] + CONTAINER_HDR_SZ);
 
 	while (left > 0) {
 		struct microcode_amd *mc;
@@ -274,8 +264,7 @@ static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
 				*new_rev = rev;
 
 				if (save_patch)
-					memcpy(patch, mc,
-					       min_t(u32, header[1], PATCH_MAX_SIZE));
+					memcpy(patch, mc, min_t(u32, header[1], PATCH_MAX_SIZE));
 			}
 		}
 
@@ -283,10 +272,10 @@ static void apply_ucode_in_initrd(void *ucode, size_t size, bool save_patch)
 		data   += offset;
 		left   -= offset;
 	}
+	return ret;
 }
 
-static bool __init load_builtin_amd_microcode(struct cpio_data *cp,
-					      unsigned int family)
+static bool get_builtin_microcode(struct cpio_data *cp, unsigned int family)
 {
 #ifdef CONFIG_X86_64
 	char fw_name[36] = "amd-ucode/microcode_amd.bin";
@@ -303,32 +292,31 @@ static bool __init load_builtin_amd_microcode(struct cpio_data *cp,
 
 void __init load_ucode_amd_bsp(unsigned int family)
 {
+	struct ucode_cpu_info *uci;
 	struct cpio_data cp;
-	bool *builtin;
-	void **data;
-	size_t *size;
+	const char *path;
+	bool use_pa;
 
-#ifdef CONFIG_X86_32
-	data =  (void **)__pa_nodebug(&ucode_cpio.data);
-	size = (size_t *)__pa_nodebug(&ucode_cpio.size);
-	builtin = (bool *)__pa_nodebug(&ucode_builtin);
-#else
-	data = &ucode_cpio.data;
-	size = &ucode_cpio.size;
-	builtin = &ucode_builtin;
-#endif
+	if (IS_ENABLED(CONFIG_X86_32)) {
+		uci	= (struct ucode_cpu_info *)__pa_nodebug(ucode_cpu_info);
+		path	= (const char *)__pa_nodebug(ucode_path);
+		use_pa	= true;
+	} else {
+		uci     = ucode_cpu_info;
+		path	= ucode_path;
+		use_pa	= false;
+	}
 
-	*builtin = load_builtin_amd_microcode(&cp, family);
-	if (!*builtin)
-		cp = find_ucode_in_initrd();
+	if (!get_builtin_microcode(&cp, family))
+		cp = find_microcode_in_initrd(path, use_pa);
 
 	if (!(cp.data && cp.size))
 		return;
 
-	*data = cp.data;
-	*size = cp.size;
+	/* Get BSP's CPUID.EAX(1), needed in load_microcode_amd() */
+	uci->cpu_sig.sig = cpuid_eax(1);
 
-	apply_ucode_in_initrd(cp.data, cp.size, true);
+	apply_microcode_early_amd(cp.data, cp.size, true);
 }
 
 #ifdef CONFIG_X86_32
@@ -342,8 +330,7 @@ void __init load_ucode_amd_bsp(unsigned int family)
 void load_ucode_amd_ap(unsigned int family)
 {
 	struct microcode_amd *mc;
-	size_t *usize;
-	void **ucode;
+	struct cpio_data cp;
 
 	mc = (struct microcode_amd *)__pa_nodebug(amd_ucode_patch);
 	if (mc->hdr.patch_id && mc->hdr.processor_rev_id) {
@@ -351,55 +338,63 @@ void load_ucode_amd_ap(unsigned int family)
 		return;
 	}
 
-	ucode = (void *)__pa_nodebug(&container);
-	usize = (size_t *)__pa_nodebug(&container_size);
+	if (!get_builtin_microcode(&cp, family))
+		cp = find_microcode_in_initrd((const char *)__pa_nodebug(ucode_path), true);
 
-	if (!*ucode || !*usize)
+	if (!(cp.data && cp.size))
 		return;
 
-	apply_ucode_in_initrd(*ucode, *usize, false);
-}
-
-static void __init collect_cpu_sig_on_bsp(void *arg)
-{
-	unsigned int cpu = smp_processor_id();
-	struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
-
-	uci->cpu_sig.sig = cpuid_eax(0x00000001);
-}
-
-static void __init get_bsp_sig(void)
-{
-	unsigned int bsp = boot_cpu_data.cpu_index;
-	struct ucode_cpu_info *uci = ucode_cpu_info + bsp;
-
-	if (!uci->cpu_sig.sig)
-		smp_call_function_single(bsp, collect_cpu_sig_on_bsp, NULL, 1);
+	/*
+	 * This would set amd_ucode_patch above so that the following APs can
+	 * use it directly instead of going down this path again.
+	 */
+	apply_microcode_early_amd(cp.data, cp.size, true);
 }
 #else
 void load_ucode_amd_ap(unsigned int family)
 {
 	struct equiv_cpu_entry *eq;
 	struct microcode_amd *mc;
-	u8 *cont = container;
 	u32 rev, eax;
 	u16 eq_id;
 
-	if (!container)
-		return;
-
-	/*
-	 * 64-bit runs with paging enabled, thus early==false.
-	 */
+	/* 64-bit runs with paging enabled, thus early==false. */
 	if (check_current_patch_level(&rev, false))
 		return;
 
-	/* Add CONFIG_RANDOMIZE_MEMORY offset. */
-	if (!ucode_builtin)
-		cont += PAGE_OFFSET - __PAGE_OFFSET_BASE;
+	/* First AP hasn't cached it yet, go through the blob. */
+	if (!cont.data) {
+		struct cpio_data cp = { NULL, 0, "" };
+
+		if (cont.size == -1)
+			return;
+
+reget:
+		if (!get_builtin_microcode(&cp, family)) {
+#ifdef CONFIG_BLK_DEV_INITRD
+			cp = find_cpio_data(ucode_path, (void *)initrd_start,
+					    initrd_end - initrd_start, NULL);
+#endif
+			if (!(cp.data && cp.size)) {
+				/*
+				 * Mark it so that other APs do not scan again
+				 * for no real reason and slow down boot
+				 * needlessly.
+				 */
+				cont.size = -1;
+				return;
+			}
+		}
+
+		cont = apply_microcode_early_amd(cp.data, cp.size, false);
+		if (!(cont.data && cont.size)) {
+			cont.size = -1;
+			return;
+		}
+	}
 
 	eax = cpuid_eax(0x00000001);
-	eq  = (struct equiv_cpu_entry *)(cont + CONTAINER_HDR_SZ);
+	eq  = (struct equiv_cpu_entry *)(cont.data + CONTAINER_HDR_SZ);
 
 	eq_id = find_equiv_id(eq, eax);
 	if (!eq_id)
@@ -414,64 +409,50 @@ void load_ucode_amd_ap(unsigned int family)
 		}
 
 	} else {
-		if (!ucode_cpio.data)
-			return;
 
 		/*
 		 * AP has a different equivalence ID than BSP, looks like
 		 * mixed-steppings silicon so go through the ucode blob anew.
 		 */
-		apply_ucode_in_initrd(ucode_cpio.data, ucode_cpio.size, false);
+		goto reget;
 	}
 }
-#endif
+#endif /* CONFIG_X86_32 */
 
 static enum ucode_state
 load_microcode_amd(int cpu, u8 family, const u8 *data, size_t size);
 
-int __init save_microcode_in_initrd_amd(unsigned int family)
+int __init save_microcode_in_initrd_amd(unsigned int fam)
 {
-	unsigned long cont;
-	int retval = 0;
 	enum ucode_state ret;
-	u8 *cont_va;
-	u32 eax;
+	int retval = 0;
+	u16 eq_id;
 
-	if (!container)
-		return -EINVAL;
+	if (!cont.data) {
+		if (IS_ENABLED(CONFIG_X86_32) && (cont.size != -1)) {
+			struct cpio_data cp = { NULL, 0, "" };
 
-#ifdef CONFIG_X86_32
-	get_bsp_sig();
-	cont	= (unsigned long)container;
-	cont_va = __va(container);
-#else
-	/*
-	 * We need the physical address of the container for both bitness since
-	 * boot_params.hdr.ramdisk_image is a physical address.
-	 */
-	cont    = __pa(container);
-	cont_va = container;
+#ifdef CONFIG_BLK_DEV_INITRD
+			cp = find_cpio_data(ucode_path, (void *)initrd_start,
+					    initrd_end - initrd_start, NULL);
 #endif
 
-	/*
-	 * Take into account the fact that the ramdisk might get relocated and
-	 * therefore we need to recompute the container's position in virtual
-	 * memory space.
-	 */
-	if (relocated_ramdisk)
-		container = (u8 *)(__va(relocated_ramdisk) +
-			     (cont - boot_params.hdr.ramdisk_image));
-	else
-		container = cont_va;
+			if (!(cp.data && cp.size)) {
+				cont.size = -1;
+				return -EINVAL;
+			}
 
-	/* Add CONFIG_RANDOMIZE_MEMORY offset. */
-	if (!ucode_builtin)
-		container += PAGE_OFFSET - __PAGE_OFFSET_BASE;
+			cont = find_proper_container(cp.data, cp.size, &eq_id);
+			if (!eq_id) {
+				cont.size = -1;
+				return -EINVAL;
+			}
 
-	eax   = cpuid_eax(0x00000001);
-	eax   = ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff);
+		} else
+			return -EINVAL;
+	}
 
-	ret = load_microcode_amd(smp_processor_id(), eax, container, container_size);
+	ret = load_microcode_amd(smp_processor_id(), fam, cont.data, cont.size);
 	if (ret != UCODE_OK)
 		retval = -EINVAL;
 
@@ -479,8 +460,8 @@ int __init save_microcode_in_initrd_amd(unsigned int family)
 	 * This will be freed any msec now, stash patches for the current
 	 * family and switch to patch cache for cpu hotplug, etc later.
 	 */
-	container = NULL;
-	container_size = 0;
+	cont.data = NULL;
+	cont.size = 0;
 
 	return retval;
 }
@@ -498,8 +479,10 @@ void reload_ucode_amd(void)
 		return;
 
 	mc = (struct microcode_amd *)amd_ucode_patch;
+	if (!mc)
+		return;
 
-	if (mc && rev < mc->hdr.patch_id) {
+	if (rev < mc->hdr.patch_id) {
 		if (!__apply_microcode_amd(mc)) {
 			ucode_new_rev = mc->hdr.patch_id;
 			pr_info("reload patch_level=0x%08x\n", ucode_new_rev);
diff --git a/arch/x86/kernel/cpu/microcode/core.c b/arch/x86/kernel/cpu/microcode/core.c
index f737039d59b2..dd3d64f07364 100644
--- a/arch/x86/kernel/cpu/microcode/core.c
+++ b/arch/x86/kernel/cpu/microcode/core.c
@@ -39,6 +39,7 @@
 #include <asm/microcode.h>
 #include <asm/processor.h>
 #include <asm/cmdline.h>
+#include <asm/setup.h>
 
 #define MICROCODE_VERSION	"2.01"
 
@@ -196,6 +197,58 @@ static int __init save_microcode_in_initrd(void)
 	return -EINVAL;
 }
 
+struct cpio_data find_microcode_in_initrd(const char *path, bool use_pa)
+{
+#ifdef CONFIG_BLK_DEV_INITRD
+	unsigned long start = 0;
+	size_t size;
+
+#ifdef CONFIG_X86_32
+	struct boot_params *params;
+
+	if (use_pa)
+		params = (struct boot_params *)__pa_nodebug(&boot_params);
+	else
+		params = &boot_params;
+
+	size = params->hdr.ramdisk_size;
+
+	/*
+	 * Set start only if we have an initrd image. We cannot use initrd_start
+	 * because it is not set that early yet.
+	 */
+	if (size)
+		start = params->hdr.ramdisk_image;
+
+# else /* CONFIG_X86_64 */
+	size  = (unsigned long)boot_params.ext_ramdisk_size << 32;
+	size |= boot_params.hdr.ramdisk_size;
+
+	if (size) {
+		start  = (unsigned long)boot_params.ext_ramdisk_image << 32;
+		start |= boot_params.hdr.ramdisk_image;
+
+		start += PAGE_OFFSET;
+	}
+# endif
+
+	/*
+	 * Did we relocate the ramdisk?
+	 *
+	 * So we possibly relocate the ramdisk *after* applying microcode on the
+	 * BSP so we rely on use_pa (use physical addresses) - even if it is not
+	 * absolutely correct - to determine whether we've done the ramdisk
+	 * relocation already.
+	 */
+	if (!use_pa && relocated_ramdisk)
+		start = initrd_start;
+
+	return find_cpio_data(path, (void *)start, size, NULL);
+#else /* !CONFIG_BLK_DEV_INITRD */
+	return (struct cpio_data){ NULL, 0, "" };
+#endif
+}
+
 void reload_early_microcode(void)
 {
 	int vendor, family;
@@ -455,7 +508,8 @@ static struct attribute_group mc_attr_group = {
 
 static void microcode_fini_cpu(int cpu)
 {
-	microcode_ops->microcode_fini_cpu(cpu);
+	if (microcode_ops->microcode_fini_cpu)
+		microcode_ops->microcode_fini_cpu(cpu);
 }
 
 static enum ucode_state microcode_resume_cpu(int cpu)
@@ -584,12 +638,7 @@ static int mc_cpu_down_prep(unsigned int cpu)
 	/* Suspend is in progress, only remove the interface */
 	sysfs_remove_group(&dev->kobj, &mc_attr_group);
 	pr_debug("CPU%d removed\n", cpu);
-	/*
-	 * When a CPU goes offline, don't free up or invalidate the copy of
-	 * the microcode in kernel memory, so that we can reuse it when the
-	 * CPU comes back online without unnecessarily requesting the userspace
-	 * for it again.
-	 */
+
 	return 0;
 }
 
diff --git a/arch/x86/kernel/cpu/microcode/intel.c b/arch/x86/kernel/cpu/microcode/intel.c
index 5e631d2cf37f..54d50c3694d8 100644
--- a/arch/x86/kernel/cpu/microcode/intel.c
+++ b/arch/x86/kernel/cpu/microcode/intel.c
@@ -39,27 +39,10 @@
 #include <asm/setup.h>
 #include <asm/msr.h>
 
-/*
- * Temporary microcode blobs pointers storage. We note here during early load
- * the pointers to microcode blobs we've got from whatever storage (detached
- * initrd, builtin). Later on, we put those into final storage
- * mc_saved_data.mc_saved.
- *
- * Important: those are offsets from the beginning of initrd or absolute
- * addresses within the kernel image when built-in.
- */
-static unsigned long mc_tmp_ptrs[MAX_UCODE_COUNT];
-
-static struct mc_saved_data {
-	unsigned int num_saved;
-	struct microcode_intel **mc_saved;
-} mc_saved_data;
+static const char ucode_path[] = "kernel/x86/microcode/GenuineIntel.bin";
 
-/* Microcode blobs within the initrd. 0 if builtin. */
-static struct ucode_blobs {
-	unsigned long start;
-	bool valid;
-} blobs;
+/* Current microcode patch used in early patching */
+struct microcode_intel *intel_ucode_patch;
 
 static inline bool cpu_signatures_match(unsigned int s1, unsigned int p1,
 					unsigned int s2, unsigned int p2)
@@ -116,103 +99,23 @@ static int has_newer_microcode(void *mc, unsigned int csig, int cpf, int new_rev
 	return find_matching_signature(mc, csig, cpf);
 }
 
-/* Go through saved patches and find the one suitable for the current CPU. */
-static enum ucode_state
-find_microcode_patch(struct microcode_intel **saved,
-		     unsigned int num_saved, struct ucode_cpu_info *uci)
-{
-	struct microcode_intel *ucode_ptr, *new_mc = NULL;
-	struct microcode_header_intel *mc_hdr;
-	int new_rev, ret, i;
-
-	new_rev = uci->cpu_sig.rev;
-
-	for (i = 0; i < num_saved; i++) {
-		ucode_ptr = saved[i];
-		mc_hdr	  = (struct microcode_header_intel *)ucode_ptr;
-
-		ret = has_newer_microcode(ucode_ptr,
-					  uci->cpu_sig.sig,
-					  uci->cpu_sig.pf,
-					  new_rev);
-		if (!ret)
-			continue;
-
-		new_rev = mc_hdr->rev;
-		new_mc  = ucode_ptr;
-	}
-
-	if (!new_mc)
-		return UCODE_NFOUND;
-
-	uci->mc = (struct microcode_intel *)new_mc;
-	return UCODE_OK;
-}
-
-static inline void
-copy_ptrs(struct microcode_intel **mc_saved, unsigned long *mc_ptrs,
-	  unsigned long off, int num_saved)
-{
-	int i;
-
-	for (i = 0; i < num_saved; i++)
-		mc_saved[i] = (struct microcode_intel *)(mc_ptrs[i] + off);
-}
-
-#ifdef CONFIG_X86_32
-static void
-microcode_phys(struct microcode_intel **mc_saved_tmp, struct mc_saved_data *mcs)
-{
-	int i;
-	struct microcode_intel ***mc_saved;
-
-	mc_saved = (struct microcode_intel ***)__pa_nodebug(&mcs->mc_saved);
-
-	for (i = 0; i < mcs->num_saved; i++) {
-		struct microcode_intel *p;
-
-		p = *(struct microcode_intel **)__pa_nodebug(mcs->mc_saved + i);
-		mc_saved_tmp[i] = (struct microcode_intel *)__pa_nodebug(p);
-	}
-}
-#endif
-
-static enum ucode_state
-load_microcode(struct mc_saved_data *mcs, unsigned long *mc_ptrs,
-	       unsigned long offset, struct ucode_cpu_info *uci)
-{
-	struct microcode_intel *mc_saved_tmp[MAX_UCODE_COUNT];
-	unsigned int count = mcs->num_saved;
-
-	if (!mcs->mc_saved) {
-		copy_ptrs(mc_saved_tmp, mc_ptrs, offset, count);
-
-		return find_microcode_patch(mc_saved_tmp, count, uci);
-	} else {
-#ifdef CONFIG_X86_32
-		microcode_phys(mc_saved_tmp, mcs);
-		return find_microcode_patch(mc_saved_tmp, count, uci);
-#else
-		return find_microcode_patch(mcs->mc_saved, count, uci);
-#endif
-	}
-}
-
 /*
  * Given CPU signature and a microcode patch, this function finds if the
  * microcode patch has matching family and model with the CPU.
+ *
+ * %true - if there's a match
+ * %false - otherwise
  */
-static enum ucode_state
-matching_model_microcode(struct microcode_header_intel *mc_header,
-			unsigned long sig)
+static bool microcode_matches(struct microcode_header_intel *mc_header,
+			      unsigned long sig)
 {
-	unsigned int fam, model;
-	unsigned int fam_ucode, model_ucode;
-	struct extended_sigtable *ext_header;
 	unsigned long total_size = get_totalsize(mc_header);
 	unsigned long data_size = get_datasize(mc_header);
-	int ext_sigcount, i;
+	struct extended_sigtable *ext_header;
+	unsigned int fam_ucode, model_ucode;
 	struct extended_signature *ext_sig;
+	unsigned int fam, model;
+	int ext_sigcount, i;
 
 	fam   = x86_family(sig);
 	model = x86_model(sig);
@@ -221,11 +124,11 @@ matching_model_microcode(struct microcode_header_intel *mc_header,
 	model_ucode = x86_model(mc_header->sig);
 
 	if (fam == fam_ucode && model == model_ucode)
-		return UCODE_OK;
+		return true;
 
 	/* Look for ext. headers: */
 	if (total_size <= data_size + MC_HEADER_SIZE)
-		return UCODE_NFOUND;
+		return false;
 
 	ext_header   = (void *) mc_header + data_size + MC_HEADER_SIZE;
 	ext_sig      = (void *)ext_header + EXT_HEADER_SIZE;
@@ -236,114 +139,69 @@ matching_model_microcode(struct microcode_header_intel *mc_header,
 		model_ucode = x86_model(ext_sig->sig);
 
 		if (fam == fam_ucode && model == model_ucode)
-			return UCODE_OK;
+			return true;
 
 		ext_sig++;
 	}
-	return UCODE_NFOUND;
+	return false;
 }
 
-static int
-save_microcode(struct mc_saved_data *mcs,
-	       struct microcode_intel **mc_saved_src,
-	       unsigned int num_saved)
+static struct ucode_patch *__alloc_microcode_buf(void *data, unsigned int size)
 {
-	int i, j;
-	struct microcode_intel **saved_ptr;
-	int ret;
-
-	if (!num_saved)
-		return -EINVAL;
-
-	/*
-	 * Copy new microcode data.
-	 */
-	saved_ptr = kcalloc(num_saved, sizeof(struct microcode_intel *), GFP_KERNEL);
-	if (!saved_ptr)
-		return -ENOMEM;
-
-	for (i = 0; i < num_saved; i++) {
-		struct microcode_header_intel *mc_hdr;
-		struct microcode_intel *mc;
-		unsigned long size;
-
-		if (!mc_saved_src[i]) {
-			ret = -EINVAL;
-			goto err;
-		}
+	struct ucode_patch *p;
 
-		mc     = mc_saved_src[i];
-		mc_hdr = &mc->hdr;
-		size   = get_totalsize(mc_hdr);
+	p = kzalloc(size, GFP_KERNEL);
+	if (!p)
+		return ERR_PTR(-ENOMEM);
 
-		saved_ptr[i] = kmemdup(mc, size, GFP_KERNEL);
-		if (!saved_ptr[i]) {
-			ret = -ENOMEM;
-			goto err;
-		}
+	p->data = kmemdup(data, size, GFP_KERNEL);
+	if (!p->data) {
+		kfree(p);
+		return ERR_PTR(-ENOMEM);
 	}
 
-	/*
-	 * Point to newly saved microcode.
-	 */
-	mcs->mc_saved  = saved_ptr;
-	mcs->num_saved = num_saved;
-
-	return 0;
-
-err:
-	for (j = 0; j <= i; j++)
-		kfree(saved_ptr[j]);
-	kfree(saved_ptr);
-
-	return ret;
+	return p;
 }
 
-/*
- * A microcode patch in ucode_ptr is saved into mc_saved
- * - if it has matching signature and newer revision compared to an existing
- *   patch mc_saved.
- * - or if it is a newly discovered microcode patch.
- *
- * The microcode patch should have matching model with CPU.
- *
- * Returns: The updated number @num_saved of saved microcode patches.
- */
-static unsigned int _save_mc(struct microcode_intel **mc_saved,
-			     u8 *ucode_ptr, unsigned int num_saved)
+static void save_microcode_patch(void *data, unsigned int size)
 {
 	struct microcode_header_intel *mc_hdr, *mc_saved_hdr;
+	struct ucode_patch *iter, *tmp, *p;
+	bool prev_found = false;
 	unsigned int sig, pf;
-	int found = 0, i;
 
-	mc_hdr = (struct microcode_header_intel *)ucode_ptr;
+	mc_hdr = (struct microcode_header_intel *)data;
 
-	for (i = 0; i < num_saved; i++) {
-		mc_saved_hdr = (struct microcode_header_intel *)mc_saved[i];
+	list_for_each_entry_safe(iter, tmp, &microcode_cache, plist) {
+		mc_saved_hdr = (struct microcode_header_intel *)iter->data;
 		sig	     = mc_saved_hdr->sig;
 		pf	     = mc_saved_hdr->pf;
 
-		if (!find_matching_signature(ucode_ptr, sig, pf))
-			continue;
-
-		found = 1;
+		if (find_matching_signature(data, sig, pf)) {
+			prev_found = true;
 
-		if (mc_hdr->rev <= mc_saved_hdr->rev)
-			continue;
+			if (mc_hdr->rev <= mc_saved_hdr->rev)
+				continue;
 
-		/*
-		 * Found an older ucode saved earlier. Replace it with
-		 * this newer one.
-		 */
-		mc_saved[i] = (struct microcode_intel *)ucode_ptr;
-		break;
+			p = __alloc_microcode_buf(data, size);
+			if (IS_ERR(p))
+				pr_err("Error allocating buffer %p\n", data);
+			else
+				list_replace(&iter->plist, &p->plist);
+		}
 	}
 
-	/* Newly detected microcode, save it to memory. */
-	if (i >= num_saved && !found)
-		mc_saved[num_saved++] = (struct microcode_intel *)ucode_ptr;
-
-	return num_saved;
+	/*
+	 * There weren't any previous patches found in the list cache; save the
+	 * newly found.
+	 */
+	if (!prev_found) {
+		p = __alloc_microcode_buf(data, size);
+		if (IS_ERR(p))
+			pr_err("Error allocating buffer for %p\n", data);
+		else
+			list_add_tail(&p->plist, &microcode_cache);
+	}
 }
 
 static int microcode_sanity_check(void *mc, int print_err)
@@ -448,78 +306,75 @@ static int microcode_sanity_check(void *mc, int print_err)
  * Get microcode matching with BSP's model. Only CPUs with the same model as
  * BSP can stay in the platform.
  */
-static enum ucode_state __init
-get_matching_model_microcode(unsigned long start, void *data, size_t size,
-			     struct mc_saved_data *mcs, unsigned long *mc_ptrs,
-			     struct ucode_cpu_info *uci)
+static struct microcode_intel *
+scan_microcode(void *data, size_t size, struct ucode_cpu_info *uci, bool save)
 {
-	struct microcode_intel *mc_saved_tmp[MAX_UCODE_COUNT];
 	struct microcode_header_intel *mc_header;
-	unsigned int num_saved = mcs->num_saved;
-	enum ucode_state state = UCODE_OK;
-	unsigned int leftover = size;
-	u8 *ucode_ptr = data;
+	struct microcode_intel *patch = NULL;
 	unsigned int mc_size;
-	int i;
-
-	while (leftover && num_saved < ARRAY_SIZE(mc_saved_tmp)) {
 
-		if (leftover < sizeof(mc_header))
+	while (size) {
+		if (size < sizeof(struct microcode_header_intel))
 			break;
 
-		mc_header = (struct microcode_header_intel *)ucode_ptr;
+		mc_header = (struct microcode_header_intel *)data;
 
 		mc_size = get_totalsize(mc_header);
-		if (!mc_size || mc_size > leftover ||
-			microcode_sanity_check(ucode_ptr, 0) < 0)
+		if (!mc_size ||
+		    mc_size > size ||
+		    microcode_sanity_check(data, 0) < 0)
 			break;
 
-		leftover -= mc_size;
+		size -= mc_size;
 
-		/*
-		 * Since APs with same family and model as the BSP may boot in
-		 * the platform, we need to find and save microcode patches
-		 * with the same family and model as the BSP.
-		 */
-		if (matching_model_microcode(mc_header, uci->cpu_sig.sig) != UCODE_OK) {
-			ucode_ptr += mc_size;
+		if (!microcode_matches(mc_header, uci->cpu_sig.sig)) {
+			data += mc_size;
 			continue;
 		}
 
-		num_saved = _save_mc(mc_saved_tmp, ucode_ptr, num_saved);
+		if (save) {
+			save_microcode_patch(data, mc_size);
+			goto next;
+		}
 
-		ucode_ptr += mc_size;
-	}
 
-	if (leftover) {
-		state = UCODE_ERROR;
-		return state;
-	}
+		if (!patch) {
+			if (!has_newer_microcode(data,
+						 uci->cpu_sig.sig,
+						 uci->cpu_sig.pf,
+						 uci->cpu_sig.rev))
+				goto next;
 
-	if (!num_saved) {
-		state = UCODE_NFOUND;
-		return state;
-	}
+		} else {
+			struct microcode_header_intel *phdr = &patch->hdr;
+
+			if (!has_newer_microcode(data,
+						 phdr->sig,
+						 phdr->pf,
+						 phdr->rev))
+				goto next;
+		}
 
-	for (i = 0; i < num_saved; i++)
-		mc_ptrs[i] = (unsigned long)mc_saved_tmp[i] - start;
+		/* We have a newer patch, save it. */
+		patch = data;
 
-	mcs->num_saved = num_saved;
+next:
+		data += mc_size;
+	}
 
-	return state;
+	if (size)
+		return NULL;
+
+	return patch;
 }
 
 static int collect_cpu_info_early(struct ucode_cpu_info *uci)
 {
 	unsigned int val[2];
 	unsigned int family, model;
-	struct cpu_signature csig;
+	struct cpu_signature csig = { 0 };
 	unsigned int eax, ebx, ecx, edx;
 
-	csig.sig = 0;
-	csig.pf = 0;
-	csig.rev = 0;
-
 	memset(uci, 0, sizeof(*uci));
 
 	eax = 0x00000001;
@@ -527,8 +382,8 @@ static int collect_cpu_info_early(struct ucode_cpu_info *uci)
 	native_cpuid(