Re: [RFT PATCH v4 3/9] RISC-V: Implement late mapping page table allocation functions

[Atish Patra]

On Thu, Jul 30, 2020 at 2:14 AM Mike Rapoport <rppt@xxxxxxxxxx> wrote:
>
> Hi,
>
> On Wed, Jul 29, 2020 at 04:36:29PM -0700, Atish Patra wrote:
> > Currently, page table setup is done during setup_va_final where fixmap can
> > be used to create the temporary mappings. The physical frame is allocated
> > from memblock_alloc_* functions. However, this won't work if page table
> > mapping needs to be created for a different mm context (i.e. efi mm) at
> > a later point of time.
>
> TBH, I'm not very happy to see pte/pmd allocations, but I don't see a
> way to reuse the existing routines in this case.
>
> As a general wild idea, maybe it's worth using something like
>
> struct pt_alloc_ops {
>         pte_t *(*get_pte_virt)(phys_addr_t pa);
>         phys_addr_t (*alloc_pte)(uintptr_t va);
>         ...
> };
>
> and add there implementations: nommu, MMU early and MMU late.
>

Yeah. That will be much cleaner. Thanks. I will change it to have
function pointers
instead of if else blocks.

> Some more comments below.
>
> > Use generic kernel page allocation function & macros for any mapping
> > after setup_vm_final.
> >
> > Signed-off-by: Atish Patra <atish.patra@xxxxxxx>
> > ---
> >  arch/riscv/mm/init.c | 29 ++++++++++++++++++++++++-----
> >  1 file changed, 24 insertions(+), 5 deletions(-)
> >
> > diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
> > index 68c608a0e91f..cba03fec08c1 100644
> > --- a/arch/riscv/mm/init.c
> > +++ b/arch/riscv/mm/init.c
> > @@ -212,6 +212,7 @@ pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
> >  pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
> >  pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
> >  static bool mmu_enabled;
> > +static bool late_mapping;
> >
> >  #define MAX_EARLY_MAPPING_SIZE       SZ_128M
> >
> > @@ -236,7 +237,9 @@ void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
> >
> >  static pte_t *__init get_pte_virt(phys_addr_t pa)
> >  {
> > -     if (mmu_enabled) {
> > +     if (late_mapping)
> > +             return (pte_t *) __va(pa);
> > +     else if (mmu_enabled) {
> >               clear_fixmap(FIX_PTE);
> >               return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
> >       } else {
> > @@ -246,13 +249,19 @@ static pte_t *__init get_pte_virt(phys_addr_t pa)
> >
> >  static phys_addr_t __init alloc_pte(uintptr_t va)
> >  {
> > +     unsigned long vaddr;
> >       /*
> >        * We only create PMD or PGD early mappings so we
> >        * should never reach here with MMU disabled.
> >        */
> >       BUG_ON(!mmu_enabled);
> > -
> > -     return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
> > +     if (late_mapping) {
> > +             vaddr = __get_free_page(GFP_KERNEL);
> > +             if (!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)))
> > +                     BUG();
>
> Is BUG() here really necessary? If I understand correctly, this would be
> used to enable mappings for EFI runtime services, so we probably want to
> propagete the error to to caller and, if really necessary, BUG() there,
> don't we?
>

If __get_free_page is failing here, something is seriously wrong and
the system should panic.
But I agree with you that this should happen in the caller rather than
the callee.
We need to change the signature of each create_*_mapping function
in order to propagate the error to the caller. Do you see any issues with that ?
As it will affect page table allocation for all three cases, should we
consolidate all the BUG()
cases in caller only ?

> > +             return __pa(vaddr);
> > +     } else
> > +             return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
> >  }
> >
> >  static void __init create_pte_mapping(pte_t *ptep,
> > @@ -281,7 +290,9 @@ pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
> >
> >  static pmd_t *__init get_pmd_virt(phys_addr_t pa)
> >  {
> > -     if (mmu_enabled) {
> > +     if (late_mapping)
> > +             return (pmd_t *) __va(pa);
> > +     else if (mmu_enabled) {
> >               clear_fixmap(FIX_PMD);
> >               return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
> >       } else {
> > @@ -292,8 +303,13 @@ static pmd_t *__init get_pmd_virt(phys_addr_t pa)
> >  static phys_addr_t __init alloc_pmd(uintptr_t va)
> >  {
> >       uintptr_t pmd_num;
> > +     unsigned long vaddr;
> >
> > -     if (mmu_enabled)
> > +     if (late_mapping) {
> > +             vaddr = __get_free_page(GFP_KERNEL);
> > +             BUG_ON(!vaddr);
> > +             return __pa(vaddr);
>
> Does nommu also need to allocate a page for pmd?
>

Not sure if I understand the question correctly. Before MMU is enabled,
we are using statically allocated early_pmd.

> > +     } else if (mmu_enabled)
> >               return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
> >
> >       pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
> > @@ -533,6 +549,9 @@ static void __init setup_vm_final(void)
> >       /* Move to swapper page table */
> >       csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
> >       local_flush_tlb_all();
> > +
> > +     /* generic page allocation function must be used to setup page table */
> > +     late_mapping = true;
> >  }
> >  #else
> >  asmlinkage void __init setup_vm(uintptr_t dtb_pa)
> > --
> > 2.24.0
> >
>
> --
> Sincerely yours,
> Mike.



-- 
Regards,
Atish