fixmap is a global resource and is used recursively in create pud mapping.
It may lead to race condition when alloc_init_pud is called concurrently.
Fox example:
alloc_init_pud is called when kernel_init. If memory hotplug
thread, which will also call alloc_init_pud, happens during
kernel_init, the race for fixmap occurs.
The race condition flow can be:
*************** begin **************
kerenl_init thread virtio-mem workqueue thread
================== ======== ==================
alloc_init_pud(...)
pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
... ...
READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
... ...
pud_clear_fixmap() //fixmap break
READ_ONCE(*pudp) //CRASH!
**************** end ***************
Hence, a spin lock is introduced to protect the fixmap during create pdg
mapping.
Signed-off-by: Jianyong Wu <[email protected]>
---
arch/arm64/mm/mmu.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index acfae9b41cc8..98ac09ae9588 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
static DEFINE_SPINLOCK(swapper_pgdir_lock);
+static DEFINE_SPINLOCK(fixmap_lock);
void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
{
@@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
}
BUG_ON(p4d_bad(p4d));
+ /*
+ * fixmap is global resource, thus it needs to be protected by a lock
+ * in case of race condition.
+ */
+ spin_lock(&fixmap_lock);
pudp = pud_set_fixmap_offset(p4dp, addr);
do {
pud_t old_pud = READ_ONCE(*pudp);
@@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
} while (pudp++, addr = next, addr != end);
pud_clear_fixmap();
+ spin_unlock(&fixmap_lock);
}
static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
--
2.17.1
On Fri, Dec 10, 2021 at 05:54:32PM +0800, Jianyong Wu wrote:
> fixmap is a global resource and is used recursively in create pud mapping.
> It may lead to race condition when alloc_init_pud is called concurrently.
>
> Fox example:
> alloc_init_pud is called when kernel_init. If memory hotplug
> thread, which will also call alloc_init_pud, happens during
> kernel_init, the race for fixmap occurs.
>
> The race condition flow can be:
>
> *************** begin **************
>
> kerenl_init thread virtio-mem workqueue thread
> ================== ======== ==================
> alloc_init_pud(...)
> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> ... ...
> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> ... ...
> pud_clear_fixmap() //fixmap break
> READ_ONCE(*pudp) //CRASH!
>
> **************** end ***************
>
> Hence, a spin lock is introduced to protect the fixmap during create pdg
> mapping.
>
> Signed-off-by: Jianyong Wu <[email protected]>
It looks fine to me:
Reviewed-by: Catalin Marinas <[email protected]>
Do we need a cc stable? Fixmap was introduced in 4.6, so usually:
Fixes: f4710445458c ("arm64: mm: use fixmap when creating page tables")
Cc: <[email protected]> # 4.6.x
but I haven't checked when memory hotplug was added to be able to
trigger the race. It may not need to go back that far.
--
Catalin
Hi Catalin,
> -----Original Message-----
> From: Catalin Marinas <[email protected]>
> Sent: Friday, December 10, 2021 7:22 PM
> To: Jianyong Wu <[email protected]>
> Cc: [email protected]; Anshuman Khandual <[email protected]>;
> [email protected]; [email protected]; linux-
> [email protected]; [email protected];
> [email protected]; [email protected]; Justin He <[email protected]>;
> nd <[email protected]>; Mark Rutland <[email protected]>
> Subject: Re: [PATCH v2] arm64/mm: avoid fixmap race condition when create
> pud mapping
>
> On Fri, Dec 10, 2021 at 05:54:32PM +0800, Jianyong Wu wrote:
> > fixmap is a global resource and is used recursively in create pud mapping.
> > It may lead to race condition when alloc_init_pud is called concurrently.
> >
> > Fox example:
> > alloc_init_pud is called when kernel_init. If memory hotplug thread,
> > which will also call alloc_init_pud, happens during kernel_init, the
> > race for fixmap occurs.
> >
> > The race condition flow can be:
> >
> > *************** begin **************
> >
> > kerenl_init thread virtio-mem workqueue thread
> > ================== ======== ==================
> > alloc_init_pud(...)
> > pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> > ... ...
> > READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> > ... ...
> > pud_clear_fixmap() //fixmap break
> > READ_ONCE(*pudp) //CRASH!
> >
> > **************** end ***************
> >
> > Hence, a spin lock is introduced to protect the fixmap during create
> > pdg mapping.
> >
> > Signed-off-by: Jianyong Wu <[email protected]>
>
> It looks fine to me:
>
> Reviewed-by: Catalin Marinas <[email protected]>
>
> Do we need a cc stable? Fixmap was introduced in 4.6, so usually:
>
> Fixes: f4710445458c ("arm64: mm: use fixmap when creating page tables")
> Cc: <[email protected]> # 4.6.x
>
> but I haven't checked when memory hotplug was added to be able to trigger
> the race. It may not need to go back that far.
>
For now, I only find the memory hotplug can conflict with it. I see that memory hotplug is enabled on arm64 in 4.20, thus what about changing "4.6" to "4.20".
Thanks
Jianyong
> --
> Catalin
On 12/10/21 3:24 PM, Jianyong Wu wrote:
> fixmap is a global resource and is used recursively in create pud mapping.
> It may lead to race condition when alloc_init_pud is called concurrently.
>
> Fox example:
> alloc_init_pud is called when kernel_init. If memory hotplug
> thread, which will also call alloc_init_pud, happens during
> kernel_init, the race for fixmap occurs.
>
> The race condition flow can be:
>
> *************** begin **************
>
> kerenl_init thread virtio-mem workqueue thread
> ================== ======== ==================
> alloc_init_pud(...)
> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> ... ...
> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> ... ...
> pud_clear_fixmap() //fixmap break
> READ_ONCE(*pudp) //CRASH!
>
> **************** end ***************
>
> Hence, a spin lock is introduced to protect the fixmap during create pdg
> mapping.
>
> Signed-off-by: Jianyong Wu <[email protected]>
> ---
> arch/arm64/mm/mmu.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index acfae9b41cc8..98ac09ae9588 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
> static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
>
> static DEFINE_SPINLOCK(swapper_pgdir_lock);
> +static DEFINE_SPINLOCK(fixmap_lock);
>
> void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
> {
> @@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> }
> BUG_ON(p4d_bad(p4d));
>
> + /*
> + * fixmap is global resource, thus it needs to be protected by a lock
> + * in case of race condition.
> + */
> + spin_lock(&fixmap_lock);
> pudp = pud_set_fixmap_offset(p4dp, addr);
> do {
> pud_t old_pud = READ_ONCE(*pudp);
> @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> } while (pudp++, addr = next, addr != end);
>
> pud_clear_fixmap();
> + spin_unlock(&fixmap_lock);
> }
>
> static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
>
As the race could only happen with memory hotplug being enabled, could
not we wrap this around with CONFIG_MEMORY_HOTPLUG, just to narrow its
scope possibly speed up other non-hotplug cases ?
> -----Original Message-----
> From: Anshuman Khandual <[email protected]>
> Sent: Monday, December 13, 2021 2:56 PM
> To: Jianyong Wu <[email protected]>; Catalin Marinas
> <[email protected]>; [email protected]; [email protected]
> Cc: [email protected]; [email protected]; linux-arm-
> [email protected]; [email protected]; [email protected]; Justin
> He <[email protected]>; nd <[email protected]>
> Subject: Re: [PATCH v2] arm64/mm: avoid fixmap race condition when create
> pud mapping
>
>
>
> On 12/10/21 3:24 PM, Jianyong Wu wrote:
> > fixmap is a global resource and is used recursively in create pud mapping.
> > It may lead to race condition when alloc_init_pud is called concurrently.
> >
> > Fox example:
> > alloc_init_pud is called when kernel_init. If memory hotplug thread,
> > which will also call alloc_init_pud, happens during kernel_init, the
> > race for fixmap occurs.
> >
> > The race condition flow can be:
> >
> > *************** begin **************
> >
> > kerenl_init thread virtio-mem workqueue thread
> > ================== ======== ==================
> > alloc_init_pud(...)
> > pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> > ... ...
> > READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> > ... ...
> > pud_clear_fixmap() //fixmap break
> > READ_ONCE(*pudp) //CRASH!
> >
> > **************** end ***************
> >
> > Hence, a spin lock is introduced to protect the fixmap during create
> > pdg mapping.
> >
> > Signed-off-by: Jianyong Wu <[email protected]>
> > ---
> > arch/arm64/mm/mmu.c | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c index
> > acfae9b41cc8..98ac09ae9588 100644
> > --- a/arch/arm64/mm/mmu.c
> > +++ b/arch/arm64/mm/mmu.c
> > @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD]
> __page_aligned_bss
> > __maybe_unused; static pud_t bm_pud[PTRS_PER_PUD]
> __page_aligned_bss
> > __maybe_unused;
> >
> > static DEFINE_SPINLOCK(swapper_pgdir_lock);
> > +static DEFINE_SPINLOCK(fixmap_lock);
> >
> > void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd) { @@ -329,6 +330,11 @@
> > static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long
> end,
> > }
> > BUG_ON(p4d_bad(p4d));
> >
> > + /*
> > + * fixmap is global resource, thus it needs to be protected by a lock
> > + * in case of race condition.
> > + */
> > + spin_lock(&fixmap_lock);
> > pudp = pud_set_fixmap_offset(p4dp, addr);
> > do {
> > pud_t old_pud = READ_ONCE(*pudp);
> > @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned
> long addr, unsigned long end,
> > } while (pudp++, addr = next, addr != end);
> >
> > pud_clear_fixmap();
> > + spin_unlock(&fixmap_lock);
> > }
> >
> > static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
> >
>
> As the race could only happen with memory hotplug being enabled, could
> not we wrap this around with CONFIG_MEMORY_HOTPLUG, just to narrow
> its scope possibly speed up other non-hotplug cases ?
I think it's better.
Thanks
Jianyong
On 13.12.21 08:27, Jianyong Wu wrote:
>
>
>> -----Original Message-----
>> From: Anshuman Khandual <[email protected]>
>> Sent: Monday, December 13, 2021 2:56 PM
>> To: Jianyong Wu <[email protected]>; Catalin Marinas
>> <[email protected]>; [email protected]; [email protected]
>> Cc: [email protected]; [email protected]; linux-arm-
>> [email protected]; [email protected]; [email protected]; Justin
>> He <[email protected]>; nd <[email protected]>
>> Subject: Re: [PATCH v2] arm64/mm: avoid fixmap race condition when create
>> pud mapping
>>
>>
>>
>> On 12/10/21 3:24 PM, Jianyong Wu wrote:
>>> fixmap is a global resource and is used recursively in create pud mapping.
>>> It may lead to race condition when alloc_init_pud is called concurrently.
>>>
>>> Fox example:
>>> alloc_init_pud is called when kernel_init. If memory hotplug thread,
>>> which will also call alloc_init_pud, happens during kernel_init, the
>>> race for fixmap occurs.
>>>
>>> The race condition flow can be:
>>>
>>> *************** begin **************
>>>
>>> kerenl_init thread virtio-mem workqueue thread
>>> ================== ======== ==================
>>> alloc_init_pud(...)
>>> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
>>> ... ...
>>> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
>>> ... ...
>>> pud_clear_fixmap() //fixmap break
>>> READ_ONCE(*pudp) //CRASH!
>>>
>>> **************** end ***************
>>>
>>> Hence, a spin lock is introduced to protect the fixmap during create
>>> pdg mapping.
>>>
>>> Signed-off-by: Jianyong Wu <[email protected]>
>>> ---
>>> arch/arm64/mm/mmu.c | 7 +++++++
>>> 1 file changed, 7 insertions(+)
>>>
>>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c index
>>> acfae9b41cc8..98ac09ae9588 100644
>>> --- a/arch/arm64/mm/mmu.c
>>> +++ b/arch/arm64/mm/mmu.c
>>> @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD]
>> __page_aligned_bss
>>> __maybe_unused; static pud_t bm_pud[PTRS_PER_PUD]
>> __page_aligned_bss
>>> __maybe_unused;
>>>
>>> static DEFINE_SPINLOCK(swapper_pgdir_lock);
>>> +static DEFINE_SPINLOCK(fixmap_lock);
>>>
>>> void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd) { @@ -329,6 +330,11 @@
>>> static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long
>> end,
>>> }
>>> BUG_ON(p4d_bad(p4d));
>>>
>>> + /*
>>> + * fixmap is global resource, thus it needs to be protected by a lock
>>> + * in case of race condition.
>>> + */
>>> + spin_lock(&fixmap_lock);
>>> pudp = pud_set_fixmap_offset(p4dp, addr);
>>> do {
>>> pud_t old_pud = READ_ONCE(*pudp);
>>> @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned
>> long addr, unsigned long end,
>>> } while (pudp++, addr = next, addr != end);
>>>
>>> pud_clear_fixmap();
>>> + spin_unlock(&fixmap_lock);
>>> }
>>>
>>> static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
>>>
>>
>> As the race could only happen with memory hotplug being enabled, could
>> not we wrap this around with CONFIG_MEMORY_HOTPLUG, just to narrow
>> its scope possibly speed up other non-hotplug cases ?
>
> I think it's better.
We better avoid using ifdef if not really necessary, it just uglifies
the code. We could add
if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
...
But should we really try to micto-optimize this code and make it harder
to read for the purpose of an unproven performance gain? (if there is no
contention, we'll most probably not even recognize that we're taking a lock)
--
Thanks,
David / dhildenb
On Mon, Dec 13, 2021 at 08:37:04AM +0100, David Hildenbrand wrote:
> On 13.12.21 08:27, Jianyong Wu wrote:
> >>> @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> >>> } while (pudp++, addr = next, addr != end);
> >>>
> >>> pud_clear_fixmap();
> >>> + spin_unlock(&fixmap_lock);
> >>> }
> >>>
> >>> static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
> >>>
> >>
> >> As the race could only happen with memory hotplug being enabled, could
> >> not we wrap this around with CONFIG_MEMORY_HOTPLUG, just to narrow
> >> its scope possibly speed up other non-hotplug cases ?
Assuming you can quantify the speed-up...
> > I think it's better.
>
> We better avoid using ifdef if not really necessary, it just uglifies
> the code. We could add
>
> if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
> ...
>
> But should we really try to micto-optimize this code and make it harder
> to read for the purpose of an unproven performance gain? (if there is no
> contention, we'll most probably not even recognize that we're taking a lock)
I agree, I don't see the point of this optimisation.
--
Catalin
On 12/10/21 3:24 PM, Jianyong Wu wrote:
> fixmap is a global resource and is used recursively in create pud mapping.
> It may lead to race condition when alloc_init_pud is called concurrently.
>
> Fox example:
> alloc_init_pud is called when kernel_init. If memory hotplug
> thread, which will also call alloc_init_pud, happens during
> kernel_init, the race for fixmap occurs.
>
> The race condition flow can be:
>
> *************** begin **************
>
> kerenl_init thread virtio-mem workqueue thread
> ================== ======== ==================
> alloc_init_pud(...)
> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> ... ...
> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> ... ...
> pud_clear_fixmap() //fixmap break
> READ_ONCE(*pudp) //CRASH!
>
> **************** end ***************
>
> Hence, a spin lock is introduced to protect the fixmap during create pdg
> mapping.
>
> Signed-off-by: Jianyong Wu <[email protected]>
> ---
> arch/arm64/mm/mmu.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index acfae9b41cc8..98ac09ae9588 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
> static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
>
> static DEFINE_SPINLOCK(swapper_pgdir_lock);
> +static DEFINE_SPINLOCK(fixmap_lock);
>
> void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
> {
> @@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> }
> BUG_ON(p4d_bad(p4d));
>
> + /*
> + * fixmap is global resource, thus it needs to be protected by a lock
> + * in case of race condition.
> + */
Small nit, format and align this comment block. I guess
could also be done while merging this patch as well.
> + spin_lock(&fixmap_lock);
> pudp = pud_set_fixmap_offset(p4dp, addr);
> do {
> pud_t old_pud = READ_ONCE(*pudp);
> @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> } while (pudp++, addr = next, addr != end);
>
> pud_clear_fixmap();
> + spin_unlock(&fixmap_lock);
> }
>
> static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
>
Reviewed-by: Anshuman Khandual <[email protected]>
On Mon, 13 Dec 2021 at 11:16, Anshuman Khandual
<[email protected]> wrote:
>
>
>
> On 12/10/21 3:24 PM, Jianyong Wu wrote:
> > fixmap is a global resource and is used recursively in create pud mapping.
> > It may lead to race condition when alloc_init_pud is called concurrently.
> >
> > Fox example:
> > alloc_init_pud is called when kernel_init. If memory hotplug
> > thread, which will also call alloc_init_pud, happens during
> > kernel_init, the race for fixmap occurs.
> >
> > The race condition flow can be:
> >
> > *************** begin **************
> >
> > kerenl_init thread virtio-mem workqueue thread
> > ================== ======== ==================
> > alloc_init_pud(...)
> > pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> > ... ...
> > READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> > ... ...
> > pud_clear_fixmap() //fixmap break
> > READ_ONCE(*pudp) //CRASH!
> >
> > **************** end ***************
> >
> > Hence, a spin lock is introduced to protect the fixmap during create pdg
> > mapping.
> >
> > Signed-off-by: Jianyong Wu <[email protected]>
> > ---
> > arch/arm64/mm/mmu.c | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> > index acfae9b41cc8..98ac09ae9588 100644
> > --- a/arch/arm64/mm/mmu.c
> > +++ b/arch/arm64/mm/mmu.c
> > @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
> > static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
> >
> > static DEFINE_SPINLOCK(swapper_pgdir_lock);
> > +static DEFINE_SPINLOCK(fixmap_lock);
> >
> > void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
> > {
> > @@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> > }
> > BUG_ON(p4d_bad(p4d));
> >
> > + /*
> > + * fixmap is global resource, thus it needs to be protected by a lock
> > + * in case of race condition.
> > + */
>
> Small nit, format and align this comment block. I guess
> could also be done while merging this patch as well.
>
> > + spin_lock(&fixmap_lock);
> > pudp = pud_set_fixmap_offset(p4dp, addr);
> > do {
> > pud_t old_pud = READ_ONCE(*pudp);
> > @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> > } while (pudp++, addr = next, addr != end);
> >
> > pud_clear_fixmap();
> > + spin_unlock(&fixmap_lock);
> > }
> >
> > static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
> >
>
> Reviewed-by: Anshuman Khandual <[email protected]>
We have different fixmap slots for different page table levels, so
'fixmap_lock' is not the right name.
But we already have swapper_pgdir_lock as well, which serializes the
use of the pgdir level fixmap slot. And we have no spinlocks
protecting the other levels.
So should we perhaps clean this up more comprehensively? Wouldn't it
be better to add a mutex to __create_pgd_mapping(), for instance?
On Mon, Dec 13, 2021 at 11:35:16AM +0100, Ard Biesheuvel wrote:
> On Mon, 13 Dec 2021 at 11:16, Anshuman Khandual
> <[email protected]> wrote:
> > On 12/10/21 3:24 PM, Jianyong Wu wrote:
> > > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> > > index acfae9b41cc8..98ac09ae9588 100644
> > > --- a/arch/arm64/mm/mmu.c
> > > +++ b/arch/arm64/mm/mmu.c
> > > @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
> > > static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
> > >
> > > static DEFINE_SPINLOCK(swapper_pgdir_lock);
> > > +static DEFINE_SPINLOCK(fixmap_lock);
> > >
> > > void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
> > > {
> > > @@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> > > }
> > > BUG_ON(p4d_bad(p4d));
> > >
> > > + /*
> > > + * fixmap is global resource, thus it needs to be protected by a lock
> > > + * in case of race condition.
> > > + */
> >
> > Small nit, format and align this comment block. I guess
> > could also be done while merging this patch as well.
> >
> > > + spin_lock(&fixmap_lock);
> > > pudp = pud_set_fixmap_offset(p4dp, addr);
> > > do {
> > > pud_t old_pud = READ_ONCE(*pudp);
> > > @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> > > } while (pudp++, addr = next, addr != end);
> > >
> > > pud_clear_fixmap();
> > > + spin_unlock(&fixmap_lock);
> > > }
> > >
> > > static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
> > >
> >
> > Reviewed-by: Anshuman Khandual <[email protected]>
>
> We have different fixmap slots for different page table levels, so
> 'fixmap_lock' is not the right name.
>
> But we already have swapper_pgdir_lock as well, which serializes the
> use of the pgdir level fixmap slot. And we have no spinlocks
> protecting the other levels.
>
> So should we perhaps clean this up more comprehensively? Wouldn't it
> be better to add a mutex to __create_pgd_mapping(), for instance?
That does sound like a better way to do things, but the simplicity of this
patch is quite attractive for backporting. Would you object to me queuing
it as-is, on the premise that I'm more than happy to take consolidation
changes on top?
Will
On Mon, 13 Dec 2021 at 14:45, Will Deacon <[email protected]> wrote:
>
> On Mon, Dec 13, 2021 at 11:35:16AM +0100, Ard Biesheuvel wrote:
> > On Mon, 13 Dec 2021 at 11:16, Anshuman Khandual
> > <[email protected]> wrote:
> > > On 12/10/21 3:24 PM, Jianyong Wu wrote:
> > > > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> > > > index acfae9b41cc8..98ac09ae9588 100644
> > > > --- a/arch/arm64/mm/mmu.c
> > > > +++ b/arch/arm64/mm/mmu.c
> > > > @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
> > > > static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
> > > >
> > > > static DEFINE_SPINLOCK(swapper_pgdir_lock);
> > > > +static DEFINE_SPINLOCK(fixmap_lock);
> > > >
> > > > void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
> > > > {
> > > > @@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> > > > }
> > > > BUG_ON(p4d_bad(p4d));
> > > >
> > > > + /*
> > > > + * fixmap is global resource, thus it needs to be protected by a lock
> > > > + * in case of race condition.
> > > > + */
> > >
> > > Small nit, format and align this comment block. I guess
> > > could also be done while merging this patch as well.
> > >
> > > > + spin_lock(&fixmap_lock);
> > > > pudp = pud_set_fixmap_offset(p4dp, addr);
> > > > do {
> > > > pud_t old_pud = READ_ONCE(*pudp);
> > > > @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> > > > } while (pudp++, addr = next, addr != end);
> > > >
> > > > pud_clear_fixmap();
> > > > + spin_unlock(&fixmap_lock);
> > > > }
> > > >
> > > > static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
> > > >
> > >
> > > Reviewed-by: Anshuman Khandual <[email protected]>
> >
> > We have different fixmap slots for different page table levels, so
> > 'fixmap_lock' is not the right name.
> >
> > But we already have swapper_pgdir_lock as well, which serializes the
> > use of the pgdir level fixmap slot. And we have no spinlocks
> > protecting the other levels.
> >
> > So should we perhaps clean this up more comprehensively? Wouldn't it
> > be better to add a mutex to __create_pgd_mapping(), for instance?
>
> That does sound like a better way to do things, but the simplicity of this
> patch is quite attractive for backporting. Would you object to me queuing
> it as-is, on the premise that I'm more than happy to take consolidation
> changes on top?
>
No objections from me.
On Fri, 10 Dec 2021 17:54:32 +0800, Jianyong Wu wrote:
> fixmap is a global resource and is used recursively in create pud mapping.
> It may lead to race condition when alloc_init_pud is called concurrently.
>
> Fox example:
> alloc_init_pud is called when kernel_init. If memory hotplug
> thread, which will also call alloc_init_pud, happens during
> kernel_init, the race for fixmap occurs.
>
> [...]
Applied to arm64 (for-next/fixes), thanks!
[1/1] arm64/mm: avoid fixmap race condition when create pud mapping
https://git.kernel.org/arm64/c/a6a6b9aa34e5
Cheers,
--
Will
https://fixes.arm64.dev
https://next.arm64.dev
https://will.arm64.dev
On Fri, Dec 10, 2021 at 05:54:32PM +0800, Jianyong Wu wrote:
> fixmap is a global resource and is used recursively in create pud mapping.
> It may lead to race condition when alloc_init_pud is called concurrently.
>
> Fox example:
> alloc_init_pud is called when kernel_init. If memory hotplug
> thread, which will also call alloc_init_pud, happens during
> kernel_init, the race for fixmap occurs.
>
> The race condition flow can be:
>
> *************** begin **************
>
> kerenl_init thread virtio-mem workqueue thread
> ================== ======== ==================
> alloc_init_pud(...)
> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> ... ...
> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> ... ...
> pud_clear_fixmap() //fixmap break
> READ_ONCE(*pudp) //CRASH!
>
> **************** end ***************
>
> Hence, a spin lock is introduced to protect the fixmap during create pdg
> mapping.
>
> Signed-off-by: Jianyong Wu <[email protected]>
I am afraid there is a problem to take a spinlock there.
node 0 deferred pages initialised in 2740ms
pgdatinit0 (176) used greatest stack depth: 59184 bytes left
devtmpfs: initialized
KASLR disabled due to lack of seed
BUG: sleeping function called from invalid context at mm/page_alloc.c:5151
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1, name: swapper/0
preempt_count: 1, expected: 0
1 lock held by swapper/0/1:
#0: ffff800009ea3278 (fixmap_lock){+.+.}-{2:2}, at: __create_pgd_mapping
alloc_init_pud at /usr/src/linux-next/arch/arm64/mm/mmu.c:340 (discriminator 4)
(inlined by) __create_pgd_mapping at /usr/src/linux-next/arch/arm64/mm/mmu.c:393 (discriminator 4)
CPU: 0 PID: 1 Comm: swapper/0 Tainted: G W 5.16.0-rc5-next-20211214
Call trace:
dump_backtrace
show_stack
dump_stack_lvl
dump_stack
__might_resched
__might_sleep
__alloc_pages
alloc_page_interleave
alloc_pages
__get_free_pages
__pgd_pgtable_alloc
__create_pgd_mapping
__phys_to_pte_val at /usr/src/linux-next/./arch/arm64/include/asm/pgtable.h:77
(inlined by) __pud_populate at /usr/src/linux-next/./arch/arm64/include/asm/pgalloc.h:25
(inlined by) alloc_init_cont_pmd at /usr/src/linux-next/arch/arm64/mm/mmu.c:277
(inlined by) alloc_init_pud at /usr/src/linux-next/arch/arm64/mm/mmu.c:358
(inlined by) __create_pgd_mapping at /usr/src/linux-next/arch/arm64/mm/mmu.c:393
map_entry_trampoline
map_entry_trampoline at /usr/src/linux-next/arch/arm64/mm/mmu.c:639
do_one_initcall
kernel_init_freeable
kernel_init
ret_from_fork
> ---
> arch/arm64/mm/mmu.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index acfae9b41cc8..98ac09ae9588 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -63,6 +63,7 @@ static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
> static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
>
> static DEFINE_SPINLOCK(swapper_pgdir_lock);
> +static DEFINE_SPINLOCK(fixmap_lock);
>
> void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
> {
> @@ -329,6 +330,11 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> }
> BUG_ON(p4d_bad(p4d));
>
> + /*
> + * fixmap is global resource, thus it needs to be protected by a lock
> + * in case of race condition.
> + */
> + spin_lock(&fixmap_lock);
> pudp = pud_set_fixmap_offset(p4dp, addr);
> do {
> pud_t old_pud = READ_ONCE(*pudp);
> @@ -359,6 +365,7 @@ static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
> } while (pudp++, addr = next, addr != end);
>
> pud_clear_fixmap();
> + spin_unlock(&fixmap_lock);
> }
>
> static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
> --
> 2.17.1
>
On Wed, Dec 15, 2021 at 09:13:37AM -0500, Qian Cai wrote:
> On Fri, Dec 10, 2021 at 05:54:32PM +0800, Jianyong Wu wrote:
> > fixmap is a global resource and is used recursively in create pud mapping.
> > It may lead to race condition when alloc_init_pud is called concurrently.
> >
> > Fox example:
> > alloc_init_pud is called when kernel_init. If memory hotplug
> > thread, which will also call alloc_init_pud, happens during
> > kernel_init, the race for fixmap occurs.
> >
> > The race condition flow can be:
> >
> > *************** begin **************
> >
> > kerenl_init thread virtio-mem workqueue thread
> > ================== ======== ==================
> > alloc_init_pud(...)
> > pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> > ... ...
> > READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> > ... ...
> > pud_clear_fixmap() //fixmap break
> > READ_ONCE(*pudp) //CRASH!
> >
> > **************** end ***************
> >
> > Hence, a spin lock is introduced to protect the fixmap during create pdg
> > mapping.
> >
> > Signed-off-by: Jianyong Wu <[email protected]>
>
> I am afraid there is a problem to take a spinlock there.
>
> node 0 deferred pages initialised in 2740ms
> pgdatinit0 (176) used greatest stack depth: 59184 bytes left
> devtmpfs: initialized
> KASLR disabled due to lack of seed
> BUG: sleeping function called from invalid context at mm/page_alloc.c:5151
Thanks for the report. Definitely a bug, GFP_KERNEL_ACCOUNT allows for
the page allocator to sleep. I'll drop the patch from the arm64
for-next/fixes branch for now and sort it out later.
--
Catalin
On 15.12.21 17:02, Catalin Marinas wrote:
> On Wed, Dec 15, 2021 at 09:13:37AM -0500, Qian Cai wrote:
>> On Fri, Dec 10, 2021 at 05:54:32PM +0800, Jianyong Wu wrote:
>>> fixmap is a global resource and is used recursively in create pud mapping.
>>> It may lead to race condition when alloc_init_pud is called concurrently.
>>>
>>> Fox example:
>>> alloc_init_pud is called when kernel_init. If memory hotplug
>>> thread, which will also call alloc_init_pud, happens during
>>> kernel_init, the race for fixmap occurs.
>>>
>>> The race condition flow can be:
>>>
>>> *************** begin **************
>>>
>>> kerenl_init thread virtio-mem workqueue thread
>>> ================== ======== ==================
>>> alloc_init_pud(...)
>>> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
>>> ... ...
>>> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
>>> ... ...
>>> pud_clear_fixmap() //fixmap break
>>> READ_ONCE(*pudp) //CRASH!
>>>
>>> **************** end ***************
>>>
>>> Hence, a spin lock is introduced to protect the fixmap during create pdg
>>> mapping.
>>>
>>> Signed-off-by: Jianyong Wu <[email protected]>
>>
>> I am afraid there is a problem to take a spinlock there.
>>
>> node 0 deferred pages initialised in 2740ms
>> pgdatinit0 (176) used greatest stack depth: 59184 bytes left
>> devtmpfs: initialized
>> KASLR disabled due to lack of seed
>> BUG: sleeping function called from invalid context at mm/page_alloc.c:5151
>
> Thanks for the report. Definitely a bug, GFP_KERNEL_ACCOUNT allows for
> the page allocator to sleep. I'll drop the patch from the arm64
> for-next/fixes branch for now and sort it out later.
>
Sounds like we need a mutex then.
--
Thanks,
David / dhildenb
Thanks Qian to report this.
I will change to mutex lock and resend it later.
Thanks
Jianyong
> -----Original Message-----
> From: David Hildenbrand <[email protected]>
> Sent: Thursday, December 16, 2021 12:04 AM
> To: Catalin Marinas <[email protected]>; Qian Cai
> <[email protected]>
> Cc: Jianyong Wu <[email protected]>; [email protected]; Anshuman
> Khandual <[email protected]>; [email protected];
> [email protected]; [email protected]; linux-arm-
> [email protected]; [email protected]; Justin He
> <[email protected]>; nd <[email protected]>
> Subject: Re: [PATCH v2] arm64/mm: avoid fixmap race condition when create
> pud mapping
>
> On 15.12.21 17:02, Catalin Marinas wrote:
> > On Wed, Dec 15, 2021 at 09:13:37AM -0500, Qian Cai wrote:
> >> On Fri, Dec 10, 2021 at 05:54:32PM +0800, Jianyong Wu wrote:
> >>> fixmap is a global resource and is used recursively in create pud mapping.
> >>> It may lead to race condition when alloc_init_pud is called concurrently.
> >>>
> >>> Fox example:
> >>> alloc_init_pud is called when kernel_init. If memory hotplug thread,
> >>> which will also call alloc_init_pud, happens during kernel_init, the
> >>> race for fixmap occurs.
> >>>
> >>> The race condition flow can be:
> >>>
> >>> *************** begin **************
> >>>
> >>> kerenl_init thread virtio-mem workqueue thread
> >>> ================== ======== ==================
> >>> alloc_init_pud(...)
> >>> pudp = pud_set_fixmap_offset(..) alloc_init_pud(...)
> >>> ... ...
> >>> READ_ONCE(*pudp) //OK! pudp = pud_set_fixmap_offset(
> >>> ... ...
> >>> pud_clear_fixmap() //fixmap break
> >>> READ_ONCE(*pudp) //CRASH!
> >>>
> >>> **************** end ***************
> >>>
> >>> Hence, a spin lock is introduced to protect the fixmap during create
> >>> pdg mapping.
> >>>
> >>> Signed-off-by: Jianyong Wu <[email protected]>
> >>
> >> I am afraid there is a problem to take a spinlock there.
> >>
> >> node 0 deferred pages initialised in 2740ms
> >> pgdatinit0 (176) used greatest stack depth: 59184 bytes left
> >> devtmpfs: initialized
> >> KASLR disabled due to lack of seed
> >> BUG: sleeping function called from invalid context at
> >> mm/page_alloc.c:5151
> >
> > Thanks for the report. Definitely a bug, GFP_KERNEL_ACCOUNT allows for
> > the page allocator to sleep. I'll drop the patch from the arm64
> > for-next/fixes branch for now and sort it out later.
> >
>
> Sounds like we need a mutex then.
>
> --
> Thanks,
>
> David / dhildenb