kexec: Allocate kernel above bzImage's pref_address

A relocatable kernel will relocate itself to pref_address if it is
loaded below pref_address. This means a booted kernel may be relocating
itself to an area with reserved memory on modern systems, potentially
clobbering arbitrary data that may be important to the system.

This is often the case, as the default value of PHYSICAL_START is
0x1000000 and kernels are typically loaded at 0x100000 or above by
bootloaders like iPXE or kexec. GRUB behaves like the approach
implemented here.

Also fixes the documentation around pref_address and PHYSICAL_START to
be accurate.

[ dhansen: changelog tweak ]

Co-developed-by: Cloud Hsu <[email protected]>
Signed-off-by: Cloud Hsu <[email protected]>
Signed-off-by: Chris Koch <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Reviewed-by: H. Peter Anvin (Intel) <[email protected]>
Link: https://lore.kernel.org/all/20231215190521.3796022-1-chrisko%40google.com

Author: hugelgupf

Documentation/arch/x86/boot.rst

@@ -878,7 +878,8 @@ Protocol:	2.10+
   address if possible.
 
   A non-relocatable kernel will unconditionally move itself and to run
-  at this address.
+  at this address. A relocatable kernel will move itself to this address if it
+  loaded below this address.
 
 ============	=======
 Field name:	init_size

arch/x86/Kconfig

@@ -2114,11 +2114,11 @@ config PHYSICAL_START
 	help
 	  This gives the physical address where the kernel is loaded.
 
-	  If kernel is a not relocatable (CONFIG_RELOCATABLE=n) then
-	  bzImage will decompress itself to above physical address and
-	  run from there. Otherwise, bzImage will run from the address where
-	  it has been loaded by the boot loader and will ignore above physical
-	  address.
+	  If the kernel is not relocatable (CONFIG_RELOCATABLE=n) then bzImage
+	  will decompress itself to above physical address and run from there.
+	  Otherwise, bzImage will run from the address where it has been loaded
+	  by the boot loader. The only exception is if it is loaded below the
+	  above physical address, in which case it will relocate itself there.
 
 	  In normal kdump cases one does not have to set/change this option
 	  as now bzImage can be compiled as a completely relocatable image

arch/x86/kernel/kexec-bzimage64.c

@@ -503,7 +503,10 @@ static void *bzImage64_load(struct kimage *image, char *kernel,
 	kbuf.bufsz =  kernel_len - kern16_size;
 	kbuf.memsz = PAGE_ALIGN(header->init_size);
 	kbuf.buf_align = header->kernel_alignment;
-	kbuf.buf_min = MIN_KERNEL_LOAD_ADDR;
+	if (header->pref_address < MIN_KERNEL_LOAD_ADDR)
+		kbuf.buf_min = MIN_KERNEL_LOAD_ADDR;
+	else
+		kbuf.buf_min = header->pref_address;
 	kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
 	ret = kexec_add_buffer(&kbuf);
 	if (ret)