On 04/07/2024 23:55, Johannes Berg wrote:
On Thu, 2024-07-04 at 21:16 +0100, Anton Ivanov wrote:
- qi->queue_depth = 0;
+ wmb(); /* Ensure that RX processing elsewhere sees the changes */
+ atomic_set(&qi->queue_depth, 0);
}
I don't understand this.
prep_queue_for_rx() is
On Thu, 2024-07-04 at 21:16 +0100, Anton Ivanov wrote:
> > > - qi->queue_depth = 0;
> > > + wmb(); /* Ensure that RX processing elsewhere sees the changes */
> > > + atomic_set(&qi->queue_depth, 0);
> > > }
> > I don't understand this.
> >
> > prep_queue_for_rx() is called by vector_mmsg_rx(),
On 04/07/2024 17:12, Johannes Berg wrote:
On Thu, 2024-07-04 at 13:22 +0100, anton.iva...@cambridgegreys.com
wrote:
@@ -675,11 +657,20 @@ static void prep_queue_for_rx(struct vector_queue *qi)
struct vector_private *vp = netdev_priv(qi->dev);
struct mmsghdr *mmsg_vector = qi->mms
From: Benjamin Berg
With the change to use execve() we can now safely clear the memory up to
STUB_START as rseq will not be trying to use memory in that region. Also,
on 64 bit the previous changes should mean that there is no usable
memory range above the stub.
Make the change and remove the co
From: Benjamin Berg
The larger memory space is useful to support more applications inside
UML. One example for this is ASAN instrumentation of userspace
applications which requires addresses that would otherwise not be
available.
Signed-off-by: Benjamin Berg
---
v7:
- Reword options and fix d
From: Benjamin Berg
Using clone will not undo features that have been enabled by libc. An
example of this already happening is rseq, which could cause the kernel
to read/write memory of the userspace process. In the future the
standard library might also use mseal by default to protect itself,
wh
From: Benjamin Berg
The calculation was wrong as it only subtracted one and then rounded
down for alignment. However, this is incorrect if host_task_size is not
already aligned.
This probably worked fine because on 64 bit the host_task_size is bigger
than returned by os_get_top_address.
Signed-
From: Benjamin Berg
When loading the UML binary, the host kernel will place the stack at the
highest possible address. It will then map the program name and
environment variables onto the start of the stack.
As such, an easy way to figure out the host_task_size is to use the
highest pointer to a
From: Benjamin Berg
We may have a TASK_SIZE from the host that is bigger than UML is able to
address with a three-level pagetable. Guard against that by clipping the
maximum TASK_SIZE to the maximum addressable area.
Signed-off-by: Benjamin Berg
---
v7: Fix integer overflow on 32 bit with 3-l
From: Benjamin Berg
The 64bit version did not have a stub_syscall1 function yet. Add it as
it will be useful to implement a static binary for stub loading.
Signed-off-by: Benjamin Berg
---
arch/x86/um/shared/sysdep/stub_64.h | 11 +++
1 file changed, 11 insertions(+)
diff --git a/arch
From: Benjamin Berg
The new version of the patchset uses execveat on a memfd instead of
cloning twice to disable rseq. This should be much more robust going
forward as it will also avoid issues with other new features like mseal.
This patchset fixes a few bugs, adds a new method of discovering t
On Thu, 2024-07-04 at 18:49 +0200, Johannes Berg wrote:
> On Thu, 2024-07-04 at 18:27 +0200, Benjamin Berg wrote:
> >
> > + /* set a nice name */
> > + stub_syscall2(__NR_prctl, PR_SET_NAME, (unsigned long)"uml-userspace");
>
> Is that even needed when you're passing it as argv[0] in execve()
On Thu, 2024-07-04 at 18:27 +0200, Benjamin Berg wrote:
> From: Benjamin Berg
>
> With the change to use execve() we can now safely clear the memory up to
> STUB_START as rseq will not be trying to use memory in that region. Also,
> on 64 bit the previous changes should mean that there is no usab
On Thu, 2024-07-04 at 18:27 +0200, Benjamin Berg wrote:
>
> + /* set a nice name */
> + stub_syscall2(__NR_prctl, PR_SET_NAME, (unsigned long)"uml-userspace");
Is that even needed when you're passing it as argv[0] in execve()? But
whatever, it's fine, just wondering.
> + /* setup sig
From: Benjamin Berg
With the change to use execve() we can now safely clear the memory up to
STUB_START as rseq will not be trying to use memory in that region. Also,
on 64 bit the previous changes should mean that there is no usable
memory range above the stub.
Make the change and remove the co
From: Benjamin Berg
The larger memory space is useful to support more applications inside
UML. One example for this is ASAN instrumentation of userspace
applications which requires addresses that would otherwise not be
available.
Signed-off-by: Benjamin Berg
---
v7:
- Reword options and fix d
From: Benjamin Berg
When loading the UML binary, the host kernel will place the stack at the
highest possible address. It will then map the program name and
environment variables onto the start of the stack.
As such, an easy way to figure out the host_task_size is to use the
highest pointer to a
From: Benjamin Berg
The calculation was wrong as it only subtracted one and then rounded
down for alignment. However, this is incorrect if host_task_size is not
already aligned.
This probably worked fine because on 64 bit the host_task_size is bigger
than returned by os_get_top_address.
Signed-
From: Benjamin Berg
We may have a TASK_SIZE from the host that is bigger than UML is able to
address with a three-level pagetable. Guard against that by clipping the
maximum TASK_SIZE to the maximum addressable area.
Signed-off-by: Benjamin Berg
---
v7: Fix integer overflow on 32 bit with 3-l
From: Benjamin Berg
The new version of the patchset uses execveat on a memfd instead of
cloning twice to disable rseq. This should be much more robust going
forward as it will also avoid issues with other new features like mseal.
This patchset fixes a few bugs, adds a new method of discovering t
From: Benjamin Berg
Using clone will not undo features that have been enabled by libc. An
example of this already happening is rseq, which could cause the kernel
to read/write memory of the userspace process. In the future the
standard library might also use mseal by default to protect itself,
wh
From: Benjamin Berg
The 64bit version did not have a stub_syscall1 function yet. Add it as
it will be useful to implement a static binary for stub loading.
Signed-off-by: Benjamin Berg
---
arch/x86/um/shared/sysdep/stub_64.h | 11 +++
1 file changed, 11 insertions(+)
diff --git a/arch
On Thu, 2024-07-04 at 13:22 +0100, anton.iva...@cambridgegreys.com
wrote:
>
> @@ -675,11 +657,20 @@ static void prep_queue_for_rx(struct vector_queue *qi)
> struct vector_private *vp = netdev_priv(qi->dev);
> struct mmsghdr *mmsg_vector = qi->mmsg_vector;
> void **skbuff_vector =
Hi Linus,
This PR fixes a few kselftests [1]. This has been in linux-next for a week and
rebased to add Mark Brown's Tested-by. The race condition found while writing
this fix is not new and seems specific to UML's hostfs (I also tested against
ext4 and btrfs without being able to trigger this i
From: Anton Ivanov
UML vector drivers use ring buffer structures which map
preallocated skbs onto mmsg vectors for use with sendmmsg
and recvmmsg. They are designed around a single consumer,
single producer pattern allowing simultaneous enqueue and
dequeue.
Lock debugging with preemption showed
Hi folks, how can I unsubscribe from this group.?
Thanks in advance.
S
> On 24 Jun 2024, at 18:08, Keith Busch wrote:
>
> On Mon, Jun 17, 2024 at 08:04:41AM +0200, Christoph Hellwig wrote:
>> -#define blk_queue_nonrot(q) test_bit(QUEUE_FLAG_NONROT, &(q)->queue_flags)
>> +#define blk_queue_nonrot
On 04/07/2024 10:55, Johannes Berg wrote:
On Thu, 2024-07-04 at 10:52 +0100, Anton Ivanov wrote:
There is an extra issue there - stats. I need to double-check the locking when
they are being fetched.
Same story - these need atomics. Otherwise we can potentially get the same ABBA
lock issue
From: Johannes Berg
When generalizing this, I was in the mindset of this being
"userspace" code, but even there we should not use variable
arrays as the kernel is moving away from allowing that.
Simply reserve (but not use) enough space for the maximum
two descriptors we might need now, and retu
On Thu, 2024-07-04 at 11:55 +0200, Johannes Berg wrote:
> On Thu, 2024-07-04 at 10:52 +0100, Anton Ivanov wrote:
> > >
> > > There is an extra issue there - stats. I need to double-check the locking
> > > when
> > > they are being fetched.
> >
> > Same story - these need atomics. Otherwise we ca
On Thu, 2024-07-04 at 10:52 +0100, Anton Ivanov wrote:
> >
> > There is an extra issue there - stats. I need to double-check the locking
> > when
> > they are being fetched.
>
> Same story - these need atomics. Otherwise we can potentially get the same
> ABBA
> lock issue when they are being fe
On Thu, 2024-07-04 at 10:45 +0100, Anton Ivanov wrote:
> > (it probably also never even executes twice unless you actually have SMP
> > or preemption?)
>
> It does. If half of the vector is at the end of the array which is used to
> imitate a ring buffer and the other half is at the beginning. Qu
On 04/07/2024 10:45, Anton Ivanov wrote:
On 04/07/2024 10:17, Johannes Berg wrote:
Hi Anton,
Thanks for taking a look! Also thanks for the other explanation.
On Thu, 2024-07-04 at 08:21 +0100, anton.iva...@cambridgegreys.com
wrote:
From: Anton Ivanov
UML vector drivers use ring buffer
On 04/07/2024 10:21, Johannes Berg wrote:
On Thu, 2024-07-04 at 07:58 +0100, Anton Ivanov wrote:
+__initcall(register_power_off);
Anton Ivanov
Heh, I thought I saw an Acked-by from you but then patchwork didn't seem
to pick it up, I guess I'll treat it as one anyway? :)
Yes. Sorry. Too hi
On 04/07/2024 10:17, Johannes Berg wrote:
Hi Anton,
Thanks for taking a look! Also thanks for the other explanation.
On Thu, 2024-07-04 at 08:21 +0100, anton.iva...@cambridgegreys.com
wrote:
From: Anton Ivanov
UML vector drivers use ring buffer structures which map
preallocated skbs onto
On Thu, 2024-07-04 at 07:58 +0100, Anton Ivanov wrote:
>
> > +__initcall(register_power_off);
>
> Anton Ivanov
Heh, I thought I saw an Acked-by from you but then patchwork didn't seem
to pick it up, I guess I'll treat it as one anyway? :)
johannes
Hi Anton,
Thanks for taking a look! Also thanks for the other explanation.
On Thu, 2024-07-04 at 08:21 +0100, anton.iva...@cambridgegreys.com
wrote:
> From: Anton Ivanov
>
> UML vector drivers use ring buffer structures which map
> preallocated skbs onto mmsg vectors for use with sendmmsg
> and
From: Anton Ivanov
UML vector drivers use ring buffer structures which map
preallocated skbs onto mmsg vectors for use with sendmmsg
and recvmmsg. They are designed around a single consumer,
single producer pattern allowing simultaneous enqueue and
dequeue.
Lock debugging with preemption showed
On 03/07/2024 17:46, Johannes Berg wrote:
From: Johannes Berg
If open fails, we have already set vp->opened, but it's
not reset so that any further attempts will just return
-ENXIO. Reset vp->opened even if close has nothing to do.
This helps e.g. with slirp4netns handling only a single
con
On 03/07/2024 17:46, Johannes Berg wrote:
From: Johannes Berg
This lock is useless, all the places that are using
it for some locking will already hold the RTNL. Just
remove it.
Signed-off-by: Johannes Berg
---
v2: remove unused flags variables
---
arch/um/drivers/vector_kern.c | 16 +---
39 matches
Mail list logo
