| Message ID | 20200530182900.159874-1-marex@denx.de |
|---|---|
| State | Accepted |
| Commit | 1e7a69f661ae792e3e4bdc0156b26d87d089f5db |
| Headers | show |
| Series | ARM: imx: hab: panic on authentication failure | expand |
Hi Marek, [Adding Breno] On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: > > Instead of hang()ing the system and thus disallowing any automated > recovery possibility from a HAB authentication failure, panic() . > The panic() function can be configured to hang() the system after > printing an error message, however the default is to reset the > system instead. > > This allows redundant boot to work correctly. In case the primary > or secondary image cannot be authenticated, the system reboots and > bootrom can try to start the other one. > > Signed-off-by: Marek Vasut <marex at denx.de> > Cc: Fabio Estevam <festevam at gmail.com> > Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> > Cc: Peng Fan <peng.fan at nxp.com> > Cc: Stefano Babic <sbabic at denx.de> This is a better behavior indeed: Reviewed-by: Fabio Estevam <festevam at gmail.com>
On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: > Hi Marek, > > [Adding Breno] > > On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: > > > > Instead of hang()ing the system and thus disallowing any automated > > recovery possibility from a HAB authentication failure, panic() . > > The panic() function can be configured to hang() the system after > > printing an error message, however the default is to reset the > > system instead. > > > > This allows redundant boot to work correctly. In case the primary > > or secondary image cannot be authenticated, the system reboots and > > bootrom can try to start the other one. > > > > Signed-off-by: Marek Vasut <marex at denx.de> > > Cc: Fabio Estevam <festevam at gmail.com> > > Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> > > Cc: Peng Fan <peng.fan at nxp.com> > > Cc: Stefano Babic <sbabic at denx.de> > > This is a better behavior indeed: > > Reviewed-by: Fabio Estevam <festevam at gmail.com> What about this? Have you ignored this patch for a reason? :/ https://marc.info/?l=u-boot&m=159069441005730&w=2
On 5/30/20 10:14 PM, Patrick Wildt wrote: > On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: >> Hi Marek, >> >> [Adding Breno] >> >> On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: >>> >>> Instead of hang()ing the system and thus disallowing any automated >>> recovery possibility from a HAB authentication failure, panic() . >>> The panic() function can be configured to hang() the system after >>> printing an error message, however the default is to reset the >>> system instead. >>> >>> This allows redundant boot to work correctly. In case the primary >>> or secondary image cannot be authenticated, the system reboots and >>> bootrom can try to start the other one. >>> >>> Signed-off-by: Marek Vasut <marex at denx.de> >>> Cc: Fabio Estevam <festevam at gmail.com> >>> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> >>> Cc: Peng Fan <peng.fan at nxp.com> >>> Cc: Stefano Babic <sbabic at denx.de> >> >> This is a better behavior indeed: >> >> Reviewed-by: Fabio Estevam <festevam at gmail.com> > > What about this? Have you ignored this patch for a reason? :/ > > https://marc.info/?l=u-boot&m=159069441005730&w=2 Yes, and the reason is I was not even aware of your patch, sorry. The CC list in this mail should cover all the interested parties, so use it when sending V2, or use patman. The patch looks fine, one nit is that you should return errno.h return value and another is that it changes the current behavior. Now that I look at this imx code, board_spl_fit_post_load() should not even be in arch/ , sigh, but that's for separate patch either way. So I think if you want to support this sort of fallback, you should make the board_spl_fit_post_load() be in board/ files, with default __weak implementation calling some arch_hab_authenticate...() which implements current content of board_spl_fit_post_load(), and let boards decide how to handle the fallback if it needs to be altered. Would that work ?
On Sat, May 30, 2020 at 10:29:19PM +0200, Marek Vasut wrote: > On 5/30/20 10:14 PM, Patrick Wildt wrote: > > On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: > >> Hi Marek, > >> > >> [Adding Breno] > >> > >> On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: > >>> > >>> Instead of hang()ing the system and thus disallowing any automated > >>> recovery possibility from a HAB authentication failure, panic() . > >>> The panic() function can be configured to hang() the system after > >>> printing an error message, however the default is to reset the > >>> system instead. > >>> > >>> This allows redundant boot to work correctly. In case the primary > >>> or secondary image cannot be authenticated, the system reboots and > >>> bootrom can try to start the other one. > >>> > >>> Signed-off-by: Marek Vasut <marex at denx.de> > >>> Cc: Fabio Estevam <festevam at gmail.com> > >>> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> > >>> Cc: Peng Fan <peng.fan at nxp.com> > >>> Cc: Stefano Babic <sbabic at denx.de> > >> > >> This is a better behavior indeed: > >> > >> Reviewed-by: Fabio Estevam <festevam at gmail.com> > > > > What about this? Have you ignored this patch for a reason? :/ > > > > https://marc.info/?l=u-boot&m=159069441005730&w=2 > > Yes, and the reason is I was not even aware of your patch, sorry. The CC > list in this mail should cover all the interested parties, so use it > when sending V2, or use patman. I already had 11 people on CC, but apparently I missed you. > The patch looks fine, one nit is that you should return errno.h return > value and another is that it changes the current behavior. Now that I > look at this imx code, board_spl_fit_post_load() should not even be in > arch/ , sigh, but that's for separate patch either way. > > So I think if you want to support this sort of fallback, you should make > the board_spl_fit_post_load() be in board/ files, with default __weak > implementation calling some arch_hab_authenticate...() which implements > current content of board_spl_fit_post_load(), and let boards decide how > to handle the fallback if it needs to be altered. > > Would that work ? I'm not sure. In comparison to the people from NXP who are paid to upstream their code and still don't do it correctly, I'm doing this in my spare time and I'm not sure I want to bikeshed all day long. I can send a V3 that replaces the -1 with EINVAL, EACCESS, EPERM or something the like. If you want to clean up after NXP, feel free to. Which errno would you like to see? Best regards, Patrick
On 5/30/20 10:53 PM, Patrick Wildt wrote: > On Sat, May 30, 2020 at 10:29:19PM +0200, Marek Vasut wrote: >> On 5/30/20 10:14 PM, Patrick Wildt wrote: >>> On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: >>>> Hi Marek, >>>> >>>> [Adding Breno] >>>> >>>> On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: >>>>> >>>>> Instead of hang()ing the system and thus disallowing any automated >>>>> recovery possibility from a HAB authentication failure, panic() . >>>>> The panic() function can be configured to hang() the system after >>>>> printing an error message, however the default is to reset the >>>>> system instead. >>>>> >>>>> This allows redundant boot to work correctly. In case the primary >>>>> or secondary image cannot be authenticated, the system reboots and >>>>> bootrom can try to start the other one. >>>>> >>>>> Signed-off-by: Marek Vasut <marex at denx.de> >>>>> Cc: Fabio Estevam <festevam at gmail.com> >>>>> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> >>>>> Cc: Peng Fan <peng.fan at nxp.com> >>>>> Cc: Stefano Babic <sbabic at denx.de> >>>> >>>> This is a better behavior indeed: >>>> >>>> Reviewed-by: Fabio Estevam <festevam at gmail.com> >>> >>> What about this? Have you ignored this patch for a reason? :/ >>> >>> https://marc.info/?l=u-boot&m=159069441005730&w=2 >> >> Yes, and the reason is I was not even aware of your patch, sorry. The CC >> list in this mail should cover all the interested parties, so use it >> when sending V2, or use patman. > > I already had 11 people on CC, but apparently I missed you. > >> The patch looks fine, one nit is that you should return errno.h return >> value and another is that it changes the current behavior. Now that I >> look at this imx code, board_spl_fit_post_load() should not even be in >> arch/ , sigh, but that's for separate patch either way. >> >> So I think if you want to support this sort of fallback, you should make >> the board_spl_fit_post_load() be in board/ files, with default __weak >> implementation calling some arch_hab_authenticate...() which implements >> current content of board_spl_fit_post_load(), and let boards decide how >> to handle the fallback if it needs to be altered. >> >> Would that work ? > > I'm not sure. In comparison to the people from NXP who are paid to > upstream their code and still don't do it correctly, I'm doing this > in my spare time and I'm not sure I want to bikeshed all day long. > > I can send a V3 that replaces the -1 with EINVAL, EACCESS, EPERM or > something the like. If you want to clean up after NXP, feel free to. In fact, what is it that you're trying to achieve with this fallback ? What are you falling back to , another fallback fitImage ?
On Sun, May 31, 2020 at 05:38:05PM +0200, Marek Vasut wrote: > On 5/30/20 10:53 PM, Patrick Wildt wrote: > > On Sat, May 30, 2020 at 10:29:19PM +0200, Marek Vasut wrote: > >> On 5/30/20 10:14 PM, Patrick Wildt wrote: > >>> On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: > >>>> Hi Marek, > >>>> > >>>> [Adding Breno] > >>>> > >>>> On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: > >>>>> > >>>>> Instead of hang()ing the system and thus disallowing any automated > >>>>> recovery possibility from a HAB authentication failure, panic() . > >>>>> The panic() function can be configured to hang() the system after > >>>>> printing an error message, however the default is to reset the > >>>>> system instead. > >>>>> > >>>>> This allows redundant boot to work correctly. In case the primary > >>>>> or secondary image cannot be authenticated, the system reboots and > >>>>> bootrom can try to start the other one. > >>>>> > >>>>> Signed-off-by: Marek Vasut <marex at denx.de> > >>>>> Cc: Fabio Estevam <festevam at gmail.com> > >>>>> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> > >>>>> Cc: Peng Fan <peng.fan at nxp.com> > >>>>> Cc: Stefano Babic <sbabic at denx.de> > >>>> > >>>> This is a better behavior indeed: > >>>> > >>>> Reviewed-by: Fabio Estevam <festevam at gmail.com> > >>> > >>> What about this? Have you ignored this patch for a reason? :/ > >>> > >>> https://marc.info/?l=u-boot&m=159069441005730&w=2 > >> > >> Yes, and the reason is I was not even aware of your patch, sorry. The CC > >> list in this mail should cover all the interested parties, so use it > >> when sending V2, or use patman. > > > > I already had 11 people on CC, but apparently I missed you. > > > >> The patch looks fine, one nit is that you should return errno.h return > >> value and another is that it changes the current behavior. Now that I > >> look at this imx code, board_spl_fit_post_load() should not even be in > >> arch/ , sigh, but that's for separate patch either way. > >> > >> So I think if you want to support this sort of fallback, you should make > >> the board_spl_fit_post_load() be in board/ files, with default __weak > >> implementation calling some arch_hab_authenticate...() which implements > >> current content of board_spl_fit_post_load(), and let boards decide how > >> to handle the fallback if it needs to be altered. > >> > >> Would that work ? > > > > I'm not sure. In comparison to the people from NXP who are paid to > > upstream their code and still don't do it correctly, I'm doing this > > in my spare time and I'm not sure I want to bikeshed all day long. > > > > I can send a V3 that replaces the -1 with EINVAL, EACCESS, EPERM or > > something the like. If you want to clean up after NXP, feel free to. > > In fact, what is it that you're trying to achieve with this fallback ? > What are you falling back to , another fallback fitImage ? Exactly. I have a device with a glued enclosure, with no external sources, apart from a serial console. If the SPL fails to load the U-Boot main image from eMMC, the only way to fix it is to destroy the case, open up the enclosure and short some lines. That's not really nice, since we'd have to get a new enclosure, new serial number label,... it's a hassle. If the SPL was gone as well, the BootROM would fall back to other sources. But with only one half of U-Boot missing, it would just hang. I'm sure that's also why you replace the hang() with a panic(). I thought that if the SPL is still fine, but only the U-Boot image was gone, why not make use of the spl_boot_list to try and boot from another source? Like yModem. For that I sent the following fix, also with many people CCd: https://marc.info/?l=u-boot&m=158893200030861&w=2 Now the board spl boot order can have eMMC as first, and yModem as second. If eMMC fails, it falls back to yModem. If none work, it though hang()s, doing puts(SPL_TPL_PROMPT "failed to boot from all boot devices\n"); hang(); Maybe you want this as panic instead? Anyway, I think this is nicer option for recovery, instead of simply hanging or rebooting. Patrick
On 5/31/20 5:53 PM, Patrick Wildt wrote: > On Sun, May 31, 2020 at 05:38:05PM +0200, Marek Vasut wrote: >> On 5/30/20 10:53 PM, Patrick Wildt wrote: >>> On Sat, May 30, 2020 at 10:29:19PM +0200, Marek Vasut wrote: >>>> On 5/30/20 10:14 PM, Patrick Wildt wrote: >>>>> On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: >>>>>> Hi Marek, >>>>>> >>>>>> [Adding Breno] >>>>>> >>>>>> On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: >>>>>>> >>>>>>> Instead of hang()ing the system and thus disallowing any automated >>>>>>> recovery possibility from a HAB authentication failure, panic() . >>>>>>> The panic() function can be configured to hang() the system after >>>>>>> printing an error message, however the default is to reset the >>>>>>> system instead. >>>>>>> >>>>>>> This allows redundant boot to work correctly. In case the primary >>>>>>> or secondary image cannot be authenticated, the system reboots and >>>>>>> bootrom can try to start the other one. >>>>>>> >>>>>>> Signed-off-by: Marek Vasut <marex at denx.de> >>>>>>> Cc: Fabio Estevam <festevam at gmail.com> >>>>>>> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> >>>>>>> Cc: Peng Fan <peng.fan at nxp.com> >>>>>>> Cc: Stefano Babic <sbabic at denx.de> >>>>>> >>>>>> This is a better behavior indeed: >>>>>> >>>>>> Reviewed-by: Fabio Estevam <festevam at gmail.com> >>>>> >>>>> What about this? Have you ignored this patch for a reason? :/ >>>>> >>>>> https://marc.info/?l=u-boot&m=159069441005730&w=2 >>>> >>>> Yes, and the reason is I was not even aware of your patch, sorry. The CC >>>> list in this mail should cover all the interested parties, so use it >>>> when sending V2, or use patman. >>> >>> I already had 11 people on CC, but apparently I missed you. >>> >>>> The patch looks fine, one nit is that you should return errno.h return >>>> value and another is that it changes the current behavior. Now that I >>>> look at this imx code, board_spl_fit_post_load() should not even be in >>>> arch/ , sigh, but that's for separate patch either way. >>>> >>>> So I think if you want to support this sort of fallback, you should make >>>> the board_spl_fit_post_load() be in board/ files, with default __weak >>>> implementation calling some arch_hab_authenticate...() which implements >>>> current content of board_spl_fit_post_load(), and let boards decide how >>>> to handle the fallback if it needs to be altered. >>>> >>>> Would that work ? >>> >>> I'm not sure. In comparison to the people from NXP who are paid to >>> upstream their code and still don't do it correctly, I'm doing this >>> in my spare time and I'm not sure I want to bikeshed all day long. >>> >>> I can send a V3 that replaces the -1 with EINVAL, EACCESS, EPERM or >>> something the like. If you want to clean up after NXP, feel free to. >> >> In fact, what is it that you're trying to achieve with this fallback ? >> What are you falling back to , another fallback fitImage ? > > Exactly. > > I have a device with a glued enclosure, with no external sources, apart > from a serial console. If the SPL fails to load the U-Boot main image > from eMMC, the only way to fix it is to destroy the case, open up the > enclosure and short some lines. That's not really nice, since we'd have > to get a new enclosure, new serial number label,... it's a hassle. Look for SRC PERSIST_SECONDARY_BOOT in your datasheet then. This will let you use two copies of SPL, two copies of U-Boot, etc. > If the SPL was gone as well, the BootROM would fall back to other > sources. But with only one half of U-Boot missing, it would just hang. > I'm sure that's also why you replace the hang() with a panic(). > > I thought that if the SPL is still fine, but only the U-Boot image was > gone, why not make use of the spl_boot_list to try and boot from another > source? Like yModem. For that I sent the following fix, also with many > people CCd: > > https://marc.info/?l=u-boot&m=158893200030861&w=2 > > Now the board spl boot order can have eMMC as first, and yModem as > second. If eMMC fails, it falls back to yModem. If none work, it > though hang()s, doing > > puts(SPL_TPL_PROMPT "failed to boot from all boot devices\n"); > hang(); > > Maybe you want this as panic instead? > > Anyway, I think this is nicer option for recovery, instead of simply > hanging or rebooting. The problem is, this only works for fitImage and not for raw uImage. But in your case, that is probably OK. So if you can just fix the errno return value to some -EINVAL (?) and send a V3, I think that would be good.
On Sun, May 31, 2020 at 06:51:14PM +0200, Marek Vasut wrote: > On 5/31/20 5:53 PM, Patrick Wildt wrote: > > On Sun, May 31, 2020 at 05:38:05PM +0200, Marek Vasut wrote: > >> On 5/30/20 10:53 PM, Patrick Wildt wrote: > >>> On Sat, May 30, 2020 at 10:29:19PM +0200, Marek Vasut wrote: > >>>> On 5/30/20 10:14 PM, Patrick Wildt wrote: > >>>>> On Sat, May 30, 2020 at 03:31:29PM -0300, Fabio Estevam wrote: > >>>>>> Hi Marek, > >>>>>> > >>>>>> [Adding Breno] > >>>>>> > >>>>>> On Sat, May 30, 2020 at 3:29 PM Marek Vasut <marex at denx.de> wrote: > >>>>>>> > >>>>>>> Instead of hang()ing the system and thus disallowing any automated > >>>>>>> recovery possibility from a HAB authentication failure, panic() . > >>>>>>> The panic() function can be configured to hang() the system after > >>>>>>> printing an error message, however the default is to reset the > >>>>>>> system instead. > >>>>>>> > >>>>>>> This allows redundant boot to work correctly. In case the primary > >>>>>>> or secondary image cannot be authenticated, the system reboots and > >>>>>>> bootrom can try to start the other one. > >>>>>>> > >>>>>>> Signed-off-by: Marek Vasut <marex at denx.de> > >>>>>>> Cc: Fabio Estevam <festevam at gmail.com> > >>>>>>> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> > >>>>>>> Cc: Peng Fan <peng.fan at nxp.com> > >>>>>>> Cc: Stefano Babic <sbabic at denx.de> > >>>>>> > >>>>>> This is a better behavior indeed: > >>>>>> > >>>>>> Reviewed-by: Fabio Estevam <festevam at gmail.com> > >>>>> > >>>>> What about this? Have you ignored this patch for a reason? :/ > >>>>> > >>>>> https://marc.info/?l=u-boot&m=159069441005730&w=2 > >>>> > >>>> Yes, and the reason is I was not even aware of your patch, sorry. The CC > >>>> list in this mail should cover all the interested parties, so use it > >>>> when sending V2, or use patman. > >>> > >>> I already had 11 people on CC, but apparently I missed you. > >>> > >>>> The patch looks fine, one nit is that you should return errno.h return > >>>> value and another is that it changes the current behavior. Now that I > >>>> look at this imx code, board_spl_fit_post_load() should not even be in > >>>> arch/ , sigh, but that's for separate patch either way. > >>>> > >>>> So I think if you want to support this sort of fallback, you should make > >>>> the board_spl_fit_post_load() be in board/ files, with default __weak > >>>> implementation calling some arch_hab_authenticate...() which implements > >>>> current content of board_spl_fit_post_load(), and let boards decide how > >>>> to handle the fallback if it needs to be altered. > >>>> > >>>> Would that work ? > >>> > >>> I'm not sure. In comparison to the people from NXP who are paid to > >>> upstream their code and still don't do it correctly, I'm doing this > >>> in my spare time and I'm not sure I want to bikeshed all day long. > >>> > >>> I can send a V3 that replaces the -1 with EINVAL, EACCESS, EPERM or > >>> something the like. If you want to clean up after NXP, feel free to. > >> > >> In fact, what is it that you're trying to achieve with this fallback ? > >> What are you falling back to , another fallback fitImage ? > > > > Exactly. > > > > I have a device with a glued enclosure, with no external sources, apart > > from a serial console. If the SPL fails to load the U-Boot main image > > from eMMC, the only way to fix it is to destroy the case, open up the > > enclosure and short some lines. That's not really nice, since we'd have > > to get a new enclosure, new serial number label,... it's a hassle. > > Look for SRC PERSIST_SECONDARY_BOOT in your datasheet then. This will > let you use two copies of SPL, two copies of U-Boot, etc. I'll have a look! > > If the SPL was gone as well, the BootROM would fall back to other > > sources. But with only one half of U-Boot missing, it would just hang. > > I'm sure that's also why you replace the hang() with a panic(). > > > > I thought that if the SPL is still fine, but only the U-Boot image was > > gone, why not make use of the spl_boot_list to try and boot from another > > source? Like yModem. For that I sent the following fix, also with many > > people CCd: > > > > https://marc.info/?l=u-boot&m=158893200030861&w=2 > > > > Now the board spl boot order can have eMMC as first, and yModem as > > second. If eMMC fails, it falls back to yModem. If none work, it > > though hang()s, doing > > > > puts(SPL_TPL_PROMPT "failed to boot from all boot devices\n"); > > hang(); > > > > Maybe you want this as panic instead? > > > > Anyway, I think this is nicer option for recovery, instead of simply > > hanging or rebooting. > > The problem is, this only works for fitImage and not for raw uImage. But > in your case, that is probably OK. > > So if you can just fix the errno return value to some -EINVAL (?) and > send a V3, I think that would be good. Ok, will do, thanks!
diff --git a/arch/arm/mach-imx/spl.c b/arch/arm/mach-imx/spl.c index 1a231c67f5..76a5f7aca6 100644 --- a/arch/arm/mach-imx/spl.c +++ b/arch/arm/mach-imx/spl.c @@ -293,8 +293,7 @@ __weak void __noreturn jump_to_image_no_args(struct spl_image_info *spl_image) CSF_PAD_SIZE, offset)) { image_entry(); } else { - puts("spl: ERROR: image authentication fail\n"); - hang(); + panic("spl: ERROR: image authentication fail\n"); } } } @@ -320,8 +319,7 @@ void board_spl_fit_post_load(ulong load_addr, size_t length) if (imx_hab_authenticate_image(load_addr, offset + IVT_SIZE + CSF_PAD_SIZE, offset)) { - puts("spl: ERROR: image authentication unsuccessful\n"); - hang(); + panic("spl: ERROR: image authentication unsuccessful\n"); } } #endif
Instead of hang()ing the system and thus disallowing any automated recovery possibility from a HAB authentication failure, panic() . The panic() function can be configured to hang() the system after printing an error message, however the default is to reset the system instead. This allows redundant boot to work correctly. In case the primary or secondary image cannot be authenticated, the system reboots and bootrom can try to start the other one. Signed-off-by: Marek Vasut <marex at denx.de> Cc: Fabio Estevam <festevam at gmail.com> Cc: NXP i.MX U-Boot Team <uboot-imx at nxp.com> Cc: Peng Fan <peng.fan at nxp.com> Cc: Stefano Babic <sbabic at denx.de> --- arch/arm/mach-imx/spl.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-)