[PATCH] arm: bootm: Disable LMB reservation for command line and board info on arm64

[Jan Kiszka]

On 21.07.21 08:08, Jan Kiszka wrote:
> On 21.07.21 01:34, Marek Vasut wrote:
>> On 7/20/21 11:08 AM, Jan Kiszka wrote:
>> [...]
>>>> diff --git a/arch/arm/lib/bootm.c b/arch/arm/lib/bootm.c
>>>> index f60ee3a7e6..23b99a541c 100644
>>>> --- a/arch/arm/lib/bootm.c
>>>> +++ b/arch/arm/lib/bootm.c
>>>> @@ -43,6 +43,7 @@ DECLARE_GLOBAL_DATA_PTR;
>>>>     static struct tag *params;
>>>>   +#ifndef CONFIG_ARM64
>>>>   static ulong get_sp(void)
>>>>   {
>>>>       ulong ret;
>>>> @@ -86,6 +87,7 @@ void arch_lmb_reserve(struct lmb *lmb)
>>>>           break;
>>>>       }
>>>>   }
>>>> +#endif
>>>>     __weak void board_quiesce_devices(void)
>>>>   {
>>>>
>>> This causes troubles for [1], but I have no clue yet, what is happening.
>>> Without the patch, we start like this:
>>>
>>> Found U-Boot script /boot/boot.scr
>>> 889 bytes read in 21 ms (41 KiB/s)
>>> ## Executing script at 83000000
>>> Loading /usr/lib/linux-image-4.19.94/ti/k3-am6548-iot2050-advanced.dtb...
>>> 78306 bytes read in 25 ms (3 MiB/s)
>>> Loading /boot/vmlinux-4.19.94...
>>> 13537288 bytes read in 107 ms (120.7 MiB/s)
>>> ## Flattened Device Tree blob at 82000000
>>>     Booting using the fdt blob at 0x82000000
>>>     Loading Device Tree to 00000000fdefa000, end 00000000fdf101e1 ... OK
>>>
>>> Starting kernel ...
>>>
>>>
>>> With the patch applied, I'm getting stuck like this:
>>>
>>> Found U-Boot script /boot/boot.scr
>>> 889 bytes read in 21 ms (41 KiB/s)
>>> ## Executing script at 83000000
>>> Loading /usr/lib/linux-image-4.19.94/ti/k3-am6548-iot2050-advanced.dtb...
>>> 78306 bytes read in 25 ms (3 MiB/s)
>>> Loading /boot/vmlinux-4.19.94...
>>> 13537288 bytes read in 109 ms (118.4 MiB/s)
>>> ## Flattened Device Tree blob at 82000000
>>>     Booting using the fdt blob at 0x82000000
>>>     Loading Device Tree to 00000000fffe9000, end 00000000fffff1e1 ...
>>>
>>>
>>> Obviously, the DT target adress changed, possibly to an
>>> unsupported/reserved address. But I do not understand the mechanics
>>> behind all this yet. Any hints welcome on what goes wrong here and
>>> whether something needs to be adjusted in our board settings.
>>
>> Can you share the output of bdinfo on this board ?
> 
> Sure (with your commit reverted for now):
> 
> IOT2050> bdinfo
> boot_params = 0x0000000000000000
> DRAM bank   = 0x0000000000000000
> -> start    = 0x0000000080000000
> -> size     = 0x0000000080000000
> flashstart  = 0x0000000000000000
> flashsize   = 0x0000000000000000
> flashoffset = 0x0000000000000000
> baudrate    = 115200 bps
> relocaddr   = 0x00000000fff2d000
> reloc off   = 0x000000007f72d000
> Build       = 64-bit
> current eth = unknown
> ethaddr     = 70:77:77:77:57:70
> IP addr     = <NULL>
> fdt_blob    = 0x00000000fdef9f10
> new_fdt     = 0x00000000fdef9f10
> fdt_size    = 0x0000000000012ea0
> lmb_dump_all:
>  memory.cnt  = 0x1
>  memory[0]      [0x80000000-0xffffffff], 0x80000000 bytes flags: 0
>  reserved.cnt  = 0x2
>  reserved[0]    [0x9e800000-0xa21fffff], 0x03a00000 bytes flags: 4
>  reserved[1]    [0xfdef86c0-0xffffffff], 0x02107940 bytes flags: 0
> arch_number = 0x0000000000000000
> TLB addr    = 0x00000000ffff0000
> irq_sp      = 0x00000000fdef9b20
> sp start    = 0x00000000fdef9b20
> Early malloc usage: 23c8 / 8000
> 
> 
> There is some "TLB" block apparently overlapping when we move the DT.
> What's this? Looking at the code, not a TLB with the usual meaning but
> rather the page table U-Boot uses for itself, right?
> 
> Were we just lucky so far with side effects of the LMB reservation on
> this platform (and I suspect that it affect k3 in general, not only our
> board)?
> 

Just rebased over master, kept Marek's change, and rerun bdinfo:

IOT2050> bdinfo
boot_params = 0x0000000000000000
DRAM bank   = 0x0000000000000000
-> start    = 0x0000000080000000
-> size     = 0x0000000040000000
flashstart  = 0x0000000000000000
flashsize   = 0x0000000000000000
flashoffset = 0x0000000000000000
baudrate    = 115200 bps
relocaddr   = 0x00000000bff41000
reloc off   = 0x000000003f741000
Build       = 64-bit
fdt_blob    = 0x00000000bdf0d750
new_fdt     = 0x00000000bdf0d750
fdt_size    = 0x0000000000013660
lmb_dump_all:
 memory.cnt  = 0x1
 memory[0]      [0x80000000-0xbfffffff], 0x40000000 bytes flags: 0
 reserved.cnt  = 0x1
 reserved[0]    [0x9e800000-0xa21fffff], 0x03a00000 bytes flags: 4
arch_number = 0x0000000000000000
TLB addr    = 0x00000000bfff0000
irq_sp      = 0x00000000bdf0d360
sp start    = 0x00000000bdf0d360
Early malloc usage: 23c8 / 8000
IOT2050> boot
switch to partitions #0, OK
mmc0 is current device
Scanning mmc 0:1...
Found U-Boot script /boot/boot.scr
911 bytes read in 28 ms (31.3 KiB/s)
## Executing script at 83000000
Loading /usr/lib/linux-image-4.19.94/ti/k3-am6528-iot2050-basic.dtb...
77409 bytes read in 34 ms (2.2 MiB/s)
Loading /boot/vmlinux-4.19.94...
13537288 bytes read in 587 ms (22 MiB/s)
## Flattened Device Tree blob at 82000000
   Booting using the fdt blob at 0x82000000
   Loading Device Tree to 00000000bffea000, end 00000000bffffe60 ...


>From that, I would definitely say that the fdt is loaded into the TLB
area, and that is likely the page table still in use at this stage.

Lokesh, Marek is overly busy and asked me to debug this further. But I
would need some indication if my interpretations are right and some hint
what is supposed to protect the TLB range from being used by other
things. And if that protection would be generic or board-specific.

I've checked the AM65x-SR1.0 EVM I have here, and I can see there as
well that mmu_setup() of U-Boot proper works with the same address range
the that boot_relocate_fdt() will later overwrite. It looks to me like
TLBs (the actual ones) are protecting us in some cases from the
consequences, at least on the EVM, not on our boards. Soon after the
overwriting, dcache and mmu will be turned off anyway for Linux.

Thanks,
Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux