2021-08-18 22:05:23

by Peilin Ye

[permalink] [raw]
Subject: [PATCH RESEND v2] docs: x86: Remove obsolete information about x86_64 vmalloc() faulting

From: Peilin Ye <[email protected]>

x86_64 vmalloc() mappings are no longer "synchronized" among page tables
via faulting since commit 6eb82f994026 ("x86/mm: Pre-allocate P4D/PUD
pages for vmalloc area"), since the corresponding P4D or PUD pages are
now preallocated at boot, by preallocate_vmalloc_pages(). Drop the
"lazily synchronized" description for less confusion.

While this file is x86_64-specific, it is worth noting that things are
different for x86_32, where vmalloc()-related changes to `init_mm.pgd` are
synchronized to all page tables in the system during runtime, via
arch_sync_kernel_mappings(). Unfortunately, this synchronization is
subject to race condition, which is further handled via faulting, see
vmalloc_fault(). See commit 4819e15f740e ("x86/mm/32: Bring back vmalloc
faulting on x86_32") for more details.

Reviewed-by: Muchun Song <[email protected]>
Signed-off-by: Peilin Ye <[email protected]>
---
Hi all,

Resending this with Muchun's Reviewed-by:.

Thanks,
Peilin Ye

Changes in v2:
- More information for x86_32 in commit message (Joerg Roedel
<[email protected]>)
- Use my new email address for work

Documentation/x86/x86_64/mm.rst | 4 ----
1 file changed, 4 deletions(-)

diff --git a/Documentation/x86/x86_64/mm.rst b/Documentation/x86/x86_64/mm.rst
index ede1875719fb..9798676bb0bf 100644
--- a/Documentation/x86/x86_64/mm.rst
+++ b/Documentation/x86/x86_64/mm.rst
@@ -140,10 +140,6 @@ The direct mapping covers all memory in the system up to the highest
memory address (this means in some cases it can also include PCI memory
holes).

-vmalloc space is lazily synchronized into the different PML4/PML5 pages of
-the processes using the page fault handler, with init_top_pgt as
-reference.
-
We map EFI runtime services in the 'efi_pgd' PGD in a 64Gb large virtual
memory window (this size is arbitrary, it can be raised later if needed).
The mappings are not part of any other kernel PGD and are only available
--
2.20.1


2021-08-19 08:03:28

by Jörg Rödel

[permalink] [raw]
Subject: Re: [PATCH RESEND v2] docs: x86: Remove obsolete information about x86_64 vmalloc() faulting

On Wed, Aug 18, 2021 at 03:01:23PM -0700, Peilin Ye wrote:
> From: Peilin Ye <[email protected]>
>
> x86_64 vmalloc() mappings are no longer "synchronized" among page tables
> via faulting since commit 6eb82f994026 ("x86/mm: Pre-allocate P4D/PUD
> pages for vmalloc area"), since the corresponding P4D or PUD pages are
> now preallocated at boot, by preallocate_vmalloc_pages(). Drop the
> "lazily synchronized" description for less confusion.
>
> While this file is x86_64-specific, it is worth noting that things are
> different for x86_32, where vmalloc()-related changes to `init_mm.pgd` are
> synchronized to all page tables in the system during runtime, via
> arch_sync_kernel_mappings(). Unfortunately, this synchronization is
> subject to race condition, which is further handled via faulting, see
> vmalloc_fault(). See commit 4819e15f740e ("x86/mm/32: Bring back vmalloc
> faulting on x86_32") for more details.
>
> Reviewed-by: Muchun Song <[email protected]>
> Signed-off-by: Peilin Ye <[email protected]>

Reviewed-by: Joerg Roedel <[email protected]>

> ---
> Hi all,
>
> Resending this with Muchun's Reviewed-by:.
>
> Thanks,
> Peilin Ye
>
> Changes in v2:
> - More information for x86_32 in commit message (Joerg Roedel
> <[email protected]>)
> - Use my new email address for work
>
> Documentation/x86/x86_64/mm.rst | 4 ----
> 1 file changed, 4 deletions(-)
>
> diff --git a/Documentation/x86/x86_64/mm.rst b/Documentation/x86/x86_64/mm.rst
> index ede1875719fb..9798676bb0bf 100644
> --- a/Documentation/x86/x86_64/mm.rst
> +++ b/Documentation/x86/x86_64/mm.rst
> @@ -140,10 +140,6 @@ The direct mapping covers all memory in the system up to the highest
> memory address (this means in some cases it can also include PCI memory
> holes).
>
> -vmalloc space is lazily synchronized into the different PML4/PML5 pages of
> -the processes using the page fault handler, with init_top_pgt as
> -reference.
> -
> We map EFI runtime services in the 'efi_pgd' PGD in a 64Gb large virtual
> memory window (this size is arbitrary, it can be raised later if needed).
> The mappings are not part of any other kernel PGD and are only available
> --
> 2.20.1
>

2021-08-20 17:16:22

by Jonathan Corbet

[permalink] [raw]
Subject: Re: [PATCH RESEND v2] docs: x86: Remove obsolete information about x86_64 vmalloc() faulting

Joerg Roedel <[email protected]> writes:

> On Wed, Aug 18, 2021 at 03:01:23PM -0700, Peilin Ye wrote:
>> From: Peilin Ye <[email protected]>
>>
>> x86_64 vmalloc() mappings are no longer "synchronized" among page tables
>> via faulting since commit 6eb82f994026 ("x86/mm: Pre-allocate P4D/PUD
>> pages for vmalloc area"), since the corresponding P4D or PUD pages are
>> now preallocated at boot, by preallocate_vmalloc_pages(). Drop the
>> "lazily synchronized" description for less confusion.
>>
>> While this file is x86_64-specific, it is worth noting that things are
>> different for x86_32, where vmalloc()-related changes to `init_mm.pgd` are
>> synchronized to all page tables in the system during runtime, via
>> arch_sync_kernel_mappings(). Unfortunately, this synchronization is
>> subject to race condition, which is further handled via faulting, see
>> vmalloc_fault(). See commit 4819e15f740e ("x86/mm/32: Bring back vmalloc
>> faulting on x86_32") for more details.
>>
>> Reviewed-by: Muchun Song <[email protected]>
>> Signed-off-by: Peilin Ye <[email protected]>
>
> Reviewed-by: Joerg Roedel <[email protected]>

Applied, thanks.

jon