On 16/08/16 09:25, Thiago Jung Bauermann wrote:

Am Montag, 15 August 2016, 17:46:34 schrieb Balbir Singh:

quoted

On Thu, Aug 11, 2016 at 08:08:09PM -0300, Thiago Jung Bauermann wrote:

quoted

+/**
+ * elf64_apply_relocate_add - apply 64 bit RELA relocations
+ * @elf_info:		Support information for the ELF binary being 

relocated.

quoted

quoted

+ * @strtab:		String table for the associated symbol 

table.

quoted

quoted

+ * @symindex:		Section header index for the associated 

symbol table.

quoted

quoted

+ * @relsec:		Section header index for the relocations to 

apply.

quoted

quoted

+ * @obj_name:		The name of the ELF binary, for information 

messages.

quoted

quoted

+ */
+int elf64_apply_relocate_add(const struct elf_info *elf_info,
+			     const char *strtab, unsigned int symindex,
+			     unsigned int relsec, const char *obj_name)
+{
+	unsigned int i;
+	Elf64_Shdr *sechdrs = elf_info->sechdrs;
+	Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
+	Elf64_Sym *sym;
+	unsigned long *location;
+	unsigned long value;
+

For the relocatable kernel we expect only

R_PPC64_RELATIVE
R_PPC64_NONE
R_PPC64_ADDR64

In the future we can use this to check/assert the usage of this
for the core kernel (vmlinux) when loaded.

Did we check elf64_apply_relocate_add with zImage and vmlinux?

kexec_file_load doesn't call call elf64_apply_relocate_add on the kernel 
image, it only uses it to relocate the purgatory. So whether it is loading a 
zImage or a vmlinux file, the function will work in the same way since the 
purgatory binary is the same regardless of the kernel image format.

Thanks for clarifying.

For the same reason, as it currently stands kexec_file_load can't check the 
relocation types used in the kernel image. But it is possible to add such a 
check/assertion in kexec_elf_64.c:build_elf_exec_info if we want.

I tested kexec_file_load on both relocatable and non-relocatable vmlinux and 
it works correctly.

I hadn't tested with zImage yet. I just did, and I had two problems:

1. For some reason, it has an INTERP segment. This patch series doesn't 
support loading program interpreters for ELF binaries, so 
kexec_elf_64.c:build_elf_exec_info refuses to load them.

2. If I disable the check for the INTERP segment, the zImage file loads 
correctly, but then I get an exception during reboot when loading the kexec 
image, right before jumping into the purgatory. I suspect this is because 
the LOAD segment has a virtual address of 0, and the first kernel is not 
coping well with that. But I still have to debug it further.

Is there a reason for the zImage ELF header to request an interpreter and to 
have a virtual address of 0?

Not that I am aware of. 

Balbir Singh