[v4,10/10] ARM: HYP/non-sec/PSCI: emit DT nodes

Generate the PSCI node in the device tree.

Also add a reserve section for the "secure" code that lives in
in normal RAM, so that the kernel knows it'd better not trip on
it.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
---
 arch/arm/cpu/armv7/Makefile  |   1 +
 arch/arm/cpu/armv7/virt-dt.c | 100 +++++++++++++++++++++++++++++++++++++++++++
 arch/arm/include/asm/armv7.h |   1 +
 arch/arm/lib/bootm-fdt.c     |  12 +++++-
 4 files changed, 112 insertions(+), 2 deletions(-)
 create mode 100644 arch/arm/cpu/armv7/virt-dt.c

I'm a U-Boot newbie so please feel free to correct how I'm reporting 
this issue..

I recently downloaded the 2014.04-rc3 snapshot to build U-Boot for my 
custom DA850-based board.  The only change was to add a new target 
"dav850evm_nand" in boards.cfg with the added parameter "USE_NAND".

The resulting AIS file was programmed into EVM-compatible NAND using 
standard sfh_OMAP-L138 method.

The board failed to boot, and stayed in a loop printing the SPL console 
message repeatedly.

After some debugging with CCS 5.5 and an XDS100v2, I found that 
incorrect code was being loaded into the 0xc108000 RAM destination. The 
da850evm.h file defines CONFIG_SYS_NAND_U_BOOT_OFFS as 0x28000, which 
corresponds to an AIS offset of 0x8000 but the u-boot header did not 
appear there in the AIS file.  A search revealed that the Makefile 
catenated u-boot immediately after the SPL without any padding.

Further investigation revealed that the target Makefile needs 
CONFIG_SPL_MAX_SIZE to be defined as 0x8000 in order for the padding to 
be performed properly; however, this constant was apparently deleted 
during a series of changes in April, 2013 to accommodate separate code 
and BSS size limits for another target.  In its place, 
CONFIG_SPL_MAX_FOOTPRINT was defined as 32768.   Unfortunately, the 
da850evm Makefile does not refer to this constant.

To solve the problem, I added the following 2 lines in my 
custom-modified da850evm.h:
#define CONFIG_SPL_PAD_TO    0x8000
#define CONFIG_SPL_MAX_SIZE  0x8000

although the first line may not be strictly required.  This solved the 
problem and allowed the board to boot.

Doesn't this mean that other similar targets may be broken?

How can I assist with contributing the patch to fix this?  One problem I 
have is that I only have an EXP, not an EVM board.  My custom target 
board shares some features with the EVM such as NOR and NAND memory, but 
other things are different like the use of a fixed PHY.  This makes it 
impossible for me to completely verify a da850evm build.

I would like to see more target configurations available for the 
DA850EVM but it seems like development for this has stopped.  NAND boot 
was a simple change, but adding USB support required me to copy & paste 
code from the da830evm target.  Are there any future plans to do this, 
or will I need to do this development for my custom board only?

Tom Taylor

On Sat, Apr 26, 2014 at 7:17 AM, Marc Zyngier <marc.zyngier@arm.com> wrote:

> diff --git a/arch/arm/cpu/armv7/virt-dt.c b/arch/arm/cpu/armv7/virt-dt.c
> new file mode 100644
> index 0000000..0b0d6a7
> --- /dev/null
> +++ b/arch/arm/cpu/armv7/virt-dt.c

> +
> +static int fdt_psci(void *fdt)
> +{
> +#ifdef CONFIG_ARMV7_PSCI
> +       int nodeoff;
> +       int tmp;
> +
> +       nodeoff = fdt_path_offset(fdt, "/cpus");
> +       if (nodeoff < 0) {
> +               printf("couldn't find /cpus\n");
> +               return nodeoff;
> +       }
> +
> +       /* add 'enable-method = "psci"' to each cpu node */
> +       for (tmp = fdt_first_subnode(fdt, nodeoff);
> +            tmp >= 0;
> +            tmp = fdt_next_subnode(fdt, tmp)) {
> +               const struct fdt_property *prop;
> +               int len;
> +
> +               prop = fdt_get_property(fdt, tmp, "device_type", &len);
> +               if (!prop)
> +                       continue;
> +               if (len < 4)
> +                       continue;
> +               if (strcmp(prop->data, "cpu"))
> +                       continue;
> +
> +               fdt_setprop_string(fdt, tmp, "enable-method", "psci");
> +       }
> +
> +       nodeoff = fdt_path_offset(fdt, "/psci");
> +       if (nodeoff < 0) {
> +               nodeoff = fdt_path_offset(fdt, "/");
> +               if (nodeoff < 0)
> +                       return nodeoff;
> +
> +               nodeoff = fdt_add_subnode(fdt, nodeoff, "psci");
> +               if (nodeoff < 0)
> +                       return nodeoff;
> +       }
> +
> +       tmp = fdt_setprop_string(fdt, nodeoff, "compatible", "arm,psci");
> +       if (tmp)
> +               return tmp;
> +       tmp = fdt_setprop_string(fdt, nodeoff, "method", "smc");
> +       if (tmp)
> +               return tmp;
> +       tmp = fdt_setprop_u32(fdt, nodeoff, "cpu_suspend", ARM_PSCI_FN_CPU_SUSPEND);
> +       if (tmp)
> +               return tmp;
> +       tmp = fdt_setprop_u32(fdt, nodeoff, "cpu_off", ARM_PSCI_FN_CPU_OFF);
> +       if (tmp)
> +               return tmp;
> +       tmp = fdt_setprop_u32(fdt, nodeoff, "cpu_on", ARM_PSCI_FN_CPU_ON);
> +       if (tmp)
> +               return tmp;
> +       tmp = fdt_setprop_u32(fdt, nodeoff, "migrate", ARM_PSCI_FN_MIGRATE);
> +       if (tmp)
> +               return tmp;
> +#endif
> +       return 0;
> +}
>

So, I wonder if it would be better to be a bit more selective or cautious
about adding these nodes and properties.  Specifically, if they are already
present in the device tree itself, perhaps they should be honored and left
alone?

I understand that U-Boot gets to define what it implements, and that if
the secure monitor code doesn't actually implement something, or for
that matter *does* implement it, it makes sense for U-Boot to be able
to state those facts in a device tree.  However, the DTS may also be
stating what it has implemented or willing to honor on the Linux side
as well.  So, yeah, there has to be agreement here.

But who gets to make the final adjustment to the device tree?  U-boot
with this code, or the DTS author who may have hand coded specific
wishes and loaded a specific device tree?

HTH,
jdl

On Sat, Apr 26, 2014 at 01:34:58PM -0400, Tom Taylor wrote:

> I'm a U-Boot newbie so please feel free to correct how I'm reporting
> this issue..
> 
> I recently downloaded the 2014.04-rc3 snapshot to build U-Boot for
> my custom DA850-based board.  The only change was to add a new
> target "dav850evm_nand" in boards.cfg with the added parameter
> "USE_NAND".
> 
> The resulting AIS file was programmed into EVM-compatible NAND using
> standard sfh_OMAP-L138 method.
> 
> The board failed to boot, and stayed in a loop printing the SPL
> console message repeatedly.
> 
> After some debugging with CCS 5.5 and an XDS100v2, I found that
> incorrect code was being loaded into the 0xc108000 RAM destination.
> The da850evm.h file defines CONFIG_SYS_NAND_U_BOOT_OFFS as 0x28000,
> which corresponds to an AIS offset of 0x8000 but the u-boot header
> did not appear there in the AIS file.  A search revealed that the
> Makefile catenated u-boot immediately after the SPL without any
> padding.
> 
> Further investigation revealed that the target Makefile needs
> CONFIG_SPL_MAX_SIZE to be defined as 0x8000 in order for the padding
> to be performed properly; however, this constant was apparently
> deleted during a series of changes in April, 2013 to accommodate
> separate code and BSS size limits for another target.  In its place,
> CONFIG_SPL_MAX_FOOTPRINT was defined as 32768.   Unfortunately, the
> da850evm Makefile does not refer to this constant.
> 
> To solve the problem, I added the following 2 lines in my
> custom-modified da850evm.h:
> #define CONFIG_SPL_PAD_TO    0x8000
> #define CONFIG_SPL_MAX_SIZE  0x8000

Thanks for looking into this.  I saw a similar issue on my am18xx EVM
but was 5 bugs deep and didn't have time to bisect it then.

I think the answer here is to post a patch fixing the boards listed in:
$ git grep -l ARM926EJS include/configs/ | xargs grep -l DAVINCI

As they all share that set of constraints.

Thanks for digging into this!

On Fri, May 02 2014 at  9:13:08 pm BST, Jon Loeliger <loeliger@gmail.com> wrote:
> On Sat, Apr 26, 2014 at 7:17 AM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>
>> diff --git a/arch/arm/cpu/armv7/virt-dt.c b/arch/arm/cpu/armv7/virt-dt.c
>> new file mode 100644
>> index 0000000..0b0d6a7
>> --- /dev/null
>> +++ b/arch/arm/cpu/armv7/virt-dt.c
>
>> +
>> +static int fdt_psci(void *fdt)
>> +{
>> +#ifdef CONFIG_ARMV7_PSCI
>> +       int nodeoff;
>> +       int tmp;
>> +
>> +       nodeoff = fdt_path_offset(fdt, "/cpus");
>> +       if (nodeoff < 0) {
>> +               printf("couldn't find /cpus\n");
>> +               return nodeoff;
>> +       }
>> +
>> +       /* add 'enable-method = "psci"' to each cpu node */
>> +       for (tmp = fdt_first_subnode(fdt, nodeoff);
>> +            tmp >= 0;
>> +            tmp = fdt_next_subnode(fdt, tmp)) {
>> +               const struct fdt_property *prop;
>> +               int len;
>> +
>> +               prop = fdt_get_property(fdt, tmp, "device_type", &len);
>> +               if (!prop)
>> +                       continue;
>> +               if (len < 4)
>> +                       continue;
>> +               if (strcmp(prop->data, "cpu"))
>> +                       continue;
>> +
>> +               fdt_setprop_string(fdt, tmp, "enable-method", "psci");
>> +       }
>> +
>> +       nodeoff = fdt_path_offset(fdt, "/psci");
>> +       if (nodeoff < 0) {
>> +               nodeoff = fdt_path_offset(fdt, "/");
>> +               if (nodeoff < 0)
>> +                       return nodeoff;
>> +
>> +               nodeoff = fdt_add_subnode(fdt, nodeoff, "psci");
>> +               if (nodeoff < 0)
>> +                       return nodeoff;
>> +       }
>> +
>> +       tmp = fdt_setprop_string(fdt, nodeoff, "compatible", "arm,psci");
>> +       if (tmp)
>> +               return tmp;
>> +       tmp = fdt_setprop_string(fdt, nodeoff, "method", "smc");
>> +       if (tmp)
>> +               return tmp;
>> + tmp = fdt_setprop_u32(fdt, nodeoff, "cpu_suspend",
>> ARM_PSCI_FN_CPU_SUSPEND);
>> +       if (tmp)
>> +               return tmp;
>> +       tmp = fdt_setprop_u32(fdt, nodeoff, "cpu_off", ARM_PSCI_FN_CPU_OFF);
>> +       if (tmp)
>> +               return tmp;
>> +       tmp = fdt_setprop_u32(fdt, nodeoff, "cpu_on", ARM_PSCI_FN_CPU_ON);
>> +       if (tmp)
>> +               return tmp;
>> +       tmp = fdt_setprop_u32(fdt, nodeoff, "migrate", ARM_PSCI_FN_MIGRATE);
>> +       if (tmp)
>> +               return tmp;
>> +#endif
>> +       return 0;
>> +}
>>

Hi Jon,

> So, I wonder if it would be better to be a bit more selective or
> cautious about adding these nodes and properties.  Specifically, if
> they are already present in the device tree itself, perhaps they
> should be honored and left alone?

Well, we have exactly two possibilities:
- PSCI is provided by the platform's firmware, and we'd better not touch
the DT at all.
- PSCI is provided U-Boot, and we *own* the PSCI related nodes.

I don't think there is an alternative to that, because either U-Boot
or the firmware will install its own secure vectors. The DT manipulation
code just reflect this situation.

> I understand that U-Boot gets to define what it implements, and that if
> the secure monitor code doesn't actually implement something, or for
> that matter *does* implement it, it makes sense for U-Boot to be able
> to state those facts in a device tree.  However, the DTS may also be
> stating what it has implemented or willing to honor on the Linux side
> as well.  So, yeah, there has to be agreement here.
>
> But who gets to make the final adjustment to the device tree?  U-boot
> with this code, or the DTS author who may have hand coded specific
> wishes and loaded a specific device tree?

We could have an environment variable named
"i_know_this_looks_silly_but_nevertheless_please_pretty_please_leave_my_cpu_nodes_alone"?

	M.