Kexec on arm64

[Mark Rutland]

On Wed, Jul 30, 2014 at 06:46:39AM +0100, Arun Chandran wrote:
> On Tue, Jul 29, 2014 at 7:05 PM, Mark Rutland <mark.rutland at arm.com> wrote:
> > [...]
> >
> >> The default code did not work.
> >>
> >> It is working with the change below
> >>
> >> ###############
> >> diff --git a/arch/arm64/kernel/machine_kexec.c
> >> b/arch/arm64/kernel/machine_kexec.c
> >> index 5632473..7c5f859 100644
> >> --- a/arch/arm64/kernel/machine_kexec.c
> >> +++ b/arch/arm64/kernel/machine_kexec.c
> >> @@ -147,12 +147,17 @@ static bool kexec_is_dtb_user(const dtb_t *dtb)
> >>  /**
> >>   * kexec_list_walk - Helper to walk the kimage page list.
> >>   */
> >> -
> >> +static int kexec_kernel_size;
> >> +#define IMG_SIZE_NONE  0
> >> +#define KERN_SIZE_FLAG 1
> >> +#define DTB_SIZE_FLAG  2
> >>  static void kexec_list_walk(void *ctx, unsigned long kimage_head,
> >>         void (*cb)(void *ctx, unsigned int flag, void *addr, void *dest))
> >>  {
> >>         void *dest;
> >>         unsigned long *entry;
> >> +       int imgsize_flag = IMG_SIZE_NONE;
> >> +
> >>
> >>         for (entry = &kimage_head, dest = NULL; ; entry++) {
> >>                 unsigned int flag = *entry & IND_FLAGS;
> >> @@ -164,10 +169,18 @@ static void kexec_list_walk(void *ctx, unsigned
> >> long kimage_head,
> >>                         cb(ctx, flag, addr, NULL);
> >>                         break;
> >>                 case IND_DESTINATION:
> >> +                       if (imgsize_flag == IMG_SIZE_NONE) {
> >> +                               kexec_kernel_size = 0;
> >> +                               imgsize_flag = KERN_SIZE_FLAG;
> >> +                       } else if (imgsize_flag == KERN_SIZE_FLAG) {
> >> +                               imgsize_flag = DTB_SIZE_FLAG;
> >> +                       }
> >>                         dest = addr;
> >>                         cb(ctx, flag, addr, NULL);
> >>                         break;
> >>                 case IND_SOURCE:
> >> +                       if (imgsize_flag == KERN_SIZE_FLAG)
> >> +                               kexec_kernel_size++;
> >>                         cb(ctx, flag, addr, dest);
> >>                         dest += PAGE_SIZE;
> >>                         break;
> >> @@ -693,5 +706,20 @@ void machine_kexec(struct kimage *image)
> >>
> >>         kexec_list_walk(NULL, image->head, kexec_list_flush_cb);
> >>
> >> +       /*
> >> +        * Make sure virtual addresses of new kernel are flushed
> >> +        * SZ_512K = TEXT_OFFSET
> >
> > TEXT_OFFSET is not guaranteed to be 512K. The TEXT_OFFSET area also
> > shouldn't need to be flushed.
> >
> > Since c218bca74eea (arm64: Relax the kernel cache requirements for
> > boot), the kernel will flush the cache for anything outside of the Image
> > that it writes to before enabling the MMU and caches (e.g. the idmap and
> > swapper page tables). Once caches are up we shouldn't care.
> >
> 
> Ok. TEXT_OFFSET macro is not exported, that's why I used SZ_512K.
> Any particular reason for not exporting it?

The TEXT_OFFSET of the kernel you intend to execute can be found in its
Image header. Take a look at Documentation/arm64/booting.txt. This may
be different from the TEXT_OFFSET of the current kernel.

As of v3.17 all fields in the header will be little-endian, and
booting.txt has been updated for this. You can check the image_size
field to determine whether the fields are all little-endian. Only when
image_size is zero can TEXT_OFFSET be assumed to be 0x80000.

> 
> > Assuming that the existing kernel code is correct, the only region we
> > should need to flush out to the PoC is the region from _text to _edata
> > (i.e. just the contents of the Image).
> >
> >> +        * kexec_kernel = kexec_kernel_size * PAGE_SIZE
> >> +        * Don't know = (SZ_4M + SZ_1M)
> >> +        * SZ_4M = not working
> >> +        * SZ_6M = working
> >> +        * SZ_8M = working
> >> +        *
> >> +        * so chose SZ_4M + SZ_1M; Don't know why this is required
> >> +        * BSS, stack ??
> >> +        *
> >> +        */
> >> +       __flush_dcache_area((void *)PAGE_OFFSET, SZ_512K +
> >> (kexec_kernel_size * PAGE_SIZE) + SZ_4M + SZ_1M);
> >> +
> >>         soft_restart(reboot_code_buffer_phys);
> >>  }
> >
> > How big exactly is the kernel Image you're trying to kexec?
> 
> For the 1st and second stage I use the same kernel as they
> are combined with intramfs final image size varies.
> 
> 1st stage
> -------------
> $ls -l arch/arm64/boot/uImage
> -rw-rw-r-- 1 arun arun 12895544 Jul 30 10:57 arch/arm64/boot/uImage
> 
> It will boot to intramfs
> 
> $ls /
> bin  dtb.dtb  home  lib    linuxrc  mnt  proc  run   share  tmp  var
> dev  etc      init  lib64  media    opt  root  sbin  sys    usr  vmlinux.strip
> 
> kexec will boot vmlinux.strip
> $ls -l vmlinux.strip
> -rwxrwxr-x 1 arun arun 8194760 Jul 30 10:55 vmlinux.strip
> 
> 2nd stage
> --------------
> $ls -l arch/arm64/boot/uImage
> -rw-rw-r-- 1 arun arun 8127800 Jul 30 10:54 arch/arm64/boot/uImage
> 
> The corresponding vmlinux is converted to vmlinux.strip
> 
> $aarch64-linux-gnu-strip vmlinux -o vmlinux.strip
> $cp vmlinux.strip /ramfs/aarch64_le_rootfs/

OK, thanks for the info.

Cheers,
Mark.