arm64: head: Switch endianness before populating the ID map

Ensure that the endianness used for populating the ID map matches the
endianness that the running kernel will be using, as this is no longer
guaranteed now that create_idmap() is invoked before init_kernel_el().

Note that doing so is only safe if the MMU is off, as switching the
endianness with the MMU on results in the active ID map to become
invalid. So also clear the M bit when toggling the EE bit in SCTLR, and
mark the MMU as disabled at boot.

Note that the same issue has resulted in preserve_boot_args() recording
the contents of registers X0 ... X3 in the wrong byte order, although
this is arguably a very minor concern.

Fixes: 32b135a7 ("arm64: head: avoid cache invalidation when entering with the MMU on")
Reported-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/r/20230125185910.962733-1-ardb@kernel.orgSigned-off-by: Catalin Marinas <catalin.marinas@arm.com>

arch/arm64/include/asm/sysreg.h

@@ -575,6 +575,7 @@
 #define SCTLR_ELx_DSSBS	(BIT(44))
 #define SCTLR_ELx_ATA	(BIT(43))
 
+#define SCTLR_ELx_EE_SHIFT	25
 #define SCTLR_ELx_ENIA_SHIFT	31
 
 #define SCTLR_ELx_ITFSB	 (BIT(37))
@@ -583,7 +584,7 @@
 #define SCTLR_ELx_LSMAOE (BIT(29))
 #define SCTLR_ELx_nTLSMD (BIT(28))
 #define SCTLR_ELx_ENDA	 (BIT(27))
-#define SCTLR_ELx_EE     (BIT(25))
+#define SCTLR_ELx_EE     (BIT(SCTLR_ELx_EE_SHIFT))
 #define SCTLR_ELx_EIS	 (BIT(22))
 #define SCTLR_ELx_IESB	 (BIT(21))
 #define SCTLR_ELx_TSCXT	 (BIT(20))

arch/arm64/kernel/head.S

@@ -129,10 +129,31 @@ SYM_CODE_START_LOCAL(record_mmu_state)
 	mrs	x19, sctlr_el1
 	b.ne	0f
 	mrs	x19, sctlr_el2
-0:	tst	x19, #SCTLR_ELx_C		// Z := (C == 0)
+0:
+CPU_LE( tbnz	x19, #SCTLR_ELx_EE_SHIFT, 1f	)
+CPU_BE( tbz	x19, #SCTLR_ELx_EE_SHIFT, 1f	)
+	tst	x19, #SCTLR_ELx_C		// Z := (C == 0)
 	and	x19, x19, #SCTLR_ELx_M		// isolate M bit
 	csel	x19, xzr, x19, eq		// clear x19 if Z
 	ret
+
+	/*
+	 * Set the correct endianness early so all memory accesses issued
+	 * before init_kernel_el() occur in the correct byte order. Note that
+	 * this means the MMU must be disabled, or the active ID map will end
+	 * up getting interpreted with the wrong byte order.
+	 */
+1:	eor	x19, x19, #SCTLR_ELx_EE
+	bic	x19, x19, #SCTLR_ELx_M
+	b.ne	2f
+	pre_disable_mmu_workaround
+	msr	sctlr_el2, x19
+	b	3f
+	pre_disable_mmu_workaround
+2:	msr	sctlr_el1, x19
+3:	isb
+	mov	x19, xzr
+	ret
 SYM_CODE_END(record_mmu_state)
 
 /*