Mapping as little as 64GB can take more than 10 seconds,
triggering issues on kernels with CONFIG_PREEMPT_NONE=y.
ib_umem_get() already splits the work in 2MB units on x86_64,
adding a cond_resched() in the long-lasting loop is enough
to solve the issue.
Note that sg_alloc_table() can still use more than 100 ms,
which is also problematic. This might be addressed later
in ib_umem_add_sg_table(), adding new blocks in sgl
on demand.
Signed-off-by: Eric Dumazet <[email protected]>
Cc: Doug Ledford <[email protected]>
Cc: Jason Gunthorpe <[email protected]>
Cc: [email protected]
---
drivers/infiniband/core/umem.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index 82455a1392f1d19c96ae956f0bd4e93e3a52d29c..831bff8d52e547834e9e04064127fbb280595126 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -261,6 +261,7 @@ struct ib_umem *ib_umem_get(struct ib_device *device, unsigned long addr,
sg = umem->sg_head.sgl;
while (npages) {
+ cond_resched();
ret = pin_user_pages_fast(cur_base,
min_t(unsigned long, npages,
PAGE_SIZE /
--
2.28.0.rc0.142.g3c755180ce-goog
On Wed, Jul 29, 2020 at 06:57:55PM -0700, Eric Dumazet wrote:
> Mapping as little as 64GB can take more than 10 seconds,
> triggering issues on kernels with CONFIG_PREEMPT_NONE=y.
>
> ib_umem_get() already splits the work in 2MB units on x86_64,
> adding a cond_resched() in the long-lasting loop is enough
> to solve the issue.
>
> Note that sg_alloc_table() can still use more than 100 ms,
> which is also problematic. This might be addressed later
> in ib_umem_add_sg_table(), adding new blocks in sgl
> on demand.
I have seen some patches in progress to do exactly this, the
motivation is to reduce the memory consumption if a lot of pages are
combined.
> Signed-off-by: Eric Dumazet <[email protected]>
> Cc: Doug Ledford <[email protected]>
> Cc: Jason Gunthorpe <[email protected]>
> Cc: [email protected]
> ---
> drivers/infiniband/core/umem.c | 1 +
> 1 file changed, 1 insertion(+)
Why [PATCH net] ?
Anyhow, applied to rdma for-next
Thanks,
Jason
On Fri, Jul 31, 2020 at 10:17 AM Jason Gunthorpe <[email protected]> wrote:
>
> On Wed, Jul 29, 2020 at 06:57:55PM -0700, Eric Dumazet wrote:
> > Mapping as little as 64GB can take more than 10 seconds,
> > triggering issues on kernels with CONFIG_PREEMPT_NONE=y.
> >
> > ib_umem_get() already splits the work in 2MB units on x86_64,
> > adding a cond_resched() in the long-lasting loop is enough
> > to solve the issue.
> >
> > Note that sg_alloc_table() can still use more than 100 ms,
> > which is also problematic. This might be addressed later
> > in ib_umem_add_sg_table(), adding new blocks in sgl
> > on demand.
>
> I have seen some patches in progress to do exactly this, the
> motivation is to reduce the memory consumption if a lot of pages are
> combined.
Nice ;)
>
> > Signed-off-by: Eric Dumazet <[email protected]>
> > Cc: Doug Ledford <[email protected]>
> > Cc: Jason Gunthorpe <[email protected]>
> > Cc: [email protected]
> > ---
> > drivers/infiniband/core/umem.c | 1 +
> > 1 file changed, 1 insertion(+)
>
> Why [PATCH net] ?
Sorry, I used a script that I normally use for net submissions, forgot
to remove this tag ;)
>
> Anyhow, applied to rdma for-next
Thanks !
>
> Thanks,
> Jason