[U-Boot] [PATCH 4/9] efi_loader: Add boot time services

[Alexander Graf]

On 15.01.16 14:02, Leif Lindholm wrote:
> On Fri, Jan 15, 2016 at 01:13:15AM +0100, Alexander Graf wrote:
>> On 26.12.15 19:09, Leif Lindholm wrote:
>>>> diff --git a/lib/efi_loader/efi_boottime.c b/lib/efi_loader/efi_boottime.c
>>>> new file mode 100644
>>>> index 0000000..ed95962
>>>> --- /dev/null
>>>> +++ b/lib/efi_loader/efi_boottime.c
>>>> @@ -0,0 +1,838 @@
>>>> +/*
>>>> + *  EFI application boot time services
>>>> + *
> ...
>>>> +
>>>> +static unsigned long efi_raise_tpl(unsigned long new_tpl)
>>>> +{
>>>> +	EFI_ENTRY("0x%lx", new_tpl);
>>>> +	return EFI_EXIT(efi_unsupported(__func__));
>>>
>>> "Unlike other UEFI interface functions, EFI_BOOT_SERVICES.RaiseTPL()
>>> does not return a status code. Instead, it returns the previous task
>>> priority level, which is to be restored later with a matching call to
>>> RestoreTPL()."
>>
>> Since we don't do TPLs (or IRQs for that matter), I'll just return 0 here.
> 
> Sure.
> 
>>>> +}
>>>> +
>>>> +static void efi_restore_tpl(unsigned long old_tpl)
>>>> +{
>>>> +	EFI_ENTRY("0x%lx", old_tpl);
>>>> +	EFI_EXIT(efi_unsupported(__func__));
>>>
>>> (void function, nothing to return)
>>
>> Yes, hence no return. EFI_EXIT deals with the gd swapping and
>> efi_unsupported() gives me a nice debug message :).
> 
> Ah, ok.
> 
>>>> +static efi_status_t efi_allocate_pages(int type, int memory_type,
>>>> +				       unsigned long pages, uint64_t *memory)
>>>> +{
>>>> +	u64 len = pages << 12;
>>>> +	efi_status_t r = EFI_SUCCESS;
>>>> +
>>>> +	EFI_ENTRY("%d, %d, 0x%lx, %p", type, memory_type, pages, memory);
>>>> +
>>>> +	switch (type) {
>>>> +	case 0:
>>>> +		/* Any page means we can go to efi_alloc */
>>>> +		*memory = (unsigned long)efi_alloc(len, memory_type);
>>>> +		break;
>>>> +	case 1:
>>>> +		/* Max address */
>>>> +		if (gd->relocaddr < *memory) {
>>>> +			*memory = (unsigned long)efi_alloc(len, memory_type);
>>>> +			break;
>>>> +		}
>>>> +		r = EFI_UNSUPPORTED;
>>>
>>> EFI_OUT_OF_RESOURCES/EFI_NOT_FOUND?
>>>
>>>> +		break;
>>>> +	case 2:
>>>> +		/* Exact address, grant it. The addr is already in *memory. */
>>>
>>> As far as I can tell, this is why GRUB works. Because it filters
>>> through the memory map manually, requesting to allocate its heap at an
>>> exact address in a region of free memory in the UEFI memory map.
>>
>> Yes.
>>
>>> The key is that EFI_LOADER_MEMORY will be used by applications loaded
>>> as well as by U-Boot to load applications into. A simple example where
>>> this could be problematic would be a large(ish) initrd loaded via initrd=
>>> on kernel (stub loader) command line rather than via GRUB.
>>
>> Ah, so here the 128MB limit on the LOADER_DATA section might bite us?
> 
> Yeah, I think so.
> 
>>>> +	runtime_start = (ulong)&__efi_runtime_start & ~0xfffULL;
>>>> +	runtime_end = ((ulong)&__efi_runtime_stop + 0xfff) & ~0xfffULL;
>>>> +	runtime_len_pages = (runtime_end - runtime_start) >> 12;
>>>> +	runtime_len = runtime_len_pages << 12;
>>>> +
>>>> +	/* Fill in where normal RAM is (up to U-Boot) */
>>>> +	efi_memory_map[0].num_pages = gd->relocaddr >> 12;
>>>
>>> U-Boot question: is gd->relocaddr always the offset from start of RAM?
>>> How does this work with gaps in memory map?
>>
>> U-Boot always relocates itself at TOM (or at least what we consider TOM
>> here). gd->relocaddr is the physical address of the start of U-Boot
>> which is right below TOM.
> 
> Still ... would need additional handling if memory has holes (like,
> fitted with multiple DIMMs smaller than the individual memory window
> reserved for them). Or even on something like Juno, which has 2GB of
> RAM at 0x00_8000_0000, and a further 6GB at 0x08_8000_0000 (I think).

Yes. I think we'll have to just implement that when we hit the first
board that does this (and has awareness in U-Boot for it).

With the UEFI entry path, Linux will ignore memory nodes in dt, right?

> 
>>>> +#ifdef CONFIG_SYS_SDRAM_BASE
>>>> +	efi_memory_map[0].physical_start = CONFIG_SYS_SDRAM_BASE;
>>>> +	efi_memory_map[0].virtual_start = CONFIG_SYS_SDRAM_BASE;
>>>> +	efi_memory_map[0].num_pages -= CONFIG_SYS_SDRAM_BASE >> 12;
>>> #else
>>> #error "..."
>>> ?
>>
>> If it's not defined, it's 0 :).
> 
> Make it
> #if CONFIG_SYS_SDRAM_BASE != 0
> for clarity?

I really don't think we'll make it more readable with more #iffery here :).


Alex