Re: [PATCH 3/3] LoongArch: mm: Add unified function populate_kernel_pte

From: bibo mao
Date: Thu Aug 10 2023 - 00:43:09 EST

Next message: Jie Luo: "Re: [PATCH v1 3/4] clk: qcom: add clock controller driver for qca8386/qca8084"
Previous message: Manikanta Pubbisetty: "Re: [PATCH] Revert "Revert "wifi: ath11k: Enable threaded NAPI"""
In reply to: Huacai Chen: "Re: [PATCH 3/3] LoongArch: mm: Add unified function populate_kernel_pte"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

在 2023/8/10 12:27, Huacai Chen 写道:
> On Thu, Aug 10, 2023 at 12:09 PM bibo mao <maobibo@xxxxxxxxxxx> wrote:
>>
>>
>>
>> 在 2023/8/2 15:25, Huacai Chen 写道:
>>> On Tue, Aug 1, 2023 at 9:22 AM bibo mao <maobibo@xxxxxxxxxxx> wrote:
>>>>
>>>>
>>>>
>>>> 在 2023/7/31 22:15, Huacai Chen 写道:
>>>>> On Wed, Jul 12, 2023 at 11:16 AM Bibo Mao <maobibo@xxxxxxxxxxx> wrote:
>>>>>>
>>>>>> Function pcpu_populate_pte and fixmap_pte are similar, they populate
>>>>>> one page from kernel address space. And there is confusion between
>>>>>> pgd and p4d in function fixmap_pte, such as pgd_none always returns
>>>>>> zero. This patch adds unified function populate_kernel_pte and replaces
>>>>>> pcpu_populate_pte and fixmap_pte.
>>>>>>
>>>>>> Signed-off-by: Bibo Mao <maobibo@xxxxxxxxxxx>
>>>>>> ---
>>>>>> arch/loongarch/include/asm/pgalloc.h | 1 +
>>>>>> arch/loongarch/kernel/numa.c | 40 +--------------------
>>>>>> arch/loongarch/mm/init.c | 52 ++++++++++++++++------------
>>>>>> 3 files changed, 32 insertions(+), 61 deletions(-)
>>>>>>
>>>>>> diff --git a/arch/loongarch/include/asm/pgalloc.h b/arch/loongarch/include/asm/pgalloc.h
>>>>>> index af1d1e4a6965..ca17b573dba6 100644
>>>>>> --- a/arch/loongarch/include/asm/pgalloc.h
>>>>>> +++ b/arch/loongarch/include/asm/pgalloc.h
>>>>>> @@ -91,4 +91,5 @@ static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long address)
>>>>>>
>>>>>> #endif /* __PAGETABLE_PUD_FOLDED */
>>>>>>
>>>>>> +extern pte_t * __init populate_kernel_pte(unsigned long addr);
>>>>>> #endif /* _ASM_PGALLOC_H */
>>>>>> diff --git a/arch/loongarch/kernel/numa.c b/arch/loongarch/kernel/numa.c
>>>>>> index 778e1c20bfb0..24a693b76873 100644
>>>>>> --- a/arch/loongarch/kernel/numa.c
>>>>>> +++ b/arch/loongarch/kernel/numa.c
>>>>>> @@ -67,46 +67,8 @@ static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
>>>>>>
>>>>>> void __init pcpu_populate_pte(unsigned long addr)
>>>>>> {
>>>>>> - pgd_t *pgd = pgd_offset_k(addr);
>>>>>> - p4d_t *p4d = p4d_offset(pgd, addr);
>>>>>> - pud_t *pud;
>>>>>> - pmd_t *pmd;
>>>>>> -
>>>>>> - if (p4d_none(*p4d)) {
>>>>>> - pud = memblock_alloc_raw(PAGE_SIZE, PAGE_SIZE);
>>>>>> - if (!pud)
>>>>>> - goto err_alloc;
>>>>>> - p4d_populate(&init_mm, p4d, pud);
>>>>>> -#ifndef __PAGETABLE_PUD_FOLDED
>>>>>> - pud_init(pud);
>>>>>> -#endif
>>>>>> - }
>>>>>> -
>>>>>> - pud = pud_offset(p4d, addr);
>>>>>> - if (pud_none(*pud)) {
>>>>>> - pmd = memblock_alloc_raw(PAGE_SIZE, PAGE_SIZE);
>>>>>> - if (!pmd)
>>>>>> - goto err_alloc;
>>>>>> - pud_populate(&init_mm, pud, pmd);
>>>>>> -#ifndef __PAGETABLE_PMD_FOLDED
>>>>>> - pmd_init(pmd);
>>>>>> -#endif
>>>>>> - }
>>>>>> -
>>>>>> - pmd = pmd_offset(pud, addr);
>>>>>> - if (!pmd_present(*pmd)) {
>>>>>> - pte_t *pte;
>>>>>> -
>>>>>> - pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
>>>>>> - if (!pte)
>>>>>> - goto err_alloc;
>>>>>> - pmd_populate_kernel(&init_mm, pmd, pte);
>>>>>> - }
>>>>>> -
>>>>>> + populate_kernel_pte(addr);
>>>>>> return;
>>>>>> -
>>>>>> -err_alloc:
>>>>>> - panic("%s: Failed to allocate memory\n", __func__);
>>>>>> }
>>>>>>
>>>>>> void __init setup_per_cpu_areas(void)
>>>>>> diff --git a/arch/loongarch/mm/init.c b/arch/loongarch/mm/init.c
>>>>>> index 3b7d8129570b..6cd2948373ae 100644
>>>>>> --- a/arch/loongarch/mm/init.c
>>>>>> +++ b/arch/loongarch/mm/init.c
>>>>>> @@ -191,46 +191,49 @@ void vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *al
>>>>>> #endif
>>>>>> #endif
>>>>>>
>>>>>> -static pte_t *fixmap_pte(unsigned long addr)
>>>>>> +pte_t * __init populate_kernel_pte(unsigned long addr)
>>>>>> {
>>>>>> - pgd_t *pgd;
>>>>>> - p4d_t *p4d;
>>>>>> + pgd_t *pgd = pgd_offset_k(addr);
>>>>>> + p4d_t *p4d = p4d_offset(pgd, addr);
>>>>>> pud_t *pud;
>>>>>> pmd_t *pmd;
>>>>>>
>>>>>> - pgd = pgd_offset_k(addr);
>>>>>> - p4d = p4d_offset(pgd, addr);
>>>>>> -
>>>>>> - if (pgd_none(*pgd)) {
>>>>>> - pud_t *new __maybe_unused;
>>>>>> -
>>>>>> - new = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
>>>>>> - pgd_populate(&init_mm, pgd, new);
>>>>>> + if (p4d_none(*p4d)) {
>>>>>> + pud = memblock_alloc_raw(PAGE_SIZE, PAGE_SIZE);
>>>>>> + if (!pud)
>>>>>> + goto err_alloc;
>>>>>> + p4d_populate(&init_mm, p4d, pud);
>>>>>> #ifndef __PAGETABLE_PUD_FOLDED
>>>>>> - pud_init(new);
>>>>>> + pud_init(pud);
>>>>>> #endif
>>>>>> }
>>>>>>
>>>>>> pud = pud_offset(p4d, addr);
>>>>>> if (pud_none(*pud)) {
>>>>>> - pmd_t *new __maybe_unused;
>>>>>> -
>>>>>> - new = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
>>>>>> - pud_populate(&init_mm, pud, new);
>>>>>> + pmd = memblock_alloc_raw(PAGE_SIZE, PAGE_SIZE);
>>>>>> + if (!pmd)
>>>>>> + goto err_alloc;
>>>>>> + pud_populate(&init_mm, pud, pmd);
>>>>>> #ifndef __PAGETABLE_PMD_FOLDED
>>>>>> - pmd_init(new);
>>>>>> + pmd_init(pmd);
>>>>>> #endif
>>>>>> }
>>>>>>
>>>>>> pmd = pmd_offset(pud, addr);
>>>>>> - if (pmd_none(*pmd)) {
>>>>>> - pte_t *new __maybe_unused;
>>>>>> + if (!pmd_present(*pmd)) {
>>>>>> + pte_t *pte;
>>>>>>
>>>>>> - new = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
>>>>>> - pmd_populate_kernel(&init_mm, pmd, new);
>>>>>> + pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
>>>>> I don't think memblock_alloc_low() here can be replaced by memblock_alloc().
>>>> Can you share me the points that pte table must be allocated with memblock_alloc_low
>>>> in this place?
>>> I forget the reason now, so if you confirm memblock_alloc() works well
>>> here, you can use it. But please don't use memblock_alloc_raw().
>> what a mess, there is more comments if there is special reason, else everyone can
>> forgot by elapsed time.
>>
>> why the function memblock_alloc_raw can not be use? there is one useless page copy.
> This is not a performance critical path, keeping consistency with
> mm/percpu.c can make life easier.
yes, it is not critical path, it influences boot speed. Can we make it better? else it
will be just so so.

Regards
Bibo Mao
>
> Huacai
>
>>
>> Regards
>> Bibo Mao
>>
>>
>>>
>>> Huacai
>>>>
>>>> Regards
>>>> Bibo Mao
>>>>>
>>>>>
>>>>> Huacai
>>>>>> + if (!pte)
>>>>>> + goto err_alloc;
>>>>>> + pmd_populate_kernel(&init_mm, pmd, pte);
>>>>>> }
>>>>>>
>>>>>> return pte_offset_kernel(pmd, addr);
>>>>>> +
>>>>>> +err_alloc:
>>>>>> + panic("%s: Failed to allocate memory\n", __func__);
>>>>>> + return NULL;
>>>>>> }
>>>>>>
>>>>>> void __init __set_fixmap(enum fixed_addresses idx,
>>>>>> @@ -241,7 +244,12 @@ void __init __set_fixmap(enum fixed_addresses idx,
>>>>>>
>>>>>> BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
>>>>>>
>>>>>> - ptep = fixmap_pte(addr);
>>>>>> + /*
>>>>>> + * Now only FIX_EARLYCON_MEM_BASE fixed map is used
>>>>>> + * __set_fixmap must be called before mem_init since function
>>>>>> + * populate_kernel_pte allocates memory with memblock_alloc method.
>>>>>> + */
>>>>>> + ptep = populate_kernel_pte(addr);
>>>>>> if (!pte_none(*ptep)) {
>>>>>> pte_ERROR(*ptep);
>>>>>> return;
>>>>>> --
>>>>>> 2.27.0
>>>>>>
>>>>
>>>>
>>
>>

Next message: Jie Luo: "Re: [PATCH v1 3/4] clk: qcom: add clock controller driver for qca8386/qca8084"
Previous message: Manikanta Pubbisetty: "Re: [PATCH] Revert "Revert "wifi: ath11k: Enable threaded NAPI"""
In reply to: Huacai Chen: "Re: [PATCH 3/3] LoongArch: mm: Add unified function populate_kernel_pte"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]