From: "Jake Freeland" <jfree@freebsd.org>
To: "Burakov, Anatoly" <anatoly.burakov@intel.com>,
"Jake Freeland" <jfree@FreeBSD.org>,
"Chenbo Xia" <chenbox@nvidia.com>,
"Nipun Gupta" <nipun.gupta@amd.com>,
"Tyler Retzlaff" <roretzla@linux.microsoft.com>,
"Bruce Richardson" <bruce.richardson@intel.com>
Cc: <dev@dpdk.org>
Subject: Re: [PATCH 1/4] bus/pci: Use force-noreplace flag when mapping PCI resources
Date: Mon, 12 May 2025 09:26:44 -0500 [thread overview]
Message-ID: <D9U9327H5W8O.EKJ2V2MNZM97@freebsd.org> (raw)
In-Reply-To: <53d50d81-75ae-4d23-89b2-bfd6e5733550@intel.com>
On Thu May 8, 2025 at 6:32 AM CDT, Anatoly Burakov wrote:
> On 5/6/2025 7:40 PM, Jake Freeland wrote:
> > When mapping PCI resources in secondary processes, use the
> > RTE_MAP_FORCE_ADDRESS_NOREPLACE flag to indicate that the
> > mapping must be made at the provided address.
> >
> > Without this flag, the kernel might return a different address for
> > the mapping, even if the requested region was available.
> >
> > Signed-off-by: Jake Freeland <jfree@FreeBSD.org>
> > ---
> > drivers/bus/pci/pci_common_uio.c | 4 +++-
> > lib/eal/common/eal_private.h | 7 ++++++-
> > lib/eal/include/rte_eal_paging.h | 7 ++++++-
> > lib/eal/unix/eal_unix_memory.c | 8 +++++++-
> > 4 files changed, 22 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
> > index 30503bd23a..71974e9f56 100644
> > --- a/drivers/bus/pci/pci_common_uio.c
> > +++ b/drivers/bus/pci/pci_common_uio.c
> > @@ -10,6 +10,7 @@
> > #include <sys/mman.h>
> >
> > #include <rte_eal.h>
> > +#include <rte_eal_paging.h>
> > #include <rte_pci.h>
> > #include <rte_bus_pci.h>
> > #include <rte_tailq.h>
> > @@ -58,7 +59,8 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
> >
> > void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
> > fd, (off_t)uio_res->maps[i].offset,
> > - (size_t)uio_res->maps[i].size, 0);
> > + (size_t)uio_res->maps[i].size,
> > + RTE_MAP_FORCE_ADDRESS_NOREPLACE);
> >
> > /* fd is not needed in secondary process, close it */
> > close(fd);
> > diff --git a/lib/eal/common/eal_private.h b/lib/eal/common/eal_private.h
> > index 04ba8ddb86..aaeb140eaf 100644
> > --- a/lib/eal/common/eal_private.h
> > +++ b/lib/eal/common/eal_private.h
> > @@ -211,7 +211,12 @@ enum eal_mem_reserve_flags {
> > * @see RTE_MAP_FORCE_ADDRESS for description of possible consequences
> > * (although implementations are not required to use it).
> > */
> > - EAL_RESERVE_FORCE_ADDRESS = 1 << 1
> > + EAL_RESERVE_FORCE_ADDRESS = 1 << 1,
> > + /**
> > + * Force reserving memory at the requested address, but fail if a
> > + * preexisting mapping collides with the request.
> > + */
> > + EAL_RESERVE_FORCE_ADDRESS_NOREPLACE = 1 << 2,
> > };
> >
> > /**
> > diff --git a/lib/eal/include/rte_eal_paging.h b/lib/eal/include/rte_eal_paging.h
> > index c60317d0f5..7b1983b615 100644
> > --- a/lib/eal/include/rte_eal_paging.h
> > +++ b/lib/eal/include/rte_eal_paging.h
> > @@ -34,7 +34,12 @@ enum rte_map_flags {
> > * may remove all other mappings in the requested region. However,
> > * it is not required to do so, thus mapping with this flag may fail.
> > */
> > - RTE_MAP_FORCE_ADDRESS = 1 << 3
> > + RTE_MAP_FORCE_ADDRESS = 1 << 3,
> > + /**
> > + * Force mapping to the requested address, but fail if a preexisting
> > + * mapping collides with the request.
> > + */
> > + RTE_MAP_FORCE_ADDRESS_NOREPLACE = 1 << 4,
> > };
> >
> > /**
> > diff --git a/lib/eal/unix/eal_unix_memory.c b/lib/eal/unix/eal_unix_memory.c
> > index c540f1e838..61e914b8db 100644
> > --- a/lib/eal/unix/eal_unix_memory.c
> > +++ b/lib/eal/unix/eal_unix_memory.c
> > @@ -17,11 +17,13 @@
> > #ifdef RTE_EXEC_ENV_LINUX
> > #define EAL_DONTDUMP MADV_DONTDUMP
> > #define EAL_DODUMP MADV_DODUMP
> > +#define EAL_FIXED_NOREPLACE MAP_FIXED_NOREPLACE
>
> Is this available in all supported Linux kernel versions?
Support for the MAP_FIXED_NOREPLACE flag was adding in Linux 4.17.
It looks like we're still supporting DPDK 22.11 until Dec 2025. When
22.11 was released, the most recent LTS kernel version was 5.15.
Additionally, the man page specifies:
"older kernels which do not recognize the MAP_FIXED_NOREPLACE flag will
typically (upon detecting a collision with a preexisting mapping) fall
back to a “non-MAP_FIXED” type of behavior"
Thanks,
Jake Freeland
next prev parent reply other threads:[~2025-05-12 14:26 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-05-06 17:40 [PATCH 0/4] BSD PCI Fixes Jake Freeland
2025-05-06 17:40 ` [PATCH 1/4] bus/pci: Use force-noreplace flag when mapping PCI resources Jake Freeland
2025-05-08 11:32 ` Burakov, Anatoly
2025-05-12 14:26 ` Jake Freeland [this message]
2025-05-12 14:42 ` Bruce Richardson
2025-05-06 17:40 ` [PATCH 2/4] bus/pci/bsd: Map resources at EAL baseaddr Jake Freeland
2025-05-08 11:31 ` Burakov, Anatoly
2025-05-06 17:40 ` [PATCH 3/4] bus/pci/bsd: Eliminate potential overflow Jake Freeland
2025-05-08 11:28 ` Burakov, Anatoly
2025-05-06 17:40 ` [PATCH 4/4] bus/pci/bsd: Fix device existence check Jake Freeland
2025-05-08 11:27 ` Burakov, Anatoly
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=D9U9327H5W8O.EKJ2V2MNZM97@freebsd.org \
--to=jfree@freebsd.org \
--cc=anatoly.burakov@intel.com \
--cc=bruce.richardson@intel.com \
--cc=chenbox@nvidia.com \
--cc=dev@dpdk.org \
--cc=nipun.gupta@amd.com \
--cc=roretzla@linux.microsoft.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).