[PATCH 1/8] x86, microcode, intel: forbid some incorrect metadata

[Henrique de Moraes Holschuh <hmh@hmh.eng.br>]

The Intel SDM vol 3A, section 9.11.1, and also table 9-6, requires that
the Data Size field be a multiple of 4 bytes.  All of the microcode
update header structures are dword-based, so the Total Size field must
also be a multiple of the dword size.

Ensure that data_size is a multiple of the dword size (4 bytes).  The
driver code assumes this to be true for both data_size and total_size,
and will not work correctly otherwise.

Futhermore, require that total_size be a multiple of 1024, as per the
Intel SDM, vol 3A, section 9.11.1, page 9-28; table 9-6, page 9-29, and
others.  Test added by request of Borislav Petkov.

Also refuse a microcode update with a microcode revision of zero.
According to the Intel SDM, vol 3A, section 9.11.7, page 9-36, a
microcode revision of zero is special:

    "CPUID returns a value in a model specific register in addition to
    its usual register return values.  The semantics of CPUID cause it
    to deposit an update ID value in the 64-bit model-specific register
    at address 08BH (IA32_BIOS_SIGN_ID).  If no update is present in the
    processor, the value in the MSR remains unmodified.  The BIOS must
    pre-load a zero into the MSR before executing CPUID.  If a read of
    the MSR at 8BH still returns zero after executing CPUID, this
    indicates that no update is present."

This effectively reserves revision zero to mean "no microcode update
installed on the processor": the microcode loader cannot differentiate
sucess from failure when updating microcode to the same revision as the
one currently installed on the processor, and this would always happen
to updates to revision zero in the BIOS/UEFI loader.

There is every reason to be paranoid about any microcode update with a
revision of zero, as Intel will never release such a microcode update.

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
---
 arch/x86/kernel/cpu/microcode/intel_lib.c |   25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/microcode/intel_lib.c b/arch/x86/kernel/cpu/microcode/intel_lib.c
index ce69320..25915e3 100644
--- a/arch/x86/kernel/cpu/microcode/intel_lib.c
+++ b/arch/x86/kernel/cpu/microcode/intel_lib.c
@@ -55,9 +55,10 @@ int microcode_sanity_check(void *mc, int print_err)
 	total_size = get_totalsize(mc_header);
 	data_size = get_datasize(mc_header);
 
-	if (data_size + MC_HEADER_SIZE > total_size) {
+	if ((data_size % DWSIZE) || (total_size % 1024) ||
+	    (data_size + MC_HEADER_SIZE > total_size)) {
 		if (print_err)
-			pr_err("error! Bad data size in microcode data file\n");
+			pr_err("error: bad data size or total size in microcode data file\n");
 		return -EINVAL;
 	}
 
@@ -83,6 +84,26 @@ int microcode_sanity_check(void *mc, int print_err)
 		ext_sigcount = ext_header->count;
 	}
 
+	/*
+	 * A version 1 loader cannot differentiate failure from success when
+	 * attempting a microcode update to the same revision as the one
+	 * currently installed.  The loader is supposed to never attempt a
+	 * same-version update (or a microcode downgrade, for that matter).
+	 *
+	 * This will always cause issues for microcode updates to revision zero
+	 * in the UEFI/BIOS microcode loader: the processor reports a revision
+	 * of zero when it is running without any microcode updates installed,
+	 * such as after a reset/power up.
+	 *
+	 * Intel will never issue a microcode update with a revision of zero
+	 * for the version 1 loader.  Reject it.
+	 */
+	if (mc_header->rev == 0) { /* reserved for "no-update-installed" */
+		if (print_err)
+			pr_err("error: restricted revision 0 in microcode data file\n");
+		return -EINVAL;
+	}
+
 	/* check extended table checksum */
 	if (ext_table_size) {
 		int ext_table_sum = 0;
-- 
1.7.10.4