linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Alex Ghiti <alex@ghiti.fr>
To: Atish Patra <atishp@atishpatra.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Anup Patel <anup@brainfault.org>,
	Atish Patra <Atish.Patra@wdc.com>,
	linux-riscv <linux-riscv@lists.infradead.org>,
	"linux-kernel@vger.kernel.org List"
	<linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 2/2] riscv: Use PUD/PGDIR entries for linear mapping when possible
Date: Sat, 20 Jun 2020 05:04:00 -0400	[thread overview]
Message-ID: <f1a5ec6e-540b-497f-a9ad-f2d1e7adfc65@ghiti.fr> (raw)
In-Reply-To: <CAOnJCU+s5JuNdPg_R-Cg2+WnMjR51DD0ekbRr84EFCig6=YyZA@mail.gmail.com>

Hi Atish,

Le 6/19/20 à 2:16 PM, Atish Patra a écrit :
> On Thu, Jun 18, 2020 at 9:28 PM Alex Ghiti <alex@ghiti.fr> wrote:
>> Hi Atish,
>>
>> Le 6/18/20 à 8:47 PM, Atish Patra a écrit :
>>> On Wed, Jun 3, 2020 at 8:38 AM Alexandre Ghiti <alex@ghiti.fr> wrote:
>>>> Improve best_map_size so that PUD or PGDIR entries are used for linear
>>>> mapping when possible as it allows better TLB utilization.
>>>>
>>>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>>>> ---
>>>>    arch/riscv/mm/init.c | 45 +++++++++++++++++++++++++++++++++-----------
>>>>    1 file changed, 34 insertions(+), 11 deletions(-)
>>>>
>>>> diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
>>>> index 9a5c97e091c1..d275f9f834cf 100644
>>>> --- a/arch/riscv/mm/init.c
>>>> +++ b/arch/riscv/mm/init.c
>>>> @@ -424,13 +424,29 @@ static void __init create_pgd_mapping(pgd_t *pgdp,
>>>>           create_pgd_next_mapping(nextp, va, pa, sz, prot);
>>>>    }
>>>>
>>>> -static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
>>>> +static bool is_map_size_ok(uintptr_t map_size, phys_addr_t base,
>>>> +                          uintptr_t base_virt, phys_addr_t size)
>>>>    {
>>>> -       /* Upgrade to PMD_SIZE mappings whenever possible */
>>>> -       if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
>>>> -               return PAGE_SIZE;
>>>> +       return !((base & (map_size - 1)) || (base_virt & (map_size - 1)) ||
>>>> +                       (size < map_size));
>>>> +}
>>>> +
>>>> +static uintptr_t __init best_map_size(phys_addr_t base, uintptr_t base_virt,
>>>> +                                     phys_addr_t size)
>>>> +{
>>>> +#ifndef __PAGETABLE_PMD_FOLDED
>>>> +       if (is_map_size_ok(PGDIR_SIZE, base, base_virt, size))
>>>> +               return PGDIR_SIZE;
>>>> +
>>>> +       if (pgtable_l4_enabled)
>>>> +               if (is_map_size_ok(PUD_SIZE, base, base_virt, size))
>>>> +                       return PUD_SIZE;
>>>> +#endif
>>>> +
>>>> +       if (is_map_size_ok(PMD_SIZE, base, base_virt, size))
>>>> +               return PMD_SIZE;
>>>>
>>>> -       return PMD_SIZE;
>>>> +       return PAGE_SIZE;
>>>>    }
>>>>
>>>>    /*
>>>> @@ -576,7 +592,7 @@ void create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size)
>>>>    asmlinkage void __init setup_vm(uintptr_t dtb_pa)
>>>>    {
>>>>           uintptr_t va, end_va;
>>>> -       uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
>>>> +       uintptr_t map_size;
>>>>
>>>>           load_pa = (uintptr_t)(&_start);
>>>>           load_sz = (uintptr_t)(&_end) - load_pa;
>>>> @@ -587,6 +603,7 @@ asmlinkage void __init setup_vm(uintptr_t dtb_pa)
>>>>
>>>>           kernel_virt_addr = KERNEL_VIRT_ADDR;
>>>>
>>>> +       map_size = best_map_size(load_pa, PAGE_OFFSET, MAX_EARLY_MAPPING_SIZE);
>>>>           va_pa_offset = PAGE_OFFSET - load_pa;
>>>>           va_kernel_pa_offset = kernel_virt_addr - load_pa;
>>>>           pfn_base = PFN_DOWN(load_pa);
>>>> @@ -700,6 +717,8 @@ static void __init setup_vm_final(void)
>>>>
>>>>           /* Map all memory banks */
>>>>           for_each_memblock(memory, reg) {
>>>> +               uintptr_t remaining_size;
>>>> +
>>>>                   start = reg->base;
>>>>                   end = start + reg->size;
>>>>
>>>> @@ -707,15 +726,19 @@ static void __init setup_vm_final(void)
>>>>                           break;
>>>>                   if (memblock_is_nomap(reg))
>>>>                           continue;
>>>> -               if (start <= __pa(PAGE_OFFSET) &&
>>>> -                   __pa(PAGE_OFFSET) < end)
>>>> -                       start = __pa(PAGE_OFFSET);
>>>>
>>>> -               map_size = best_map_size(start, end - start);
>>>> -               for (pa = start; pa < end; pa += map_size) {
>>>> +               pa = start;
>>>> +               remaining_size = reg->size;
>>>> +
>>>> +               while (remaining_size) {
>>>>                           va = (uintptr_t)__va(pa);
>>>> +                       map_size = best_map_size(pa, va, remaining_size);
>>>> +
>>>>                           create_pgd_mapping(swapper_pg_dir, va, pa,
>>>>                                              map_size, PAGE_KERNEL);
>>>> +
>>>> +                       pa += map_size;
>>>> +                       remaining_size -= map_size;
>>>>                   }
>>>>           }
>>>>
>>> This may not work in the RV32 with 2G memory  and if the map_size is
>>> determined to be a page size
>>> for the last memblock. Both pa & remaining_size will overflow and the
>>> loop will try to map memory from zero again.
>> I'm not sure I understand: if pa starts at 0x8000_0000 and size is 2G,
>> then pa will overflow in the last iteration, but remaining_size will
>> then be equal to 0 right ?
>>
> Not unless the remaining_size is at least page size aligned. The last
> remaining size would "fff".
> It will overflow as well after subtracting the map_size.
>
>> And by the way, I realize that this loop only handles sizes that are
>> aligned on map_size.
>>
> Yeah.


Thanks for noticing, I send a v2.

Alex


>
>> Thanks,
>>
>> Alex
>>
>>
>>>> --
>>>> 2.20.1
>>>>
>>>>
>
>

  reply	other threads:[~2020-06-20  9:04 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-06-03 15:36 [PATCH 0/2] PUD/PGDIR entries for linear mapping Alexandre Ghiti
2020-06-03 15:36 ` [PATCH 1/2] riscv: Get memory below load_pa while ensuring linear mapping is PMD aligned Alexandre Ghiti
2020-06-03 15:36 ` [PATCH 2/2] riscv: Use PUD/PGDIR entries for linear mapping when possible Alexandre Ghiti
2020-06-19  0:47   ` Atish Patra
2020-06-19  4:28     ` Alex Ghiti
2020-06-19 18:16       ` Atish Patra
2020-06-20  9:04         ` Alex Ghiti [this message]
2020-06-21  9:39           ` Alex Ghiti
2020-06-22 19:11             ` Atish Patra
2020-06-29 14:42               ` Alex Ghiti
2020-06-10 18:32 ` [PATCH 0/2] PUD/PGDIR entries for linear mapping Atish Patra
2020-06-11  6:51   ` Alex Ghiti
2020-06-11 17:29     ` Atish Patra
2020-06-12 12:59       ` Alex Ghiti
2020-06-12 13:17         ` Alex Ghiti
2020-06-12 17:43           ` Atish Patra
2020-06-15  5:35             ` Alex Ghiti
2020-06-16 21:52               ` Atish Patra
2020-06-29 14:46 ` Alex Ghiti

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=f1a5ec6e-540b-497f-a9ad-f2d1e7adfc65@ghiti.fr \
    --to=alex@ghiti.fr \
    --cc=Atish.Patra@wdc.com \
    --cc=anup@brainfault.org \
    --cc=atishp@atishpatra.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-riscv@lists.infradead.org \
    --cc=palmer@dabbelt.com \
    --cc=paul.walmsley@sifive.com \
    --subject='Re: [PATCH 2/2] riscv: Use PUD/PGDIR entries for linear mapping when possible' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).