Re: [PATCH v3 4/6] arm64: Utilize phys_initrd_start/phys_initrd_size

From: Ard Biesheuvel
Date: Mon Nov 05 2018 - 16:07:37 EST


On 5 November 2018 at 22:05, Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
> On 11/5/18 1:00 PM, Ard Biesheuvel wrote:
>> On 5 November 2018 at 21:51, Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
>>> On 11/5/18 12:44 PM, Ard Biesheuvel wrote:
>>>> On 5 November 2018 at 21:41, Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
>>>>> On 11/5/18 12:39 PM, Ard Biesheuvel wrote:
>>>>>> Hi Florian,
>>>>>>
>>>>>> On 31 October 2018 at 20:28, Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
>>>>>>> ARM64 is the only architecture that re-defines
>>>>>>> __early_init_dt_declare_initrd() in order for that function to populate
>>>>>>> initrd_start/initrd_end with physical addresses instead of virtual
>>>>>>> addresses. Instead of having an override we can leverage
>>>>>>> drivers/of/fdt.c populating phys_initrd_start/phys_initrd_size to
>>>>>>> populate those variables for us.
>>>>>>>
>>>>>>> Signed-off-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
>>>>>>> ---
>>>>>>> arch/arm64/mm/init.c | 19 +++++++++----------
>>>>>>> 1 file changed, 9 insertions(+), 10 deletions(-)
>>>>>>>
>>>>>>> diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
>>>>>>> index 3cf87341859f..00ef2166bb73 100644
>>>>>>> --- a/arch/arm64/mm/init.c
>>>>>>> +++ b/arch/arm64/mm/init.c
>>>>>>> @@ -72,8 +72,8 @@ static int __init early_initrd(char *p)
>>>>>>> if (*endp == ',') {
>>>>>>> size = memparse(endp + 1, NULL);
>>>>>>>
>>>>>>> - initrd_start = start;
>>>>>>> - initrd_end = start + size;
>>>>>>> + phys_initrd_start = start;
>>>>>>> + phys_initrd_size = size;
>>>>>>> }
>>>>>>> return 0;
>>>>>>> }
>>>>>>> @@ -408,14 +408,14 @@ void __init arm64_memblock_init(void)
>>>>>>> memblock_add(__pa_symbol(_text), (u64)(_end - _text));
>>>>>>> }
>>>>>>>
>>>>>>> - if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
>>>>>>> + if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
>>>>>>> /*
>>>>>>> * Add back the memory we just removed if it results in the
>>>>>>> * initrd to become inaccessible via the linear mapping.
>>>>>>> * Otherwise, this is a no-op
>>>>>>> */
>>>>>>> - u64 base = initrd_start & PAGE_MASK;
>>>>>>> - u64 size = PAGE_ALIGN(initrd_end) - base;
>>>>>>> + u64 base = phys_initrd_start & PAGE_MASK;
>>>>>>> + u64 size = PAGE_ALIGN(phys_initrd_size);
>>>>>>>
>>>>>>> /*
>>>>>>> * We can only add back the initrd memory if we don't end up
>>>>>>> @@ -460,12 +460,11 @@ void __init arm64_memblock_init(void)
>>>>>>> */
>>>>>>> memblock_reserve(__pa_symbol(_text), _end - _text);
>>>>>>> #ifdef CONFIG_BLK_DEV_INITRD
>>>>>>> - if (initrd_start) {
>>>>>>> - memblock_reserve(initrd_start, initrd_end - initrd_start);
>>>>>>> -
>>>>>>> + if (phys_initrd_size) {
>>>>>>> /* the generic initrd code expects virtual addresses */
>>>>>>> - initrd_start = __phys_to_virt(initrd_start);
>>>>>>> - initrd_end = __phys_to_virt(initrd_end);
>>>>>>> + initrd_start = __phys_to_virt(phys_initrd_start);
>>>>>>> + initrd_end = initrd_start + phys_initrd_size;
>>>>>>> + initrd_below_start_ok = 0;
>>>>>>
>>>>>> Where is this assignment coming from?
>>>>>
>>>>> __early_init_dt_declare_initrd() sets initrd_below_start_ok to 1 though
>>>>> after patch #5 this is not necessary any more.
>>>>
>>>> Yes, but why? The original arm64 version of
>>>> __early_init_dt_declare_initrd() does not set it but now you set to 1
>>>> in the IS_ENABLED(CONFIG_ARM64) section in the generic code and set it
>>>> back to 0 here.
>>>
>>> Humm, it is an if (!IS_ENABLED(CONFIG_ARM64)) condition, so we would not
>>> be taking that branch on an ARM64 kernel.
>>>
>>
>> Right. So now that we are not setting it to 1 on arm64, there is no
>> longer a reason to set it to 0 again, no?
>
> Correct, and in fact, this is not a problem either at patch #4 (which
> has the custom __early_init_dt_declare_initrd()) or #5 (which removes
> it), any other feedback you would like me to address before addressing
> Rob's suggestion?
>

No, I think this is ok. The conditional on arm64 in generic code is a
bit nasty, and it would be nicer generally if we only have to record a
single memory range, but if this fixes the issue you were addressing
originally, I'm fine with it.