diff mbox series

arm64: treat PF_IO_WORKER like PF_KTHREAD for mitigations

Message ID dfc03be2-e246-ed23-6efb-fbd54a8e7781@kernel.dk (mailing list archive)
State New, archived
Headers show
Series arm64: treat PF_IO_WORKER like PF_KTHREAD for mitigations | expand

Commit Message

Jens Axboe Jan. 25, 2023, 4:43 p.m. UTC 
Like PF_KTHREAD, PF_IO_WORKER never exit to userspace. They exist
entirely within the kernel, and hence don't need any task mitigations
applied.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

---

Comments

Will Deacon Jan. 26, 2023, 2 p.m. UTC | #1 
On Wed, Jan 25, 2023 at 09:43:34AM -0700, Jens Axboe wrote:
> Like PF_KTHREAD, PF_IO_WORKER never exit to userspace. They exist
> entirely within the kernel, and hence don't need any task mitigations
> applied.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> ---
> 
> diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
> index fca9cc6f5581..25a21c3d446c 100644
> --- a/arch/arm64/kernel/proton-pack.c
> +++ b/arch/arm64/kernel/proton-pack.c
> @@ -654,7 +654,7 @@ static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
>  void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
>  {
>  	struct pt_regs *regs = task_pt_regs(tsk);
> -	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
> +	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);

Hmm, the other two uses of PF_KTHREAD in arch/arm64 also look pretty
suspect in light of this proposal. Should we also update
ssbs_thread_switch() and access_ok()? If not, then a comment would be
handy to say why PF_KTHREAD is sufficient there.

Will
Jens Axboe Jan. 26, 2023, 2:07 p.m. UTC | #2 
On 1/26/23 7:00 AM, Will Deacon wrote:
> On Wed, Jan 25, 2023 at 09:43:34AM -0700, Jens Axboe wrote:
>> Like PF_KTHREAD, PF_IO_WORKER never exit to userspace. They exist
>> entirely within the kernel, and hence don't need any task mitigations
>> applied.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>
>> ---
>>
>> diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
>> index fca9cc6f5581..25a21c3d446c 100644
>> --- a/arch/arm64/kernel/proton-pack.c
>> +++ b/arch/arm64/kernel/proton-pack.c
>> @@ -654,7 +654,7 @@ static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
>>  void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
>>  {
>>  	struct pt_regs *regs = task_pt_regs(tsk);
>> -	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
>> +	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);
> 
> Hmm, the other two uses of PF_KTHREAD in arch/arm64 also look pretty
> suspect in light of this proposal. Should we also update
> ssbs_thread_switch() and access_ok()? If not, then a comment would be
> handy to say why PF_KTHREAD is sufficient there.

The uaccess one looks like, PF_IO_WORKER threads are just normal userspace
threads. The only difference is that they never exit to userspace, they
remain in the kernel. But everything else is just like a thread.

But yes, the ssbs_thread_switch() should have this check too. I'll send
out an updated patch.
Catalin Marinas Jan. 26, 2023, 3:04 p.m. UTC | #3 
On Thu, Jan 26, 2023 at 07:07:45AM -0700, Jens Axboe wrote:
> On 1/26/23 7:00 AM, Will Deacon wrote:
> > On Wed, Jan 25, 2023 at 09:43:34AM -0700, Jens Axboe wrote:
> >> Like PF_KTHREAD, PF_IO_WORKER never exit to userspace. They exist
> >> entirely within the kernel, and hence don't need any task mitigations
> >> applied.
> >>
> >> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> >>
> >> ---
> >>
> >> diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
> >> index fca9cc6f5581..25a21c3d446c 100644
> >> --- a/arch/arm64/kernel/proton-pack.c
> >> +++ b/arch/arm64/kernel/proton-pack.c
> >> @@ -654,7 +654,7 @@ static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
> >>  void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
> >>  {
> >>  	struct pt_regs *regs = task_pt_regs(tsk);
> >> -	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
> >> +	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);
> > 
> > Hmm, the other two uses of PF_KTHREAD in arch/arm64 also look pretty
> > suspect in light of this proposal. Should we also update
> > ssbs_thread_switch() and access_ok()? If not, then a comment would be
> > handy to say why PF_KTHREAD is sufficient there.
> 
> The uaccess one looks like, PF_IO_WORKER threads are just normal userspace
> threads. The only difference is that they never exit to userspace, they
> remain in the kernel. But everything else is just like a thread.

IIUC these threads are cloned from a user thread and inherit the TIF
flags. If the user already opted in to the tagged addr ABI (usually
early by libc), TIF_TAGGED_ADDR would be set for all threads and
inherited by io_worker threads so we get away with this. But it doesn't
hurt to extend the access_ok check to PF_IO_WORKER just in case the user
plays with the late enabling of TIF_TAGGED_ADDR.
Will Deacon March 28, 2023, 2:14 p.m. UTC | #4 
Hey Jens,

On Thu, Jan 26, 2023 at 07:07:45AM -0700, Jens Axboe wrote:
> On 1/26/23 7:00 AM, Will Deacon wrote:
> > On Wed, Jan 25, 2023 at 09:43:34AM -0700, Jens Axboe wrote:
> >> Like PF_KTHREAD, PF_IO_WORKER never exit to userspace. They exist
> >> entirely within the kernel, and hence don't need any task mitigations
> >> applied.
> >>
> >> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> >>
> >> ---
> >>
> >> diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
> >> index fca9cc6f5581..25a21c3d446c 100644
> >> --- a/arch/arm64/kernel/proton-pack.c
> >> +++ b/arch/arm64/kernel/proton-pack.c
> >> @@ -654,7 +654,7 @@ static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
> >>  void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
> >>  {
> >>  	struct pt_regs *regs = task_pt_regs(tsk);
> >> -	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
> >> +	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);
> > 
> > Hmm, the other two uses of PF_KTHREAD in arch/arm64 also look pretty
> > suspect in light of this proposal. Should we also update
> > ssbs_thread_switch() and access_ok()? If not, then a comment would be
> > handy to say why PF_KTHREAD is sufficient there.
> 
> The uaccess one looks like, PF_IO_WORKER threads are just normal userspace
> threads. The only difference is that they never exit to userspace, they
> remain in the kernel. But everything else is just like a thread.
> 
> But yes, the ssbs_thread_switch() should have this check too. I'll send
> out an updated patch.

I think this one slipped through the cracks, as I don't see any usage of
PF_IO_WORKER in arch/arm64/. Do you plan to update the patch?

Cheers,

Will
Jens Axboe March 28, 2023, 4:47 p.m. UTC | #5 
On 3/28/23 8:14?AM, Will Deacon wrote:
> Hey Jens,
> 
> On Thu, Jan 26, 2023 at 07:07:45AM -0700, Jens Axboe wrote:
>> On 1/26/23 7:00?AM, Will Deacon wrote:
>>> On Wed, Jan 25, 2023 at 09:43:34AM -0700, Jens Axboe wrote:
>>>> Like PF_KTHREAD, PF_IO_WORKER never exit to userspace. They exist
>>>> entirely within the kernel, and hence don't need any task mitigations
>>>> applied.
>>>>
>>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>>
>>>> ---
>>>>
>>>> diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
>>>> index fca9cc6f5581..25a21c3d446c 100644
>>>> --- a/arch/arm64/kernel/proton-pack.c
>>>> +++ b/arch/arm64/kernel/proton-pack.c
>>>> @@ -654,7 +654,7 @@ static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
>>>>  void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
>>>>  {
>>>>  	struct pt_regs *regs = task_pt_regs(tsk);
>>>> -	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
>>>> +	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);
>>>
>>> Hmm, the other two uses of PF_KTHREAD in arch/arm64 also look pretty
>>> suspect in light of this proposal. Should we also update
>>> ssbs_thread_switch() and access_ok()? If not, then a comment would be
>>> handy to say why PF_KTHREAD is sufficient there.
>>
>> The uaccess one looks like, PF_IO_WORKER threads are just normal userspace
>> threads. The only difference is that they never exit to userspace, they
>> remain in the kernel. But everything else is just like a thread.
>>
>> But yes, the ssbs_thread_switch() should have this check too. I'll send
>> out an updated patch.
> 
> I think this one slipped through the cracks, as I don't see any usage of
> PF_IO_WORKER in arch/arm64/. Do you plan to update the patch?
> 

Oops, yes looks like it did. Should look something like this, added a
comment as well. I'll send it out properly.


diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
index 71d59b5abede..9ce614b2177e 100644
--- a/arch/arm64/kernel/process.c
+++ b/arch/arm64/kernel/process.c
@@ -445,9 +445,11 @@ static void ssbs_thread_switch(struct task_struct *next)
 {
 	/*
 	 * Nothing to do for kernel threads, but 'regs' may be junk
-	 * (e.g. idle task) so check the flags and bail early.
+	 * (e.g. idle task) so check the flags and bail early. Nothing to do
+	 * for IO worker threads either, as they never transition between
+	 * kernel and userspace.
 	 */
-	if (unlikely(next->flags & PF_KTHREAD))
+	if (unlikely(next->flags & (PF_KTHREAD | PF_IO_WORKER)))
 		return;
 
 	/*
diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
index fca9cc6f5581..25a21c3d446c 100644
--- a/arch/arm64/kernel/proton-pack.c
+++ b/arch/arm64/kernel/proton-pack.c
@@ -654,7 +654,7 @@ static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
 void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
 {
 	struct pt_regs *regs = task_pt_regs(tsk);
-	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
+	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);
 
 	if (spectre_v4_mitigations_off())
 		ssbs = true;
diff mbox series

Patch

diff --git a/arch/arm64/kernel/proton-pack.c b/arch/arm64/kernel/proton-pack.c
index fca9cc6f5581..25a21c3d446c 100644
--- a/arch/arm64/kernel/proton-pack.c
+++ b/arch/arm64/kernel/proton-pack.c
@@ -654,7 +654,7 @@  static void __update_pstate_ssbs(struct pt_regs *regs, bool state)
 void spectre_v4_enable_task_mitigation(struct task_struct *tsk)
 {
 	struct pt_regs *regs = task_pt_regs(tsk);
-	bool ssbs = false, kthread = tsk->flags & PF_KTHREAD;
+	bool ssbs = false, kthread = tsk->flags & (PF_KTHREAD | PF_IO_WORKER);
 
 	if (spectre_v4_mitigations_off())
 		ssbs = true;