* Luck, Tony <tony.l...@intel.com>:
> On Sat, Sep 16, 2017 at 12:53:42PM +0900, Sergey Senozhatsky wrote:
> > Hello
> >
> > RFC
> >
> > On some arches C function pointers are indirect and point to
> > a function descriptor, which contains the actual pointer to the code.
> > This mostly doesn't matter, except for cases when people want to print
> > out function pointers in symbolic format, because the usual '%pS/%ps'
> > does not work on those arches as expected. That's the reason why we
> > have '%pF/%pf', but since it's here because of a subtle ABI detail
> > specific to some arches (ppc64/ia64/parisc64) it's easy to misuse
> > '%pF/%pf' and '%pS/%ps' (see [1], for example).
>
> A few new warnings when building on ia64:
>
> arch/ia64/kernel/module.c:931: warning: passing argument 1 of
> 'dereference_function_descriptor' makes pointer from integer without a cast
> arch/ia64/kernel/module.c:931: warning: return makes integer from pointer
> without a cast
> kernel/kallsyms.c:325: warning: assignment makes integer from pointer without
> a cast
> kernel/kallsyms.c:325: warning: passing argument 1 of
> 'dereference_kernel_function_descriptor' makes pointer from integer without a
> cast
I got similiar warnings on parisc.
This patch on top of yours fixed those:
diff --git a/arch/parisc/kernel/module.c b/arch/parisc/kernel/module.c
index bc2eae8..4f34b46 100644
--- a/arch/parisc/kernel/module.c
+++ b/arch/parisc/kernel/module.c
@@ -66,6 +66,7 @@
#include <asm/pgtable.h>
#include <asm/unwind.h>
+#include <asm/sections.h>
#if 0
#define DEBUGP printk
@@ -959,12 +960,12 @@ void module_arch_cleanup(struct module *mod)
unsigned long dereference_module_function_descriptor(struct module *mod,
unsigned long addr)
{
- void *opd_sz = mod->arch.fdesc_offset +
+ unsigned long opd_sz = mod->arch.fdesc_offset +
mod->arch.fdesc_max * sizeof(Elf64_Fdesc);
if (addr < mod->arch.fdesc_offset || opd_sz < addr)
return addr;
- return dereference_function_descriptor(addr);
+ return (unsigned long) dereference_function_descriptor((void *) addr);
}
#endif
diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c
index e2fc09e..76f4de6 100644
--- a/kernel/kallsyms.c
+++ b/kernel/kallsyms.c
@@ -322,7 +322,7 @@ const char *kallsyms_lookup(unsigned long addr,
if (is_ksym_addr(addr)) {
unsigned long pos;
- addr = dereference_kernel_function_descriptor(addr);
+ addr = dereference_kernel_function_descriptor((void *) addr);
pos = get_symbol_pos(addr, symbolsize, offset);
/* Grab name */
kallsyms_expand_symbol(get_symbol_offset(pos),
I did tried your testcases too.
"echo 1 > /proc/sys/vm/drop_caches" gave correct output:
printk#1 schedule_timeout+0x0/0x4a8
printk#2 schedule_timeout+0x0/0x4a8
printk#3 proc_sys_call_handler+0x120/0x180
printk#4 proc_sys_call_handler+0x120/0x180
printk#5 proc_sys_call_handler+0x120/0x180
printk#6 proc_sys_call_handler+0x120/0x180
and here is "modprobe zram":
printk#7 __UNIQUE_ID_vermagic8+0xb9a4/0xbd04 [zram]
printk#8 __UNIQUE_ID_vermagic8+0xb9a4/0xbd04 [zram]
printk#9 do_one_initcall+0x194/0x290
printk#10 do_one_initcall+0x194/0x290
printk#11 do_one_initcall+0x194/0x290
printk#12 do_one_initcall+0x194/0x290
printk#13 zram_init+0x22c/0x2a0 [zram]
printk#14 zram_init+0x22c/0x2a0 [zram]
printk#15 zram_init+0x22c/0x2a0 [zram]
printk#16 zram_init+0x22c/0x2a0 [zram]
I wonder why printk#7 and printk#8 don't show "zram_init"...
Regarding your patches:
In arch/parisc/kernel/process.c:
+void *dereference_kernel_function_descriptor(void *ptr)
+{
+ if (ptr < (void *)__start_opd || (void *)__end_opd < ptr)
This needs to be (__end_opd is outside):
+ if (ptr < (void *)__start_opd || (void *)__end_opd <= ptr)
The same is true for the checks in the other arches.
I'd suggest to move the various
extern char __start_opd[], __end_opd[];
out of arch/<arch>/include/asm/sections.h and into <asm-generic/sections.h>
I'll continue to test.
Helge
