| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Fri, 24 May 2024 11:47:57 +0000
From: Christophe Leroy <christophe.leroy@...roup.eu>
To: Oscar Salvador <osalvador@...e.de>
CC: Andrew Morton <akpm@...ux-foundation.org>, Jason Gunthorpe
<jgg@...dia.com>, Peter Xu <peterx@...hat.com>, Michael Ellerman
<mpe@...erman.id.au>, Nicholas Piggin <npiggin@...il.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"linux-mm@...ck.org" <linux-mm@...ck.org>, "linuxppc-dev@...ts.ozlabs.org"
<linuxppc-dev@...ts.ozlabs.org>
Subject: Re: [RFC PATCH v2 07/20] powerpc/8xx: Rework support for 8M pages
using contiguous PTE entries
Le 24/05/2024 à 12:02, Oscar Salvador a écrit :
> On Fri, May 17, 2024 at 09:00:01PM +0200, Christophe Leroy wrote:
>> In order to fit better with standard Linux page tables layout, add
>> support for 8M pages using contiguous PTE entries in a standard
>> page table. Page tables will then be populated with 1024 similar
>> entries and two PMD entries will point to that page table.
>>
>> The PMD entries also get a flag to tell it is addressing an 8M page,
>> this is required for the HW tablewalk assistance.
>>
>> Signed-off-by: Christophe Leroy <christophe.leroy@...roup.eu>
>
> I guess that this will slightly change if you remove patch#1 and patch#2
> as you said you will.
> So I will not comment on the overall design because I do not know how it will
> look afterwards, but just some things that caught my eye
Sure. I should send-out a v3 today or tomorrow, once I've done a few
more tests.
>
>> --- a/arch/powerpc/include/asm/hugetlb.h
>> +++ b/arch/powerpc/include/asm/hugetlb.h
>> @@ -41,7 +41,16 @@ void hugetlb_free_pgd_range(struct mmu_gather *tlb, unsigned long addr,
>> static inline pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
>> unsigned long addr, pte_t *ptep)
>> {
>> - return __pte(pte_update(mm, addr, ptep, ~0UL, 0, 1));
>> + pmd_t *pmdp = (pmd_t *)ptep;
>> + pte_t pte;
>> +
>> + if (IS_ENABLED(CONFIG_PPC_8xx) && pmdp == pmd_off(mm, ALIGN_DOWN(addr, SZ_8M))) {
>
> There are quite some places where you do the "pmd_off" to check whether that
> is a 8MB entry.
I refactored the code, now I have only two places with it: pte_update()
and huge_ptep_get()
By the way it doesn't check that PMD is 8M, it checks that the ptep
points to the first PMD entry matching the said address.
> I think it would make somse sense to have some kind of macro/function to make
> more clear what we are checking against.
> e.g:
>
> #define pmd_is_SZ_8M(mm, addr, pmdp) (pmdp == pmd_off(mm, ALIGN_DOWN(addr, SZ_8M)))
> (or whatever name you see fit)
>
> then you would just need
>
> if (IS_ENABLED(CONFIG_PPC_8xx && pmd_is_SZ_8M(mm, addr, pdmp))
>
> Because I see that is also scaterred in 8xx code.
>
>
>> + pte = __pte(pte_update(mm, addr, pte_offset_kernel(pmdp, 0), ~0UL, 0, 1));
>> + pte_update(mm, addr, pte_offset_kernel(pmdp + 1, 0), ~0UL, 0, 1);
>
> I have this fresh one because I recently read about 8xx pagetables, but not sure
> how my memory will survive this, so maybe throw a little comment in there that
> we are pointing the two pmds to the area.
The two PMD are now pointing to there own areas, we are not anymore in
the hugepd case where the PMD was pointing to a single HUGEPD containing
a single HUGEPTE.
>
> Also, the way we pass the parameters here to pte_update() is a bit awkward.
> Ideally we should be using some meaningful names?
>
> clr_all_bits = ~0UL
> set_bits = 0
> bool is_huge = true
>
> pte_update(mm, addr, pte_offset_kernel(pmdp + 1, 0), clr_all_bits, set_bits, is_huge)
>
> or something along those lines
Well, with my refactoring those functions are not modified anymore so I
won't change them.
>
>> -static inline int check_and_get_huge_psize(int shift)
>> -{
>> - return shift_to_mmu_psize(shift);
>> + if (pmdp == pmd_off(mm, ALIGN_DOWN(addr, SZ_8M)))
>
> Here you could also use the pmd_is_SZ_8M()
Yes, may do that.
>
>> + ptep = pte_offset_kernel(pmdp, 0);
>> + return ptep_get(ptep);
>> }
>>
>> #define __HAVE_ARCH_HUGE_SET_HUGE_PTE_AT
>> @@ -53,7 +33,14 @@ void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep,
>> static inline void huge_pte_clear(struct mm_struct *mm, unsigned long addr,
>> pte_t *ptep, unsigned long sz)
>> {
>> - pte_update(mm, addr, ptep, ~0UL, 0, 1);
>> + pmd_t *pmdp = (pmd_t *)ptep;
>> +
>> + if (pmdp == pmd_off(mm, ALIGN_DOWN(addr, SZ_8M))) {
>> + pte_update(mm, addr, pte_offset_kernel(pmdp, 0), ~0UL, 0, 1);
>> + pte_update(mm, addr, pte_offset_kernel(pmdp + 1, 0), ~0UL, 0, 1);
>> + } else {
>> + pte_update(mm, addr, ptep, ~0UL, 0, 1);
>> + }
>
> Could we not leverage this in huge_ptep_get_and_clear()?
I'm not modifying that anymore
> AFAICS,
>
> huge_pet_get_and_clear(mm, addr, pte_t *p)
> {
> pte_t pte = pte_val(*p);
>
> huge_pte_clear(mm, addr, p);
> return pte;
> }
>
> Or maybe it is not that easy if different powerpc platforms provide their own.
> It might be worth checking though.
>
>> }
>>
>> #define __HAVE_ARCH_HUGE_PTEP_SET_WRPROTECT
>> @@ -63,7 +50,14 @@ static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
>> unsigned long clr = ~pte_val(pte_wrprotect(__pte(~0)));
>> unsigned long set = pte_val(pte_wrprotect(__pte(0)));
>>
>> - pte_update(mm, addr, ptep, clr, set, 1);
>> + pmd_t *pmdp = (pmd_t *)ptep;
>> +
>> + if (pmdp == pmd_off(mm, ALIGN_DOWN(addr, SZ_8M))) {
>> + pte_update(mm, addr, pte_offset_kernel(pmdp, 0), clr, set, 1);
>> + pte_update(mm, addr, pte_offset_kernel(pmdp + 1, 0), clr, set, 1);
>> + } else {
>> + pte_update(mm, addr, ptep, clr, set, 1);
>
> I would replace the "1" with "is_huge" or "huge", as being done in
> __ptep_set_access_flags , something that makes it more clear without the need
> to check pte_update().
It's not modified anymore
>
>
>> #endif /* _ASM_POWERPC_PGALLOC_32_H */
>> diff --git a/arch/powerpc/include/asm/nohash/32/pte-8xx.h b/arch/powerpc/include/asm/nohash/32/pte-8xx.h
>> index 07df6b664861..b05cc4f87713 100644
>> --- a/arch/powerpc/include/asm/nohash/32/pte-8xx.h
>> +++ b/arch/powerpc/include/asm/nohash/32/pte-8xx.h
> ...
>> - * For other page sizes, we have a single entry in the table.
>> + * For 8M pages, we have 1024 entries as if it was
>> + * 4M pages, but they are flagged as 8M pages for the hardware.
>
> Maybe drop a comment that a single PMD entry is worth 4MB, so
Ok, added that the 4M is indeed PMD_SIZE
>
>> + * For 4k pages, we have a single entry in the table.
>> */
>> -static pmd_t *pmd_off(struct mm_struct *mm, unsigned long addr);
>> -static int hugepd_ok(hugepd_t hpd);
>> -
>> static inline int number_of_cells_per_pte(pmd_t *pmd, pte_basic_t val, int huge)
>> {
>> if (!huge)
>> return PAGE_SIZE / SZ_4K;
>> - else if (hugepd_ok(*((hugepd_t *)pmd)))
>> - return 1;
>> + else if ((pmd_val(*pmd) & _PMD_PAGE_MASK) == _PMD_PAGE_8M)
>> + return SZ_4M / SZ_4K;
>
> this becomes more intuitive.
>
>
>> +static inline void pmd_populate_kernel_size(struct mm_struct *mm, pmd_t *pmdp,
>> + pte_t *pte, unsigned long sz)
>> +{
>> + if (sz == SZ_8M)
>> + *pmdp = __pmd(__pa(pte) | _PMD_PRESENT | _PMD_PAGE_8M);
>> + else
>> + *pmdp = __pmd(__pa(pte) | _PMD_PRESENT);
>> +}
>> +
>> +static inline void pmd_populate_size(struct mm_struct *mm, pmd_t *pmdp,
>> + pgtable_t pte_page, unsigned long sz)
>> +{
>> + if (sz == SZ_8M)
>> + *pmdp = __pmd(__pa(pte_page) | _PMD_USER | _PMD_PRESENT | _PMD_PAGE_8M);
>> + else
>> + *pmdp = __pmd(__pa(pte_page) | _PMD_USER | _PMD_PRESENT);
>> +}
>
> In patch#1 you mentioned this will change with the removal of patch#1
> and patch#2.
Yes this goes away.
>
>> --- a/arch/powerpc/mm/hugetlbpage.c
>> +++ b/arch/powerpc/mm/hugetlbpage.c
>> @@ -183,9 +183,6 @@ pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
>> if (!hpdp)
>> return NULL;
>>
>> - if (IS_ENABLED(CONFIG_PPC_8xx) && pshift < PMD_SHIFT)
>> - return pte_alloc_huge(mm, (pmd_t *)hpdp, addr, sz);
>> -
>> BUG_ON(!hugepd_none(*hpdp) && !hugepd_ok(*hpdp));
>>
>> if (hugepd_none(*hpdp) && __hugepte_alloc(mm, hpdp, addr,
>> @@ -198,10 +195,18 @@ pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
>> pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
>> unsigned long addr, unsigned long sz)
>> {
>> + pmd_t *pmd = pmd_off(mm, addr);
>> +
>> if (sz < PMD_SIZE)
>> - return pte_alloc_huge(mm, pmd_off(mm, addr), addr, sz);
>> + return pte_alloc_huge(mm, pmd, addr, sz);
>>
>> - return NULL;
>> + if (sz != SZ_8M)
>> + return NULL;
>> + if (!pte_alloc_huge(mm, pmd, addr, sz))
>> + return NULL;
>> + if (!pte_alloc_huge(mm, pmd + 1, addr, sz))
>> + return NULL;
>> + return (pte_t *)pmd;
>
> I think that having the check for invalid huge page sizes upfront would
> make more sense, maybe just a matter of taste.
Well, it would make it less easy when we go one step further to support
e500 and book3s/64. I prefer to do it that way to keep it as flat as
possible and avoid a deep if ... if ... if
By the way I have now squashed patch 11 into patch 5.
>
> /* Unsupported size */
> if (sz > PMD_SIZE && sz = SZ_8M)
> return NULL;
>
> if (sz < PMD_SIZE)
> ...
> /* 8MB huge pages */
> ...
>
> return (pte_t *) pmd;
>
> Also, I am not a big fan of the two separate pte_alloc_huge() for pmd#0+pmd#1,
> and I am thinking we might want to hide that within a function and drop a
> comment in there explaining why we are updatng both pmds.
Now changed to:
+ for (i = 0; i < sz / PMD_SIZE; i++) {
+ if (!pte_alloc_huge(mm, pmd + i, addr))
+ return NULL;
+ }
>
>
>
>> diff --git a/arch/powerpc/mm/nohash/8xx.c b/arch/powerpc/mm/nohash/8xx.c
>> index d93433e26ded..99f656b3f9f3 100644
>> --- a/arch/powerpc/mm/nohash/8xx.c
>> +++ b/arch/powerpc/mm/nohash/8xx.c
>> @@ -48,20 +48,6 @@ unsigned long p_block_mapped(phys_addr_t pa)
>> return 0;
>> }
>>
>> -static pte_t __init *early_hugepd_alloc_kernel(hugepd_t *pmdp, unsigned long va)
>> -{
>> - if (hpd_val(*pmdp) == 0) {
>> - pte_t *ptep = memblock_alloc(sizeof(pte_basic_t), SZ_4K);
>> -
>> - if (!ptep)
>> - return NULL;
>> -
>> - hugepd_populate_kernel((hugepd_t *)pmdp, ptep, PAGE_SHIFT_8M);
>> - hugepd_populate_kernel((hugepd_t *)pmdp + 1, ptep, PAGE_SHIFT_8M);
>> - }
>> - return hugepte_offset(*(hugepd_t *)pmdp, va, PGDIR_SHIFT);
>> -}
>> -
>> static int __ref __early_map_kernel_hugepage(unsigned long va, phys_addr_t pa,
>> pgprot_t prot, int psize, bool new)
>
> Am I blind or do we never use the 'new' parameter?
> I checked the tree and it seems we always pass it 'true'.
You must be blind :)
$ git grep mmu_mapin_ram_chunk
arch/powerpc/mm/nohash/8xx.c:static int mmu_mapin_ram_chunk(unsigned
long offset, unsigned long top,
arch/powerpc/mm/nohash/8xx.c: mmu_mapin_ram_chunk(0, boundary,
PAGE_KERNEL_TEXT, true);
arch/powerpc/mm/nohash/8xx.c: mmu_mapin_ram_chunk(boundary,
einittext8, PAGE_KERNEL_TEXT, true);
arch/powerpc/mm/nohash/8xx.c: mmu_mapin_ram_chunk(einittext8,
top, PAGE_KERNEL, true);
arch/powerpc/mm/nohash/8xx.c: err =
mmu_mapin_ram_chunk(boundary, einittext8, PAGE_KERNEL, false);
arch/powerpc/mm/nohash/8xx.c: err = mmu_mapin_ram_chunk(0, sinittext,
PAGE_KERNEL_ROX, false);
>
> arch/powerpc/mm/nohash/8xx.c: err = __early_map_kernel_hugepage(v, p, prot, MMU_PAGE_512K, new);
> arch/powerpc/mm/nohash/8xx.c: err = __early_map_kernel_hugepage(v, p, prot, MMU_PAGE_8M, new);
> arch/powerpc/mm/nohash/8xx.c: err = __early_map_kernel_hugepage(v, p, prot, MMU_PAGE_512K, new);
> arch/powerpc/mm/nohash/8xx.c:
> __early_map_kernel_hugepage(VIRT_IMMR_BASE, PHYS_IMMR_BASE, PAGE_KERNEL_NCG, MMU_PAGE_512K, true);
>
> I think we can drop the 'new' and the block code that tries to handle
> it?
>
>> diff --git a/arch/powerpc/mm/pgtable.c b/arch/powerpc/mm/pgtable.c
>> index acdf64c9b93e..59f0d7706d2f 100644
>> --- a/arch/powerpc/mm/pgtable.c
>> +++ b/arch/powerpc/mm/pgtable.c
>
>> +void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep,
>> + pte_t pte, unsigned long sz)
>> +{
>> + pmd_t *pmdp = pmd_off(mm, addr);
>> +
>> + pte = set_pte_filter(pte, addr);
>> +
>> + if (sz == SZ_8M) {
>> + __set_huge_pte_at(pmdp, pte_offset_kernel(pmdp, 0), pte_val(pte));
>> + __set_huge_pte_at(pmdp, pte_offset_kernel(pmdp + 1, 0), pte_val(pte) + SZ_4M);
>
> You also mentioned that this would slightly change after you drop
> patch#0 and patch#1.
> The only comment I have right know would be to add a little comment
> explaining the layout (the replication of 1024 entries), or just
> something like "see comment from number_of_cells_per_pte".
>
Done.
Thanks for the rewiew.
Christophe
Powered by blists - more mailing lists