mbox series

[v6,00/12] mm/debug_vm_pgtable: Enhancements

Message ID 20210809092631.1888748-1-gshan@redhat.com (mailing list archive)
Headers show
Series mm/debug_vm_pgtable: Enhancements | expand

Message

Gavin Shan Aug. 9, 2021, 9:26 a.m. UTC
There are couple of issues with current implementations and this series
tries to resolve the issues:

  (a) All needed information are scattered in variables, passed to various
      test functions. The code is organized in pretty much relaxed fashion.

  (b) The page isn't allocated from buddy during page table entry modifying
      tests. The page can be invalid, conflicting to the implementations
      of set_xxx_at() on ARM64. The target page is accessed so that the
      iCache can be flushed when execution permission is given on ARM64.
      Besides, the target page can be unmapped and accessing to it causes
      kernel crash.

"struct pgtable_debug_args" is introduced to address issue (a). For issue
(b), the used page is allocated from buddy in page table entry modifying
tests. The corresponding tets will be skipped if we fail to allocate the
(huge) page. For other test cases, the original page around to kernel
symbol (@start_kernel) is still used.

The patches are organized as below. PATCH[2-10] could be combined to one
patch, but it will make the review harder:

  PATCH[1] introduces "struct pgtable_debug_args" as place holder of all
           needed information. With it, the old and new implementation
           can coexist.
  PATCH[2-10] uses "struct pgtable_debug_args" in various test functions.
  PATCH[11] removes the unused code for old implementation.
  PATCH[12] fixes the issue of corrupted page flag for ARM64

Changelog
=========
v6:
   * Populate saved page table entry pointers after
     they're allocated in init_args()                        (Anshuman)
   * Fix imbalanced preemption count issue by replacing
     pte_alloc_mmap() with pte_alloc() in init_args()        (syzbot)
v5:
   * Pick r-bs from Anshuman and rebase to 5.14.rc4          (Gavin)
   * Use args->start_p4dp to free p4d entries                (Anshuman)
   * Introduce helper to allocate huge page in init_arg()    (Anshuman)
   * Bail early if the allocated page doesn't exist in
     swap_migration_tests() and correct the comments         (Anshuman)
   * Add fixes tag to PATCH[v4 12/12]                        (Christophe)
   * Address misc comments                                   (Anshuman)
v4:
   * Determine the page allocation method according to
     MAX_ORDER                                               (Anshuman)
   * Move existing comments to init_args()                   (Anshuman)
   * Code refactoring as suggested by Anshuman               (Anshuman)
   * Improved commit log and add comments for flush_dcache_page()
     in PATCH[v4 12/12]                                      (Anshuman)
   * Address misc comments                                   (Anshuman)
v3:
   * Fix the warning caused by allocating more pages than
     (1 << (MAX_ORDER - 1)) in init_args()                   (syzbot)
   * Fix build warning by dropping unused variables in separate
     patches                                                 (0-day)
   * Missed "WARN_ON(!pud_none(pud))" in pud_huge_tests() in
     PATCH[v2 09/12]                                         (0-day)
   * Fix the subjects for PATCH[05/12] and PATCH[09/12]      (Gavin)
v2:
   * Rename struct vm_pgtable_debug to struct pgtable_debug_args.
     The parameter name to various test functions are renamed
     to "@args"                                              (Anshuman)
   * Code changes as suggested by Anshuman                   (Anshuman)

Gavin Shan (12):
  mm/debug_vm_pgtable: Introduce struct pgtable_debug_args
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in basic tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in leaf and
    savewrite tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in protnone and
    devmap tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in soft_dirty and
    swap tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in migration and
    thp tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in PTE modifying
    tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in PMD modifying
    tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in PUD modifying
    tests
  mm/debug_vm_pgtable: Use struct pgtable_debug_args in PGD and P4D
    modifying tests
  mm/debug_vm_pgtable: Remove unused code
  mm/debug_vm_pgtable: Fix corrupted page flag

 mm/debug_vm_pgtable.c | 918 +++++++++++++++++++++++++-----------------
 1 file changed, 559 insertions(+), 359 deletions(-)

Comments

Christophe Leroy Aug. 9, 2021, 10:08 a.m. UTC | #1
Le 09/08/2021 à 11:26, Gavin Shan a écrit :
> There are couple of issues with current implementations and this series
> tries to resolve the issues:
> 
>    (a) All needed information are scattered in variables, passed to various
>        test functions. The code is organized in pretty much relaxed fashion.
> 
>    (b) The page isn't allocated from buddy during page table entry modifying
>        tests. The page can be invalid, conflicting to the implementations
>        of set_xxx_at() on ARM64. The target page is accessed so that the
>        iCache can be flushed when execution permission is given on ARM64.
>        Besides, the target page can be unmapped and accessing to it causes
>        kernel crash.
> 
> "struct pgtable_debug_args" is introduced to address issue (a). For issue
> (b), the used page is allocated from buddy in page table entry modifying
> tests. The corresponding tets will be skipped if we fail to allocate the
> (huge) page. For other test cases, the original page around to kernel
> symbol (@start_kernel) is still used.
> 
> The patches are organized as below. PATCH[2-10] could be combined to one
> patch, but it will make the review harder:
> 
>    PATCH[1] introduces "struct pgtable_debug_args" as place holder of all
>             needed information. With it, the old and new implementation
>             can coexist.
>    PATCH[2-10] uses "struct pgtable_debug_args" in various test functions.
>    PATCH[11] removes the unused code for old implementation.
>    PATCH[12] fixes the issue of corrupted page flag for ARM64


I tested this series on powerpc 8xx, no failure reported.

Christophe

> 
> Changelog
> =========
> v6:
>     * Populate saved page table entry pointers after
>       they're allocated in init_args()                        (Anshuman)
>     * Fix imbalanced preemption count issue by replacing
>       pte_alloc_mmap() with pte_alloc() in init_args()        (syzbot)
> v5:
>     * Pick r-bs from Anshuman and rebase to 5.14.rc4          (Gavin)
>     * Use args->start_p4dp to free p4d entries                (Anshuman)
>     * Introduce helper to allocate huge page in init_arg()    (Anshuman)
>     * Bail early if the allocated page doesn't exist in
>       swap_migration_tests() and correct the comments         (Anshuman)
>     * Add fixes tag to PATCH[v4 12/12]                        (Christophe)
>     * Address misc comments                                   (Anshuman)
> v4:
>     * Determine the page allocation method according to
>       MAX_ORDER                                               (Anshuman)
>     * Move existing comments to init_args()                   (Anshuman)
>     * Code refactoring as suggested by Anshuman               (Anshuman)
>     * Improved commit log and add comments for flush_dcache_page()
>       in PATCH[v4 12/12]                                      (Anshuman)
>     * Address misc comments                                   (Anshuman)
> v3:
>     * Fix the warning caused by allocating more pages than
>       (1 << (MAX_ORDER - 1)) in init_args()                   (syzbot)
>     * Fix build warning by dropping unused variables in separate
>       patches                                                 (0-day)
>     * Missed "WARN_ON(!pud_none(pud))" in pud_huge_tests() in
>       PATCH[v2 09/12]                                         (0-day)
>     * Fix the subjects for PATCH[05/12] and PATCH[09/12]      (Gavin)
> v2:
>     * Rename struct vm_pgtable_debug to struct pgtable_debug_args.
>       The parameter name to various test functions are renamed
>       to "@args"                                              (Anshuman)
>     * Code changes as suggested by Anshuman                   (Anshuman)
> 
> Gavin Shan (12):
>    mm/debug_vm_pgtable: Introduce struct pgtable_debug_args
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in basic tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in leaf and
>      savewrite tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in protnone and
>      devmap tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in soft_dirty and
>      swap tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in migration and
>      thp tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in PTE modifying
>      tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in PMD modifying
>      tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in PUD modifying
>      tests
>    mm/debug_vm_pgtable: Use struct pgtable_debug_args in PGD and P4D
>      modifying tests
>    mm/debug_vm_pgtable: Remove unused code
>    mm/debug_vm_pgtable: Fix corrupted page flag
> 
>   mm/debug_vm_pgtable.c | 918 +++++++++++++++++++++++++-----------------
>   1 file changed, 559 insertions(+), 359 deletions(-)
>
Gerald Schaefer Aug. 10, 2021, 4:05 p.m. UTC | #2
On Mon,  9 Aug 2021 17:26:19 +0800
Gavin Shan <gshan@redhat.com> wrote:

> There are couple of issues with current implementations and this series
> tries to resolve the issues:
> 
>   (a) All needed information are scattered in variables, passed to various
>       test functions. The code is organized in pretty much relaxed fashion.
> 
>   (b) The page isn't allocated from buddy during page table entry modifying
>       tests. The page can be invalid, conflicting to the implementations
>       of set_xxx_at() on ARM64. The target page is accessed so that the
>       iCache can be flushed when execution permission is given on ARM64.
>       Besides, the target page can be unmapped and accessing to it causes
>       kernel crash.
> 
> "struct pgtable_debug_args" is introduced to address issue (a). For issue
> (b), the used page is allocated from buddy in page table entry modifying
> tests. The corresponding tets will be skipped if we fail to allocate the
> (huge) page. For other test cases, the original page around to kernel
> symbol (@start_kernel) is still used.
> 
> The patches are organized as below. PATCH[2-10] could be combined to one
> patch, but it will make the review harder:
> 
>   PATCH[1] introduces "struct pgtable_debug_args" as place holder of all
>            needed information. With it, the old and new implementation
>            can coexist.
>   PATCH[2-10] uses "struct pgtable_debug_args" in various test functions.
>   PATCH[11] removes the unused code for old implementation.
>   PATCH[12] fixes the issue of corrupted page flag for ARM64

Tested on s390, all seems to work fine.

Gerald
Anshuman Khandual Aug. 12, 2021, 11:11 a.m. UTC | #3
On 8/9/21 2:56 PM, Gavin Shan wrote:
> There are couple of issues with current implementations and this series
> tries to resolve the issues:
> 
>   (a) All needed information are scattered in variables, passed to various
>       test functions. The code is organized in pretty much relaxed fashion.
> 
>   (b) The page isn't allocated from buddy during page table entry modifying
>       tests. The page can be invalid, conflicting to the implementations
>       of set_xxx_at() on ARM64. The target page is accessed so that the
>       iCache can be flushed when execution permission is given on ARM64.
>       Besides, the target page can be unmapped and accessing to it causes
>       kernel crash.
> 
> "struct pgtable_debug_args" is introduced to address issue (a). For issue
> (b), the used page is allocated from buddy in page table entry modifying
> tests. The corresponding tets will be skipped if we fail to allocate the
> (huge) page. For other test cases, the original page around to kernel
> symbol (@start_kernel) is still used.
> 
> The patches are organized as below. PATCH[2-10] could be combined to one
> patch, but it will make the review harder:
> 
>   PATCH[1] introduces "struct pgtable_debug_args" as place holder of all
>            needed information. With it, the old and new implementation
>            can coexist.
>   PATCH[2-10] uses "struct pgtable_debug_args" in various test functions.
>   PATCH[11] removes the unused code for old implementation.
>   PATCH[12] fixes the issue of corrupted page flag for ARM64
> 
> Changelog
> =========
> v6:
>    * Populate saved page table entry pointers after
>      they're allocated in init_args()                        (Anshuman)
>    * Fix imbalanced preemption count issue by replacing
>      pte_alloc_mmap() with pte_alloc() in init_args()        (syzbot)

+ vgupta@kernel.org

Hello Gavin/Vineet,

This series still need to be tested on ARC ?

- Anshuman
Gavin Shan Aug. 13, 2021, 9:25 a.m. UTC | #4
Hi Anshuman and Vineet,

On 8/12/21 9:11 PM, Anshuman Khandual wrote:
> On 8/9/21 2:56 PM, Gavin Shan wrote:
>> There are couple of issues with current implementations and this series
>> tries to resolve the issues:
>>
>>    (a) All needed information are scattered in variables, passed to various
>>        test functions. The code is organized in pretty much relaxed fashion.
>>
>>    (b) The page isn't allocated from buddy during page table entry modifying
>>        tests. The page can be invalid, conflicting to the implementations
>>        of set_xxx_at() on ARM64. The target page is accessed so that the
>>        iCache can be flushed when execution permission is given on ARM64.
>>        Besides, the target page can be unmapped and accessing to it causes
>>        kernel crash.
>>
>> "struct pgtable_debug_args" is introduced to address issue (a). For issue
>> (b), the used page is allocated from buddy in page table entry modifying
>> tests. The corresponding tets will be skipped if we fail to allocate the
>> (huge) page. For other test cases, the original page around to kernel
>> symbol (@start_kernel) is still used.
>>
>> The patches are organized as below. PATCH[2-10] could be combined to one
>> patch, but it will make the review harder:
>>
>>    PATCH[1] introduces "struct pgtable_debug_args" as place holder of all
>>             needed information. With it, the old and new implementation
>>             can coexist.
>>    PATCH[2-10] uses "struct pgtable_debug_args" in various test functions.
>>    PATCH[11] removes the unused code for old implementation.
>>    PATCH[12] fixes the issue of corrupted page flag for ARM64
>>
>> Changelog
>> =========
>> v6:
>>     * Populate saved page table entry pointers after
>>       they're allocated in init_args()                        (Anshuman)
>>     * Fix imbalanced preemption count issue by replacing
>>       pte_alloc_mmap() with pte_alloc() in init_args()        (syzbot)
> 
> + vgupta@kernel.org
> 
> Hello Gavin/Vineet,
> 
> This series still need to be tested on ARC ?
> 

Yes, I'm unable to test ARC as it's not supported by QEMU yet.
It would great if Vineet can give it a try on ARC :)

Thanks,
Gavin
Vineet Gupta Aug. 13, 2021, 7:43 p.m. UTC | #5
On 8/13/21 2:25 AM, Gavin Shan wrote:
> Hi Anshuman and Vineet,
>
> On 8/12/21 9:11 PM, Anshuman Khandual wrote:
>> On 8/9/21 2:56 PM, Gavin Shan wrote:
>>> There are couple of issues with current implementations and this series
>>> tries to resolve the issues:
>>>
>>>    (a) All needed information are scattered in variables, passed to 
>>> various
>>>        test functions. The code is organized in pretty much relaxed 
>>> fashion.
>>>
>>>    (b) The page isn't allocated from buddy during page table entry 
>>> modifying
>>>        tests. The page can be invalid, conflicting to the 
>>> implementations
>>>        of set_xxx_at() on ARM64. The target page is accessed so that 
>>> the
>>>        iCache can be flushed when execution permission is given on 
>>> ARM64.
>>>        Besides, the target page can be unmapped and accessing to it 
>>> causes
>>>        kernel crash.
>>>
>>> "struct pgtable_debug_args" is introduced to address issue (a). For 
>>> issue
>>> (b), the used page is allocated from buddy in page table entry 
>>> modifying
>>> tests. The corresponding tets will be skipped if we fail to allocate 
>>> the
>>> (huge) page. For other test cases, the original page around to kernel
>>> symbol (@start_kernel) is still used.
>>>
>>> The patches are organized as below. PATCH[2-10] could be combined to 
>>> one
>>> patch, but it will make the review harder:
>>>
>>>    PATCH[1] introduces "struct pgtable_debug_args" as place holder 
>>> of all
>>>             needed information. With it, the old and new implementation
>>>             can coexist.
>>>    PATCH[2-10] uses "struct pgtable_debug_args" in various test 
>>> functions.
>>>    PATCH[11] removes the unused code for old implementation.
>>>    PATCH[12] fixes the issue of corrupted page flag for ARM64
>>>
>>> Changelog
>>> =========
>>> v6:
>>>     * Populate saved page table entry pointers after
>>>       they're allocated in init_args() (Anshuman)
>>>     * Fix imbalanced preemption count issue by replacing
>>>       pte_alloc_mmap() with pte_alloc() in init_args() (syzbot)
>>
>> + vgupta@kernel.org

Please also keep linux-snps-arc CC'ed for ARC changes.

>>
>> Hello Gavin/Vineet,
>>
>> This series still need to be tested on ARC ?
>>
>
> Yes, I'm unable to test ARC as it's not supported by QEMU yet.
> It would great if Vineet can give it a try on ARC :)

We do have a working QEMU (in the process of being upstreamed) you could try
https://github.com/foss-for-synopsys-dwc-arc-processors/qemu #master

Is this code in some shared git repo we can pull - instead of apply 8 files.

Thx,
-Vineet
Gavin Shan Aug. 14, 2021, 2:57 a.m. UTC | #6
Hi Vineet,

On 8/14/21 5:43 AM, Vineet Gupta wrote:
> On 8/13/21 2:25 AM, Gavin Shan wrote:
>> On 8/12/21 9:11 PM, Anshuman Khandual wrote:
>>> On 8/9/21 2:56 PM, Gavin Shan wrote:
>>>> There are couple of issues with current implementations and this series
>>>> tries to resolve the issues:
>>>>
>>>>    (a) All needed information are scattered in variables, passed to various
>>>>        test functions. The code is organized in pretty much relaxed fashion.
>>>>
>>>>    (b) The page isn't allocated from buddy during page table entry modifying
>>>>        tests. The page can be invalid, conflicting to the implementations
>>>>        of set_xxx_at() on ARM64. The target page is accessed so that the
>>>>        iCache can be flushed when execution permission is given on ARM64.
>>>>        Besides, the target page can be unmapped and accessing to it causes
>>>>        kernel crash.
>>>>
>>>> "struct pgtable_debug_args" is introduced to address issue (a). For issue
>>>> (b), the used page is allocated from buddy in page table entry modifying
>>>> tests. The corresponding tets will be skipped if we fail to allocate the
>>>> (huge) page. For other test cases, the original page around to kernel
>>>> symbol (@start_kernel) is still used.
>>>>
>>>> The patches are organized as below. PATCH[2-10] could be combined to one
>>>> patch, but it will make the review harder:
>>>>
>>>>    PATCH[1] introduces "struct pgtable_debug_args" as place holder of all
>>>>             needed information. With it, the old and new implementation
>>>>             can coexist.
>>>>    PATCH[2-10] uses "struct pgtable_debug_args" in various test functions.
>>>>    PATCH[11] removes the unused code for old implementation.
>>>>    PATCH[12] fixes the issue of corrupted page flag for ARM64
>>>>
>>>> Changelog
>>>> =========
>>>> v6:
>>>>     * Populate saved page table entry pointers after
>>>>       they're allocated in init_args() (Anshuman)
>>>>     * Fix imbalanced preemption count issue by replacing
>>>>       pte_alloc_mmap() with pte_alloc() in init_args() (syzbot)
>>>
>>> + vgupta@kernel.org
> 
> Please also keep linux-snps-arc CC'ed for ARC changes.
> 

Sure.

>>>
>>> Hello Gavin/Vineet,
>>>
>>> This series still need to be tested on ARC ?
>>>
>>
>> Yes, I'm unable to test ARC as it's not supported by QEMU yet.
>> It would great if Vineet can give it a try on ARC :)
> 
> We do have a working QEMU (in the process of being upstreamed) you could try
> https://github.com/foss-for-synopsys-dwc-arc-processors/qemu #master
> 
> Is this code in some shared git repo we can pull - instead of apply 8 files.
> 

Here is repo with v6 series included. Please help to have
a try.

    https://github.com/gwshan/linux  # kvm/arm64_debug

I tried to build the cross-compiler with buildroot with following
2 repos. Unfortunately, none of them worked for me. The QEMU
binary can be built successfully with the provided repo though.


[1] git://github.com/foss-for-synopsys-dwc-arc-processors/buildroot.git
     (branch: master)

     There are build errors after choosing little-endian ARM from
     "make menuconfig"

[2] https://github.com/buildroot/buildroot
     (branch: master)

     The build is successful, but the kernel can't be built because
     of the following errors:
    
     # make ARCH=arc CROSS_COMPILE=arc-linux- axs103_smp_defconfig
     # make -j 40
        :
     LD      vmlinux.o
     arc-linux-ld: error: attempting to link /home/gavin/sandbox/buildroot/buildroot.arc/output/host/lib
     /gcc/arc-buildroot-linux-uclibc/10.2.0/libgcc.a(_muldi3.o) with a binary vmlinux.o of different
     architecture

Thanks,
Gavin