From: "Michael R. Hines" <mrhines@linux.vnet.ibm.com>
To: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Roland Dreier <roland@kernel.org>,
qemu-devel@nongnu.org,
"linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
Yishai Hadas <yishaih@mellanox.com>,
LKML <linux-kernel@vger.kernel.org>,
Hal Rosenstock <hal.rosenstock@gmail.com>,
Jason Gunthorpe <jgunthorpe@obsidianresearch.com>,
Sean Hefty <sean.hefty@intel.com>,
Christoph Lameter <cl@linux.com>
Subject: Re: [Qemu-devel] [PATCHv2] rdma: add a new IB_ACCESS_GIFT flag
Date: Fri, 05 Apr 2013 16:17:36 -0400 [thread overview]
Message-ID: <515F3160.4020007@linux.vnet.ibm.com> (raw)
In-Reply-To: <20130324155153.GA8597@redhat.com>
The userland part of the patch was missing (IBV_ACCESS_GIFT).
I added flag that to /usr/include in addition to this patch and did a
test RDMA migrate and it seems to work without any problems.
I also removed the IBV_*_WRITE flags on the sender-side and activated
cgroups with the "memory.memsw.limit_in_bytes" activated and the
migration with RDMA also succeeded without any problems (both with *and*
without GIFT also worked).
Any additional tests you would like?
- Michael
On 03/24/2013 11:51 AM, Michael S. Tsirkin wrote:
> At the moment registering an MR breaks COW. This breaks memory
> overcommit for users such as KVM: we have a lot of COW pages, e.g.
> instances of the zero page or pages shared using KSM.
>
> If the application does not care that adapter sees stale data (for
> example, it tracks writes reregisters and resends), it can use a new
> IBV_ACCESS_GIFT flag to prevent registration from breaking COW.
>
> The semantics are similar to that of SPLICE_F_GIFT thus the name.
>
> Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> ---
>
> Please review and consider for 3.10.
>
> Changes from v1:
> rename APP_READONLY to _GIFT: similar to vmsplice's F_GIFT.
>
> drivers/infiniband/core/umem.c | 21 ++++++++++++---------
> include/rdma/ib_verbs.h | 9 ++++++++-
> 2 files changed, 20 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
> index a841123..5dee86d 100644
> --- a/drivers/infiniband/core/umem.c
> +++ b/drivers/infiniband/core/umem.c
> @@ -89,6 +89,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
> int ret;
> int off;
> int i;
> + bool gift, writable;
> DEFINE_DMA_ATTRS(attrs);
>
> if (dmasync)
> @@ -96,6 +97,15 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
>
> if (!can_do_mlock())
> return ERR_PTR(-EPERM);
> + /*
> + * We ask for writable memory if any access flags other than
> + * "remote read" or "gift" are set. "Local write" and "remote write"
> + * obviously require write access. "Remote atomic" can do
> + * things like fetch and add, which will modify memory, and
> + * "MW bind" can change permissions by binding a window.
> + */
> + gift = access & IB_ACCESS_GIFT;
> + writable = access & ~(IB_ACCESS_REMOTE_READ | IB_ACCESS_GIFT);
>
> umem = kmalloc(sizeof *umem, GFP_KERNEL);
> if (!umem)
> @@ -105,14 +115,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
> umem->length = size;
> umem->offset = addr & ~PAGE_MASK;
> umem->page_size = PAGE_SIZE;
> - /*
> - * We ask for writable memory if any access flags other than
> - * "remote read" are set. "Local write" and "remote write"
> - * obviously require write access. "Remote atomic" can do
> - * things like fetch and add, which will modify memory, and
> - * "MW bind" can change permissions by binding a window.
> - */
> - umem->writable = !!(access & ~IB_ACCESS_REMOTE_READ);
> + umem->writable = writable;
>
> /* We assume the memory is from hugetlb until proved otherwise */
> umem->hugetlb = 1;
> @@ -152,7 +155,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
> ret = get_user_pages(current, current->mm, cur_base,
> min_t(unsigned long, npages,
> PAGE_SIZE / sizeof (struct page *)),
> - 1, !umem->writable, page_list, vma_list);
> + !gift, !umem->writable, page_list, vma_list);
>
> if (ret < 0)
> goto out;
> diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
> index 98cc4b2..2e6e13c 100644
> --- a/include/rdma/ib_verbs.h
> +++ b/include/rdma/ib_verbs.h
> @@ -871,7 +871,14 @@ enum ib_access_flags {
> IB_ACCESS_REMOTE_READ = (1<<2),
> IB_ACCESS_REMOTE_ATOMIC = (1<<3),
> IB_ACCESS_MW_BIND = (1<<4),
> - IB_ZERO_BASED = (1<<5)
> + IB_ZERO_BASED = (1<<5),
> + /*
> + * IB_ACCESS_GIFT: This memory is a gift to the adapter. If memory is
> + * modified after registration, the local version and data seen by the
> + * adapter through this region rkey may differ.
> + * Only legal with IB_ACCESS_REMOTE_READ or no permissions.
> + */
> + IB_ACCESS_GIFT = (1<<6)
> };
>
> struct ib_phys_buf {
next prev parent reply other threads:[~2013-04-05 20:18 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-03-24 15:51 [Qemu-devel] [PATCHv2] rdma: add a new IB_ACCESS_GIFT flag Michael S. Tsirkin
2013-04-02 15:51 ` Michael S. Tsirkin
2013-04-02 16:57 ` Roland Dreier
2013-04-02 17:05 ` Michael S. Tsirkin
2013-04-02 22:17 ` Michael R. Hines
2013-04-03 16:07 ` Michael S. Tsirkin
2013-04-05 20:17 ` Michael R. Hines [this message]
2013-04-05 20:43 ` Roland Dreier
2013-04-05 20:51 ` Michael R. Hines
2013-04-05 21:03 ` Roland Dreier
2013-04-05 21:32 ` Michael R. Hines
2013-04-09 19:03 ` Michael S. Tsirkin
2013-04-10 1:26 ` Michael R. Hines
2013-04-10 3:24 ` Michael S. Tsirkin
2013-04-10 4:32 ` Michael R. Hines
2013-04-10 5:32 ` Michael S. Tsirkin
2013-04-10 15:48 ` Michael R. Hines
2013-04-10 15:05 ` Michael S. Tsirkin
2013-04-10 16:28 ` Michael R. Hines
2013-04-09 19:09 ` Michael S. Tsirkin
2013-04-05 20:54 ` Michael R. Hines
2013-04-09 16:39 ` Michael S. Tsirkin
2013-04-09 17:56 ` Michael R. Hines
2013-04-09 19:00 ` Michael S. Tsirkin
2013-04-09 19:12 ` Michael S. Tsirkin
2013-04-09 20:34 ` Michael S. Tsirkin
2013-04-09 20:37 ` Michael S. Tsirkin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=515F3160.4020007@linux.vnet.ibm.com \
--to=mrhines@linux.vnet.ibm.com \
--cc=cl@linux.com \
--cc=hal.rosenstock@gmail.com \
--cc=jgunthorpe@obsidianresearch.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=mst@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=roland@kernel.org \
--cc=sean.hefty@intel.com \
--cc=yishaih@mellanox.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).