Re: [PATCH bpf-next v2 0/5] execmem_alloc for BPF programs

[Christophe Leroy <christophe.leroy@csgroup.eu>]

Le 07/12/2022 à 22:04, Thomas Gleixner a écrit :
> Christophe,
> 
> On Wed, Dec 07 2022 at 16:53, Christophe Leroy wrote:
>> Le 07/12/2022 à 16:36, Thomas Gleixner a écrit :
>>> The "use free space in existing mappings" mechanism is not required to
>>> be PMD_SIZE based, right?
> 
> Correct. I just used it for the example.
> 
>>> Large page size, strict separation:
>>>
>>> struct mod_alloc_type_params {
>>>      	[MOD_ALLOC_TYPE_TEXT] = {
>>>           	.mapto_type	= MOD_ALLOC_TYPE_TEXT,
>>>                   .flags		= FLAG_SHARED_PMD | FLAG_SECOND_ADDRESS_SPACE,
>>>                   .granularity	= PMD_SIZE,
>>>                   .alignment	= MOD_ARCH_ALIGNMENT,
>>>                   .start[0]	= MODULES_VADDR,
>>>                   .end[0]		= MODULES_END,
>>>                   .start[1]	= MODULES_VADDR_2ND,
>>>                   .end[1]		= MODULES_END_2ND,
>>>                   .pgprot		= PAGE_KERNEL_EXEC,
>>>                   .fill		= text_poke,
>>>                   .invalidate	= text_poke_invalidate,
>>> 	},
>>
>> Don't restrict implementation to PMD_SIZE only.
>>
>> On powerpc 8xx:
>> - PMD_SIZE is 4 Mbytes
>> - Large pages are 512 kbytes and 8 Mbytes.
>>
>> It even has large pages of size 16 kbytes when build for 4k normal page
>> size.
> 
> @granularity takes any size which is valid from the architecture side
> and can handle the @pgprot distinctions.
> 
> That's why I separated functionality and configuration.
> 
> Note, it's not strict compile time configuration. You can either build
> it completely dynamic at boot or have a static configuration structure
> as compile time default.
> 
> That static default might be __initconst for architectures where there
> is no boot time detection and change required, but can be __initdata for
> those which need to adjust it to the needs of the detected CPU/platform
> before feeding it into the module allocator init function.
> 
> Does that answer your question?
> 

Yes it does, thanks.

Christophe