Merge tag 'x86_microcode_for_v6.19_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 microcode loading updates from Borislav Petkov:

 - Add microcode staging support on Intel: it moves the sole microcode
   blobs loading to a non-critical path so that microcode loading
   latencies are kept at minimum. The actual "directing" the hardware to
   load microcode is the only step which is done on the critical path.

   This scheme is also opportunistic as in: on a failure, the machinery
   falls back to normal loading

 - Add the capability to the AMD side of the loader to select one of two
   per-family/model/stepping patches: one is pre-Entrysign and the other
   is post-Entrysign; with the goal to take care of machines which
   haven't updated their BIOS yet - something they should absolutely do
   as this is the only proper Entrysign fix

 - Other small cleanups and fixlets

* tag 'x86_microcode_for_v6.19_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/microcode: Mark early_parse_cmdline() as __init
  x86/microcode/AMD: Select which microcode patch to load
  x86/microcode/intel: Enable staging when available
  x86/microcode/intel: Support mailbox transfer
  x86/microcode/intel: Implement staging handler
  x86/microcode/intel: Define staging state struct
  x86/microcode/intel: Establish staging control logic
  x86/microcode: Introduce staging step to reduce late-loading time
  x86/cpu/topology: Make primary thread mask available with SMP=n

9 files changed, 468 insertions, 56 deletions

diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 65cc528fbad8..3d0a0950d20a 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -166,6 +166,10 @@
 						 * Processor MMIO stale data
 						 * vulnerabilities.
 						 */
+#define ARCH_CAP_MCU_ENUM		BIT(16) /*
+						 * Indicates the presence of microcode update
+						 * feature enumeration and status information.
+						 */
 #define ARCH_CAP_FB_CLEAR		BIT(17)	/*
 						 * VERW clears CPU fill buffer
 						 * even on MDS_NO CPUs.
@@ -949,6 +953,10 @@
 #define MSR_IA32_APICBASE_BASE		(0xfffff<<12)
 
 #define MSR_IA32_UCODE_WRITE		0x00000079
+
+#define MSR_IA32_MCU_ENUMERATION	0x0000007b
+#define MCU_STAGING			BIT(4)
+
 #define MSR_IA32_UCODE_REV		0x0000008b
 
 /* Intel SGX Launch Enclave Public Key Hash MSRs */
@@ -1246,6 +1254,8 @@
 #define MSR_IA32_VMX_VMFUNC             0x00000491
 #define MSR_IA32_VMX_PROCBASED_CTLS3	0x00000492
 
+#define MSR_IA32_MCU_STAGING_MBOX_ADDR	0x000007a5
+
 /* Resctrl MSRs: */
 /* - Intel: */
 #define MSR_IA32_L3_QOS_CFG		0xc81
diff --git a/arch/x86/include/asm/topology.h b/arch/x86/include/asm/topology.h
index df943881f15b..1fadf0cf520c 100644
--- a/arch/x86/include/asm/topology.h
+++ b/arch/x86/include/asm/topology.h
@@ -218,6 +218,12 @@ static inline unsigned int topology_amd_nodes_per_pkg(void)
 	return __amd_nodes_per_pkg;
 }
 
+#else /* CONFIG_SMP */
+static inline int topology_phys_to_logical_pkg(unsigned int pkg) { return 0; }
+static inline int topology_max_smt_threads(void) { return 1; }
+static inline unsigned int topology_amd_nodes_per_pkg(void) { return 1; }
+#endif /* !CONFIG_SMP */
+
 extern struct cpumask __cpu_primary_thread_mask;
 #define cpu_primary_thread_mask ((const struct cpumask *)&__cpu_primary_thread_mask)
 
@@ -241,12 +247,6 @@ static inline bool topology_is_core_online(unsigned int cpu)
 }
 #define topology_is_core_online topology_is_core_online
 
-#else /* CONFIG_SMP */
-static inline int topology_phys_to_logical_pkg(unsigned int pkg) { return 0; }
-static inline int topology_max_smt_threads(void) { return 1; }
-static inline unsigned int topology_amd_nodes_per_pkg(void) { return 1; }
-#endif /* !CONFIG_SMP */
-
 static inline void arch_fix_phys_package_id(int num, u32 slot)
 {
 }
diff --git a/arch/x86/kernel/cpu/microcode/amd.c b/arch/x86/kernel/cpu/microcode/amd.c
index a881bf4c2011..3821a985f4ff 100644
--- a/arch/x86/kernel/cpu/microcode/amd.c
+++ b/arch/x86/kernel/cpu/microcode/amd.c
@@ -186,50 +186,61 @@ static u32 cpuid_to_ucode_rev(unsigned int val)
 	return p.ucode_rev;
 }
 
+static u32 get_cutoff_revision(u32 rev)
+{
+	switch (rev >> 8) {
+	case 0x80012: return 0x8001277; break;
+	case 0x80082: return 0x800820f; break;
+	case 0x83010: return 0x830107c; break;
+	case 0x86001: return 0x860010e; break;
+	case 0x86081: return 0x8608108; break;
+	case 0x87010: return 0x8701034; break;
+	case 0x8a000: return 0x8a0000a; break;
+	case 0xa0010: return 0xa00107a; break;
+	case 0xa0011: return 0xa0011da; break;
+	case 0xa0012: return 0xa001243; break;
+	case 0xa0082: return 0xa00820e; break;
+	case 0xa1011: return 0xa101153; break;
+	case 0xa1012: return 0xa10124e; break;
+	case 0xa1081: return 0xa108109; break;
+	case 0xa2010: return 0xa20102f; break;
+	case 0xa2012: return 0xa201212; break;
+	case 0xa4041: return 0xa404109; break;
+	case 0xa5000: return 0xa500013; break;
+	case 0xa6012: return 0xa60120a; break;
+	case 0xa7041: return 0xa704109; break;
+	case 0xa7052: return 0xa705208; break;
+	case 0xa7080: return 0xa708009; break;
+	case 0xa70c0: return 0xa70C009; break;
+	case 0xaa001: return 0xaa00116; break;
+	case 0xaa002: return 0xaa00218; break;
+	case 0xb0021: return 0xb002146; break;
+	case 0xb0081: return 0xb008111; break;
+	case 0xb1010: return 0xb101046; break;
+	case 0xb2040: return 0xb204031; break;
+	case 0xb4040: return 0xb404031; break;
+	case 0xb4041: return 0xb404101; break;
+	case 0xb6000: return 0xb600031; break;
+	case 0xb6080: return 0xb608031; break;
+	case 0xb7000: return 0xb700031; break;
+	default: break;
+
+	}
+	return 0;
+}
+
 static bool need_sha_check(u32 cur_rev)
 {
+	u32 cutoff;
+
 	if (!cur_rev) {
 		cur_rev = cpuid_to_ucode_rev(bsp_cpuid_1_eax);
 		pr_info_once("No current revision, generating the lowest one: 0x%x\n", cur_rev);
 	}
 
-	switch (cur_rev >> 8) {
-	case 0x80012: return cur_rev <= 0x8001277; break;
-	case 0x80082: return cur_rev <= 0x800820f; break;
-	case 0x83010: return cur_rev <= 0x830107c; break;
-	case 0x86001: return cur_rev <= 0x860010e; break;
-	case 0x86081: return cur_rev <= 0x8608108; break;
-	case 0x87010: return cur_rev <= 0x8701034; break;
-	case 0x8a000: return cur_rev <= 0x8a0000a; break;
-	case 0xa0010: return cur_rev <= 0xa00107a; break;
-	case 0xa0011: return cur_rev <= 0xa0011da; break;
-	case 0xa0012: return cur_rev <= 0xa001243; break;
-	case 0xa0082: return cur_rev <= 0xa00820e; break;
-	case 0xa1011: return cur_rev <= 0xa101153; break;
-	case 0xa1012: return cur_rev <= 0xa10124e; break;
-	case 0xa1081: return cur_rev <= 0xa108109; break;
-	case 0xa2010: return cur_rev <= 0xa20102f; break;
-	case 0xa2012: return cur_rev <= 0xa201212; break;
-	case 0xa4041: return cur_rev <= 0xa404109; break;
-	case 0xa5000: return cur_rev <= 0xa500013; break;
-	case 0xa6012: return cur_rev <= 0xa60120a; break;
-	case 0xa7041: return cur_rev <= 0xa704109; break;
-	case 0xa7052: return cur_rev <= 0xa705208; break;
-	case 0xa7080: return cur_rev <= 0xa708009; break;
-	case 0xa70c0: return cur_rev <= 0xa70C009; break;
-	case 0xaa001: return cur_rev <= 0xaa00116; break;
-	case 0xaa002: return cur_rev <= 0xaa00218; break;
-	case 0xb0021: return cur_rev <= 0xb002146; break;
-	case 0xb0081: return cur_rev <= 0xb008111; break;
-	case 0xb1010: return cur_rev <= 0xb101046; break;
-	case 0xb2040: return cur_rev <= 0xb204031; break;
-	case 0xb4040: return cur_rev <= 0xb404031; break;
-	case 0xb4041: return cur_rev <= 0xb404101; break;
-	case 0xb6000: return cur_rev <= 0xb600031; break;
-	case 0xb6080: return cur_rev <= 0xb608031; break;
-	case 0xb7000: return cur_rev <= 0xb700031; break;
-	default: break;
-	}
+	cutoff = get_cutoff_revision(cur_rev);
+	if (cutoff)
+		return cur_rev <= cutoff;
 
 	pr_info("You should not be seeing this. Please send the following couple of lines to x86-<at>-kernel.org\n");
 	pr_info("CPUID(1).EAX: 0x%x, current revision: 0x%x\n", bsp_cpuid_1_eax, cur_rev);
@@ -494,6 +505,7 @@ static int verify_patch(const u8 *buf, size_t buf_size, u32 *patch_size)
 {
 	u8 family = x86_family(bsp_cpuid_1_eax);
 	struct microcode_header_amd *mc_hdr;
+	u32 cur_rev, cutoff, patch_rev;
 	u32 sh_psize;
 	u16 proc_id;
 	u8 patch_fam;
@@ -533,11 +545,32 @@ static int verify_patch(const u8 *buf, size_t buf_size, u32 *patch_size)
 	proc_id	= mc_hdr->processor_rev_id;
 	patch_fam = 0xf + (proc_id >> 12);
 
-	ucode_dbg("Patch-ID 0x%08x: family: 0x%x\n", mc_hdr->patch_id, patch_fam);
-
 	if (patch_fam != family)
 		return 1;
 
+	cur_rev = get_patch_level();
+
+	/* No cutoff revision means old/unaffected by signing algorithm weakness => matches */
+	cutoff = get_cutoff_revision(cur_rev);
+	if (!cutoff)
+		goto ok;
+
+	patch_rev = mc_hdr->patch_id;
+
+	ucode_dbg("cur_rev: 0x%x, cutoff: 0x%x, patch_rev: 0x%x\n",
+		  cur_rev, cutoff, patch_rev);
+
+	if (cur_rev <= cutoff && patch_rev <= cutoff)
+		goto ok;
+
+	if (cur_rev > cutoff && patch_rev > cutoff)
+		goto ok;
+
+	return 1;
+
+ok:
+	ucode_dbg("Patch-ID 0x%08x: family: 0x%x\n", mc_hdr->patch_id, patch_fam);
+
 	return 0;
 }
 
@@ -606,8 +639,6 @@ static size_t parse_container(u8 *ucode, size_t size, struct cont_desc *desc)
 
 		mc = (struct microcode_amd *)(buf + SECTION_HDR_SIZE);
 
-		ucode_dbg("patch_id: 0x%x\n", mc->hdr.patch_id);
-
 		if (mc_patch_matches(mc, eq_id)) {
 			desc->psize = patch_size;
 			desc->mc = mc;
diff --git a/arch/x86/kernel/cpu/microcode/core.c b/arch/x86/kernel/cpu/microcode/core.c
index f75c140906d0..ccc83b0bf63c 100644
--- a/arch/x86/kernel/cpu/microcode/core.c
+++ b/arch/x86/kernel/cpu/microcode/core.c
@@ -136,7 +136,7 @@ bool __init microcode_loader_disabled(void)
 	return dis_ucode_ldr;
 }
 
-static void early_parse_cmdline(void)
+static void __init early_parse_cmdline(void)
 {
 	char cmd_buf[64] = {};
 	char *s, *p = cmd_buf;
@@ -589,6 +589,17 @@ static int load_late_stop_cpus(bool is_safe)
 		pr_err("You should switch to early loading, if possible.\n");
 	}
 
+	/*
+	 * Pre-load the microcode image into a staging device. This
+	 * process is preemptible and does not require stopping CPUs.
+	 * Successful staging simplifies the subsequent late-loading
+	 * process, reducing rendezvous time.
+	 *
+	 * Even if the transfer fails, the update will proceed as usual.
+	 */
+	if (microcode_ops->use_staging)
+		microcode_ops->stage_microcode();
+
 	atomic_set(&late_cpus_in, num_online_cpus());
 	atomic_set(&offline_in_nmi, 0);
 	loops_per_usec = loops_per_jiffy / (TICK_NSEC / 1000);
diff --git a/arch/x86/kernel/cpu/microcode/intel.c b/arch/x86/kernel/cpu/microcode/intel.c
index 371ca6eac00e..8744f3adc2a0 100644
--- a/arch/x86/kernel/cpu/microcode/intel.c
+++ b/arch/x86/kernel/cpu/microcode/intel.c
@@ -13,12 +13,15 @@
 #define pr_fmt(fmt) "microcode: " fmt
 #include <linux/earlycpio.h>
 #include <linux/firmware.h>
+#include <linux/pci_ids.h>
 #include <linux/uaccess.h>
 #include <linux/initrd.h>
 #include <linux/kernel.h>
+#include <linux/delay.h>
 #include <linux/slab.h>
 #include <linux/cpu.h>
 #include <linux/uio.h>
+#include <linux/io.h>
 #include <linux/mm.h>
 
 #include <asm/cpu_device_id.h>
@@ -33,6 +36,38 @@ static const char ucode_path[] = "kernel/x86/microcode/GenuineIntel.bin";
 
 #define UCODE_BSP_LOADED	((struct microcode_intel *)0x1UL)
 
+/* Defines for the microcode staging mailbox interface */
+#define MBOX_REG_NUM		4
+#define MBOX_REG_SIZE		sizeof(u32)
+
+#define MBOX_CONTROL_OFFSET	0x0
+#define MBOX_STATUS_OFFSET	0x4
+#define MBOX_WRDATA_OFFSET	0x8
+#define MBOX_RDDATA_OFFSET	0xc
+
+#define MASK_MBOX_CTRL_ABORT	BIT(0)
+#define MASK_MBOX_CTRL_GO	BIT(31)
+
+#define MASK_MBOX_STATUS_ERROR	BIT(2)
+#define MASK_MBOX_STATUS_READY	BIT(31)
+
+#define MASK_MBOX_RESP_SUCCESS	BIT(0)
+#define MASK_MBOX_RESP_PROGRESS	BIT(1)
+#define MASK_MBOX_RESP_ERROR	BIT(2)
+
+#define MBOX_CMD_LOAD		0x3
+#define MBOX_OBJ_STAGING	0xb
+#define MBOX_HEADER(size)	((PCI_VENDOR_ID_INTEL)    | \
+				 (MBOX_OBJ_STAGING << 16) | \
+				 ((u64)((size) / sizeof(u32)) << 32))
+
+/* The size of each mailbox header */
+#define MBOX_HEADER_SIZE	sizeof(u64)
+/* The size of staging hardware response */
+#define MBOX_RESPONSE_SIZE	sizeof(u64)
+
+#define MBOX_XACTION_TIMEOUT_MS	(10 * MSEC_PER_SEC)
+
 /* Current microcode patch used in early patching on the APs. */
 static struct microcode_intel *ucode_patch_va __read_mostly;
 static struct microcode_intel *ucode_patch_late __read_mostly;
@@ -54,6 +89,23 @@ struct extended_sigtable {
 	struct extended_signature	sigs[];
 };
 
+/**
+ * struct staging_state - Track the current staging process state
+ *
+ * @mmio_base:		MMIO base address for staging
+ * @ucode_len:		Total size of the microcode image
+ * @chunk_size:		Size of each data piece
+ * @bytes_sent:		Total bytes transmitted so far
+ * @offset:		Current offset in the microcode image
+ */
+struct staging_state {
+	void __iomem		*mmio_base;
+	unsigned int		ucode_len;
+	unsigned int		chunk_size;
+	unsigned int		bytes_sent;
+	unsigned int		offset;
+};
+
 #define DEFAULT_UCODE_TOTALSIZE (DEFAULT_UCODE_DATASIZE + MC_HEADER_SIZE)
 #define EXT_HEADER_SIZE		(sizeof(struct extended_sigtable))
 #define EXT_SIGNATURE_SIZE	(sizeof(struct extended_signature))
@@ -299,6 +351,298 @@ static __init struct microcode_intel *scan_microcode(void *data, size_t size,
 	return size ? NULL : patch;
 }
 
+static inline u32 read_mbox_dword(void __iomem *mmio_base)
+{
+	u32 dword = readl(mmio_base + MBOX_RDDATA_OFFSET);
+
+	/* Acknowledge read completion to the staging hardware */
+	writel(0, mmio_base + MBOX_RDDATA_OFFSET);
+	return dword;
+}
+
+static inline void write_mbox_dword(void __iomem *mmio_base, u32 dword)
+{
+	writel(dword, mmio_base + MBOX_WRDATA_OFFSET);
+}
+
+static inline u64 read_mbox_header(void __iomem *mmio_base)
+{
+	u32 high, low;
+
+	low  = read_mbox_dword(mmio_base);
+	high = read_mbox_dword(mmio_base);
+
+	return ((u64)high << 32) | low;
+}
+
+static inline void write_mbox_header(void __iomem *mmio_base, u64 value)
+{
+	write_mbox_dword(mmio_base, value);
+	write_mbox_dword(mmio_base, value >> 32);
+}
+
+static void write_mbox_data(void __iomem *mmio_base, u32 *chunk, unsigned int chunk_bytes)
+{
+	int i;
+
+	/*
+	 * The MMIO space is mapped as Uncached (UC). Each write arrives
+	 * at the device as an individual transaction in program order.
+	 * The device can then reassemble the sequence accordingly.
+	 */
+	for (i = 0; i < chunk_bytes / sizeof(u32); i++)
+		write_mbox_dword(mmio_base, chunk[i]);
+}
+
+/*
+ * Prepare for a new microcode transfer: reset hardware and record the
+ * image size.
+ */
+static void init_stage(struct staging_state *ss)
+{
+	ss->ucode_len = get_totalsize(&ucode_patch_late->hdr);
+
+	/*
+	 * Abort any ongoing process, effectively resetting the device.
+	 * Unlike regular mailbox data processing requests, this
+	 * operation does not require a status check.
+	 */
+	writel(MASK_MBOX_CTRL_ABORT, ss->mmio_base + MBOX_CONTROL_OFFSET);
+}
+
+/*
+ * Update the chunk size and decide whether another chunk can be sent.
+ * This accounts for remaining data and retry limits.
+ */
+static bool can_send_next_chunk(struct staging_state *ss, int *err)
+{
+	/* A page size or remaining bytes if this is the final chunk */
+	ss->chunk_size = min(PAGE_SIZE, ss->ucode_len - ss->offset);
+
+	/*
+	 * Each microcode image is divided into chunks, each at most
+	 * one page size. A 10-chunk image would typically require 10
+	 * transactions.
+	 *
+	 * However, the hardware managing the mailbox has limited
+	 * resources and may not cache the entire image, potentially
+	 * requesting the same chunk multiple times.
+	 *
+	 * To tolerate this behavior, allow up to twice the expected
+	 * number of transactions (i.e., a 10-chunk image can take up to
+	 * 20 attempts).
+	 *
+	 * If the number of attempts exceeds this limit, treat it as
+	 * exceeding the maximum allowed transfer size.
+	 */
+	if (ss->bytes_sent + ss->chunk_size > ss->ucode_len * 2) {
+		*err = -EMSGSIZE;
+		return false;
+	}
+
+	*err = 0;
+	return true;
+}
+
+/*
+ * The hardware indicates completion by returning a sentinel end offset.
+ */
+static inline bool is_end_offset(u32 offset)
+{
+	return offset == UINT_MAX;
+}
+
+/*
+ * Determine whether staging is complete: either the hardware signaled
+ * the end offset, or no more transactions are permitted (retry limit
+ * reached).
+ */
+static inline bool staging_is_complete(struct staging_state *ss, int *err)
+{
+	return is_end_offset(ss->offset) || !can_send_next_chunk(ss, err);
+}
+
+/*
+ * Wait for the hardware to complete a transaction.
+ * Return 0 on success, or an error code on failure.
+ */
+static int wait_for_transaction(struct staging_state *ss)
+{
+	u32 timeout, status;
+
+	/* Allow time for hardware to complete the operation: */
+	for (timeout = 0; timeout < MBOX_XACTION_TIMEOUT_MS; timeout++) {
+		msleep(1);
+
+		status = readl(ss->mmio_base + MBOX_STATUS_OFFSET);
+		/* Break out early if the hardware is ready: */
+		if (status & MASK_MBOX_STATUS_READY)
+			break;
+	}
+
+	/* Check for explicit error response */
+	if (status & MASK_MBOX_STATUS_ERROR)
+		return -EIO;
+
+	/*
+	 * Hardware has neither responded to the action nor signaled any
+	 * error. Treat this as a timeout.
+	 */
+	if (!(status & MASK_MBOX_STATUS_READY))
+		return -ETIMEDOUT;
+
+	return 0;
+}
+
+/*
+ * Transmit a chunk of the microcode image to the hardware.
+ * Return 0 on success, or an error code on failure.
+ */
+static int send_data_chunk(struct staging_state *ss, void *ucode_ptr)
+{
+	u32 *src_chunk = ucode_ptr + ss->offset;
+	u16 mbox_size;
+
+	/*
+	 * Write a 'request' mailbox object in this order:
+	 *  1. Mailbox header includes total size
+	 *  2. Command header specifies the load operation
+	 *  3. Data section contains a microcode chunk
+	 *
+	 * Thus, the mailbox size is two headers plus the chunk size.
+	 */
+	mbox_size = MBOX_HEADER_SIZE * 2 + ss->chunk_size;
+	write_mbox_header(ss->mmio_base, MBOX_HEADER(mbox_size));
+	write_mbox_header(ss->mmio_base, MBOX_CMD_LOAD);
+	write_mbox_data(ss->mmio_base, src_chunk, ss->chunk_size);
+	ss->bytes_sent += ss->chunk_size;
+
+	/* Notify the hardware that the mailbox is ready for processing. */
+	writel(MASK_MBOX_CTRL_GO, ss->mmio_base + MBOX_CONTROL_OFFSET);
+
+	return wait_for_transaction(ss);
+}
+
+/*
+ * Retrieve the next offset from the hardware response.
+ * Return 0 on success, or an error code on failure.
+ */
+static int fetch_next_offset(struct staging_state *ss)
+{
+	const u64 expected_header = MBOX_HEADER(MBOX_HEADER_SIZE + MBOX_RESPONSE_SIZE);
+	u32 offset, status;
+	u64 header;
+
+	/*
+	 * The 'response' mailbox returns three fields, in order:
+	 *  1. Header
+	 *  2. Next offset in the microcode image
+	 *  3. Status flags
+	 */
+	header = read_mbox_header(ss->mmio_base);
+	offset = read_mbox_dword(ss->mmio_base);
+	status = read_mbox_dword(ss->mmio_base);
+
+	/* All valid responses must start with the expected header. */
+	if (header != expected_header) {
+		pr_err_once("staging: invalid response header (0x%llx)\n", header);
+		return -EBADR;
+	}
+
+	/*
+	 * Verify the offset: If not at the end marker, it must not
+	 * exceed the microcode image length.
+	 */
+	if (!is_end_offset(offset) && offset > ss->ucode_len) {
+		pr_err_once("staging: invalid offset (%u) past the image end (%u)\n",
+			    offset, ss->ucode_len);
+		return -EINVAL;
+	}
+
+	/* Hardware may report errors explicitly in the status field */
+	if (status & MASK_MBOX_RESP_ERROR)
+		return -EPROTO;
+
+	ss->offset = offset;
+	return 0;
+}
+
+/*
+ * Handle the staging process using the mailbox MMIO interface. The
+ * microcode image is transferred in chunks until completion.
+ * Return 0 on success or an error code on failure.
+ */
+static int do_stage(u64 mmio_pa)
+{
+	struct staging_state ss = {};
+	int err;
+
+	ss.mmio_base = ioremap(mmio_pa, MBOX_REG_NUM * MBOX_REG_SIZE);
+	if (WARN_ON_ONCE(!ss.mmio_base))
+		return -EADDRNOTAVAIL;
+
+	init_stage(&ss);
+
+	/* Perform the staging process while within the retry limit */
+	while (!staging_is_complete(&ss, &err)) {
+		/* Send a chunk of microcode each time: */
+		err = send_data_chunk(&ss, ucode_patch_late);
+		if (err)
+			break;
+		/*
+		 * Then, ask the hardware which piece of the image it
+		 * needs next. The same piece may be sent more than once.
+		 */
+		err = fetch_next_offset(&ss);
+		if (err)
+			break;
+	}
+
+	iounmap(ss.mmio_base);
+
+	return err;
+}
+
+static void stage_microcode(void)
+{
+	unsigned int pkg_id = UINT_MAX;
+	int cpu, err;
+	u64 mmio_pa;
+
+	if (!IS_ALIGNED(get_totalsize(&ucode_patch_late->hdr), sizeof(u32))) {
+		pr_err("Microcode image 32-bit misaligned (0x%x), staging failed.\n",
+			get_totalsize(&ucode_patch_late->hdr));
+		return;
+	}
+
+	lockdep_assert_cpus_held();
+
+	/*
+	 * The MMIO address is unique per package, and all the SMT
+	 * primary threads are online here. Find each MMIO space by
+	 * their package IDs to avoid duplicate staging.
+	 */
+	for_each_cpu(cpu, cpu_primary_thread_mask) {
+		if (topology_logical_package_id(cpu) == pkg_id)
+			continue;
+
+		pkg_id = topology_logical_package_id(cpu);
+
+		err = rdmsrq_on_cpu(cpu, MSR_IA32_MCU_STAGING_MBOX_ADDR, &mmio_pa);
+		if (WARN_ON_ONCE(err))
+			return;
+
+		err = do_stage(mmio_pa);
+		if (err) {
+			pr_err("Error: staging failed (%d) for CPU%d at package %u.\n",
+			       err, cpu, pkg_id);
+			return;
+		}
+	}
+
+	pr_info("Staging of patch revision 0x%x succeeded.\n", ucode_patch_late->hdr.rev);
+}
+
 static enum ucode_state __apply_microcode(struct ucode_cpu_info *uci,
 					  struct microcode_intel *mc,
 					  u32 *cur_rev)
@@ -627,6 +971,7 @@ static struct microcode_ops microcode_intel_ops = {
 	.collect_cpu_info	= collect_cpu_info,
 	.apply_microcode	= apply_microcode_late,
 	.finalize_late_load	= finalize_late_load,
+	.stage_microcode	= stage_microcode,
 	.use_nmi		= IS_ENABLED(CONFIG_X86_64),
 };
 
@@ -638,6 +983,18 @@ static __init void calc_llc_size_per_core(struct cpuinfo_x86 *c)
 	llc_size_per_core = (unsigned int)llc_size;
 }
 
+static __init bool staging_available(void)
+{
+	u64 val;
+
+	val = x86_read_arch_cap_msr();
+	if (!(val & ARCH_CAP_MCU_ENUM))
+		return false;
+
+	rdmsrq(MSR_IA32_MCU_ENUMERATION, val);
+	return !!(val & MCU_STAGING);
+}
+
 struct microcode_ops * __init init_intel_microcode(void)
 {
 	struct cpuinfo_x86 *c = &boot_cpu_data;
@@ -648,6 +1005,11 @@ struct microcode_ops * __init init_intel_microcode(void)
 		return NULL;
 	}
 
+	if (staging_available()) {
+		microcode_intel_ops.use_staging = true;
+		pr_info("Enabled staging feature.\n");
+	}
+
 	calc_llc_size_per_core(c);
 
 	return &microcode_intel_ops;
diff --git a/arch/x86/kernel/cpu/microcode/internal.h b/arch/x86/kernel/cpu/microcode/internal.h
index ae8dbc2b908d..a10b547eda1e 100644
--- a/arch/x86/kernel/cpu/microcode/internal.h
+++ b/arch/x86/kernel/cpu/microcode/internal.h
@@ -31,10 +31,12 @@ struct microcode_ops {
 	 * See also the "Synchronization" section in microcode_core.c.
 	 */
 	enum ucode_state	(*apply_microcode)(int cpu);
+	void			(*stage_microcode)(void);
 	int			(*collect_cpu_info)(int cpu, struct cpu_signature *csig);
 	void			(*finalize_late_load)(int result);
 	unsigned int		nmi_safe	: 1,
-				use_nmi		: 1;
+				use_nmi		: 1,
+				use_staging	: 1;
 };
 
 struct early_load_data {
diff --git a/arch/x86/kernel/cpu/topology.c b/arch/x86/kernel/cpu/topology.c
index 6073a16628f9..f55ea3cdbf88 100644
--- a/arch/x86/kernel/cpu/topology.c
+++ b/arch/x86/kernel/cpu/topology.c
@@ -75,15 +75,11 @@ bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
 	return phys_id == (u64)cpuid_to_apicid[cpu];
 }
 
-#ifdef CONFIG_SMP
 static void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid)
 {
 	if (!(apicid & (__max_threads_per_core - 1)))
 		cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
 }
-#else
-static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
-#endif
 
 /*
  * Convert the APIC ID to a domain level ID by masking out the low bits
diff --git a/arch/x86/kernel/cpu/topology_common.c b/arch/x86/kernel/cpu/topology_common.c
index b5a5e1411469..71625795d711 100644
--- a/arch/x86/kernel/cpu/topology_common.c
+++ b/arch/x86/kernel/cpu/topology_common.c
@@ -16,6 +16,9 @@ EXPORT_SYMBOL_GPL(x86_topo_system);
 unsigned int __amd_nodes_per_pkg __ro_after_init;
 EXPORT_SYMBOL_GPL(__amd_nodes_per_pkg);
 
+/* CPUs which are the primary SMT threads */
+struct cpumask __cpu_primary_thread_mask __read_mostly;
+
 void topology_set_dom(struct topo_scan *tscan, enum x86_topology_domains dom,
 		      unsigned int shift, unsigned int ncpus)
 {
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index c2107047dc14..5cd6950ab672 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -103,9 +103,6 @@ EXPORT_PER_CPU_SYMBOL(cpu_core_map);
 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_die_map);
 EXPORT_PER_CPU_SYMBOL(cpu_die_map);
 
-/* CPUs which are the primary SMT threads */
-struct cpumask __cpu_primary_thread_mask __read_mostly;
-
 /* Representing CPUs for which sibling maps can be computed */
 static cpumask_var_t cpu_sibling_setup_mask;