* [PATCH 4/5] ISDN-Gigaset: Release memory in gigaset_initcs() after an allocation failure
From: SF Markus Elfring @ 2016-09-26 15:43 UTC (permalink / raw)
To: gigaset307x-common, netdev, Karsten Keil, Paul Bolle
Cc: LKML, kernel-janitors, Julia Lawall
In-Reply-To: <6d4bbb77-914f-19b8-a1a9-2731d1158612@users.sourceforge.net>
From: Markus Elfring <elfring@users.sourceforge.net>
Date: Mon, 26 Sep 2016 16:30:50 +0200
Memory was not released (as it would be expected) when one call
of further resource reservations failed.
* Split a condition check for memory allocation failures so that
each pointer from these function calls will be checked immediately.
See also background information:
Topic "CWE-754: Improper check for unusual or exceptional conditions"
Link: https://cwe.mitre.org/data/definitions/754.html
* Adjust jump targets according to the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
---
drivers/isdn/gigaset/common.c | 20 ++++++++++++--------
1 file changed, 12 insertions(+), 8 deletions(-)
diff --git a/drivers/isdn/gigaset/common.c b/drivers/isdn/gigaset/common.c
index c05a2a4..2e9382f 100644
--- a/drivers/isdn/gigaset/common.c
+++ b/drivers/isdn/gigaset/common.c
@@ -710,10 +710,13 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
cs->mode = M_UNKNOWN;
cs->mstate = MS_UNINITIALIZED;
cs->bcs = kmalloc_array(channels, sizeof(*cs->bcs), GFP_KERNEL);
+ if (!cs->bcs)
+ goto report_failure;
+
cs->inbuf = kmalloc(sizeof(*cs->inbuf), GFP_KERNEL);
- if (!cs->bcs || !cs->inbuf) {
- goto error;
- }
+ if (!cs->inbuf)
+ goto free_bcs;
+
++cs->cs_init;
gig_dbg(DEBUG_INIT, "setting up at_state");
@@ -737,14 +740,14 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
gig_dbg(DEBUG_INIT, "setting up iif");
if (gigaset_isdn_regdev(cs, modulename) < 0) {
pr_err("error registering ISDN device\n");
- goto error;
+ goto free_bcs;
}
make_valid(cs, VALID_ID);
++cs->cs_init;
gig_dbg(DEBUG_INIT, "setting up hw");
if (cs->ops->initcshw(cs) < 0)
- goto error;
+ goto free_bcs;
++cs->cs_init;
@@ -759,7 +762,7 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
gig_dbg(DEBUG_INIT, "setting up bcs[%d]", i);
if (gigaset_initbcs(cs->bcs + i, cs, i) < 0) {
pr_err("could not allocate channel %d data\n", i);
- goto error;
+ goto free_bcs;
}
}
@@ -772,8 +775,9 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
gig_dbg(DEBUG_INIT, "cs initialized");
return cs;
-
-error:
+free_bcs:
+ kfree(cs->bcs);
+report_failure:
gig_dbg(DEBUG_INIT, "failed");
gigaset_freecs(cs);
return NULL;
--
2.10.0
^ permalink raw reply related
* [PATCH 3/5] ISDN-Gigaset: Delete an error message for a failed memory allocation
From: SF Markus Elfring @ 2016-09-26 15:42 UTC (permalink / raw)
To: gigaset307x-common, netdev, Karsten Keil, Paul Bolle
Cc: LKML, kernel-janitors, Julia Lawall, Wolfram Sang
In-Reply-To: <6d4bbb77-914f-19b8-a1a9-2731d1158612@users.sourceforge.net>
From: Markus Elfring <elfring@users.sourceforge.net>
Date: Mon, 26 Sep 2016 15:35:47 +0200
Omit an extra message for a memory allocation failure in this function.
Link: http://events.linuxfoundation.org/sites/events/files/slides/LCJ16-Refactor_Strings-WSang_0.pdf
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
---
drivers/isdn/gigaset/common.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/drivers/isdn/gigaset/common.c b/drivers/isdn/gigaset/common.c
index f505b42..c05a2a4 100644
--- a/drivers/isdn/gigaset/common.c
+++ b/drivers/isdn/gigaset/common.c
@@ -712,7 +712,6 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
cs->bcs = kmalloc_array(channels, sizeof(*cs->bcs), GFP_KERNEL);
cs->inbuf = kmalloc(sizeof(*cs->inbuf), GFP_KERNEL);
if (!cs->bcs || !cs->inbuf) {
- pr_err("out of memory\n");
goto error;
}
++cs->cs_init;
--
2.10.0
^ permalink raw reply related
* [PATCH 2/5] ISDN-Gigaset: Improve another size determination in gigaset_initcs()
From: SF Markus Elfring @ 2016-09-26 15:40 UTC (permalink / raw)
To: gigaset307x-common, netdev, Karsten Keil, Paul Bolle
Cc: LKML, kernel-janitors, Julia Lawall
In-Reply-To: <6d4bbb77-914f-19b8-a1a9-2731d1158612@users.sourceforge.net>
From: Markus Elfring <elfring@users.sourceforge.net>
Date: Sun, 25 Sep 2016 22:32:14 +0200
Replace the specification of a data structure by a pointer dereference
as the parameter for the operator "sizeof" to make the corresponding size
determination a bit safer according to the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
---
drivers/isdn/gigaset/common.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/isdn/gigaset/common.c b/drivers/isdn/gigaset/common.c
index cecbb6a..f505b42 100644
--- a/drivers/isdn/gigaset/common.c
+++ b/drivers/isdn/gigaset/common.c
@@ -710,7 +710,7 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
cs->mode = M_UNKNOWN;
cs->mstate = MS_UNINITIALIZED;
cs->bcs = kmalloc_array(channels, sizeof(*cs->bcs), GFP_KERNEL);
- cs->inbuf = kmalloc(sizeof(struct inbuf_t), GFP_KERNEL);
+ cs->inbuf = kmalloc(sizeof(*cs->inbuf), GFP_KERNEL);
if (!cs->bcs || !cs->inbuf) {
pr_err("out of memory\n");
goto error;
--
2.10.0
^ permalink raw reply related
* [PATCH 1/5] ISDN-Gigaset: Use kmalloc_array() in two functions
From: SF Markus Elfring @ 2016-09-26 15:38 UTC (permalink / raw)
To: gigaset307x-common, netdev, Karsten Keil, Paul Bolle
Cc: LKML, kernel-janitors, Julia Lawall
In-Reply-To: <6d4bbb77-914f-19b8-a1a9-2731d1158612@users.sourceforge.net>
From: Markus Elfring <elfring@users.sourceforge.net>
Date: Sun, 25 Sep 2016 22:22:04 +0200
* Multiplications for the size determination of memory allocations
indicated that array data structures should be processed.
Thus use the corresponding function "kmalloc_array".
This issue was detected by using the Coccinelle software.
* Replace the specification of a data structure by a pointer dereference
to make the corresponding size determination a bit safer according to
the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
---
drivers/isdn/gigaset/common.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/drivers/isdn/gigaset/common.c b/drivers/isdn/gigaset/common.c
index 7c78144..cecbb6a 100644
--- a/drivers/isdn/gigaset/common.c
+++ b/drivers/isdn/gigaset/common.c
@@ -709,8 +709,7 @@ struct cardstate *gigaset_initcs(struct gigaset_driver *drv, int channels,
cs->mode = M_UNKNOWN;
cs->mstate = MS_UNINITIALIZED;
-
- cs->bcs = kmalloc(channels * sizeof(struct bc_state), GFP_KERNEL);
+ cs->bcs = kmalloc_array(channels, sizeof(*cs->bcs), GFP_KERNEL);
cs->inbuf = kmalloc(sizeof(struct inbuf_t), GFP_KERNEL);
if (!cs->bcs || !cs->inbuf) {
pr_err("out of memory\n");
@@ -1089,8 +1088,7 @@ struct gigaset_driver *gigaset_initdriver(unsigned minor, unsigned minors,
drv->ops = ops;
drv->owner = owner;
INIT_LIST_HEAD(&drv->list);
-
- drv->cs = kmalloc(minors * sizeof *drv->cs, GFP_KERNEL);
+ drv->cs = kmalloc_array(minors, sizeof(*drv->cs), GFP_KERNEL);
if (!drv->cs)
goto error;
--
2.10.0
^ permalink raw reply related
* Re: [PATCHv2 0/2] net: fec: Reset ethernet PHY whenever the enet_out clock is being enabled
From: Jörg Krause @ 2016-09-26 15:37 UTC (permalink / raw)
To: Uwe Kleine-König, Lothar Waßmann
Cc: Mark Rutland, Andrew Lunn, Stefan Agner, Greg Ungerer,
Nimrod Andy, Jeff Kirsher, Sascha Hauer, devicetree, Pawel Moll,
Ian Campbell, Richard Cochran, Rob Herring,
Russell King - ARM Linux, linux-arm-kernel, Fabio Estevam,
Kevin Hao, netdev, linux-kernel, Sascha Hauer, Kumar Gala,
Lucas Stach, Shawn Guo, David S. Miller, Philippe Reynes
In-Reply-To: <20160112203002.GZ24441@pengutronix.de>
Hi all,
On Tue, 2016-01-12 at 21:30 +0100, Uwe Kleine-König wrote:
> Hello,
>
> On Tue, Jan 12, 2016 at 05:04:44PM +0100, Lothar Waßmann wrote:
> >
> > >
> > > On Tue, Jan 12, 2016 at 04:17:54PM +0100, Lothar Waßmann wrote:
> > > >
> > > > This patchset fixes a regression introduced by
> > > > commit e8fcfcd5684a ("net: fec: optimize the clock management
> > > > to save power")
> > > > for ethernet PHYs that are using ENET_OUT as reference clock
> > > > (on i.MX6 or i.MX28)
> > > >
> > > > Changes vs. v1:
> > > > - fixed reference to the commit that introduced the regression.
> > > > - dropped patch to use gpiod framework. This should be added
> > > > later,
> > > > after the affected DTBs have been updated to specify the
> > > > correct
> > > > gpio_flags.
> > > >
> > > > Patch overview:
> > > > 1. cleanup patch to remove redundant NULL checks
> > > > 2. call fec_reset_phy() after the ENET_OUT clock has been
> > > > enabled
> > >
> > > I definitely want to test these on my SolidRun boards before
> > > these get
> > > merged: the AR8035 on there is configured via pin-straps, and
> > > then
> > > further tweaked with PHY quirks. Resetting with the iMX6 in the
> > > wrong state may result in the AR8035 being reconfigured (even
> > > jumping
> > > to a different MDIO address) and certainly would need the PHY
> > > quirks
> > > re-running.
> > >
> > As far as I can tell, all SolidRun boards do not specify the
> > enet_out
> > clock in the dtb, so the PHY reset behaviour should be unaffected
> > by
> > this patch on those boards, since the additional fec_reset_phy()
> > call is
> > framed by:
> > if (fep->clk_enet_out) {
> > ...
> > }
> >
> > But verifying this explicitly is of course a good idea.
>
> If the SolidRun boards don't do this, this doesn't mean it's safe in
> general. The problem is real, isn't it?
Anything new on this topic? I am facing the same issue with Linux
Kernel 4.7 and the MCSC LAN8720A on an i.MX28 board.
I found this patch on patchwork and it works for me.
Best regard
Jörg Krause
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply
* [PATCH 0/5] ISDN-Gigaset: Fine-tuning for three function implementations
From: SF Markus Elfring @ 2016-09-26 15:37 UTC (permalink / raw)
To: gigaset307x-common, netdev, Karsten Keil, Paul Bolle
Cc: LKML, kernel-janitors, Julia Lawall
In-Reply-To: <566ABCD9.1060404@users.sourceforge.net>
From: Markus Elfring <elfring@users.sourceforge.net>
Date: Mon, 26 Sep 2016 17:27:17 +0200
Some update suggestions were taken into account
from static source code analysis.
Markus Elfring (5):
Use kmalloc_array() in two functions
Improve another size determination in gigaset_initcs()
Delete an error message for a failed memory allocation
Release memory in gigaset_initcs() after an allocation failure
Enclose two expressions for the sizeof operator by parentheses
drivers/isdn/gigaset/common.c | 31 ++++++++++++++++---------------
1 file changed, 16 insertions(+), 15 deletions(-)
--
2.10.0
^ permalink raw reply
* Re: [PATCH v2 3/7] ipv6 addrconf: rtr_solicits == -1 means unlimited
From: Lorenzo Colitti @ 2016-09-26 15:34 UTC (permalink / raw)
To: Maciej Żenczykowski
Cc: Maciej Żenczykowski, David S . Miller,
netdev@vger.kernel.org, Erik Kline
In-Reply-To: <1474797596-13320-3-git-send-email-zenczykowski@gmail.com>
On Sun, Sep 25, 2016 at 6:59 PM, Maciej Żenczykowski
<zenczykowski@gmail.com> wrote:
> + ifp->idev->cnf.rtr_solicits != 0 &&
Given that some of this patch checks for == -1, and some of it checks
for != 0... is it possible that setting the value to something
unexpected like -3 will cause any issues to the stack? (Other than
just rendering IPv6 unusable on this interface, which seems like a
given.)
^ permalink raw reply
* Re: [PATCH net-next 4/4] net/sched: act_mirred: Implement ingress actions
From: Shmulik Ladkani @ 2016-09-26 15:26 UTC (permalink / raw)
To: Hannes Frederic Sowa
Cc: Florian Westphal, Jamal Hadi Salim, David S. Miller, WANG Cong,
Eric Dumazet, netdev, Daniel Borkmann
In-Reply-To: <54aa5404-f4c8-03e1-0b62-8c070bd6d65b@stressinduktion.org>
Hi,
On Mon, 26 Sep 2016 16:43:16 +0200 Hannes Frederic Sowa <hannes@stressinduktion.org> wrote:
> On 26.09.2016 03:35, Florian Westphal wrote:
> >
> > Yes, but I think we get same issue when we deal with stacked
> > interfaces, and redirect is to e.g. vlan on top of physical device.
>
> We do have the adjacent upper lists in all netdevices, calculating if a
> mirred actions would insert the skb on a stacked device above us should
> be as easy as querying netdev_has_upper_dev and should be possible to
> check that during config time.
This does not seem the right direction:
- We should NOT ban egress redirect to an "upper device", this limits
flexibility available today:
One may validly redirect to an "upper device" according to a very
specific criteria such that the skb will NOT be caught again by the
filter when re-iterated down the stack.
- And even if we did, at "config time" as you say, one can always stack
the target device ontop the filtered device at a LATER time.
And obviously, testing "is upper" while executing action is a bad idea.
- Moreover, as Daniel noted, this is just a band-aid which limitely
addresses only few of the already existing troubles with stacked
virtual netdevices (with or without act_mirred).
Frankly, I think the loop-detection suggestions overload act_mirred
with AI that might imply limitations not well assessed (even on existing
usecases).
If needed, I'd go with the "recursion detection" suggested by Daniel,
which is simple and aids with the issues of "egress" redirect that
already exist today.
Moreover, as the patch series is all about "ingress redirect" support,
the "recursion detection" deserves a series of its own, we shouldn't
bundle these two distinct features.
Regards,
Shmulik
^ permalink raw reply
* Re: [PATCH v3 net-next] net: phy: Add Edge-rate driver for Microsemi PHYs.
From: Andrew Lunn @ 2016-09-26 15:12 UTC (permalink / raw)
To: Raju Lakkaraju
Cc: netdev-u79uwXL29TY76Z2rM5mHXA, devicetree-u79uwXL29TY76Z2rM5mHXA,
f.fainelli-Re5JQEeQqe8AvxtiuMwx3w,
Allan.Nielsen-dzo6w/eZyo2tG0bUXCXiUA
In-Reply-To: <1474897705-30802-1-git-send-email-Raju.Lakkaraju-dzo6w/eZyo2tG0bUXCXiUA@public.gmane.org>
On Mon, Sep 26, 2016 at 07:18:25PM +0530, Raju Lakkaraju wrote:
> Hi,
>
> Thank you for review the code and valuable comments.
> Summary of changes:
> V1: Initial version of Edge-rate driver add by using IOCTL.
> V2: Changed edge-rate parameter to Device Tree with magic number.
> V3: Added Device Tree documentatoin and edge-rate parameter table.
> Added probe function initialize the vsc8531 private datastructure.
> Thanks,
> Raju.
>
> From: Raju Lakkaraju <Raju.Lakkaraju-dzo6w/eZyo2tG0bUXCXiUA@public.gmane.org>
>
> ---
Please read SubmittingPatches about what the --- means, and what
should be above it and what should be below it.
Andrew
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* Re: [PATCH net-next 4/4] net/sched: act_mirred: Implement ingress actions
From: Hannes Frederic Sowa @ 2016-09-26 15:12 UTC (permalink / raw)
To: Daniel Borkmann, Florian Westphal, Jamal Hadi Salim
Cc: Shmulik Ladkani, David S. Miller, WANG Cong, Eric Dumazet, netdev
In-Reply-To: <57E9366D.6020906@iogearbox.net>
On 26.09.2016 16:53, Daniel Borkmann wrote:
> On 09/26/2016 04:43 PM, Hannes Frederic Sowa wrote:
>> On 26.09.2016 03:35, Florian Westphal wrote:
>>> Jamal Hadi Salim <jhs@mojatatu.com> wrote:
>>>> On 16-09-25 02:31 PM, Florian Westphal wrote:
>>>>> Shmulik Ladkani <shmulik.ladkani@gmail.com> wrote:
>>>>>> We can later address any loop-detection improvements in mirred.
>>>>>> WDYT?
>>>>>
>>>>> You can address this after fixing infamous spinlock recursion hard
>>>>> lockup (which has existed forever):
>>>>>
>>>>> tc qdisc add dev eth0 root handle 1: prio
>>>>> tc filter add dev eth0 parent 1: protocol ip u32 match u32 0 0 flowid
>>>>> 1:2 action mirred egress redirect dev eth0
>>>>>
>>>>> (only do this on toy vm)
>>>>
>>>> Realize didnt respond to this. Seems very simple to fix:
>>>> if skb->dev->ifindex and m->tcfm_dev->ifindex are the
>>>> same, then you can drop the packet.
>>>
>>> Yes, but I think we get same issue when we deal with stacked
>>> interfaces, and redirect is to e.g. vlan on top of physical device.
>>
>> We do have the adjacent upper lists in all netdevices, calculating if a
>> mirred actions would insert the skb on a stacked device above us should
>> be as easy as querying netdev_has_upper_dev and should be possible to
>> check that during config time.
>
> But that would still not be enough, no? In the sense that with above
> scenario, you could redirect to some arbitrary device that redirects
> this back to the original device if on purpose configured as such,
> thus they don't necessarily need to have a stacked relationship.
Yes, it would only help with the scenario Florian described above.
Personally, I would only try to fix and warn against the easy to detect
cases. It is easy enough to just create a loop with your local attached
L2 which brings your box into a endless loop processing the same packet
again and again. Because it is out of control of the kernel you cannot
do anything at all.
I would just care that we sometimes reschedule and don't do everything
in one stack so we don't corrupt the machine and an admin has still a
chance to solve the problem.
Bye,
Hannes
^ permalink raw reply
* Re: [2/3] hostap: Delete an unnecessary jump label in prism2_ioctl_priv_hostapd()
From: Kalle Valo @ 2016-09-26 15:06 UTC (permalink / raw)
To: SF Markus Elfring
Cc: linux-wireless, netdev, Jouni Malinen, LKML, kernel-janitors,
Julia Lawall
In-Reply-To: <65ed528f-7ec5-a0f2-6219-3e63a78ca0ee@users.sourceforge.net>
SF Markus Elfring <elfring@users.sourceforge.net> wrote:
> From: Markus Elfring <elfring@users.sourceforge.net>
> Date: Sat, 20 Aug 2016 18:21:29 +0200
>
> Remove a jump label which is unneeded in this function at the end.
>
> Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
2 patches set to Rejected.
9291771 [2/3] hostap: Delete an unnecessary jump label in prism2_ioctl_priv_hostapd()
9291775 [3/3] hostap: Delete unnecessary initialisations for the variable "ret"
Reason: The benefit is not clear.
--
https://patchwork.kernel.org/patch/9291771/
Documentation about submitting wireless patches and checking status
from patchwork:
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
^ permalink raw reply
* RE: [PATCH v2] fs/select: add vmalloc fallback for select(2)
From: David Laight @ 2016-09-26 15:02 UTC (permalink / raw)
To: 'Vlastimil Babka', Eric Dumazet
Cc: Alexander Viro, Andrew Morton, linux-fsdevel@vger.kernel.org,
linux-kernel@vger.kernel.org, linux-mm@kvack.org, Michal Hocko,
netdev@vger.kernel.org, Linux API, linux-man@vger.kernel.org
In-Reply-To: <5bb958c9-542e-e86b-779c-e3d93dc01632@suse.cz>
From: Vlastimil Babka
> Sent: 26 September 2016 11:02
> On 09/23/2016 03:35 PM, David Laight wrote:
> > From: Vlastimil Babka
> >> Sent: 23 September 2016 10:59
> > ...
> >> > I suspect that fdt->max_fds is an upper bound for the highest fd the
> >> > process has open - not the RLIMIT_NOFILE value.
> >>
> >> I gathered that the highest fd effectively limits the number of files,
> >> so it's the same. I might be wrong.
> >
> > An application can reduce RLIMIT_NOFILE below that of an open file.
>
> OK, I did some more digging in the code, and my understanding is that:
>
> - fdt->max_fds is the current size of the fdtable, which isn't allocated upfront
> to match the limit, but grows as needed. This means it's OK for
> core_sys_select() to silently cap nfds, as it knows there are no fd's with
> higher number in the fdtable, so it's a performance optimization.
Not entirely, if any bits are set for fd above fdt->max_fds then select()
call should fail - fd not open.
> However, to
> match what the manpage says, there should be another check against RLIMIT_NOFILE
> to return -EINVAL, which there isn't, AFAICS.
>
> - fdtable is expanded (and fdt->max_fds bumped) by
> expand_files()->expand_fdtable() which checks against fs.nr_open sysctl, which
> seems to be 1048576 where I checked.
>
> - callers of expand_files(), such as dup(), check the rlimit(RLIMIT_NOFILE) to
> limit the expansion.
>
> So yeah, application can reduce RLIMIT_NOFILE, but it has no effect on fdtable
> and fdt->max_fds that is already above the limit. Select syscall would have to
> check the rlimit to conform to the manpage. Or (rather?) we should fix the manpage.
I think the manpage should be fixed (delete that clause).
Then add code to the system call to scan the high bit sets (above fdt->max_fds)
for any non-zero bytes. This can be done into a small buffer.
> As for the original vmalloc() flood concern, I still think we're safe, as
> ordinary users are limited by RLIMIT_NOFILE way below sizes that would need
> vmalloc(), and root has many other options to DOS the system (or worse).
Some processes need very high numbers of fd.
Likely they don't use select() on them, but trashing performance if they
do is a bit silly.
Trying to slit the 3 masks first seems sensible.
David
^ permalink raw reply
* [PATCH net-next] bnx2x: free the mac filter group list before freeing the cmd
From: Jason Baron @ 2016-09-26 15:00 UTC (permalink / raw)
To: davem; +Cc: Yuval.Mintz, Ariel.Elior, netdev
The group list must be freed prior to freeing the command otherwise
we have a use-after-free.
Signed-off-by: Jason Baron <jbaron@akamai.com>
Cc: Yuval Mintz <Yuval.Mintz@qlogic.com>
Cc: Ariel Elior <Ariel.Elior@qlogic.com>
---
drivers/net/ethernet/broadcom/bnx2x/bnx2x_sp.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_sp.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_sp.c
index 4947a9c..cea6bdc 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_sp.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_sp.c
@@ -2714,8 +2714,8 @@ static int bnx2x_mcast_enqueue_cmd(struct bnx2x *bp,
elem_group = (struct bnx2x_mcast_elem_group *)
__get_free_page(GFP_ATOMIC | __GFP_ZERO);
if (!elem_group) {
- kfree(new_cmd);
bnx2x_free_groups(&new_cmd->group_head);
+ kfree(new_cmd);
return -ENOMEM;
}
total_elems -= MCAST_MAC_ELEMS_PER_PG;
--
1.9.1
^ permalink raw reply related
* Re: [PATCH] brcmfmac: implement more accurate skb tracking
From: Dan Williams @ 2016-09-26 14:59 UTC (permalink / raw)
To: Rafał Miłecki, Arend Van Spriel
Cc: Kalle Valo, Franky Lin, Hante Meuleman, Pieter-Paul Giesberts,
Franky Lin, linux-wireless@vger.kernel.org,
open list:BROADCOM BRCM80211 IEEE802.11n WIRELESS DRIVER,
Network Development, Linux Kernel Mailing List,
Rafał Miłecki
In-Reply-To: <CACna6rwnLb3sEPmdx6uhYMstVn04WbP2xzznsjnwgguEXn04=w@mail.gmail.com>
On Mon, 2016-09-26 at 14:13 +0200, Rafał Miłecki wrote:
> On 26 September 2016 at 13:46, Arend Van Spriel
> <arend.vanspriel@broadcom.com> wrote:
> >
> > On 26-9-2016 12:23, Rafał Miłecki wrote:
> > >
> > > From: Rafał Miłecki <rafal@milecki.pl>
> > >
> > > We need to track 802.1x packets to know if there are any pending
> > > ones
> > > for transmission. This is required for performing key update in
> > > the
> > > firmware.
> >
> > The problem we are trying to solve is a pretty old one. The problem
> > is
> > that wpa_supplicant uses two separate code paths: EAPOL messaging
> > through data path and key configuration though nl80211.
>
> Can I find it described/reported somewhere?
If I understand the issue correctly, you can find all this in the
supplicant code. Once the supplicant has done whatever it wants to do
with the data frames that just happen to be EAPOL it then sends the
keys down to the driver with nl80211.
But it sounds like, instead of sniffing EAPOL frames in the driver skb
tracking and sniffing ETH_P_PAE, you should probably implement support
for NL80211_CMD_CRIT_PROTOCOL_START/NL80211_CMD_CRIT_PROTOCOL_STOP and
key off the passed-in NL80211_CRIT_PROTO_EAPOL. At least at the
beginning of connection setup only EAPOL packets will be allowed
anyway.
It doesn't seem like the supplicant uses NL80211_CRIT_PROTO_EAPOL yet,
but that should also be fixed in the supplicant itself. You should
probably get some comments from Jouni on how he'd like to see all this
work. But generally the less specific sniffing of frames in drivers,
likely the better.
Dan
>
> >
> > >
> > > Unfortunately our old tracking code wasn't very accurate. It was
> > > treating skb as pending as soon as it was passed by the netif.
> > > Actual
> > > handling packet to the firmware was happening later as brcmfmac
> > > internally queues them and uses its own worker(s).
> >
> > That does not seem right. As soon as we get a 1x packet we need to
> > wait
> > with key configuration regardless whether it is still in the driver
> > or
> > handed over to firmware already.
>
> OK, thanks.
^ permalink raw reply
* Re: [PATCH net-next 4/4] net/sched: act_mirred: Implement ingress actions
From: Daniel Borkmann @ 2016-09-26 14:53 UTC (permalink / raw)
To: Hannes Frederic Sowa, Florian Westphal, Jamal Hadi Salim
Cc: Shmulik Ladkani, David S. Miller, WANG Cong, Eric Dumazet, netdev
In-Reply-To: <54aa5404-f4c8-03e1-0b62-8c070bd6d65b@stressinduktion.org>
On 09/26/2016 04:43 PM, Hannes Frederic Sowa wrote:
> On 26.09.2016 03:35, Florian Westphal wrote:
>> Jamal Hadi Salim <jhs@mojatatu.com> wrote:
>>> On 16-09-25 02:31 PM, Florian Westphal wrote:
>>>> Shmulik Ladkani <shmulik.ladkani@gmail.com> wrote:
>>>>> We can later address any loop-detection improvements in mirred.
>>>>> WDYT?
>>>>
>>>> You can address this after fixing infamous spinlock recursion hard
>>>> lockup (which has existed forever):
>>>>
>>>> tc qdisc add dev eth0 root handle 1: prio
>>>> tc filter add dev eth0 parent 1: protocol ip u32 match u32 0 0 flowid
>>>> 1:2 action mirred egress redirect dev eth0
>>>>
>>>> (only do this on toy vm)
>>>
>>> Realize didnt respond to this. Seems very simple to fix:
>>> if skb->dev->ifindex and m->tcfm_dev->ifindex are the
>>> same, then you can drop the packet.
>>
>> Yes, but I think we get same issue when we deal with stacked
>> interfaces, and redirect is to e.g. vlan on top of physical device.
>
> We do have the adjacent upper lists in all netdevices, calculating if a
> mirred actions would insert the skb on a stacked device above us should
> be as easy as querying netdev_has_upper_dev and should be possible to
> check that during config time.
But that would still not be enough, no? In the sense that with above
scenario, you could redirect to some arbitrary device that redirects
this back to the original device if on purpose configured as such,
thus they don't necessarily need to have a stacked relationship.
>> And we have such loops even without tc, for instance when placing
>> both veth ends in same bridge :-(
>
> We can't fix that without a ttl in the sk_buff struct.
>
> Bye,
> Hannes
^ permalink raw reply
* Re: [PATCH v3] bpf: Set register type according to is_valid_access()
From: Daniel Borkmann @ 2016-09-26 14:49 UTC (permalink / raw)
To: Mickaël Salaün, linux-kernel
Cc: Alexei Starovoitov, Andy Lutomirski, Kees Cook, Sargun Dhillon,
Tejun Heo, netdev
In-Reply-To: <20160924180150.23620-1-mic@digikod.net>
On 09/24/2016 08:01 PM, Mickaël Salaün wrote:
> This prevent future potential pointer leaks when an unprivileged eBPF
> program will read a pointer value from its context. Even if
> is_valid_access() returns a pointer type, the eBPF verifier replace it
> with UNKNOWN_VALUE. The register value that contains a kernel address is
> then allowed to leak. Moreover, this fix allows unprivileged eBPF
> programs to use functions with (legitimate) pointer arguments.
>
> Not an issue currently since reg_type is only set for PTR_TO_PACKET or
> PTR_TO_PACKET_END in XDP and TC programs that can only be loaded as
> privileged. For now, the only unprivileged eBPF program allowed is for
> socket filtering and all the types from its context are UNKNOWN_VALUE.
> However, this fix is important for future unprivileged eBPF programs
> which could use pointers in their context.
>
> Signed-off-by: Mickaël Salaün <mic@digikod.net>
> Cc: Alexei Starovoitov <ast@kernel.org>
> Cc: Daniel Borkmann <daniel@iogearbox.net>
Seems okay to me:
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
^ permalink raw reply
* Re: [PATCH net-next 4/4] net/sched: act_mirred: Implement ingress actions
From: Hannes Frederic Sowa @ 2016-09-26 14:43 UTC (permalink / raw)
To: Florian Westphal, Jamal Hadi Salim
Cc: Shmulik Ladkani, David S. Miller, WANG Cong, Eric Dumazet, netdev,
Daniel Borkmann
In-Reply-To: <20160926013504.GA1959@breakpoint.cc>
On 26.09.2016 03:35, Florian Westphal wrote:
> Jamal Hadi Salim <jhs@mojatatu.com> wrote:
>> On 16-09-25 02:31 PM, Florian Westphal wrote:
>>> Shmulik Ladkani <shmulik.ladkani@gmail.com> wrote:
>>>> We can later address any loop-detection improvements in mirred.
>>>> WDYT?
>>>
>>> You can address this after fixing infamous spinlock recursion hard
>>> lockup (which has existed forever):
>>>
>>> tc qdisc add dev eth0 root handle 1: prio
>>> tc filter add dev eth0 parent 1: protocol ip u32 match u32 0 0 flowid
>>> 1:2 action mirred egress redirect dev eth0
>>>
>>> (only do this on toy vm)
>>>
>>
>> Realize didnt respond to this. Seems very simple to fix:
>> if skb->dev->ifindex and m->tcfm_dev->ifindex are the
>> same, then you can drop the packet.
>
> Yes, but I think we get same issue when we deal with stacked
> interfaces, and redirect is to e.g. vlan on top of physical device.
We do have the adjacent upper lists in all netdevices, calculating if a
mirred actions would insert the skb on a stacked device above us should
be as easy as querying netdev_has_upper_dev and should be possible to
check that during config time.
> And we have such loops even without tc, for instance when placing
> both veth ends in same bridge :-(
We can't fix that without a ttl in the sk_buff struct.
Bye,
Hannes
^ permalink raw reply
* Re: [1/3] hostap: Use memdup_user() rather than duplicating its implementation
From: Kalle Valo @ 2016-09-26 14:19 UTC (permalink / raw)
To: SF Markus Elfring
Cc: linux-wireless-u79uwXL29TY76Z2rM5mHXA,
netdev-u79uwXL29TY76Z2rM5mHXA, Jouni Malinen, LKML,
kernel-janitors-u79uwXL29TY76Z2rM5mHXA, Julia Lawall
In-Reply-To: <efa66c16-7998-9ceb-0be6-d62dd249a2dc-Rn4VEauK+AKRv+LV9MX5uipxlwaOVQ5f@public.gmane.org>
SF Markus Elfring <elfring-Rn4VEauK+AKRv+LV9MX5uipxlwaOVQ5f@public.gmane.org> wrote:
> From: Markus Elfring <elfring-Rn4VEauK+AKRv+LV9MX5uipxlwaOVQ5f@public.gmane.org>
> Date: Sat, 20 Aug 2016 18:19:43 +0200
>
> Reuse existing functionality from memdup_user() instead of keeping
> duplicate source code.
>
> This issue was detected by using the Coccinelle software.
>
> Signed-off-by: Markus Elfring <elfring-Rn4VEauK+AKRv+LV9MX5uipxlwaOVQ5f@public.gmane.org>
Patch set to Rejected.
[1/3] hostap: Use memdup_user() rather than duplicating i... 2016-08-20 SF Markus El Rejected
Reason: A similar patch is already applied.
Applying: hostap: Use memdup_user() rather than duplicating its implementation
Using index info to reconstruct a base tree...
Falling back to patching base and 3-way merge...
Auto-merging drivers/net/wireless/intersil/hostap/hostap_ioctl.c
CONFLICT (content): Merge conflict in drivers/net/wireless/intersil/hostap/hostap_ioctl.c
Failed to merge in the changes.
Patch failed at 0001 hostap: Use memdup_user() rather than duplicating its implementation
--
https://patchwork.kernel.org/patch/9306999/
Documentation about submitting wireless patches and checking status
from patchwork:
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
^ permalink raw reply
* [PATCH 11/11] qedr: Add events support and register IB device
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem, dledford
Cc: Ariel.Elior, Michal.Kalderon, Yuval.Mintz, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani@cavium.com>
Add error handling support.
Register ib device with ib stack.
Signed-off-by: Rajesh Borundia <rajesh.borundia@cavium.com>
Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
---
drivers/infiniband/hw/qedr/main.c | 114 ++++++++++++++++++++++++++++++++++++-
drivers/infiniband/hw/qedr/verbs.c | 37 ++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 9 +++
include/linux/qed/common_hsi.h | 1 +
4 files changed, 159 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 735bc70..fc2afd0 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -79,10 +79,25 @@ static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str,
(fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
}
+static struct net_device *qedr_get_netdev(struct ib_device *dev, u8 port_num)
+{
+ struct qedr_dev *qdev;
+
+ qdev = get_qedr_dev(dev);
+ dev_hold(qdev->ndev);
+
+ /* The HW vendor's device driver must guarantee
+ * that this function returns NULL before the net device reaches
+ * NETDEV_UNREGISTER_FINAL state.
+ */
+ return qdev->ndev;
+}
+
static int qedr_register_device(struct qedr_dev *dev)
{
strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
+ dev->ibdev.node_guid = dev->attr.node_guid;
memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
dev->ibdev.owner = THIS_MODULE;
dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
@@ -151,11 +166,15 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.post_send = qedr_post_send;
dev->ibdev.post_recv = qedr_post_recv;
+ dev->ibdev.process_mad = qedr_process_mad;
+ dev->ibdev.get_port_immutable = qedr_port_immutable;
+ dev->ibdev.get_netdev = qedr_get_netdev;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
- return 0;
+ return ib_register_device(&dev->ibdev, NULL);
}
/* This function allocates fast-path status block memory */
@@ -530,6 +549,92 @@ static int qedr_set_device_attr(struct qedr_dev *dev)
return 0;
}
+void qedr_unaffiliated_event(void *context,
+ u8 event_code)
+{
+ pr_err("unaffiliated event not implemented yet\n");
+}
+
+
+void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
+{
+#define EVENT_TYPE_NOT_DEFINED 0
+#define EVENT_TYPE_CQ 1
+#define EVENT_TYPE_QP 2
+ struct qedr_dev *dev = (struct qedr_dev *)context;
+ union event_ring_data *data = fw_handle;
+ u64 roce_handle64 = ((u64)data->roce_handle.hi << 32) +
+ data->roce_handle.lo;
+ u8 event_type = EVENT_TYPE_NOT_DEFINED;
+ struct ib_event event;
+ struct ib_cq *ibcq;
+ struct ib_qp *ibqp;
+ struct qedr_cq *cq;
+ struct qedr_qp *qp;
+
+ switch (e_code) {
+ case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
+ event.event = IB_EVENT_CQ_ERR;
+ event_type = EVENT_TYPE_CQ;
+ break;
+ case ROCE_ASYNC_EVENT_SQ_DRAINED:
+ event.event = IB_EVENT_SQ_DRAINED;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
+ event.event = IB_EVENT_QP_FATAL;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
+ event.event = IB_EVENT_QP_REQ_ERR;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
+ event.event = IB_EVENT_QP_ACCESS_ERR;
+ event_type = EVENT_TYPE_QP;
+ break;
+ default:
+ DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
+ roce_handle64);
+ }
+
+ switch (event_type) {
+ case EVENT_TYPE_CQ:
+ cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
+ if (cq) {
+ ibcq = &cq->ibcq;
+ if (ibcq->event_handler) {
+ event.device = ibcq->device;
+ event.element.cq = ibcq;
+ ibcq->event_handler(&event, ibcq->cq_context);
+ }
+ } else {
+ WARN(1,
+ "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
+ roce_handle64);
+ }
+ DP_ERR(dev, "CQ event %d on hanlde %p\n", e_code, cq);
+ break;
+ case EVENT_TYPE_QP:
+ qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
+ if (qp) {
+ ibqp = &qp->ibqp;
+ if (ibqp->event_handler) {
+ event.device = ibqp->device;
+ event.element.qp = ibqp;
+ ibqp->event_handler(&event, ibqp->qp_context);
+ }
+ } else {
+ WARN(1,
+ "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
+ roce_handle64);
+ }
+ DP_ERR(dev, "QP event %d on hanlde %p\n", e_code, qp);
+ break;
+ default:
+ break;
+ }
+}
static int qedr_init_hw(struct qedr_dev *dev)
{
@@ -559,6 +664,8 @@ static int qedr_init_hw(struct qedr_dev *dev)
cur_pbl->pbl_ptr = (u64)p_phys_table;
}
+ events.affiliated_event = qedr_affiliated_event;
+ events.unaffiliated_event = qedr_unaffiliated_event;
events.context = dev;
in_params->events = &events;
@@ -657,11 +764,13 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
- goto reg_err;
+ goto sysfs_err;
DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
return dev;
+sysfs_err:
+ ib_unregister_device(&dev->ibdev);
reg_err:
qedr_sync_free_irqs(dev);
irq_err:
@@ -681,6 +790,7 @@ static void qedr_remove(struct qedr_dev *dev)
* of the registered clients.
*/
qedr_remove_sysfiles(dev);
+ ib_unregister_device(&dev->ibdev);
qedr_stop_hw(dev);
qedr_sync_free_irqs(dev);
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 9f0a72f..36424ec 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -3522,3 +3522,40 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
spin_unlock_irqrestore(&cq->cq_lock, flags);
return done;
}
+
+int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags,
+ u8 port_num,
+ const struct ib_wc *in_wc,
+ const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *mad_hdr,
+ size_t in_mad_size, struct ib_mad_hdr *out_mad,
+ size_t *out_mad_size, u16 *out_mad_pkey_index)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+
+ DP_DEBUG(dev, QEDR_MSG_GSI,
+ "QEDR_PROCESS_MAD in_mad %x %x %x %x %x %x %x %x\n",
+ mad_hdr->attr_id, mad_hdr->base_version, mad_hdr->attr_mod,
+ mad_hdr->class_specific, mad_hdr->class_version,
+ mad_hdr->method, mad_hdr->mgmt_class, mad_hdr->status);
+ return IB_MAD_RESULT_SUCCESS;
+}
+
+int qedr_port_immutable(struct ib_device *ibdev, u8 port_num,
+ struct ib_port_immutable *immutable)
+{
+ struct ib_port_attr attr;
+ int err;
+
+ err = qedr_query_port(ibdev, port_num, &attr);
+ if (err)
+ return err;
+
+ immutable->pkey_tbl_len = attr.pkey_tbl_len;
+ immutable->gid_tbl_len = attr.gid_tbl_len;
+ immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
+ RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
+ immutable->max_mad_size = IB_MGMT_MAD_SIZE;
+
+ return 0;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index 9760fd1..89c21a9 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -89,4 +89,13 @@ int qedr_post_send(struct ib_qp *, struct ib_send_wr *,
struct ib_send_wr **bad_wr);
int qedr_post_recv(struct ib_qp *, struct ib_recv_wr *,
struct ib_recv_wr **bad_wr);
+int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags,
+ u8 port_num, const struct ib_wc *in_wc,
+ const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *in_mad,
+ size_t in_mad_size, struct ib_mad_hdr *out_mad,
+ size_t *out_mad_size, u16 *out_mad_pkey_index);
+
+int qedr_port_immutable(struct ib_device *ibdev, u8 port_num,
+ struct ib_port_immutable *immutable);
#endif
diff --git a/include/linux/qed/common_hsi.h b/include/linux/qed/common_hsi.h
index 1902763..734deb0 100644
--- a/include/linux/qed/common_hsi.h
+++ b/include/linux/qed/common_hsi.h
@@ -674,6 +674,7 @@ union event_ring_data {
struct iscsi_eqe_data iscsi_info;
struct malicious_vf_eqe_data malicious_vf;
struct initial_cleanup_eqe_data vf_init_cleanup;
+ struct regpair roce_handle;
};
/* Event Ring Entry */
--
1.8.3.1
^ permalink raw reply related
* [PATCH 07/11] qedr: Add support for memory registeration verbs
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem, dledford
Cc: Ariel.Elior, Michal.Kalderon, Yuval.Mintz, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani@cavium.com>
Add support for user, dma and memory regions registration.
Signed-off-by: Rajesh Borundia <rajesh.borundia@cavium.com>
Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
---
drivers/infiniband/hw/qedr/main.c | 11 +-
drivers/infiniband/hw/qedr/qedr.h | 40 ++++
drivers/infiniband/hw/qedr/verbs.c | 365 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 14 ++
drivers/net/ethernet/qlogic/qed/qed_roce.c | 240 +++++++++++++++++++
include/linux/qed/qed_roce_if.h | 6 +
6 files changed, 674 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 4669b26..d86ba2c 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -100,7 +100,9 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(CREATE_QP) |
QEDR_UVERBS(MODIFY_QP) |
QEDR_UVERBS(QUERY_QP) |
- QEDR_UVERBS(DESTROY_QP);
+ QEDR_UVERBS(DESTROY_QP) |
+ QEDR_UVERBS(REG_MR) |
+ QEDR_UVERBS(DEREG_MR);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -133,8 +135,13 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.query_pkey = qedr_query_pkey;
+ dev->ibdev.get_dma_mr = qedr_get_dma_mr;
+ dev->ibdev.dereg_mr = qedr_dereg_mr;
+ dev->ibdev.reg_user_mr = qedr_reg_user_mr;
+ dev->ibdev.alloc_mr = qedr_alloc_mr;
+ dev->ibdev.map_mr_sg = qedr_map_mr_sg;
+
dev->ibdev.dma_device = &dev->pdev->dev;
-
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index fb0831a..cfb6860 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -377,6 +377,41 @@ struct qedr_qp {
struct qedr_userq urq;
};
+struct qedr_ah {
+ struct ib_ah ibah;
+ struct ib_ah_attr attr;
+};
+
+enum qedr_mr_type {
+ QEDR_MR_USER,
+ QEDR_MR_KERNEL,
+ QEDR_MR_DMA,
+ QEDR_MR_FRMR
+};
+
+struct mr_info {
+ struct qedr_pbl *pbl_table;
+ struct qedr_pbl_info pbl_info;
+ struct list_head free_pbl_list;
+ struct list_head inuse_pbl_list;
+ u32 completed;
+ u32 completed_handled;
+};
+
+struct qedr_mr {
+ struct ib_mr ibmr;
+ struct ib_umem *umem;
+
+ struct qed_rdma_register_tid_in_params hw_mr;
+ enum qedr_mr_type type;
+
+ struct qedr_dev *dev;
+ struct mr_info info;
+
+ u64 *pages;
+ u32 npages;
+};
+
static inline int qedr_get_dmac(struct qedr_dev *dev,
struct ib_ah_attr *ah_attr, u8 *mac_addr)
{
@@ -421,4 +456,9 @@ static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp)
{
return container_of(ibqp, struct qedr_qp, ibqp);
}
+
+static inline struct qedr_mr *get_qedr_mr(struct ib_mr *ibmr)
+{
+ return container_of(ibmr, struct qedr_mr, ibmr);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 4f3c84f..3e7681d 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -2089,3 +2089,368 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
return rc;
}
+
+static void free_mr_info(struct qedr_dev *dev, struct mr_info *info)
+{
+ struct qedr_pbl *pbl, *tmp;
+
+ if (info->pbl_table)
+ list_add_tail(&info->pbl_table->list_entry,
+ &info->free_pbl_list);
+
+ if (!list_empty(&info->inuse_pbl_list))
+ list_splice(&info->inuse_pbl_list, &info->free_pbl_list);
+
+ list_for_each_entry_safe(pbl, tmp, &info->free_pbl_list, list_entry) {
+ list_del(&pbl->list_entry);
+ qedr_free_pbl(dev, &info->pbl_info, pbl);
+ }
+}
+
+static int init_mr_info(struct qedr_dev *dev, struct mr_info *info,
+ size_t page_list_len, bool two_layered)
+{
+ struct qedr_pbl *tmp;
+ int rc;
+
+ INIT_LIST_HEAD(&info->free_pbl_list);
+ INIT_LIST_HEAD(&info->inuse_pbl_list);
+
+ rc = qedr_prepare_pbl_tbl(dev, &info->pbl_info,
+ page_list_len, two_layered);
+ if (rc)
+ goto done;
+
+ info->pbl_table = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
+ if (!info->pbl_table) {
+ rc = -ENOMEM;
+ goto done;
+ }
+
+ DP_DEBUG(dev, QEDR_MSG_MR, "pbl_table_pa = %pa\n",
+ &info->pbl_table->pa);
+
+ /* in usual case we use 2 PBLs, so we add one to free
+ * list and allocating another one
+ */
+ tmp = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
+ if (!tmp) {
+ DP_DEBUG(dev, QEDR_MSG_MR, "Extra PBL is not allocated\n");
+ goto done;
+ }
+
+ list_add_tail(&tmp->list_entry, &info->free_pbl_list);
+
+ DP_DEBUG(dev, QEDR_MSG_MR, "extra pbl_table_pa = %pa\n", &tmp->pa);
+
+done:
+ if (rc)
+ free_mr_info(dev, info);
+
+ return rc;
+}
+
+struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len,
+ u64 usr_addr, int acc, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_mr *mr;
+ struct qedr_pd *pd;
+ int rc = -ENOMEM;
+
+ pd = get_qedr_pd(ibpd);
+ DP_DEBUG(dev, QEDR_MSG_MR,
+ "qedr_register user mr pd = %d start = %lld, len = %lld, usr_addr = %lld, acc = %d\n",
+ pd->pd_id, start, len, usr_addr, acc);
+
+ if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE))
+ return ERR_PTR(-EINVAL);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(rc);
+
+ mr->type = QEDR_MR_USER;
+
+ mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0);
+ if (IS_ERR(mr->umem)) {
+ rc = -EFAULT;
+ goto err0;
+ }
+
+ rc = init_mr_info(dev, &mr->info, ib_umem_page_count(mr->umem), 1);
+ if (rc)
+ goto err1;
+
+ qedr_populate_pbls(dev, mr->umem, mr->info.pbl_table,
+ &mr->info.pbl_info);
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ /* Index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR;
+ mr->hw_mr.key = 0;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+ mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+ mr->hw_mr.mw_bind = false;
+ mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa;
+ mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
+ mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
+ mr->hw_mr.page_size_log = ilog2(mr->umem->page_size);
+ mr->hw_mr.fbo = ib_umem_offset(mr->umem);
+ mr->hw_mr.length = len;
+ mr->hw_mr.vaddr = usr_addr;
+ mr->hw_mr.zbva = false;
+ mr->hw_mr.phy_mr = false;
+ mr->hw_mr.dma_mr = false;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err2;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
+ mr->hw_mr.remote_atomic)
+ mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+
+ DP_DEBUG(dev, QEDR_MSG_MR, "register user mr lkey: %x\n",
+ mr->ibmr.lkey);
+ return &mr->ibmr;
+
+err2:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err1:
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+err0:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
+
+int qedr_dereg_mr(struct ib_mr *ib_mr)
+{
+ struct qedr_mr *mr = get_qedr_mr(ib_mr);
+ struct qedr_dev *dev = get_qedr_dev(ib_mr->device);
+ int rc = 0;
+
+ rc = dev->ops->rdma_deregister_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ if (rc)
+ return rc;
+
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+
+ if ((mr->type != QEDR_MR_DMA) && (mr->type != QEDR_MR_FRMR))
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+
+ /* it could be user registered memory. */
+ if (mr->umem)
+ ib_umem_release(mr->umem);
+
+ kfree(mr);
+
+ return rc;
+}
+
+struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int max_page_list_len)
+{
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_mr *mr;
+ int rc = -ENOMEM;
+
+ DP_DEBUG(dev, QEDR_MSG_MR,
+ "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd->pd_id,
+ max_page_list_len);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(rc);
+
+ mr->dev = dev;
+ mr->type = QEDR_MR_FRMR;
+
+ rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
+ if (rc)
+ goto err0;
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err0;
+ }
+
+ /* Index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
+ mr->hw_mr.key = 0;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = 0;
+ mr->hw_mr.remote_read = 0;
+ mr->hw_mr.remote_write = 0;
+ mr->hw_mr.remote_atomic = 0;
+ mr->hw_mr.mw_bind = false;
+ mr->hw_mr.pbl_ptr = 0;
+ mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
+ mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
+ mr->hw_mr.fbo = 0;
+ mr->hw_mr.length = 0;
+ mr->hw_mr.vaddr = 0;
+ mr->hw_mr.zbva = false;
+ mr->hw_mr.phy_mr = true;
+ mr->hw_mr.dma_mr = false;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ mr->ibmr.rkey = mr->ibmr.lkey;
+
+ DP_DEBUG(dev, QEDR_MSG_MR, "alloc frmr: %x\n", mr->ibmr.lkey);
+ return mr;
+
+err1:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err0:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
+
+struct ib_mr *qedr_alloc_mr(struct ib_pd *ibpd,
+ enum ib_mr_type mr_type, u32 max_num_sg)
+{
+ struct qedr_dev *dev;
+ struct qedr_mr *mr;
+
+ if (mr_type != IB_MR_TYPE_MEM_REG)
+ return ERR_PTR(-EINVAL);
+
+ mr = __qedr_alloc_mr(ibpd, max_num_sg);
+
+ if (IS_ERR(mr))
+ return ERR_PTR(-EINVAL);
+
+ dev = mr->dev;
+
+ return &mr->ibmr;
+}
+
+static int qedr_set_page(struct ib_mr *ibmr, u64 addr)
+{
+ struct qedr_mr *mr = get_qedr_mr(ibmr);
+ struct qedr_pbl *pbl_table;
+ struct regpair *pbe;
+ u32 pbes_in_page;
+
+ if (unlikely(mr->npages == mr->info.pbl_info.num_pbes)) {
+ DP_ERR(mr->dev, "qedr_set_page failes when %d\n", mr->npages);
+ return -ENOMEM;
+ }
+
+ DP_DEBUG(mr->dev, QEDR_MSG_MR, "qedr_set_page pages[%d] = 0x%llx\n",
+ mr->npages, addr);
+
+ pbes_in_page = mr->info.pbl_info.pbl_size / sizeof(u64);
+ pbl_table = mr->info.pbl_table + (mr->npages / pbes_in_page);
+ pbe = (struct regpair *)pbl_table->va;
+ pbe += mr->npages % pbes_in_page;
+ pbe->lo = cpu_to_le32((u32)addr);
+ pbe->hi = cpu_to_le32((u32)upper_32_bits(addr));
+
+ mr->npages++;
+
+ return 0;
+}
+
+static void handle_completed_mrs(struct qedr_dev *dev, struct mr_info *info)
+{
+ int work = info->completed - info->completed_handled - 1;
+
+ DP_DEBUG(dev, QEDR_MSG_MR, "Special FMR work = %d\n", work);
+ while (work-- > 0 && !list_empty(&info->inuse_pbl_list)) {
+ struct qedr_pbl *pbl;
+
+ /* Free all the page list that are possible to be freed
+ * (all the ones that were invalidated), under the assumption
+ * that if an FMR was completed successfully that means that
+ * if there was an invalidate operation before it also ended
+ */
+ pbl = list_first_entry(&info->inuse_pbl_list,
+ struct qedr_pbl, list_entry);
+ list_del(&pbl->list_entry);
+ list_add_tail(&pbl->list_entry, &info->free_pbl_list);
+ info->completed_handled++;
+ }
+}
+
+int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset)
+{
+ struct qedr_mr *mr = get_qedr_mr(ibmr);
+
+ mr->npages = 0;
+
+ handle_completed_mrs(mr->dev, &mr->info);
+ return ib_sg_to_pages(ibmr, sg, sg_nents, NULL, qedr_set_page);
+}
+
+struct ib_mr *qedr_get_dma_mr(struct ib_pd *ibpd, int acc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct qedr_mr *mr;
+ int rc;
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(-ENOMEM);
+
+ mr->type = QEDR_MR_DMA;
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ /* index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+ mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+ mr->hw_mr.dma_mr = true;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err2;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
+ mr->hw_mr.remote_atomic)
+ mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+
+ DP_DEBUG(dev, QEDR_MSG_MR, "get dma mr: lkey = %x\n", mr->ibmr.lkey);
+ return &mr->ibmr;
+
+err2:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err1:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index 14b4980..efa9ea2 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -70,4 +70,18 @@ int qedr_query_qp(struct ib_qp *, struct ib_qp_attr *qp_attr,
int qp_attr_mask, struct ib_qp_init_attr *);
int qedr_destroy_qp(struct ib_qp *ibqp);
+struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr);
+int qedr_destroy_ah(struct ib_ah *ibah);
+
+int qedr_dereg_mr(struct ib_mr *);
+struct ib_mr *qedr_get_dma_mr(struct ib_pd *, int acc);
+
+struct ib_mr *qedr_reg_user_mr(struct ib_pd *, u64 start, u64 length,
+ u64 virt, int acc, struct ib_udata *);
+
+int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset);
+
+struct ib_mr *qedr_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+ u32 max_num_sg);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index cf92ff7..04e33ee 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -690,6 +690,17 @@ struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
return p_hwfn->p_rdma_info->dev;
}
+void qed_rdma_free_tid(void *rdma_cxt, u32 itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tid_map, itid);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
{
struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
@@ -2321,6 +2332,231 @@ int qed_rdma_modify_qp(void *rdma_cxt,
return rc;
}
+int qed_rdma_register_tid(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_register_tid_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ enum rdma_tid_type tid_type;
+ u8 fw_return_code;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", params->itid);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_REGISTER_MR,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (p_hwfn->p_rdma_info->last_tid < params->itid)
+ p_hwfn->p_rdma_info->last_tid = params->itid;
+
+ p_ramrod = &p_ent->ramrod.rdma_register_tid;
+
+ p_ramrod->flags = 0;
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_TWO_LEVEL_PBL,
+ params->pbl_two_level);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_ZERO_BASED, params->zbva);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_PHY_MR, params->phy_mr);
+
+ /* Don't initialize D/C field, as it may override other bits. */
+ if (!(params->tid_type == QED_RDMA_TID_FMR) && !(params->dma_mr))
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_PAGE_SIZE_LOG,
+ params->page_size_log - 12);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_MAX_ID,
+ p_hwfn->p_rdma_info->last_tid);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_READ,
+ params->remote_read);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_WRITE,
+ params->remote_write);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_ATOMIC,
+ params->remote_atomic);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_WRITE,
+ params->local_write);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_READ, params->local_read);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_ENABLE_MW_BIND,
+ params->mw_bind);
+
+ SET_FIELD(p_ramrod->flags1,
+ RDMA_REGISTER_TID_RAMROD_DATA_PBL_PAGE_SIZE_LOG,
+ params->pbl_page_size_log - 12);
+
+ SET_FIELD(p_ramrod->flags2,
+ RDMA_REGISTER_TID_RAMROD_DATA_DMA_MR, params->dma_mr);
+
+ switch (params->tid_type) {
+ case QED_RDMA_TID_REGISTERED_MR:
+ tid_type = RDMA_TID_REGISTERED_MR;
+ break;
+ case QED_RDMA_TID_FMR:
+ tid_type = RDMA_TID_FMR;
+ break;
+ case QED_RDMA_TID_MW_TYPE1:
+ tid_type = RDMA_TID_MW_TYPE1;
+ break;
+ case QED_RDMA_TID_MW_TYPE2A:
+ tid_type = RDMA_TID_MW_TYPE2A;
+ break;
+ default:
+ rc = -EINVAL;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+ SET_FIELD(p_ramrod->flags1,
+ RDMA_REGISTER_TID_RAMROD_DATA_TID_TYPE, tid_type);
+
+ p_ramrod->itid = cpu_to_le32(params->itid);
+ p_ramrod->key = params->key;
+ p_ramrod->pd = cpu_to_le16(params->pd);
+ p_ramrod->length_hi = (u8)(params->length >> 32);
+ p_ramrod->length_lo = DMA_LO_LE(params->length);
+ if (params->zbva) {
+ /* Lower 32 bits of the registered MR address.
+ * In case of zero based MR, will hold FBO
+ */
+ p_ramrod->va.hi = 0;
+ p_ramrod->va.lo = cpu_to_le32(params->fbo);
+ } else {
+ DMA_REGPAIR_LE(p_ramrod->va, params->vaddr);
+ }
+ DMA_REGPAIR_LE(p_ramrod->pbl_base, params->pbl_ptr);
+
+ /* DIF */
+ if (params->dif_enabled) {
+ SET_FIELD(p_ramrod->flags2,
+ RDMA_REGISTER_TID_RAMROD_DATA_DIF_ON_HOST_FLG, 1);
+ DMA_REGPAIR_LE(p_ramrod->dif_error_addr,
+ params->dif_error_addr);
+ DMA_REGPAIR_LE(p_ramrod->dif_runt_addr, params->dif_runt_addr);
+ }
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Register TID, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_deregister_tid(void *rdma_cxt, u32 itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_deregister_tid_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ struct qed_ptt *p_ptt;
+ u8 fw_return_code;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_DEREGISTER_MR,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.rdma_deregister_tid;
+ p_ramrod->itid = cpu_to_le32(itid);
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (fw_return_code == RDMA_RETURN_DEREGISTER_MR_BAD_STATE_ERR) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ return -EINVAL;
+ } else if (fw_return_code == RDMA_RETURN_NIG_DRAIN_REQ) {
+ /* Bit indicating that the TID is in use and a nig drain is
+ * required before sending the ramrod again
+ */
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ rc = -EBUSY;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to acquire PTT\n");
+ return rc;
+ }
+
+ rc = qed_mcp_drain(p_hwfn, p_ptt);
+ if (rc) {
+ qed_ptt_release(p_hwfn, p_ptt);
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Drain failed\n");
+ return rc;
+ }
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ /* Resend the ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_DEREGISTER_MR,
+ p_hwfn->p_rdma_info->proto,
+ &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to init sp-element\n");
+ return rc;
+ }
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Ramrod failed\n");
+ return rc;
+ }
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n",
+ fw_return_code);
+ return rc;
+ }
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "De-registered TID, rc = %d\n", rc);
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -2419,6 +2655,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_modify_qp = &qed_rdma_modify_qp,
.rdma_query_qp = &qed_rdma_query_qp,
.rdma_destroy_qp = &qed_rdma_destroy_qp,
+ .rdma_alloc_tid = &qed_rdma_alloc_tid,
+ .rdma_free_tid = &qed_rdma_free_tid,
+ .rdma_register_tid = &qed_rdma_register_tid,
+ .rdma_deregister_tid = &qed_rdma_deregister_tid,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 02321e3..0b6df6e 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -512,6 +512,12 @@ struct qed_rdma_ops {
int (*rdma_query_qp)(void *rdma_cxt, struct qed_rdma_qp *qp,
struct qed_rdma_query_qp_out_params *oparams);
int (*rdma_destroy_qp)(void *rdma_cxt, struct qed_rdma_qp *qp);
+ int
+ (*rdma_register_tid)(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *iparams);
+ int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
+ int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
+ void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related
* [PATCH 04/11] qedr: Add support for user context verbs
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem, dledford
Cc: Ariel.Elior, Michal.Kalderon, Yuval.Mintz, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani@cavium.com>
Add support for ucontext, query port, add and del gid verbs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@cavium.com>
Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
---
drivers/infiniband/hw/qedr/Makefile | 2 +-
drivers/infiniband/hw/qedr/main.c | 26 ++
drivers/infiniband/hw/qedr/qedr.h | 34 +++
drivers/infiniband/hw/qedr/verbs.c | 454 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 52 ++++
drivers/net/ethernet/qlogic/qed/qed.h | 2 +-
drivers/net/ethernet/qlogic/qed/qed_dev.c | 1 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 19 +-
drivers/net/ethernet/qlogic/qed/qed_roce.h | 1 +
include/linux/qed/qed_roce_if.h | 1 +
include/linux/qed/rdma_common.h | 1 +
include/uapi/rdma/qedr-abi.h | 53 ++++
12 files changed, 643 insertions(+), 3 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/verbs.c
create mode 100644 drivers/infiniband/hw/qedr/verbs.h
create mode 100644 include/uapi/rdma/qedr-abi.h
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
index 3a5b7a2..b10f2b1 100644
--- a/drivers/infiniband/hw/qedr/Makefile
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
-qedr-y := main.o
+qedr-y := main.o verbs.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 0b61c8d..c131d5f 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -32,6 +32,7 @@
#include <linux/module.h>
#include <rdma/ib_verbs.h>
#include <rdma/ib_addr.h>
+#include <rdma/ib_user_verbs.h>
#include <linux/netdevice.h>
#include <linux/iommu.h>
#include <net/addrconf.h>
@@ -39,6 +40,8 @@
#include <linux/qed/qed_chain.h>
#include <linux/qed/qed_if.h>
#include "qedr.h"
+#include "verbs.h"
+#include <rdma/qedr-abi.h>
MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
MODULE_AUTHOR("QLogic Corporation");
@@ -80,7 +83,30 @@ static int qedr_register_device(struct qedr_dev *dev)
memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
dev->ibdev.owner = THIS_MODULE;
+ dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
+ dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
+ QEDR_UVERBS(QUERY_DEVICE) |
+ QEDR_UVERBS(QUERY_PORT);
+
+ dev->ibdev.phys_port_cnt = 1;
+ dev->ibdev.num_comp_vectors = dev->num_cnq;
+ dev->ibdev.node_type = RDMA_NODE_IB_CA;
+
+ dev->ibdev.query_device = qedr_query_device;
+ dev->ibdev.query_port = qedr_query_port;
+ dev->ibdev.modify_port = qedr_modify_port;
+
+ dev->ibdev.query_gid = qedr_query_gid;
+ dev->ibdev.add_gid = qedr_add_gid;
+ dev->ibdev.del_gid = qedr_del_gid;
+
+ dev->ibdev.alloc_ucontext = qedr_alloc_ucontext;
+ dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext;
+ dev->ibdev.mmap = qedr_mmap;
+
+ dev->ibdev.dma_device = &dev->pdev->dev;
+
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index fbb9202..172d514 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -51,6 +51,7 @@ do { \
} while (0)
#define QEDR_MSG_INIT "INIT"
+#define QEDR_MSG_MISC "MISC"
struct qedr_dev;
@@ -178,6 +179,39 @@ struct qedr_dev {
#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+#define QEDR_ROCE_PKEY_MAX 1
+#define QEDR_ROCE_PKEY_TABLE_LEN 1
+#define QEDR_ROCE_PKEY_DEFAULT 0xffff
+
+struct qedr_ucontext {
+ struct ib_ucontext ibucontext;
+ struct qedr_dev *dev;
+ struct qedr_pd *pd;
+ u64 dpi_addr;
+ u64 dpi_phys_addr;
+ u32 dpi_size;
+ u16 dpi;
+
+ struct list_head mm_head;
+
+ /* Lock to protect mm list */
+ struct mutex mm_list_lock;
+};
+
+struct qedr_mm {
+ struct {
+ u64 phy_addr;
+ unsigned long len;
+ } key;
+ struct list_head entry;
+};
+
+static inline
+struct qedr_ucontext *get_qedr_ucontext(struct ib_ucontext *ibucontext)
+{
+ return container_of(ibucontext, struct qedr_ucontext, ibucontext);
+}
+
static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev)
{
return container_of(ibdev, struct qedr_dev, ibdev);
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
new file mode 100644
index 0000000..08465d5
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -0,0 +1,454 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/crc32.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/udp.h>
+#include <linux/iommu.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include "qedr_hsi.h"
+#include <linux/qed/qed_if.h>
+#include "qedr.h"
+#include "verbs.h"
+#include <rdma/qedr-abi.h>
+
+int qedr_query_gid(struct ib_device *ibdev, u8 port, int index,
+ union ib_gid *sgid)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ int rc = 0;
+
+ if (!rdma_cap_roce_gid_table(ibdev, port))
+ return -ENODEV;
+
+ rc = ib_get_cached_gid(ibdev, port, index, sgid, NULL);
+ if (rc == -EAGAIN) {
+ memcpy(sgid, &zgid, sizeof(*sgid));
+ return 0;
+ }
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "query gid: index=%d %llx:%llx\n", index,
+ sgid->global.interface_id, sgid->global.subnet_prefix);
+
+ return rc;
+}
+
+int qedr_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context)
+{
+ if (!rdma_cap_roce_gid_table(device, port_num))
+ return -EINVAL;
+
+ if (port_num > QEDR_MAX_PORT)
+ return -EINVAL;
+
+ if (!context)
+ return -EINVAL;
+
+ return 0;
+}
+
+int qedr_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context)
+{
+ if (!rdma_cap_roce_gid_table(device, port_num))
+ return -EINVAL;
+
+ if (port_num > QEDR_MAX_PORT)
+ return -EINVAL;
+
+ if (!context)
+ return -EINVAL;
+
+ return 0;
+}
+
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qedr_device_attr *qattr = &dev->attr;
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev,
+ "qedr_query_device called with invalid params rdma_ctx=%p\n",
+ dev->rdma_ctx);
+ return -EINVAL;
+ }
+
+ memset(attr, 0, sizeof(*attr));
+
+ attr->fw_ver = qattr->fw_ver;
+ attr->sys_image_guid = qattr->sys_image_guid;
+ attr->max_mr_size = qattr->max_mr_size;
+ attr->page_size_cap = qattr->page_size_caps;
+ attr->vendor_id = qattr->vendor_id;
+ attr->vendor_part_id = qattr->vendor_part_id;
+ attr->hw_ver = qattr->hw_ver;
+ attr->max_qp = qattr->max_qp;
+ attr->max_qp_wr = max_t(u32, qattr->max_sqe, qattr->max_rqe);
+ attr->device_cap_flags = IB_DEVICE_CURR_QP_STATE_MOD |
+ IB_DEVICE_RC_RNR_NAK_GEN |
+ IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_MGT_EXTENSIONS;
+
+ attr->max_sge = qattr->max_sge;
+ attr->max_sge_rd = qattr->max_sge;
+ attr->max_cq = qattr->max_cq;
+ attr->max_cqe = qattr->max_cqe;
+ attr->max_mr = qattr->max_mr;
+ attr->max_mw = qattr->max_mw;
+ attr->max_pd = qattr->max_pd;
+ attr->atomic_cap = dev->atomic_cap;
+ attr->max_fmr = qattr->max_fmr;
+ attr->max_map_per_fmr = 16;
+ attr->max_qp_init_rd_atom =
+ 1 << (fls(qattr->max_qp_req_rd_atomic_resc) - 1);
+ attr->max_qp_rd_atom =
+ min(1 << (fls(qattr->max_qp_resp_rd_atomic_resc) - 1),
+ attr->max_qp_init_rd_atom);
+
+ attr->max_srq = qattr->max_srq;
+ attr->max_srq_sge = qattr->max_srq_sge;
+ attr->max_srq_wr = qattr->max_srq_wr;
+
+ attr->local_ca_ack_delay = qattr->dev_ack_delay;
+ attr->max_fast_reg_page_list_len = qattr->max_mr / 8;
+ attr->max_pkeys = QEDR_ROCE_PKEY_MAX;
+ attr->max_ah = qattr->max_ah;
+
+ return 0;
+}
+
+#define QEDR_SPEED_SDR (1)
+#define QEDR_SPEED_DDR (2)
+#define QEDR_SPEED_QDR (4)
+#define QEDR_SPEED_FDR10 (8)
+#define QEDR_SPEED_FDR (16)
+#define QEDR_SPEED_EDR (32)
+
+static inline void get_link_speed_and_width(int speed, u8 *ib_speed,
+ u8 *ib_width)
+{
+ switch (speed) {
+ case 1000:
+ *ib_speed = QEDR_SPEED_SDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+ case 10000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+
+ case 20000:
+ *ib_speed = QEDR_SPEED_DDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 25000:
+ *ib_speed = QEDR_SPEED_EDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+
+ case 40000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 50000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 100000:
+ *ib_speed = QEDR_SPEED_EDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ default:
+ /* Unsupported */
+ *ib_speed = QEDR_SPEED_SDR;
+ *ib_width = IB_WIDTH_1X;
+ }
+}
+
+int qedr_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *attr)
+{
+ struct qedr_dev *dev;
+ struct qed_rdma_port *rdma_port;
+
+ dev = get_qedr_dev(ibdev);
+ if (port > 1) {
+ DP_ERR(dev, "invalid_port=0x%x\n", port);
+ return -EINVAL;
+ }
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev, "rdma_ctx is NULL\n");
+ return -EINVAL;
+ }
+
+ rdma_port = dev->ops->rdma_query_port(dev->rdma_ctx);
+ memset(attr, 0, sizeof(*attr));
+
+ if (rdma_port->port_state == QED_RDMA_PORT_UP) {
+ attr->state = IB_PORT_ACTIVE;
+ attr->phys_state = 5;
+ } else {
+ attr->state = IB_PORT_DOWN;
+ attr->phys_state = 3;
+ }
+ attr->max_mtu = IB_MTU_4096;
+ attr->active_mtu = iboe_get_mtu(dev->ndev->mtu);
+ attr->lid = 0;
+ attr->lmc = 0;
+ attr->sm_lid = 0;
+ attr->sm_sl = 0;
+ attr->port_cap_flags = IB_PORT_IP_BASED_GIDS;
+ attr->gid_tbl_len = QEDR_MAX_SGID;
+ attr->pkey_tbl_len = QEDR_ROCE_PKEY_TABLE_LEN;
+ attr->bad_pkey_cntr = rdma_port->pkey_bad_counter;
+ attr->qkey_viol_cntr = 0;
+ get_link_speed_and_width(rdma_port->link_speed,
+ &attr->active_speed, &attr->active_width);
+ attr->max_msg_sz = rdma_port->max_msg_size;
+ attr->max_vl_num = 4;
+
+ return 0;
+}
+
+int qedr_modify_port(struct ib_device *ibdev, u8 port, int mask,
+ struct ib_port_modify *props)
+{
+ struct qedr_dev *dev;
+
+ dev = get_qedr_dev(ibdev);
+ if (port > 1) {
+ DP_ERR(dev, "invalid_port=0x%x\n", port);
+ return -EINVAL;
+ }
+ return 0;
+}
+
+static int qedr_add_mmap(struct qedr_ucontext *uctx, u64 phy_addr,
+ unsigned long len)
+{
+ struct qedr_mm *mm;
+
+ mm = kzalloc(sizeof(*mm), GFP_KERNEL);
+ if (!mm)
+ return -ENOMEM;
+ mm->key.phy_addr = phy_addr;
+ /* This function might be called with a length which is not a multiple
+ * of PAGE_SIZE, while the mapping is PAGE_SIZE grained and the kernel
+ * forces this granularity by increasing the requested size if needed.
+ * When qedr_mmap is called, it will search the list with the updated
+ * length as a key. To prevent search failures, the length is rounded up
+ * in advance to PAGE_SIZE.
+ */
+ mm->key.len = roundup(len, PAGE_SIZE);
+ INIT_LIST_HEAD(&mm->entry);
+
+ mutex_lock(&uctx->mm_list_lock);
+ list_add(&mm->entry, &uctx->mm_head);
+ mutex_unlock(&uctx->mm_list_lock);
+
+ DP_DEBUG(uctx->dev, QEDR_MSG_MISC,
+ "added (addr=0x%llx,len=0x%lx) for ctx=%p\n",
+ (unsigned long long)mm->key.phy_addr,
+ (unsigned long)mm->key.len, uctx);
+
+ return 0;
+}
+
+static bool qedr_search_mmap(struct qedr_ucontext *uctx, u64 phy_addr,
+ unsigned long len)
+{
+ bool found = false;
+ struct qedr_mm *mm;
+
+ mutex_lock(&uctx->mm_list_lock);
+ list_for_each_entry(mm, &uctx->mm_head, entry) {
+ if (len != mm->key.len || phy_addr != mm->key.phy_addr)
+ continue;
+
+ found = true;
+ break;
+ }
+ mutex_unlock(&uctx->mm_list_lock);
+ DP_DEBUG(uctx->dev, QEDR_MSG_MISC,
+ "searched for (addr=0x%llx,len=0x%lx) for ctx=%p, result=%d\n",
+ mm->key.phy_addr, mm->key.len, uctx, found);
+
+ return found;
+}
+
+struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev,
+ struct ib_udata *udata)
+{
+ int rc;
+ struct qedr_ucontext *ctx;
+ struct qedr_alloc_ucontext_resp uresp;
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qed_rdma_add_user_out_params oparams;
+
+ if (!udata)
+ return ERR_PTR(-EFAULT);
+
+ ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+ if (!ctx)
+ return ERR_PTR(-ENOMEM);
+
+ rc = dev->ops->rdma_add_user(dev->rdma_ctx, &oparams);
+ if (rc) {
+ DP_ERR(dev,
+ "failed to allocate a DPI for a new RoCE application, rc=%d. To overcome this consider to increase the number of DPIs, increase the doorbell BAR size or just close unnecessary RoCE applications. In order to increase the number of DPIs consult the qedr readme\n",
+ rc);
+ goto err;
+ }
+
+ ctx->dpi = oparams.dpi;
+ ctx->dpi_addr = oparams.dpi_addr;
+ ctx->dpi_phys_addr = oparams.dpi_phys_addr;
+ ctx->dpi_size = oparams.dpi_size;
+ INIT_LIST_HEAD(&ctx->mm_head);
+ mutex_init(&ctx->mm_list_lock);
+
+ memset(&uresp, 0, sizeof(uresp));
+
+ uresp.db_pa = ctx->dpi_phys_addr;
+ uresp.db_size = ctx->dpi_size;
+ uresp.max_send_wr = dev->attr.max_sqe;
+ uresp.max_recv_wr = dev->attr.max_rqe;
+ uresp.max_srq_wr = dev->attr.max_srq_wr;
+ uresp.sges_per_send_wr = QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ uresp.sges_per_recv_wr = QEDR_MAX_RQE_ELEMENTS_PER_RQE;
+ uresp.sges_per_srq_wr = dev->attr.max_srq_sge;
+ uresp.max_cqes = QEDR_MAX_CQES;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ goto err;
+
+ ctx->dev = dev;
+
+ rc = qedr_add_mmap(ctx, ctx->dpi_phys_addr, ctx->dpi_size);
+ if (rc)
+ goto err;
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "Allocating user context %p\n",
+ &ctx->ibucontext);
+ return &ctx->ibucontext;
+
+err:
+ kfree(ctx);
+ return ERR_PTR(rc);
+}
+
+int qedr_dealloc_ucontext(struct ib_ucontext *ibctx)
+{
+ struct qedr_ucontext *uctx = get_qedr_ucontext(ibctx);
+ struct qedr_mm *mm, *tmp;
+ int status = 0;
+
+ DP_DEBUG(uctx->dev, QEDR_MSG_INIT, "Deallocating user context %p\n",
+ uctx);
+ uctx->dev->ops->rdma_remove_user(uctx->dev->rdma_ctx, uctx->dpi);
+
+ list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) {
+ DP_DEBUG(uctx->dev, QEDR_MSG_MISC,
+ "deleted (addr=0x%llx,len=0x%lx) for ctx=%p\n",
+ mm->key.phy_addr, mm->key.len, uctx);
+ list_del(&mm->entry);
+ kfree(mm);
+ }
+
+ kfree(uctx);
+ return status;
+}
+
+int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+ struct qedr_ucontext *ucontext = get_qedr_ucontext(context);
+ struct qedr_dev *dev = get_qedr_dev(context->device);
+ unsigned long vm_page = vma->vm_pgoff << PAGE_SHIFT;
+ u64 unmapped_db = dev->db_phys_addr;
+ unsigned long len = (vma->vm_end - vma->vm_start);
+ int rc = 0;
+ bool found;
+
+ DP_DEBUG(dev, QEDR_MSG_INIT,
+ "qedr_mmap called vm_page=0x%lx vm_pgoff=0x%lx unmapped_db=0x%llx db_size=%x, len=%lx\n",
+ vm_page, vma->vm_pgoff, unmapped_db, dev->db_size, len);
+ if (vma->vm_start & (PAGE_SIZE - 1)) {
+ DP_ERR(dev, "Vma_start not page aligned = %ld\n",
+ vma->vm_start);
+ return -EINVAL;
+ }
+
+ found = qedr_search_mmap(ucontext, vm_page, len);
+ if (!found) {
+ DP_ERR(dev, "Vma_pgoff not found in mapped array = %ld\n",
+ vma->vm_pgoff);
+ return -EINVAL;
+ }
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n");
+
+ if ((vm_page >= unmapped_db) && (vm_page <= (unmapped_db +
+ dev->db_size))) {
+ DP_DEBUG(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n");
+ if (vma->vm_flags & VM_READ) {
+ DP_ERR(dev, "Trying to map doorbell bar for read\n");
+ return -EPERM;
+ }
+
+ vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
+ rc = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
+ PAGE_SIZE, vma->vm_page_prot);
+ } else {
+ DP_DEBUG(dev, QEDR_MSG_INIT, "Mapping chains\n");
+ rc = remap_pfn_range(vma, vma->vm_start,
+ vma->vm_pgoff, len, vma->vm_page_prot);
+ }
+ DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_mmap return code: %d\n", rc);
+ return rc;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
new file mode 100644
index 0000000..b36678f
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -0,0 +1,52 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QEDR_VERBS_H__
+#define __QEDR_VERBS_H__
+
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata);
+int qedr_query_port(struct ib_device *, u8 port, struct ib_port_attr *props);
+int qedr_modify_port(struct ib_device *, u8 port, int mask,
+ struct ib_port_modify *props);
+
+int qedr_query_gid(struct ib_device *, u8 port, int index, union ib_gid *gid);
+
+struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *);
+int qedr_dealloc_ucontext(struct ib_ucontext *);
+
+int qedr_mmap(struct ib_ucontext *, struct vm_area_struct *vma);
+int qedr_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context);
+int qedr_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context);
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index 74052f5..a92aa80 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -53,7 +53,7 @@ enum qed_mcp_protocol_type;
static inline u32 qed_db_addr(u32 cid, u32 DEMS)
{
u32 db_addr = FIELD_VALUE(DB_LEGACY_ADDR_DEMS, DEMS) |
- FIELD_VALUE(DB_LEGACY_ADDR_ICID, cid);
+ (cid * QED_PF_DEMS_SIZE);
return db_addr;
}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index eb6fbfc..3708bdd 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -1480,6 +1480,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
resc_num[QED_RDMA_CNQ_RAM] = NUM_OF_CMDQS_CQS / num_funcs;
+ resc_num[QED_RDMA_STATS_QUEUE] = RDMA_NUM_STATISTIC_COUNTERS_BB / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index e65a130..9cdb472 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -158,7 +158,7 @@ static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
return rc;
}
- p_rdma_info = p_rdma_info;
+ p_hwfn->p_rdma_info = p_rdma_info;
p_rdma_info->proto = PROTOCOLID_ROCE;
num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
@@ -664,6 +664,22 @@ int qed_rdma_add_user(void *rdma_cxt,
return rc;
}
+struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_port *p_port = p_hwfn->p_rdma_info->port;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA Query port\n");
+
+ /* Link may have changed */
+ p_port->port_state = p_hwfn->mcp_info->link_output.link_up ?
+ QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
+
+ p_port->link_speed = p_hwfn->mcp_info->link_output.speed;
+
+ return p_port;
+}
+
struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
{
struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
@@ -872,6 +888,7 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_add_user = &qed_rdma_add_user,
.rdma_remove_user = &qed_rdma_remove_user,
.rdma_stop = &qed_rdma_stop,
+ .rdma_query_port = &qed_rdma_query_port,
.rdma_query_device = &qed_rdma_query_device,
.rdma_get_start_sb = &qed_rdma_get_sb_start,
.rdma_get_rdma_int = &qed_rdma_get_int,
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 46f14b2..c94c39e 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -102,6 +102,7 @@ int qed_rdma_alloc_tid(void *rdma_cxt, u32 *tid);
int qed_rdma_deregister_tid(void *rdma_cxt, u32 tid);
void qed_rdma_free_tid(void *rdma_cxt, u32 tid);
struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt);
+struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt);
int
qed_rdma_register_tid(void *rdma_cxt,
struct qed_rdma_register_tid_in_params *params);
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 0f7d527..e99470c 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -332,6 +332,7 @@ struct qed_rdma_ops {
void (*rdma_remove_user)(void *rdma_cxt, u16 dpi);
int (*rdma_stop)(void *rdma_cxt);
struct qed_rdma_device* (*rdma_query_device)(void *rdma_cxt);
+ struct qed_rdma_port* (*rdma_query_port)(void *rdma_cxt);
int (*rdma_get_start_sb)(struct qed_dev *cdev);
int (*rdma_get_min_cnq_msix)(struct qed_dev *cdev);
void (*rdma_cnq_prod_update)(void *rdma_cxt, u8 cnq_index, u16 prod);
diff --git a/include/linux/qed/rdma_common.h b/include/linux/qed/rdma_common.h
index 187991c..7663725 100644
--- a/include/linux/qed/rdma_common.h
+++ b/include/linux/qed/rdma_common.h
@@ -28,6 +28,7 @@
#define RDMA_MAX_PDS (64 * 1024)
#define RDMA_NUM_STATISTIC_COUNTERS MAX_NUM_VPORTS
+#define RDMA_NUM_STATISTIC_COUNTERS_BB MAX_NUM_VPORTS_BB
#define RDMA_TASK_TYPE (PROTOCOLID_ROCE)
diff --git a/include/uapi/rdma/qedr-abi.h b/include/uapi/rdma/qedr-abi.h
new file mode 100644
index 0000000..f7c7fff
--- /dev/null
+++ b/include/uapi/rdma/qedr-abi.h
@@ -0,0 +1,53 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QEDR_USER_H__
+#define __QEDR_USER_H__
+
+#include <linux/types.h>
+
+#define QEDR_ABI_VERSION (8)
+
+/* user kernel communication data structures. */
+
+struct qedr_alloc_ucontext_resp {
+ __u64 db_pa;
+ __u32 db_size;
+
+ __u32 max_send_wr;
+ __u32 max_recv_wr;
+ __u32 max_srq_wr;
+ __u32 sges_per_send_wr;
+ __u32 sges_per_recv_wr;
+ __u32 sges_per_srq_wr;
+ __u32 max_cqes;
+};
+#endif /* __QEDR_USER_H__ */
--
1.8.3.1
^ permalink raw reply related
* [PATCH 03/11] qedr: Add support for RoCE HW init
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem, dledford
Cc: Ariel.Elior, Michal.Kalderon, Yuval.Mintz, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani@cavium.com>
Allocate and setup RoCE resources, interrupts and completion queues.
Adds device attributes.
Signed-off-by: Rajesh Borundia <rajesh.borundia@cavium.com>
Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
---
drivers/infiniband/hw/qedr/main.c | 418 +++++++++++-
drivers/infiniband/hw/qedr/qedr.h | 122 ++++
drivers/infiniband/hw/qedr/qedr_hsi.h | 56 ++
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 96 +++
drivers/net/ethernet/qlogic/qed/Makefile | 1 +
drivers/net/ethernet/qlogic/qed/qed.h | 26 +-
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 6 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 6 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 155 +++++
drivers/net/ethernet/qlogic/qed/qed_main.c | 44 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 8 +-
drivers/net/ethernet/qlogic/qed/qed_roce.c | 887 +++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 117 ++++
drivers/net/ethernet/qlogic/qed/qed_sp.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
include/linux/qed/qed_if.h | 5 +-
include/linux/qed/qed_roce_if.h | 345 ++++++++++
17 files changed, 2294 insertions(+), 7 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
create mode 100644 include/linux/qed/qed_roce_if.h
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 7387d02..0b61c8d 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -36,6 +36,8 @@
#include <linux/iommu.h>
#include <net/addrconf.h>
#include <linux/qed/qede_roce.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_if.h>
#include "qedr.h"
MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
@@ -61,6 +63,17 @@ static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
return IB_LINK_LAYER_ETHERNET;
}
+static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str,
+ size_t str_len)
+{
+ struct qedr_dev *qedr = get_qedr_dev(ibdev);
+ u32 fw_ver = (u32)qedr->attr.fw_ver;
+
+ snprintf(str, str_len, "%d. %d. %d. %d",
+ (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
+ (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
+}
+
static int qedr_register_device(struct qedr_dev *dev)
{
strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
@@ -69,10 +82,141 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.owner = THIS_MODULE;
dev->ibdev.get_link_layer = qedr_link_layer;
+ dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
return 0;
}
+/* This function allocates fast-path status block memory */
+static int qedr_alloc_mem_sb(struct qedr_dev *dev,
+ struct qed_sb_info *sb_info, u16 sb_id)
+{
+ struct status_block *sb_virt;
+ dma_addr_t sb_phys;
+ int rc;
+
+ sb_virt = dma_alloc_coherent(&dev->pdev->dev,
+ sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
+ if (!sb_virt) {
+ pr_err("Status block allocation failed\n");
+ return -ENOMEM;
+ }
+
+ rc = dev->ops->common->sb_init(dev->cdev, sb_info,
+ sb_virt, sb_phys, sb_id,
+ QED_SB_TYPE_CNQ);
+ if (rc) {
+ pr_err("Status block initialization failed\n");
+ dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
+ sb_virt, sb_phys);
+ return rc;
+ }
+
+ return 0;
+}
+
+static void qedr_free_mem_sb(struct qedr_dev *dev,
+ struct qed_sb_info *sb_info, int sb_id)
+{
+ if (sb_info->sb_virt) {
+ dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
+ dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
+ (void *)sb_info->sb_virt, sb_info->sb_phys);
+ }
+}
+
+static void qedr_free_resources(struct qedr_dev *dev)
+{
+ int i;
+
+ for (i = 0; i < dev->num_cnq; i++) {
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
+ }
+
+ kfree(dev->cnq_array);
+ kfree(dev->sb_array);
+ kfree(dev->sgid_tbl);
+}
+
+static int qedr_alloc_resources(struct qedr_dev *dev)
+{
+ struct qedr_cnq *cnq;
+ __le16 *cons_pi;
+ u16 n_entries;
+ int i, rc;
+
+ dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
+ QEDR_MAX_SGID, GFP_KERNEL);
+ if (!dev->sgid_tbl)
+ return -ENOMEM;
+
+ spin_lock_init(&dev->sgid_lock);
+
+ /* Allocate Status blocks for CNQ */
+ dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
+ GFP_KERNEL);
+ if (!dev->sb_array) {
+ rc = -ENOMEM;
+ goto err1;
+ }
+
+ dev->cnq_array = kcalloc(dev->num_cnq,
+ sizeof(*dev->cnq_array), GFP_KERNEL);
+ if (!dev->cnq_array) {
+ rc = -ENOMEM;
+ goto err2;
+ }
+
+ dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
+
+ /* Allocate CNQ PBLs */
+ n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
+ for (i = 0; i < dev->num_cnq; i++) {
+ cnq = &dev->cnq_array[i];
+
+ rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
+ dev->sb_start + i);
+ if (rc)
+ goto err3;
+
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ n_entries,
+ sizeof(struct regpair *),
+ &cnq->pbl);
+ if (rc)
+ goto err4;
+
+ cnq->dev = dev;
+ cnq->sb = &dev->sb_array[i];
+ cons_pi = dev->sb_array[i].sb_virt->pi_array;
+ cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
+ cnq->index = i;
+ sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
+ i, qed_chain_get_cons_idx(&cnq->pbl));
+ }
+
+ return 0;
+err4:
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+err3:
+ for (--i; i >= 0; i--) {
+ dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ }
+ kfree(dev->cnq_array);
+err2:
+ kfree(dev->sb_array);
+err1:
+ kfree(dev->sgid_tbl);
+ return rc;
+}
+
/* QEDR sysfs interface */
static ssize_t show_rev(struct device *device, struct device_attribute *attr,
char *buf)
@@ -146,9 +290,237 @@ static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
}
}
+static const struct qed_rdma_ops *qed_ops;
+
+#define HILO_U64(hi, lo) ((((u64)(hi)) << 32) + (lo))
+
+static irqreturn_t qedr_irq_handler(int irq, void *handle)
+{
+ u16 hw_comp_cons, sw_comp_cons;
+ struct qedr_cnq *cnq = handle;
+
+ qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
+
+ qed_sb_update_sb_idx(cnq->sb);
+
+ hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
+ sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
+
+ /* Align protocol-index and chain reads */
+ rmb();
+
+ while (sw_comp_cons != hw_comp_cons) {
+ sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
+ cnq->n_comp++;
+ }
+
+ qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
+ sw_comp_cons);
+
+ qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
+
+ return IRQ_HANDLED;
+}
+
+static void qedr_sync_free_irqs(struct qedr_dev *dev)
+{
+ u32 vector;
+ int i;
+
+ for (i = 0; i < dev->int_info.used_cnt; i++) {
+ if (dev->int_info.msix_cnt) {
+ vector = dev->int_info.msix[i * dev->num_hwfns].vector;
+ synchronize_irq(vector);
+ free_irq(vector, &dev->cnq_array[i]);
+ }
+ }
+
+ dev->int_info.used_cnt = 0;
+}
+
+static int qedr_req_msix_irqs(struct qedr_dev *dev)
+{
+ int i, rc = 0;
+
+ if (dev->num_cnq > dev->int_info.msix_cnt) {
+ DP_ERR(dev,
+ "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
+ dev->num_cnq, dev->int_info.msix_cnt);
+ return -EINVAL;
+ }
+
+ for (i = 0; i < dev->num_cnq; i++) {
+ rc = request_irq(dev->int_info.msix[i * dev->num_hwfns].vector,
+ qedr_irq_handler, 0, dev->cnq_array[i].name,
+ &dev->cnq_array[i]);
+ if (rc) {
+ DP_ERR(dev, "Request cnq %d irq failed\n", i);
+ qedr_sync_free_irqs(dev);
+ } else {
+ DP_DEBUG(dev, QEDR_MSG_INIT,
+ "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
+ dev->cnq_array[i].name, i,
+ &dev->cnq_array[i]);
+ dev->int_info.used_cnt++;
+ }
+ }
+
+ return rc;
+}
+
+static int qedr_setup_irqs(struct qedr_dev *dev)
+{
+ int rc;
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
+
+ /* Learn Interrupt configuration */
+ rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
+ if (rc < 0)
+ return rc;
+ rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
+ if (rc) {
+ DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
+ return rc;
+ }
+
+ if (dev->int_info.msix_cnt) {
+ DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
+ dev->int_info.msix_cnt);
+ rc = qedr_req_msix_irqs(dev);
+ if (rc)
+ return rc;
+ }
+ DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
+ return 0;
+}
+
+static int qedr_set_device_attr(struct qedr_dev *dev)
+{
+ struct qed_rdma_device *qed_attr;
+ struct qedr_device_attr *attr;
+ u32 page_size;
+
+ /* Part 1 - query core capabilities */
+ qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
+
+ /* Part 2 - check capabilities */
+ page_size = ~dev->attr.page_size_caps + 1;
+ if (page_size > PAGE_SIZE) {
+ DP_ERR(dev,
+ "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
+ PAGE_SIZE, page_size);
+ return -ENODEV;
+ }
+
+ /* Part 3 - copy and update capabilities */
+ attr = &dev->attr;
+ attr->vendor_id = qed_attr->vendor_id;
+ attr->vendor_part_id = qed_attr->vendor_part_id;
+ attr->hw_ver = qed_attr->hw_ver;
+ attr->fw_ver = qed_attr->fw_ver;
+ attr->node_guid = qed_attr->node_guid;
+ attr->sys_image_guid = qed_attr->sys_image_guid;
+ attr->max_cnq = qed_attr->max_cnq;
+ attr->max_sge = qed_attr->max_sge;
+ attr->max_inline = qed_attr->max_inline;
+ attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
+ attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
+ attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
+ attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
+ attr->max_dev_resp_rd_atomic_resc =
+ qed_attr->max_dev_resp_rd_atomic_resc;
+ attr->max_cq = qed_attr->max_cq;
+ attr->max_qp = qed_attr->max_qp;
+ attr->max_mr = qed_attr->max_mr;
+ attr->max_mr_size = qed_attr->max_mr_size;
+ attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
+ attr->max_mw = qed_attr->max_mw;
+ attr->max_fmr = qed_attr->max_fmr;
+ attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
+ attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
+ attr->max_pd = qed_attr->max_pd;
+ attr->max_ah = qed_attr->max_ah;
+ attr->max_pkey = qed_attr->max_pkey;
+ attr->max_srq = qed_attr->max_srq;
+ attr->max_srq_wr = qed_attr->max_srq_wr;
+ attr->dev_caps = qed_attr->dev_caps;
+ attr->page_size_caps = qed_attr->page_size_caps;
+ attr->dev_ack_delay = qed_attr->dev_ack_delay;
+ attr->reserved_lkey = qed_attr->reserved_lkey;
+ attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
+ attr->max_stats_queues = qed_attr->max_stats_queues;
+
+ return 0;
+}
+
+static int qedr_init_hw(struct qedr_dev *dev)
+{
+ struct qed_rdma_add_user_out_params out_params;
+ struct qed_rdma_start_in_params *in_params;
+ struct qed_rdma_cnq_params *cur_pbl;
+ struct qed_rdma_events events;
+ dma_addr_t p_phys_table;
+ u32 page_cnt;
+ int rc = 0;
+ int i;
+
+ in_params = kzalloc(sizeof(*in_params), GFP_KERNEL);
+ if (!in_params) {
+ rc = -ENOMEM;
+ goto out;
+ }
+
+ in_params->desired_cnq = dev->num_cnq;
+ for (i = 0; i < dev->num_cnq; i++) {
+ cur_pbl = &in_params->cnq_pbl_list[i];
+
+ page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
+ cur_pbl->num_pbl_pages = page_cnt;
+
+ p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
+ cur_pbl->pbl_ptr = (u64)p_phys_table;
+ }
+
+ events.context = dev;
+
+ in_params->events = &events;
+ in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
+ in_params->max_mtu = dev->ndev->mtu;
+ ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
+
+ rc = dev->ops->rdma_init(dev->cdev, in_params);
+ if (rc)
+ goto out;
+
+ rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
+ if (rc)
+ goto out;
+
+ dev->db_addr = (void *)(uintptr_t)out_params.dpi_addr;
+ dev->db_phys_addr = out_params.dpi_phys_addr;
+ dev->db_size = out_params.dpi_size;
+ dev->dpi = out_params.dpi;
+
+ rc = qedr_set_device_attr(dev);
+out:
+ kfree(in_params);
+ if (rc)
+ DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
+
+ return rc;
+}
+
+void qedr_stop_hw(struct qedr_dev *dev)
+{
+ dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
+ dev->ops->rdma_stop(dev->rdma_ctx);
+}
+
static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
struct net_device *ndev)
{
+ struct qed_dev_rdma_info dev_info;
struct qedr_dev *dev;
int rc = 0, i;
@@ -164,21 +536,60 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
dev->ndev = ndev;
dev->cdev = cdev;
+ qed_ops = qed_get_rdma_ops();
+ if (!qed_ops) {
+ DP_ERR(dev, "Failed to get qed roce operations\n");
+ goto init_err;
+ }
+
+ dev->ops = qed_ops;
+ rc = qed_ops->fill_dev_info(cdev, &dev_info);
+ if (rc)
+ goto init_err;
+
+ dev->num_hwfns = dev_info.common.num_hwfns;
+ dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
+
+ dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
+ if (!dev->num_cnq) {
+ DP_ERR(dev, "not enough CNQ resources.\n");
+ goto init_err;
+ }
+
+
qedr_pci_set_atomic(dev, pdev);
+ rc = qedr_alloc_resources(dev);
+ if (rc)
+ goto init_err;
+
+ rc = qedr_init_hw(dev);
+ if (rc)
+ goto alloc_err;
+
+ rc = qedr_setup_irqs(dev);
+ if (rc)
+ goto irq_err;
+
rc = qedr_register_device(dev);
if (rc) {
DP_ERR(dev, "Unable to allocate register device\n");
- goto init_err;
+ goto reg_err;
}
for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
- goto init_err;
+ goto reg_err;
DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
return dev;
+reg_err:
+ qedr_sync_free_irqs(dev);
+irq_err:
+ qedr_stop_hw(dev);
+alloc_err:
+ qedr_free_resources(dev);
init_err:
ib_dealloc_device(&dev->ibdev);
DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
@@ -193,6 +604,9 @@ static void qedr_remove(struct qedr_dev *dev)
*/
qedr_remove_sysfiles(dev);
+ qedr_stop_hw(dev);
+ qedr_sync_free_irqs(dev);
+ qedr_free_resources(dev);
ib_dealloc_device(&dev->ibdev);
}
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 5f85903..fbb9202 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -35,7 +35,10 @@
#include <linux/pci.h>
#include <rdma/ib_addr.h>
#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_roce_if.h>
#include <linux/qed/qede_roce.h>
+#include "qedr_hsi.h"
#define QEDR_MODULE_VERSION "8.10.10.0"
#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
@@ -49,6 +52,59 @@ do { \
#define QEDR_MSG_INIT "INIT"
+struct qedr_dev;
+
+struct qedr_cnq {
+ struct qedr_dev *dev;
+ struct qed_chain pbl;
+ struct qed_sb_info *sb;
+ char name[32];
+ u64 n_comp;
+ __le16 *hw_cons_ptr;
+ u8 index;
+};
+
+#define QEDR_MAX_SGID 128
+
+struct qedr_device_attr {
+ u32 vendor_id;
+ u32 vendor_part_id;
+ u32 hw_ver;
+ u64 fw_ver;
+ u64 node_guid;
+ u64 sys_image_guid;
+ u8 max_cnq;
+ u8 max_sge;
+ u16 max_inline;
+ u32 max_sqe;
+ u32 max_rqe;
+ u8 max_qp_resp_rd_atomic_resc;
+ u8 max_qp_req_rd_atomic_resc;
+ u64 max_dev_resp_rd_atomic_resc;
+ u32 max_cq;
+ u32 max_qp;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_cqe;
+ u32 max_mw;
+ u32 max_fmr;
+ u32 max_mr_mw_fmr_pbl;
+ u64 max_mr_mw_fmr_size;
+ u32 max_pd;
+ u32 max_ah;
+ u8 max_pkey;
+ u32 max_srq;
+ u32 max_srq_wr;
+ u8 max_srq_sge;
+ u8 max_stats_queues;
+ u32 dev_caps;
+
+ u64 page_size_caps;
+ u8 dev_ack_delay;
+ u32 reserved_lkey;
+ u32 bad_pkey_counter;
+ struct qed_rdma_events events;
+};
struct qedr_dev {
struct ib_device ibdev;
@@ -58,7 +114,73 @@ struct qedr_dev {
enum ib_atomic_cap atomic_cap;
+ void *rdma_ctx;
+ struct qedr_device_attr attr;
+
+ const struct qed_rdma_ops *ops;
+ struct qed_int_info int_info;
+
+ struct qed_sb_info *sb_array;
+ struct qedr_cnq *cnq_array;
+ int num_cnq;
+ int sb_start;
+
+ void __iomem *db_addr;
+ u64 db_phys_addr;
+ u32 db_size;
+ u16 dpi;
+
+ union ib_gid *sgid_tbl;
+
+ /* Lock for sgid table */
+ spinlock_t sgid_lock;
+
+ u64 guid;
+
u32 dp_module;
u8 dp_level;
+ u8 num_hwfns;
};
+
+#define QEDR_MAX_SQ_PBL (0x8000)
+#define QEDR_MAX_SQ_PBL_ENTRIES (0x10000 / sizeof(void *))
+#define QEDR_SQE_ELEMENT_SIZE (sizeof(struct rdma_sq_sge))
+#define QEDR_MAX_SQE_ELEMENTS_PER_SQE (ROCE_REQ_MAX_SINGLE_SQ_WQE_SIZE / \
+ QEDR_SQE_ELEMENT_SIZE)
+#define QEDR_MAX_SQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \
+ QEDR_SQE_ELEMENT_SIZE)
+#define QEDR_MAX_SQE ((QEDR_MAX_SQ_PBL_ENTRIES) *\
+ (RDMA_RING_PAGE_SIZE) / \
+ (QEDR_SQE_ELEMENT_SIZE) /\
+ (QEDR_MAX_SQE_ELEMENTS_PER_SQE))
+/* RQ */
+#define QEDR_MAX_RQ_PBL (0x2000)
+#define QEDR_MAX_RQ_PBL_ENTRIES (0x10000 / sizeof(void *))
+#define QEDR_RQE_ELEMENT_SIZE (sizeof(struct rdma_rq_sge))
+#define QEDR_MAX_RQE_ELEMENTS_PER_RQE (RDMA_MAX_SGE_PER_RQ_WQE)
+#define QEDR_MAX_RQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \
+ QEDR_RQE_ELEMENT_SIZE)
+#define QEDR_MAX_RQE ((QEDR_MAX_RQ_PBL_ENTRIES) *\
+ (RDMA_RING_PAGE_SIZE) / \
+ (QEDR_RQE_ELEMENT_SIZE) /\
+ (QEDR_MAX_RQE_ELEMENTS_PER_RQE))
+
+#define QEDR_CQE_SIZE (sizeof(union rdma_cqe))
+#define QEDR_MAX_CQE_PBL_SIZE (512 * 1024)
+#define QEDR_MAX_CQE_PBL_ENTRIES (((QEDR_MAX_CQE_PBL_SIZE) / \
+ sizeof(u64)) - 1)
+#define QEDR_MAX_CQES ((u32)((QEDR_MAX_CQE_PBL_ENTRIES) * \
+ (QED_CHAIN_PAGE_SIZE) / QEDR_CQE_SIZE))
+
+#define QEDR_ROCE_MAX_CNQ_SIZE (0x4000)
+
+#define QEDR_MAX_PORT (1)
+
+#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+
+static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev)
+{
+ return container_of(ibdev, struct qedr_dev, ibdev);
+}
+
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi.h b/drivers/infiniband/hw/qedr/qedr_hsi.h
new file mode 100644
index 0000000..66d2752
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_hsi.h
@@ -0,0 +1,56 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QED_HSI_ROCE__
+#define __QED_HSI_ROCE__
+
+#include <linux/qed/common_hsi.h>
+#include <linux/qed/roce_common.h>
+#include "qedr_hsi_rdma.h"
+
+/* Affiliated asynchronous events / errors enumeration */
+enum roce_async_events_type {
+ ROCE_ASYNC_EVENT_NONE = 0,
+ ROCE_ASYNC_EVENT_COMM_EST = 1,
+ ROCE_ASYNC_EVENT_SQ_DRAINED,
+ ROCE_ASYNC_EVENT_SRQ_LIMIT,
+ ROCE_ASYNC_EVENT_LAST_WQE_REACHED,
+ ROCE_ASYNC_EVENT_CQ_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_CATASTROPHIC_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR,
+ ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR,
+ ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR,
+ ROCE_ASYNC_EVENT_SRQ_EMPTY,
+ MAX_ROCE_ASYNC_EVENTS_TYPE
+};
+
+#endif /* __QED_HSI_ROCE__ */
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
new file mode 100644
index 0000000..3e508fb
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -0,0 +1,96 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QED_HSI_RDMA__
+#define __QED_HSI_RDMA__
+
+#include <linux/qed/rdma_common.h>
+
+/* rdma completion notification queue element */
+struct rdma_cnqe {
+ struct regpair cq_handle;
+};
+
+struct rdma_cqe_responder {
+ struct regpair srq_wr_id;
+ struct regpair qp_handle;
+ __le32 imm_data_or_inv_r_Key;
+ __le32 length;
+ __le32 imm_data_hi;
+ __le16 rq_cons;
+ u8 flags;
+};
+
+struct rdma_cqe_requester {
+ __le16 sq_cons;
+ __le16 reserved0;
+ __le32 reserved1;
+ struct regpair qp_handle;
+ struct regpair reserved2;
+ __le32 reserved3;
+ __le16 reserved4;
+ u8 flags;
+ u8 status;
+};
+
+struct rdma_cqe_common {
+ struct regpair reserved0;
+ struct regpair qp_handle;
+ __le16 reserved1[7];
+ u8 flags;
+ u8 status;
+};
+
+/* rdma completion queue element */
+union rdma_cqe {
+ struct rdma_cqe_responder resp;
+ struct rdma_cqe_requester req;
+ struct rdma_cqe_common cmn;
+};
+
+struct rdma_sq_sge {
+ __le32 length;
+ struct regpair addr;
+ __le32 l_key;
+};
+
+struct rdma_rq_sge {
+ struct regpair addr;
+ __le32 length;
+ __le32 flags;
+};
+
+struct rdma_srq_sge {
+ struct regpair addr;
+ __le32 length;
+ __le32 l_key;
+};
+#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/net/ethernet/qlogic/qed/Makefile b/drivers/net/ethernet/qlogic/qed/Makefile
index e067098..cda0af7 100644
--- a/drivers/net/ethernet/qlogic/qed/Makefile
+++ b/drivers/net/ethernet/qlogic/qed/Makefile
@@ -5,3 +5,4 @@ qed-y := qed_cxt.o qed_dev.o qed_hw.o qed_init_fw_funcs.o qed_init_ops.o \
qed_selftest.o qed_dcbx.o qed_debug.o
qed-$(CONFIG_QED_SRIOV) += qed_sriov.o qed_vf.o
qed-$(CONFIG_QED_LL2) += qed_ll2.o
+qed-$(CONFIG_INFINIBAND_QEDR) += qed_roce.o
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index 91b571a..74052f5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -35,6 +35,9 @@ extern const struct qed_common_ops qed_common_ops_pass;
#define QED_WFQ_UNIT 100
+#define QED_WID_SIZE (1024)
+#define QED_PF_DEMS_SIZE (4)
+
/* cau states */
enum qed_coalescing_mode {
QED_COAL_MODE_DISABLE,
@@ -152,14 +155,17 @@ enum QED_RESOURCES {
QED_RL,
QED_MAC,
QED_VLAN,
+ QED_RDMA_CNQ_RAM,
QED_ILT,
QED_LL2_QUEUE,
+ QED_RDMA_STATS_QUEUE,
QED_MAX_RESC,
};
enum QED_FEATURE {
QED_PF_L2_QUE,
QED_VF,
+ QED_RDMA_CNQ,
QED_MAX_FEATURES,
};
@@ -364,6 +370,7 @@ struct qed_hwfn {
/* Protocol related */
bool using_ll2;
struct qed_ll2_info *p_ll2_info;
+ struct qed_rdma_info *p_rdma_info;
struct qed_pf_params pf_params;
bool b_rdma_enabled_in_prs;
@@ -402,6 +409,17 @@ struct qed_hwfn {
struct dbg_tools_data dbg_info;
+ /* PWM region specific data */
+ u32 dpi_size;
+ u32 dpi_count;
+
+ /* This is used to calculate the doorbell address */
+ u32 dpi_start_offset;
+
+ /* If one of the following is set then EDPM shouldn't be used */
+ u8 dcbx_no_edpm;
+ u8 db_bar_no_edpm;
+
struct qed_simd_fp_handler simd_proto_handler[64];
#ifdef CONFIG_QED_SRIOV
@@ -435,6 +453,8 @@ struct qed_int_params {
bool fp_initialized;
u8 fp_msix_base;
u8 fp_msix_cnt;
+ u8 rdma_msix_base;
+ u8 rdma_msix_cnt;
};
struct qed_dbg_feature {
@@ -541,7 +561,6 @@ struct qed_dev {
bool b_is_vf;
u32 drv_type;
-
struct qed_eth_stats *reset_stats;
struct qed_fw_data *fw_data;
@@ -574,6 +593,10 @@ struct qed_dev {
#endif
const struct firmware *firmware;
+
+ u32 rdma_max_sge;
+ u32 rdma_max_inline;
+ u32 rdma_max_srq_sge;
};
#define NUM_OF_VFS(dev) MAX_NUM_VFS_BB
@@ -641,5 +664,6 @@ void qed_get_protocol_stats(struct qed_dev *cdev,
enum qed_mcp_protocol_type type,
union qed_mcp_protocol_stats *stats);
int qed_slowpath_irq_req(struct qed_hwfn *hwfn);
+void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
#endif /* _QED_H */
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.c b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
index d9bea2a..82370a1 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
@@ -48,7 +48,13 @@
#define TM_ELEM_SIZE 4
/* ILT constants */
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+/* For RoCE we configure to 64K to cover for RoCE max tasks 256K purpose. */
+#define ILT_DEFAULT_HW_P_SIZE 4
+#else
#define ILT_DEFAULT_HW_P_SIZE 3
+#endif
+
#define ILT_PAGE_IN_BYTES(hw_p_size) (1U << ((hw_p_size) + 12))
#define ILT_CFG_REG(cli, reg) PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.h b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
index c6f6f2e..d00ad05 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
@@ -170,6 +170,12 @@ int qed_qm_reconf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
*/
void qed_cxt_release_cid(struct qed_hwfn *p_hwfn,
u32 cid);
+int qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
+ enum qed_cxt_elem_type elem_type, u32 iid);
+u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
+ enum protocol_type type);
+u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
+ enum protocol_type type);
#define QED_CTX_WORKING_MEM 0
#define QED_CTX_FL_MEM 1
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 9a8e153..eb6fbfc 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -38,6 +38,9 @@
static DEFINE_SPINLOCK(qm_lock);
+#define QED_MIN_DPIS (4)
+#define QED_MIN_PWM_REGION (QED_WID_SIZE * QED_MIN_DPIS)
+
/* API common to all protocols */
enum BAR_ID {
BAR_ID_0, /* used for GRC */
@@ -787,6 +790,140 @@ static int qed_hw_init_common(struct qed_hwfn *p_hwfn,
return rc;
}
+static int
+qed_hw_init_dpi_size(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u32 pwm_region_size, u32 n_cpus)
+{
+ u32 dpi_page_size_1, dpi_page_size_2, dpi_page_size;
+ u32 dpi_bit_shift, dpi_count;
+ u32 min_dpis;
+
+ /* Calculate DPI size */
+ dpi_page_size_1 = QED_WID_SIZE * n_cpus;
+ dpi_page_size_2 = max_t(u32, QED_WID_SIZE, PAGE_SIZE);
+ dpi_page_size = max_t(u32, dpi_page_size_1, dpi_page_size_2);
+ dpi_page_size = roundup_pow_of_two(dpi_page_size);
+ dpi_bit_shift = ilog2(dpi_page_size / 4096);
+
+ dpi_count = pwm_region_size / dpi_page_size;
+
+ min_dpis = p_hwfn->pf_params.rdma_pf_params.min_dpis;
+ min_dpis = max_t(u32, QED_MIN_DPIS, min_dpis);
+
+ /* Update hwfn */
+ p_hwfn->dpi_size = dpi_page_size;
+ p_hwfn->dpi_count = dpi_count;
+
+ /* Update registers */
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPI_BIT_SHIFT, dpi_bit_shift);
+
+ if (dpi_count < min_dpis)
+ return -EINVAL;
+
+ return 0;
+}
+
+enum QED_ROCE_EDPM_MODE {
+ QED_ROCE_EDPM_MODE_ENABLE = 0,
+ QED_ROCE_EDPM_MODE_FORCE_ON = 1,
+ QED_ROCE_EDPM_MODE_DISABLE = 2,
+};
+
+static int
+qed_hw_init_pf_doorbell_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 pwm_regsize, norm_regsize;
+ u32 non_pwm_conn, min_addr_reg1;
+ u32 db_bar_size, n_cpus;
+ u32 roce_edpm_mode;
+ u32 pf_dems_shift;
+ int rc = 0;
+ u8 cond;
+
+ db_bar_size = qed_hw_bar_size(p_hwfn, BAR_ID_1);
+ if (p_hwfn->cdev->num_hwfns > 1)
+ db_bar_size /= 2;
+
+ /* Calculate doorbell regions */
+ non_pwm_conn = qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_CORE) +
+ qed_cxt_get_proto_cid_count(p_hwfn, PROTOCOLID_CORE,
+ NULL) +
+ qed_cxt_get_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
+ NULL);
+ norm_regsize = roundup(QED_PF_DEMS_SIZE * non_pwm_conn, 4096);
+ min_addr_reg1 = norm_regsize / 4096;
+ pwm_regsize = db_bar_size - norm_regsize;
+
+ /* Check that the normal and PWM sizes are valid */
+ if (db_bar_size < norm_regsize) {
+ DP_ERR(p_hwfn->cdev,
+ "Doorbell BAR size 0x%x is too small (normal region is 0x%0x )\n",
+ db_bar_size, norm_regsize);
+ return -EINVAL;
+ }
+
+ if (pwm_regsize < QED_MIN_PWM_REGION) {
+ DP_ERR(p_hwfn->cdev,
+ "PWM region size 0x%0x is too small. Should be at least 0x%0x (Doorbell BAR size is 0x%x and normal region size is 0x%0x)\n",
+ pwm_regsize,
+ QED_MIN_PWM_REGION, db_bar_size, norm_regsize);
+ return -EINVAL;
+ }
+
+ /* Calculate number of DPIs */
+ roce_edpm_mode = p_hwfn->pf_params.rdma_pf_params.roce_edpm_mode;
+ if ((roce_edpm_mode == QED_ROCE_EDPM_MODE_ENABLE) ||
+ ((roce_edpm_mode == QED_ROCE_EDPM_MODE_FORCE_ON))) {
+ /* Either EDPM is mandatory, or we are attempting to allocate a
+ * WID per CPU.
+ */
+ n_cpus = num_active_cpus();
+ rc = qed_hw_init_dpi_size(p_hwfn, p_ptt, pwm_regsize, n_cpus);
+ }
+
+ cond = ((rc) && (roce_edpm_mode == QED_ROCE_EDPM_MODE_ENABLE)) ||
+ (roce_edpm_mode == QED_ROCE_EDPM_MODE_DISABLE);
+ if (cond || p_hwfn->dcbx_no_edpm) {
+ /* Either EDPM is disabled from user configuration, or it is
+ * disabled via DCBx, or it is not mandatory and we failed to
+ * allocated a WID per CPU.
+ */
+ n_cpus = 1;
+ rc = qed_hw_init_dpi_size(p_hwfn, p_ptt, pwm_regsize, n_cpus);
+
+ if (cond)
+ qed_rdma_dpm_bar(p_hwfn, p_ptt);
+ }
+
+ DP_INFO(p_hwfn,
+ "doorbell bar: normal_region_size=%d, pwm_region_size=%d, dpi_size=%d, dpi_count=%d, roce_edpm=%s\n",
+ norm_regsize,
+ pwm_regsize,
+ p_hwfn->dpi_size,
+ p_hwfn->dpi_count,
+ ((p_hwfn->dcbx_no_edpm) || (p_hwfn->db_bar_no_edpm)) ?
+ "disabled" : "enabled");
+
+ if (rc) {
+ DP_ERR(p_hwfn,
+ "Failed to allocate enough DPIs. Allocated %d but the current minimum is %d.\n",
+ p_hwfn->dpi_count,
+ p_hwfn->pf_params.rdma_pf_params.min_dpis);
+ return -EINVAL;
+ }
+
+ /* Update hwfn */
+ p_hwfn->dpi_start_offset = norm_regsize;
+
+ /* Update registers */
+ /* DEMS size is configured log2 of DWORDs, hence the division by 4 */
+ pf_dems_shift = ilog2(QED_PF_DEMS_SIZE / 4);
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_ICID_BIT_SHIFT_NORM, pf_dems_shift);
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_MIN_ADDR_REG1, min_addr_reg1);
+
+ return 0;
+}
+
static int qed_hw_init_port(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, int hw_mode)
{
@@ -860,6 +997,10 @@ static int qed_hw_init_pf(struct qed_hwfn *p_hwfn,
/* Pure runtime initializations - directly to the HW */
qed_int_igu_init_pure_rt(p_hwfn, p_ptt, true, true);
+ rc = qed_hw_init_pf_doorbell_bar(p_hwfn, p_ptt);
+ if (rc)
+ return rc;
+
if (b_hw_start) {
/* enable interrupts */
qed_int_igu_enable(p_hwfn, p_ptt, int_mode);
@@ -1284,6 +1425,19 @@ static void qed_hw_set_feat(struct qed_hwfn *p_hwfn)
u32 *feat_num = p_hwfn->hw_info.feat_num;
int num_features = 1;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ /* Roce CNQ require each: 1 status block. 1 CNQ, we divide the
+ * status blocks equally between L2 / RoCE but with consideration as
+ * to how many l2 queues / cnqs we have
+ */
+ if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) {
+ num_features++;
+
+ feat_num[QED_RDMA_CNQ] =
+ min_t(u32, RESC_NUM(p_hwfn, QED_SB) / num_features,
+ RESC_NUM(p_hwfn, QED_RDMA_CNQ_RAM));
+ }
+#endif
feat_num[QED_PF_L2_QUE] = min_t(u32, RESC_NUM(p_hwfn, QED_SB) /
num_features,
RESC_NUM(p_hwfn, QED_L2_QUEUE));
@@ -1325,6 +1479,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
num_funcs;
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
+ resc_num[QED_RDMA_CNQ_RAM] = NUM_OF_CMDQS_CQS / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_main.c b/drivers/net/ethernet/qlogic/qed/qed_main.c
index 48cdf62..4ee3151 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_main.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_main.c
@@ -33,6 +33,11 @@
#include "qed_hw.h"
#include "qed_selftest.h"
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+#define QED_ROCE_QPS (8192)
+#define QED_ROCE_DPIS (8)
+#endif
+
static char version[] =
"QLogic FastLinQ 4xxxx Core Module qed " DRV_MODULE_VERSION "\n";
@@ -206,8 +211,8 @@ int qed_fill_dev_info(struct qed_dev *cdev,
dev_info->pci_mem_start = cdev->pci_params.mem_start;
dev_info->pci_mem_end = cdev->pci_params.mem_end;
dev_info->pci_irq = cdev->pci_params.irq;
- dev_info->rdma_supported =
- (cdev->hwfns[0].hw_info.personality == QED_PCI_ETH_ROCE);
+ dev_info->rdma_supported = (cdev->hwfns[0].hw_info.personality ==
+ QED_PCI_ETH_ROCE);
dev_info->is_mf_default = IS_MF_DEFAULT(&cdev->hwfns[0]);
ether_addr_copy(dev_info->hw_mac, cdev->hwfns[0].hw_info.hw_mac_addr);
@@ -677,6 +682,9 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
enum qed_int_mode int_mode)
{
struct qed_sb_cnt_info sb_cnt_info;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ int num_l2_queues;
+#endif
int rc;
int i;
@@ -707,6 +715,31 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
cdev->int_params.fp_msix_cnt = cdev->int_params.out.num_vectors -
cdev->num_hwfns;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ num_l2_queues = 0;
+ for_each_hwfn(cdev, i)
+ num_l2_queues += FEAT_NUM(&cdev->hwfns[i], QED_PF_L2_QUE);
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA,
+ "cdev->int_params.fp_msix_cnt=%d num_l2_queues=%d\n",
+ cdev->int_params.fp_msix_cnt, num_l2_queues);
+
+ if (cdev->int_params.fp_msix_cnt > num_l2_queues) {
+ cdev->int_params.rdma_msix_cnt =
+ (cdev->int_params.fp_msix_cnt - num_l2_queues)
+ / cdev->num_hwfns;
+ cdev->int_params.rdma_msix_base =
+ cdev->int_params.fp_msix_base + num_l2_queues;
+ cdev->int_params.fp_msix_cnt = num_l2_queues;
+ } else {
+ cdev->int_params.rdma_msix_cnt = 0;
+ }
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA, "roce_msix_cnt=%d roce_msix_base=%d\n",
+ cdev->int_params.rdma_msix_cnt,
+ cdev->int_params.rdma_msix_base);
+#endif
+
return 0;
}
@@ -810,6 +843,13 @@ static void qed_update_pf_params(struct qed_dev *cdev,
{
int i;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ params->rdma_pf_params.num_qps = QED_ROCE_QPS;
+ params->rdma_pf_params.min_dpis = QED_ROCE_DPIS;
+ /* divide by 3 the MRs to avoid MF ILT overflow */
+ params->rdma_pf_params.num_mrs = RDMA_MAX_TIDS;
+ params->rdma_pf_params.gl_pi = QED_ROCE_PROTOCOL_INDEX;
+#endif
for (i = 0; i < cdev->num_hwfns; i++) {
struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
diff --git a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
index e75738d..b414a05 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
@@ -1448,5 +1448,11 @@
0x620000UL
#define PHY_PCIE_REG_PHY1 \
0x624000UL
-
+#define NIG_REG_ROCE_DUPLICATE_TO_HOST 0x5088f0UL
+#define PRS_REG_LIGHT_L2_ETHERTYPE_EN 0x1f0968UL
+#define NIG_REG_LLH_ENG_CLS_ENG_ID_TBL 0x501b90UL
+#define DORQ_REG_PF_DPM_ENABLE 0x100510UL
+#define DORQ_REG_PF_ICID_BIT_SHIFT_NORM 0x100448UL
+#define DORQ_REG_PF_MIN_ADDR_REG1 0x100400UL
+#define DORQ_REG_PF_DPI_BIT_SHIFT 0x100450UL
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
new file mode 100644
index 0000000..e65a130
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -0,0 +1,887 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/types.h>
+#include <asm/byteorder.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/errno.h>
+#include <linux/etherdevice.h>
+#include <linux/if_ether.h>
+#include <linux/if_vlan.h>
+#include <linux/io.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/tcp.h>
+#include <linux/bitops.h>
+#include <linux/qed/qed_roce_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qed.h"
+#include "qed_cxt.h"
+#include "qed_hsi.h"
+#include "qed_hw.h"
+#include "qed_init_ops.h"
+#include "qed_int.h"
+#include "qed_ll2.h"
+#include "qed_mcp.h"
+#include "qed_reg_addr.h"
+#include "qed_sp.h"
+#include "qed_roce.h"
+
+void qed_async_roce_event(struct qed_hwfn *p_hwfn,
+ struct event_ring_entry *p_eqe)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+
+ p_rdma_info->events.affiliated_event(p_rdma_info->events.context,
+ p_eqe->opcode, &p_eqe->data);
+}
+
+static int qed_rdma_bmap_alloc(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 max_count)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "max_count = %08x\n", max_count);
+
+ bmap->max_count = max_count;
+
+ bmap->bitmap = kzalloc(BITS_TO_LONGS(max_count) * sizeof(long),
+ GFP_KERNEL);
+ if (!bmap->bitmap) {
+ DP_NOTICE(p_hwfn,
+ "qed bmap alloc failed: cannot allocate memory (bitmap)\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocated bitmap %p\n",
+ bmap->bitmap);
+ return 0;
+}
+
+static int qed_rdma_bmap_alloc_id(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 *id_num)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "bmap = %p\n", bmap);
+
+ *id_num = find_first_zero_bit(bmap->bitmap, bmap->max_count);
+
+ if (*id_num >= bmap->max_count) {
+ DP_NOTICE(p_hwfn, "no id available max_count=%d\n",
+ bmap->max_count);
+ return -EINVAL;
+ }
+
+ __set_bit(*id_num, bmap->bitmap);
+
+ return 0;
+}
+
+static void qed_bmap_release_id(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 id_num)
+{
+ bool b_acquired;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "id_num = %08x", id_num);
+ if (id_num >= bmap->max_count)
+ return;
+
+ b_acquired = test_and_clear_bit(id_num, bmap->bitmap);
+ if (!b_acquired) {
+ DP_NOTICE(p_hwfn, "ID %d already released\n", id_num);
+ return;
+ }
+}
+
+u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id)
+{
+ /* First sb id for RoCE is after all the l2 sb */
+ return FEAT_NUM((struct qed_hwfn *)p_hwfn, QED_PF_L2_QUE) + rel_sb_id;
+}
+
+u32 qed_rdma_query_cau_timer_res(void *rdma_cxt)
+{
+ return QED_CAU_DEF_RX_TIMER_RES;
+}
+
+static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_info *p_rdma_info;
+ u32 num_cons, num_tasks;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocating RDMA\n");
+
+ /* Allocate a struct with current pf rdma info */
+ p_rdma_info = kzalloc(sizeof(*p_rdma_info), GFP_KERNEL);
+ if (!p_rdma_info) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ p_rdma_info = p_rdma_info;
+ p_rdma_info->proto = PROTOCOLID_ROCE;
+
+ num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
+
+ p_rdma_info->num_qps = num_cons / 2;
+
+ num_tasks = qed_cxt_get_proto_tid_count(p_hwfn, PROTOCOLID_ROCE);
+
+ /* Each MR uses a single task */
+ p_rdma_info->num_mrs = num_tasks;
+
+ /* Queue zone lines are shared between RoCE and L2 in such a way that
+ * they can be used by each without obstructing the other.
+ */
+ p_rdma_info->queue_zone_base = (u16)FEAT_NUM(p_hwfn, QED_L2_QUEUE);
+
+ /* Allocate a struct with device params and fill it */
+ p_rdma_info->dev = kzalloc(sizeof(*p_rdma_info->dev), GFP_KERNEL);
+ if (!p_rdma_info->dev) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info dev). rc = %d\n",
+ rc);
+ goto free_rdma_info;
+ }
+
+ /* Allocate a struct with port params and fill it */
+ p_rdma_info->port = kzalloc(sizeof(*p_rdma_info->port), GFP_KERNEL);
+ if (!p_rdma_info->port) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info port). rc = %d\n",
+ rc);
+ goto free_rdma_dev;
+ }
+
+ /* Allocate bit map for pd's */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->pd_map, RDMA_MAX_PDS);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate pd_map, rc = %d\n",
+ rc);
+ goto free_rdma_port;
+ }
+
+ /* Allocate DPI bitmap */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->dpi_map,
+ p_hwfn->dpi_count);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate DPI bitmap, rc = %d\n", rc);
+ goto free_pd_map;
+ }
+
+ /* Allocate bitmap for cq's. The maximum number of CQs is bounded to
+ * twice the number of QPs.
+ */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cq_map,
+ p_rdma_info->num_qps * 2);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate cq bitmap, rc = %d\n", rc);
+ goto free_dpi_map;
+ }
+
+ /* Allocate bitmap for toggle bit for cq icids
+ * We toggle the bit every time we create or resize cq for a given icid.
+ * The maximum number of CQs is bounded to twice the number of QPs.
+ */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->toggle_bits,
+ p_rdma_info->num_qps * 2);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate toogle bits, rc = %d\n", rc);
+ goto free_cq_map;
+ }
+
+ /* Allocate bitmap for itids */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->tid_map,
+ p_rdma_info->num_mrs);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate itids bitmaps, rc = %d\n", rc);
+ goto free_toggle_map;
+ }
+
+ /* Allocate bitmap for cids used for qps. */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cid_map, num_cons);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate cid bitmap, rc = %d\n", rc);
+ goto free_tid_map;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocation successful\n");
+ return 0;
+
+free_tid_map:
+ kfree(p_rdma_info->tid_map.bitmap);
+free_toggle_map:
+ kfree(p_rdma_info->toggle_bits.bitmap);
+free_cq_map:
+ kfree(p_rdma_info->cq_map.bitmap);
+free_dpi_map:
+ kfree(p_rdma_info->dpi_map.bitmap);
+free_pd_map:
+ kfree(p_rdma_info->pd_map.bitmap);
+free_rdma_port:
+ kfree(p_rdma_info->port);
+free_rdma_dev:
+ kfree(p_rdma_info->dev);
+free_rdma_info:
+ kfree(p_rdma_info);
+
+ return rc;
+}
+
+void qed_rdma_resc_free(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+
+ kfree(p_rdma_info->cid_map.bitmap);
+ kfree(p_rdma_info->tid_map.bitmap);
+ kfree(p_rdma_info->toggle_bits.bitmap);
+ kfree(p_rdma_info->cq_map.bitmap);
+ kfree(p_rdma_info->dpi_map.bitmap);
+ kfree(p_rdma_info->pd_map.bitmap);
+
+ kfree(p_rdma_info->port);
+ kfree(p_rdma_info->dev);
+
+ kfree(p_rdma_info);
+}
+
+static void qed_rdma_free(struct qed_hwfn *p_hwfn)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Freeing RDMA\n");
+
+ qed_rdma_resc_free(p_hwfn);
+}
+
+static void qed_rdma_get_guid(struct qed_hwfn *p_hwfn, u8 *guid)
+{
+ guid[0] = p_hwfn->hw_info.hw_mac_addr[0] ^ 2;
+ guid[1] = p_hwfn->hw_info.hw_mac_addr[1];
+ guid[2] = p_hwfn->hw_info.hw_mac_addr[2];
+ guid[3] = 0xff;
+ guid[4] = 0xfe;
+ guid[5] = p_hwfn->hw_info.hw_mac_addr[3];
+ guid[6] = p_hwfn->hw_info.hw_mac_addr[4];
+ guid[7] = p_hwfn->hw_info.hw_mac_addr[5];
+}
+
+static void qed_rdma_init_events(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_events *events;
+
+ events = &p_hwfn->p_rdma_info->events;
+
+ events->unaffiliated_event = params->events->unaffiliated_event;
+ events->affiliated_event = params->events->affiliated_event;
+ events->context = params->events->context;
+}
+
+static void qed_rdma_init_devinfo(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ u32 pci_status_control;
+ u32 num_qps;
+
+ /* Vendor specific information */
+ dev->vendor_id = cdev->vendor_id;
+ dev->vendor_part_id = cdev->device_id;
+ dev->hw_ver = 0;
+ dev->fw_ver = (FW_MAJOR_VERSION << 24) | (FW_MINOR_VERSION << 16) |
+ (FW_REVISION_VERSION << 8) | (FW_ENGINEERING_VERSION);
+
+ qed_rdma_get_guid(p_hwfn, (u8 *)&dev->sys_image_guid);
+ dev->node_guid = dev->sys_image_guid;
+
+ dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
+ RDMA_MAX_SGE_PER_RQ_WQE);
+
+ if (cdev->rdma_max_sge)
+ dev->max_sge = min_t(u32, cdev->rdma_max_sge, dev->max_sge);
+
+ dev->max_inline = ROCE_REQ_MAX_INLINE_DATA_SIZE;
+
+ dev->max_inline = (cdev->rdma_max_inline) ?
+ min_t(u32, cdev->rdma_max_inline, dev->max_inline) :
+ dev->max_inline;
+
+ dev->max_wqe = QED_RDMA_MAX_WQE;
+ dev->max_cnq = (u8)FEAT_NUM(p_hwfn, QED_RDMA_CNQ);
+
+ /* The number of QPs may be higher than QED_ROCE_MAX_QPS. because
+ * it is up-aligned to 16 and then to ILT page size within qed cxt.
+ * This is OK in terms of ILT but we don't want to configure the FW
+ * above its abilities
+ */
+ num_qps = ROCE_MAX_QPS;
+ num_qps = min_t(u64, num_qps, p_hwfn->p_rdma_info->num_qps);
+ dev->max_qp = num_qps;
+
+ /* CQs uses the same icids that QPs use hence they are limited by the
+ * number of icids. There are two icids per QP.
+ */
+ dev->max_cq = num_qps * 2;
+
+ /* The number of mrs is smaller by 1 since the first is reserved */
+ dev->max_mr = p_hwfn->p_rdma_info->num_mrs - 1;
+ dev->max_mr_size = QED_RDMA_MAX_MR_SIZE;
+
+ /* The maximum CQE capacity per CQ supported.
+ * max number of cqes will be in two layer pbl,
+ * 8 is the pointer size in bytes
+ * 32 is the size of cq element in bytes
+ */
+ if (params->cq_mode == QED_RDMA_CQ_MODE_32_BITS)
+ dev->max_cqe = QED_RDMA_MAX_CQE_32_BIT;
+ else
+ dev->max_cqe = QED_RDMA_MAX_CQE_16_BIT;
+
+ dev->max_mw = 0;
+ dev->max_fmr = QED_RDMA_MAX_FMR;
+ dev->max_mr_mw_fmr_pbl = (PAGE_SIZE / 8) * (PAGE_SIZE / 8);
+ dev->max_mr_mw_fmr_size = dev->max_mr_mw_fmr_pbl * PAGE_SIZE;
+ dev->max_pkey = QED_RDMA_MAX_P_KEY;
+
+ dev->max_qp_resp_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
+ (RDMA_RESP_RD_ATOMIC_ELM_SIZE * 2);
+ dev->max_qp_req_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
+ RDMA_REQ_RD_ATOMIC_ELM_SIZE;
+ dev->max_dev_resp_rd_atomic_resc = dev->max_qp_resp_rd_atomic_resc *
+ p_hwfn->p_rdma_info->num_qps;
+ dev->page_size_caps = QED_RDMA_PAGE_SIZE_CAPS;
+ dev->dev_ack_delay = QED_RDMA_ACK_DELAY;
+ dev->max_pd = RDMA_MAX_PDS;
+ dev->max_ah = p_hwfn->p_rdma_info->num_qps;
+ dev->max_stats_queues = (u8)RESC_NUM(p_hwfn, QED_RDMA_STATS_QUEUE);
+
+ /* Set capablities */
+ dev->dev_caps = 0;
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RNR_NAK, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RESIZE_CQ, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_MEMORY_EXT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_QUEUE_EXT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ZBVA, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_LOCAL_INV_FENCE, 1);
+
+ /* Check atomic operations support in PCI configuration space. */
+ pci_read_config_dword(cdev->pdev,
+ cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
+ &pci_status_control);
+
+ if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ATOMIC_OP, 1);
+}
+
+static void qed_rdma_init_port(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_port *port = p_hwfn->p_rdma_info->port;
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+
+ port->port_state = p_hwfn->mcp_info->link_output.link_up ?
+ QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
+
+ port->max_msg_size = min_t(u64,
+ (dev->max_mr_mw_fmr_size *
+ p_hwfn->cdev->rdma_max_sge),
+ BIT(31));
+
+ port->pkey_bad_counter = 0;
+}
+
+static int qed_rdma_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 ll2_ethertype_en;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW\n");
+ p_hwfn->b_rdma_enabled_in_prs = false;
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
+
+ p_hwfn->rdma_prs_search_reg = PRS_REG_SEARCH_ROCE;
+
+ /* We delay writing to this reg until first cid is allocated. See
+ * qed_cxt_dynamic_ilt_alloc function for more details
+ */
+ ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
+ qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
+ (ll2_ethertype_en | 0x01));
+
+ if (qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_ROCE) % 2) {
+ DP_NOTICE(p_hwfn, "The first RoCE's cid should be even\n");
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW - Done\n");
+ return 0;
+}
+
+static int qed_rdma_start_fw(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params,
+ struct qed_ptt *p_ptt)
+{
+ struct rdma_init_func_ramrod_data *p_ramrod;
+ struct qed_rdma_cnq_params *p_cnq_pbl_list;
+ struct rdma_init_func_hdr *p_params_header;
+ struct rdma_cnq_params *p_cnq_params;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ u32 cnq_id, sb_id;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Starting FW\n");
+
+ /* Save the number of cnqs for the function close ramrod */
+ p_hwfn->p_rdma_info->num_cnqs = params->desired_cnq;
+
+ /* Start Roce */
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_INIT,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.roce_init_func.rdma;
+
+ p_params_header = &p_ramrod->params_header;
+ p_params_header->cnq_start_offset = (u8)RESC_START(p_hwfn,
+ QED_RDMA_CNQ_RAM);
+ p_params_header->num_cnqs = params->desired_cnq;
+
+ if (params->cq_mode == QED_RDMA_CQ_MODE_16_BITS)
+ p_params_header->cq_ring_mode = 1;
+ else
+ p_params_header->cq_ring_mode = 0;
+
+ for (cnq_id = 0; cnq_id < params->desired_cnq; cnq_id++) {
+ sb_id = qed_rdma_get_sb_id(p_hwfn, cnq_id);
+ p_cnq_params = &p_ramrod->cnq_params[cnq_id];
+ p_cnq_pbl_list = ¶ms->cnq_pbl_list[cnq_id];
+ p_cnq_params->sb_num =
+ cpu_to_le16(p_hwfn->sbs_info[sb_id]->igu_sb_id);
+
+ p_cnq_params->sb_index = p_hwfn->pf_params.rdma_pf_params.gl_pi;
+ p_cnq_params->num_pbl_pages = p_cnq_pbl_list->num_pbl_pages;
+
+ DMA_REGPAIR_LE(p_cnq_params->pbl_base_addr,
+ p_cnq_pbl_list->pbl_ptr);
+
+ /* we assume here that cnq_id and qz_offset are the same */
+ p_cnq_params->queue_zone_num =
+ cpu_to_le16(p_hwfn->p_rdma_info->queue_zone_base +
+ cnq_id);
+ }
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_rdma_reserve_lkey(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+
+ /* The first DPI is reserved for the Kernel */
+ __set_bit(0, p_hwfn->p_rdma_info->dpi_map.bitmap);
+
+ /* Tid 0 will be used as the key for "reserved MR".
+ * The driver should allocate memory for it so it can be loaded but no
+ * ramrod should be passed on it.
+ */
+ qed_rdma_alloc_tid(p_hwfn, &dev->reserved_lkey);
+ if (dev->reserved_lkey != RDMA_RESERVED_LKEY) {
+ DP_NOTICE(p_hwfn,
+ "Reserved lkey should be equal to RDMA_RESERVED_LKEY\n");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int qed_rdma_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_rdma_start_in_params *params)
+{
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA setup\n");
+
+ spin_lock_init(&p_hwfn->p_rdma_info->lock);
+
+ qed_rdma_init_devinfo(p_hwfn, params);
+ qed_rdma_init_port(p_hwfn);
+ qed_rdma_init_events(p_hwfn, params);
+
+ rc = qed_rdma_reserve_lkey(p_hwfn);
+ if (rc)
+ return rc;
+
+ rc = qed_rdma_init_hw(p_hwfn, p_ptt);
+ if (rc)
+ return rc;
+
+ return qed_rdma_start_fw(p_hwfn, params, p_ptt);
+}
+
+int qed_rdma_stop(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_close_func_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ struct qed_ptt *p_ptt;
+ u32 ll2_ethertype_en;
+ int rc = -EBUSY;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop\n");
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Failed to acquire PTT\n");
+ return rc;
+ }
+
+ /* Disable RoCE search */
+ qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 0);
+ p_hwfn->b_rdma_enabled_in_prs = false;
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
+
+ ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
+ (ll2_ethertype_en & 0xFFFE));
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Stop RoCE */
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_CLOSE,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto out;
+
+ p_ramrod = &p_ent->ramrod.rdma_close_func;
+
+ p_ramrod->num_cnqs = p_hwfn->p_rdma_info->num_cnqs;
+ p_ramrod->cnq_start_offset = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+out:
+ qed_rdma_free(p_hwfn);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop done, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_add_user(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ u32 dpi_start_offset;
+ u32 returned_id = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding User\n");
+
+ /* Allocate DPI */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map,
+ &returned_id);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ out_params->dpi = (u16)returned_id;
+
+ /* Calculate the corresponding DPI address */
+ dpi_start_offset = p_hwfn->dpi_start_offset;
+
+ out_params->dpi_addr = (u64)((u8 __iomem *)p_hwfn->doorbells +
+ dpi_start_offset +
+ ((out_params->dpi) * p_hwfn->dpi_size));
+
+ out_params->dpi_phys_addr = p_hwfn->cdev->db_phys_addr +
+ dpi_start_offset +
+ ((out_params->dpi) * p_hwfn->dpi_size);
+
+ out_params->dpi_size = p_hwfn->dpi_size;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding user - done, rc = %d\n", rc);
+ return rc;
+}
+
+struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query device\n");
+
+ /* Return struct with device parameters */
+ return p_hwfn->p_rdma_info->dev;
+}
+
+int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID\n");
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_hwfn->p_rdma_info->tid_map, itid);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+ if (rc)
+ goto out;
+
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_TASK, *itid);
+out:
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID - done, rc = %d\n", rc);
+ return rc;
+}
+
+void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 qz_offset, u16 prod)
+{
+ struct qed_hwfn *p_hwfn;
+ u16 qz_num;
+ u32 addr;
+
+ p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ qz_num = p_hwfn->p_rdma_info->queue_zone_base + qz_offset;
+ addr = GTT_BAR0_MAP_REG_USDM_RAM +
+ USTORM_COMMON_QUEUE_CONS_OFFSET(qz_num);
+
+ REG_WR16(p_hwfn, addr, prod);
+
+ /* keep prod updates ordered */
+ wmb();
+}
+
+static int qed_fill_rdma_dev_info(struct qed_dev *cdev,
+ struct qed_dev_rdma_info *info)
+{
+ memset(info, 0, sizeof(*info));
+
+ info->rdma_type = QED_RDMA_TYPE_ROCE;
+
+ qed_fill_dev_info(cdev, &info->common);
+
+ return 0;
+}
+
+static int qed_rdma_get_sb_start(struct qed_dev *cdev)
+{
+ int feat_num;
+
+ if (cdev->num_hwfns > 1)
+ feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE);
+ else
+ feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE) *
+ cdev->num_hwfns;
+
+ return feat_num;
+}
+
+static int qed_rdma_get_min_cnq_msix(struct qed_dev *cdev)
+{
+ int n_cnq = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_RDMA_CNQ);
+ int n_msix = cdev->int_params.rdma_msix_cnt;
+
+ return min_t(int, n_cnq, n_msix);
+}
+
+static int qed_rdma_set_int(struct qed_dev *cdev, u16 cnt)
+{
+ int limit = 0;
+
+ /* Mark the fastpath as free/used */
+ cdev->int_params.fp_initialized = cnt ? true : false;
+
+ if (cdev->int_params.out.int_mode != QED_INT_MODE_MSIX) {
+ DP_ERR(cdev,
+ "qed roce supports only MSI-X interrupts (detected %d).\n",
+ cdev->int_params.out.int_mode);
+ return -EINVAL;
+ } else if (cdev->int_params.fp_msix_cnt) {
+ limit = cdev->int_params.rdma_msix_cnt;
+ }
+
+ if (!limit)
+ return -ENOMEM;
+
+ return min_t(int, cnt, limit);
+}
+
+static int qed_rdma_get_int(struct qed_dev *cdev, struct qed_int_info *info)
+{
+ memset(info, 0, sizeof(*info));
+
+ if (!cdev->int_params.fp_initialized) {
+ DP_INFO(cdev,
+ "Protocol driver requested interrupt information, but its support is not yet configured\n");
+ return -EINVAL;
+ }
+
+ if (cdev->int_params.out.int_mode == QED_INT_MODE_MSIX) {
+ int msix_base = cdev->int_params.rdma_msix_base;
+
+ info->msix_cnt = cdev->int_params.rdma_msix_cnt;
+ info->msix = &cdev->int_params.msix_table[msix_base];
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA, "msix_cnt = %d msix_base=%d\n",
+ info->msix_cnt, msix_base);
+ }
+
+ return 0;
+}
+
+static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
+{
+ return QED_LEADING_HWFN(cdev);
+}
+
+static void qed_rdma_dpm_conf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 val;
+
+ val = ((p_hwfn->dcbx_no_edpm) || (p_hwfn->db_bar_no_edpm)) ? 0 : 1;
+
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPM_ENABLE, val);
+ DP_VERBOSE(p_hwfn, (QED_MSG_DCB | QED_MSG_RDMA),
+ "Changing DPM_EN state to %d (DCBX=%d, DB_BAR=%d)\n",
+ val, p_hwfn->dcbx_no_edpm, p_hwfn->db_bar_no_edpm);
+}
+
+void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ p_hwfn->db_bar_no_edpm = true;
+
+ qed_rdma_dpm_conf(p_hwfn, p_ptt);
+}
+
+int qed_rdma_start(void *rdma_cxt, struct qed_rdma_start_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_ptt *p_ptt;
+ int rc = -EBUSY;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "desired_cnq = %08x\n", params->desired_cnq);
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt)
+ goto err;
+
+ rc = qed_rdma_alloc(p_hwfn, p_ptt, params);
+ if (rc)
+ goto err1;
+
+ rc = qed_rdma_setup(p_hwfn, p_ptt, params);
+ if (rc)
+ goto err2;
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ return rc;
+
+err2:
+ qed_rdma_free(p_hwfn);
+err1:
+ qed_ptt_release(p_hwfn, p_ptt);
+err:
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA start - error, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_rdma_init(struct qed_dev *cdev,
+ struct qed_rdma_start_in_params *params)
+{
+ return qed_rdma_start(QED_LEADING_HWFN(cdev), params);
+}
+
+void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "dpi = %08x\n", dpi);
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map, dpi);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
+static const struct qed_rdma_ops qed_rdma_ops_pass = {
+ .common = &qed_common_ops_pass,
+ .fill_dev_info = &qed_fill_rdma_dev_info,
+ .rdma_get_rdma_ctx = &qed_rdma_get_rdma_ctx,
+ .rdma_init = &qed_rdma_init,
+ .rdma_add_user = &qed_rdma_add_user,
+ .rdma_remove_user = &qed_rdma_remove_user,
+ .rdma_stop = &qed_rdma_stop,
+ .rdma_query_device = &qed_rdma_query_device,
+ .rdma_get_start_sb = &qed_rdma_get_sb_start,
+ .rdma_get_rdma_int = &qed_rdma_get_int,
+ .rdma_set_rdma_int = &qed_rdma_set_int,
+ .rdma_get_min_cnq_msix = &qed_rdma_get_min_cnq_msix,
+ .rdma_cnq_prod_update = &qed_rdma_cnq_prod_update,
+};
+
+const struct qed_rdma_ops *qed_get_rdma_ops()
+{
+ return &qed_rdma_ops_pass;
+}
+EXPORT_SYMBOL(qed_get_rdma_ops);
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
new file mode 100644
index 0000000..46f14b2
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -0,0 +1,117 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QED_ROCE_H
+#define _QED_ROCE_H
+#include <linux/types.h>
+#include <linux/bitops.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qed.h"
+#include "qed_dev_api.h"
+#include "qed_hsi.h"
+
+#define QED_RDMA_MAX_FMR (RDMA_MAX_TIDS)
+#define QED_RDMA_MAX_P_KEY (1)
+#define QED_RDMA_MAX_WQE (0x7FFF)
+#define QED_RDMA_MAX_SRQ_WQE_ELEM (0x7FFF)
+#define QED_RDMA_PAGE_SIZE_CAPS (0xFFFFF000)
+#define QED_RDMA_ACK_DELAY (15)
+#define QED_RDMA_MAX_MR_SIZE (0x10000000000ULL)
+#define QED_RDMA_MAX_CQS (RDMA_MAX_CQS)
+#define QED_RDMA_MAX_MRS (RDMA_MAX_TIDS)
+/* Add 1 for header element */
+#define QED_RDMA_MAX_SRQ_ELEM_PER_WQE (RDMA_MAX_SGE_PER_RQ_WQE + 1)
+#define QED_RDMA_MAX_SGE_PER_SRQ_WQE (RDMA_MAX_SGE_PER_RQ_WQE)
+#define QED_RDMA_SRQ_WQE_ELEM_SIZE (16)
+#define QED_RDMA_MAX_SRQS (32 * 1024)
+
+#define QED_RDMA_MAX_CQE_32_BIT (0x7FFFFFFF - 1)
+#define QED_RDMA_MAX_CQE_16_BIT (0x7FFF - 1)
+
+enum qed_rdma_toggle_bit {
+ QED_RDMA_TOGGLE_BIT_CLEAR = 0,
+ QED_RDMA_TOGGLE_BIT_SET = 1
+};
+
+struct qed_bmap {
+ unsigned long *bitmap;
+ u32 max_count;
+};
+
+struct qed_rdma_info {
+ /* spin lock to protect bitmaps */
+ spinlock_t lock;
+
+ struct qed_bmap cq_map;
+ struct qed_bmap pd_map;
+ struct qed_bmap tid_map;
+ struct qed_bmap qp_map;
+ struct qed_bmap srq_map;
+ struct qed_bmap cid_map;
+ struct qed_bmap dpi_map;
+ struct qed_bmap toggle_bits;
+ struct qed_rdma_events events;
+ struct qed_rdma_device *dev;
+ struct qed_rdma_port *port;
+ u32 last_tid;
+ u8 num_cnqs;
+ u32 num_qps;
+ u32 num_mrs;
+ u16 queue_zone_base;
+ enum protocol_type proto;
+};
+
+int
+qed_rdma_add_user(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *out_params);
+int qed_rdma_alloc_pd(void *rdma_cxt, u16 *pd);
+int qed_rdma_alloc_tid(void *rdma_cxt, u32 *tid);
+int qed_rdma_deregister_tid(void *rdma_cxt, u32 tid);
+void qed_rdma_free_tid(void *rdma_cxt, u32 tid);
+struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt);
+int
+qed_rdma_register_tid(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *params);
+void qed_rdma_remove_user(void *rdma_cxt, u16 dpi);
+int qed_rdma_start(void *p_hwfn, struct qed_rdma_start_in_params *params);
+int qed_rdma_stop(void *rdma_cxt);
+u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id);
+u32 qed_rdma_query_cau_timer_res(void *p_hwfn);
+void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 cnq_index, u16 prod);
+void qed_rdma_resc_free(struct qed_hwfn *p_hwfn);
+void qed_async_roce_event(struct qed_hwfn *p_hwfn,
+ struct event_ring_entry *p_eqe);
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_sp.h b/drivers/net/ethernet/qlogic/qed/qed_sp.h
index a3c539f..652c908 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_sp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_sp.h
@@ -85,6 +85,7 @@ union ramrod_data {
struct rdma_srq_create_ramrod_data rdma_create_srq;
struct rdma_srq_destroy_ramrod_data rdma_destroy_srq;
struct rdma_srq_modify_ramrod_data rdma_modify_srq;
+ struct roce_init_func_ramrod_data roce_init_func;
struct iscsi_slow_path_hdr iscsi_empty;
struct iscsi_init_ramrod_params iscsi_init;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_spq.c b/drivers/net/ethernet/qlogic/qed/qed_spq.c
index 349af18..caff415 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_spq.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_spq.c
@@ -28,6 +28,9 @@
#include "qed_reg_addr.h"
#include "qed_sp.h"
#include "qed_sriov.h"
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+#include "qed_roce.h"
+#endif
/***************************************************************************
* Structures & Definitions
@@ -237,6 +240,11 @@ qed_async_event_completion(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe)
{
switch (p_eqe->protocol_id) {
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ case PROTOCOLID_ROCE:
+ qed_async_roce_event(p_hwfn, p_eqe);
+ return 0;
+#endif
case PROTOCOLID_COMMON:
return qed_sriov_eqe_event(p_hwfn,
p_eqe->opcode,
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index e313742..f9ae903 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -34,6 +34,8 @@ enum dcbx_protocol_type {
DCBX_MAX_PROTOCOL_TYPE
};
+#define QED_ROCE_PROTOCOL_INDEX (3)
+
#ifdef CONFIG_DCB
#define QED_LLDP_CHASSIS_ID_STAT_LEN 4
#define QED_LLDP_PORT_ID_STAT_LEN 4
@@ -268,6 +270,7 @@ struct qed_dev_info {
enum qed_sb_type {
QED_SB_TYPE_L2_QUEUE,
+ QED_SB_TYPE_CNQ,
};
enum qed_protocol {
@@ -628,7 +631,7 @@ enum DP_MODULE {
QED_MSG_CXT = 0x800000,
QED_MSG_LL2 = 0x1000000,
QED_MSG_ILT = 0x2000000,
- QED_MSG_ROCE = 0x4000000,
+ QED_MSG_RDMA = 0x4000000,
QED_MSG_DEBUG = 0x8000000,
/* to be added...up to 0x8000000 */
};
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
new file mode 100644
index 0000000..0f7d527
--- /dev/null
+++ b/include/linux/qed/qed_roce_if.h
@@ -0,0 +1,345 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QED_ROCE_IF_H
+#define _QED_ROCE_IF_H
+#include <linux/types.h>
+#include <linux/delay.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_ll2_if.h>
+
+#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
+
+/* rdma interface */
+enum qed_rdma_tid_type {
+ QED_RDMA_TID_REGISTERED_MR,
+ QED_RDMA_TID_FMR,
+ QED_RDMA_TID_MW_TYPE1,
+ QED_RDMA_TID_MW_TYPE2A
+};
+
+struct qed_rdma_events {
+ void *context;
+ void (*affiliated_event)(void *context, u8 fw_event_code,
+ void *fw_handle);
+ void (*unaffiliated_event)(void *context, u8 event_code);
+};
+
+struct qed_rdma_device {
+ u32 vendor_id;
+ u32 vendor_part_id;
+ u32 hw_ver;
+ u64 fw_ver;
+
+ u64 node_guid;
+ u64 sys_image_guid;
+
+ u8 max_cnq;
+ u8 max_sge;
+ u8 max_srq_sge;
+ u16 max_inline;
+ u32 max_wqe;
+ u32 max_srq_wqe;
+ u8 max_qp_resp_rd_atomic_resc;
+ u8 max_qp_req_rd_atomic_resc;
+ u64 max_dev_resp_rd_atomic_resc;
+ u32 max_cq;
+ u32 max_qp;
+ u32 max_srq;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_cqe;
+ u32 max_mw;
+ u32 max_fmr;
+ u32 max_mr_mw_fmr_pbl;
+ u64 max_mr_mw_fmr_size;
+ u32 max_pd;
+ u32 max_ah;
+ u8 max_pkey;
+ u16 max_srq_wr;
+ u8 max_stats_queues;
+ u32 dev_caps;
+
+ /* Abilty to support RNR-NAK generation */
+
+#define QED_RDMA_DEV_CAP_RNR_NAK_MASK 0x1
+#define QED_RDMA_DEV_CAP_RNR_NAK_SHIFT 0
+ /* Abilty to support shutdown port */
+#define QED_RDMA_DEV_CAP_SHUTDOWN_PORT_MASK 0x1
+#define QED_RDMA_DEV_CAP_SHUTDOWN_PORT_SHIFT 1
+ /* Abilty to support port active event */
+#define QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT_MASK 0x1
+#define QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT_SHIFT 2
+ /* Abilty to support port change event */
+#define QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT_MASK 0x1
+#define QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT_SHIFT 3
+ /* Abilty to support system image GUID */
+#define QED_RDMA_DEV_CAP_SYS_IMAGE_MASK 0x1
+#define QED_RDMA_DEV_CAP_SYS_IMAGE_SHIFT 4
+ /* Abilty to support bad P_Key counter support */
+#define QED_RDMA_DEV_CAP_BAD_PKEY_CNT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BAD_PKEY_CNT_SHIFT 5
+ /* Abilty to support atomic operations */
+#define QED_RDMA_DEV_CAP_ATOMIC_OP_MASK 0x1
+#define QED_RDMA_DEV_CAP_ATOMIC_OP_SHIFT 6
+#define QED_RDMA_DEV_CAP_RESIZE_CQ_MASK 0x1
+#define QED_RDMA_DEV_CAP_RESIZE_CQ_SHIFT 7
+ /* Abilty to support modifying the maximum number of
+ * outstanding work requests per QP
+ */
+#define QED_RDMA_DEV_CAP_RESIZE_MAX_WR_MASK 0x1
+#define QED_RDMA_DEV_CAP_RESIZE_MAX_WR_SHIFT 8
+ /* Abilty to support automatic path migration */
+#define QED_RDMA_DEV_CAP_AUTO_PATH_MIG_MASK 0x1
+#define QED_RDMA_DEV_CAP_AUTO_PATH_MIG_SHIFT 9
+ /* Abilty to support the base memory management extensions */
+#define QED_RDMA_DEV_CAP_BASE_MEMORY_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BASE_MEMORY_EXT_SHIFT 10
+#define QED_RDMA_DEV_CAP_BASE_QUEUE_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BASE_QUEUE_EXT_SHIFT 11
+ /* Abilty to support multipile page sizes per memory region */
+#define QED_RDMA_DEV_CAP_MULTI_PAGE_PER_MR_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_MULTI_PAGE_PER_MR_EXT_SHIFT 12
+ /* Abilty to support block list physical buffer list */
+#define QED_RDMA_DEV_CAP_BLOCK_MODE_MASK 0x1
+#define QED_RDMA_DEV_CAP_BLOCK_MODE_SHIFT 13
+ /* Abilty to support zero based virtual addresses */
+#define QED_RDMA_DEV_CAP_ZBVA_MASK 0x1
+#define QED_RDMA_DEV_CAP_ZBVA_SHIFT 14
+ /* Abilty to support local invalidate fencing */
+#define QED_RDMA_DEV_CAP_LOCAL_INV_FENCE_MASK 0x1
+#define QED_RDMA_DEV_CAP_LOCAL_INV_FENCE_SHIFT 15
+ /* Abilty to support Loopback on QP */
+#define QED_RDMA_DEV_CAP_LB_INDICATOR_MASK 0x1
+#define QED_RDMA_DEV_CAP_LB_INDICATOR_SHIFT 16
+ u64 page_size_caps;
+ u8 dev_ack_delay;
+ u32 reserved_lkey;
+ u32 bad_pkey_counter;
+ struct qed_rdma_events events;
+};
+
+enum qed_port_state {
+ QED_RDMA_PORT_UP,
+ QED_RDMA_PORT_DOWN,
+};
+
+enum qed_roce_capability {
+ QED_ROCE_V1 = 1 << 0,
+ QED_ROCE_V2 = 1 << 1,
+};
+
+struct qed_rdma_port {
+ enum qed_port_state port_state;
+ int link_speed;
+ u64 max_msg_size;
+ u8 source_gid_table_len;
+ void *source_gid_table_ptr;
+ u8 pkey_table_len;
+ void *pkey_table_ptr;
+ u32 pkey_bad_counter;
+ enum qed_roce_capability capability;
+};
+
+struct qed_rdma_cnq_params {
+ u8 num_pbl_pages;
+ u64 pbl_ptr;
+};
+
+/* The CQ Mode affects the CQ doorbell transaction size.
+ * 64/32 bit machines should configure to 32/16 bits respectively.
+ */
+enum qed_rdma_cq_mode {
+ QED_RDMA_CQ_MODE_16_BITS,
+ QED_RDMA_CQ_MODE_32_BITS,
+};
+
+struct qed_roce_dcqcn_params {
+ u8 notification_point;
+ u8 reaction_point;
+
+ /* fields for notification point */
+ u32 cnp_send_timeout;
+
+ /* fields for reaction point */
+ u32 rl_bc_rate;
+ u16 rl_max_rate;
+ u16 rl_r_ai;
+ u16 rl_r_hai;
+ u16 dcqcn_g;
+ u32 dcqcn_k_us;
+ u32 dcqcn_timeout_us;
+};
+
+struct qed_rdma_start_in_params {
+ struct qed_rdma_events *events;
+ struct qed_rdma_cnq_params cnq_pbl_list[128];
+ u8 desired_cnq;
+ enum qed_rdma_cq_mode cq_mode;
+ struct qed_roce_dcqcn_params dcqcn_params;
+ u16 max_mtu;
+ u8 mac_addr[ETH_ALEN];
+ u8 iwarp_flags;
+};
+
+struct qed_rdma_add_user_out_params {
+ u16 dpi;
+ u64 dpi_addr;
+ u64 dpi_phys_addr;
+ u32 dpi_size;
+};
+
+enum roce_mode {
+ ROCE_V1,
+ ROCE_V2_IPV4,
+ ROCE_V2_IPV6,
+ MAX_ROCE_MODE
+};
+
+union qed_gid {
+ u8 bytes[16];
+ u16 words[8];
+ u32 dwords[4];
+ u64 qwords[2];
+ u32 ipv4_addr;
+};
+
+struct qed_rdma_register_tid_in_params {
+ u32 itid;
+ enum qed_rdma_tid_type tid_type;
+ u8 key;
+ u16 pd;
+ bool local_read;
+ bool local_write;
+ bool remote_read;
+ bool remote_write;
+ bool remote_atomic;
+ bool mw_bind;
+ u64 pbl_ptr;
+ bool pbl_two_level;
+ u8 pbl_page_size_log;
+ u8 page_size_log;
+ u32 fbo;
+ u64 length;
+ u64 vaddr;
+ bool zbva;
+ bool phy_mr;
+ bool dma_mr;
+
+ bool dif_enabled;
+ u64 dif_error_addr;
+ u64 dif_runt_addr;
+};
+
+struct qed_rdma_create_srq_in_params {
+ u64 pbl_base_addr;
+ u64 prod_pair_addr;
+ u16 num_pages;
+ u16 pd_id;
+ u16 page_size;
+};
+
+struct qed_rdma_create_srq_out_params {
+ u16 srq_id;
+};
+
+struct qed_rdma_destroy_srq_in_params {
+ u16 srq_id;
+};
+
+struct qed_rdma_modify_srq_in_params {
+ u32 wqe_limit;
+ u16 srq_id;
+};
+
+struct qed_rdma_stats_out_params {
+ u64 sent_bytes;
+ u64 sent_pkts;
+ u64 rcv_bytes;
+ u64 rcv_pkts;
+};
+
+struct qed_rdma_counters_out_params {
+ u64 pd_count;
+ u64 max_pd;
+ u64 dpi_count;
+ u64 max_dpi;
+ u64 cq_count;
+ u64 max_cq;
+ u64 qp_count;
+ u64 max_qp;
+ u64 tid_count;
+ u64 max_tid;
+};
+
+#define QED_ROCE_TX_HEAD_FAILURE (1)
+#define QED_ROCE_TX_FRAG_FAILURE (2)
+
+enum qed_rdma_type {
+ QED_RDMA_TYPE_ROCE,
+};
+
+struct qed_dev_rdma_info {
+ struct qed_dev_info common;
+ enum qed_rdma_type rdma_type;
+};
+
+struct qed_rdma_ops {
+ const struct qed_common_ops *common;
+
+ int (*fill_dev_info)(struct qed_dev *cdev,
+ struct qed_dev_rdma_info *info);
+ void *(*rdma_get_rdma_ctx)(struct qed_dev *cdev);
+
+ int (*rdma_init)(struct qed_dev *dev,
+ struct qed_rdma_start_in_params *iparams);
+
+ int (*rdma_add_user)(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *oparams);
+
+ void (*rdma_remove_user)(void *rdma_cxt, u16 dpi);
+ int (*rdma_stop)(void *rdma_cxt);
+ struct qed_rdma_device* (*rdma_query_device)(void *rdma_cxt);
+ int (*rdma_get_start_sb)(struct qed_dev *cdev);
+ int (*rdma_get_min_cnq_msix)(struct qed_dev *cdev);
+ void (*rdma_cnq_prod_update)(void *rdma_cxt, u8 cnq_index, u16 prod);
+ int (*rdma_get_rdma_int)(struct qed_dev *cdev,
+ struct qed_int_info *info);
+ int (*rdma_set_rdma_int)(struct qed_dev *cdev, u16 cnt);
+};
+
+const struct qed_rdma_ops *qed_get_rdma_ops(void);
+
+#endif
--
1.8.3.1
^ permalink raw reply related
* [PATCH 02/11] qedr: Add RoCE driver framework
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem, dledford
Cc: Ariel.Elior, Michal.Kalderon, Yuval.Mintz, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani@cavium.com>
Adds a skeletal implementation of the qed* RoCE driver -
basically the ability to communicate with the qede driver and
receive notifications from it regarding various init/exit events.
Signed-off-by: Rajesh Borundia <rajesh.borundia@cavium.com>
Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
---
drivers/infiniband/Kconfig | 2 +
drivers/infiniband/hw/Makefile | 1 +
drivers/infiniband/hw/qedr/Kconfig | 7 +
drivers/infiniband/hw/qedr/Makefile | 3 +
drivers/infiniband/hw/qedr/main.c | 254 ++++++++++++++++++++++
drivers/infiniband/hw/qedr/qedr.h | 64 ++++++
drivers/net/ethernet/qlogic/qede/Makefile | 1 +
drivers/net/ethernet/qlogic/qede/qede.h | 9 +
drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 +++++++++++++++++++++++++++
include/linux/qed/qed_if.h | 3 +-
include/linux/qed/qede_roce.h | 88 ++++++++
include/uapi/linux/pci_regs.h | 3 +
13 files changed, 768 insertions(+), 11 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/Kconfig
create mode 100644 drivers/infiniband/hw/qedr/Makefile
create mode 100644 drivers/infiniband/hw/qedr/main.c
create mode 100644 drivers/infiniband/hw/qedr/qedr.h
create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
create mode 100644 include/linux/qed/qede_roce.h
diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index e9b7dc0..77ab0f3 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -88,4 +88,6 @@ source "drivers/infiniband/sw/rxe/Kconfig"
source "drivers/infiniband/hw/hfi1/Kconfig"
+source "drivers/infiniband/hw/qedr/Kconfig"
+
endif # INFINIBAND
diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile
index c0c7cf8..c51e464 100644
--- a/drivers/infiniband/hw/Makefile
+++ b/drivers/infiniband/hw/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_INFINIBAND_NES) += nes/
obj-$(CONFIG_INFINIBAND_OCRDMA) += ocrdma/
obj-$(CONFIG_INFINIBAND_USNIC) += usnic/
obj-$(CONFIG_INFINIBAND_HFI1) += hfi1/
+obj-$(CONFIG_INFINIBAND_QEDR) += qedr/
diff --git a/drivers/infiniband/hw/qedr/Kconfig b/drivers/infiniband/hw/qedr/Kconfig
new file mode 100644
index 0000000..58b584f
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/Kconfig
@@ -0,0 +1,7 @@
+config INFINIBAND_QEDR
+ tristate "QLogic RoCE driver"
+ depends on NETDEVICES && ETHERNET && PCI && 64BIT && QEDE
+ select QED_LL2
+ ---help---
+ This driver provides low-level InfiniBand over Ethernet
+ support for QLogic QED host channel adapters (HCAs).
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
new file mode 100644
index 0000000..3a5b7a2
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
+
+qedr-y := main.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
new file mode 100644
index 0000000..7387d02
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -0,0 +1,254 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/module.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_addr.h>
+#include <linux/netdevice.h>
+#include <linux/iommu.h>
+#include <net/addrconf.h>
+#include <linux/qed/qede_roce.h>
+#include "qedr.h"
+
+MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
+MODULE_AUTHOR("QLogic Corporation");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(QEDR_MODULE_VERSION);
+
+void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
+ enum ib_event_type type)
+{
+ struct ib_event ibev;
+
+ ibev.device = &dev->ibdev;
+ ibev.element.port_num = port_num;
+ ibev.event = type;
+
+ ib_dispatch_event(&ibev);
+}
+
+static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
+ u8 port_num)
+{
+ return IB_LINK_LAYER_ETHERNET;
+}
+
+static int qedr_register_device(struct qedr_dev *dev)
+{
+ strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
+
+ memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
+ dev->ibdev.owner = THIS_MODULE;
+
+ dev->ibdev.get_link_layer = qedr_link_layer;
+
+ return 0;
+}
+
+/* QEDR sysfs interface */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+ char *buf)
+{
+ struct qedr_dev *dev = dev_get_drvdata(device);
+
+ return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
+}
+
+static ssize_t show_hca_type(struct device *device,
+ struct device_attribute *attr, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
+}
+
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
+
+static struct device_attribute *qedr_attributes[] = {
+ &dev_attr_hw_rev,
+ &dev_attr_hca_type
+};
+
+static void qedr_remove_sysfiles(struct qedr_dev *dev)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
+ device_remove_file(&dev->ibdev.dev, qedr_attributes[i]);
+}
+
+static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
+{
+ struct pci_dev *bridge;
+ u32 val;
+
+ dev->atomic_cap = IB_ATOMIC_NONE;
+
+ bridge = pdev->bus->self;
+ if (!bridge)
+ return;
+
+ /* Check whether we are connected directly or via a switch */
+ while (bridge && bridge->bus->parent) {
+ DP_DEBUG(dev, QEDR_MSG_INIT,
+ "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n",
+ bridge->bus->number, bridge->bus->primary);
+ /* Need to check Atomic Op Routing Supported all the way to
+ * root complex.
+ */
+ pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
+ if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) {
+ pcie_capability_clear_word(pdev,
+ PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ return;
+ }
+ bridge = bridge->bus->parent->self;
+ }
+ bridge = pdev->bus->self;
+
+ /* according to bridge capability */
+ pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
+ if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) {
+ pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ dev->atomic_cap = IB_ATOMIC_GLOB;
+ } else {
+ pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ }
+}
+
+static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
+ struct net_device *ndev)
+{
+ struct qedr_dev *dev;
+ int rc = 0, i;
+
+ dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
+ if (!dev) {
+ pr_err("Unable to allocate ib device\n");
+ return NULL;
+ }
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
+
+ dev->pdev = pdev;
+ dev->ndev = ndev;
+ dev->cdev = cdev;
+
+ qedr_pci_set_atomic(dev, pdev);
+
+ rc = qedr_register_device(dev);
+ if (rc) {
+ DP_ERR(dev, "Unable to allocate register device\n");
+ goto init_err;
+ }
+
+ for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
+ if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
+ goto init_err;
+
+ DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
+ return dev;
+
+init_err:
+ ib_dealloc_device(&dev->ibdev);
+ DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
+
+ return NULL;
+}
+
+static void qedr_remove(struct qedr_dev *dev)
+{
+ /* First unregister with stack to stop all the active traffic
+ * of the registered clients.
+ */
+ qedr_remove_sysfiles(dev);
+
+ ib_dealloc_device(&dev->ibdev);
+}
+
+static int qedr_close(struct qedr_dev *dev)
+{
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
+
+ return 0;
+}
+
+static void qedr_shutdown(struct qedr_dev *dev)
+{
+ qedr_close(dev);
+ qedr_remove(dev);
+}
+
+/* event handling via NIC driver ensures that all the NIC specific
+ * initialization done before RoCE driver notifies
+ * event to stack.
+ */
+static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event)
+{
+ switch (event) {
+ case QEDE_UP:
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
+ break;
+ case QEDE_DOWN:
+ qedr_close(dev);
+ break;
+ case QEDE_CLOSE:
+ qedr_shutdown(dev);
+ break;
+ case QEDE_CHANGE_ADDR:
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+ break;
+ default:
+ pr_err("Event not supported\n");
+ }
+}
+
+static struct qedr_driver qedr_drv = {
+ .name = "qedr_driver",
+ .add = qedr_add,
+ .remove = qedr_remove,
+ .notify = qedr_notify,
+};
+
+static int __init qedr_init_module(void)
+{
+ return qede_roce_register_driver(&qedr_drv);
+}
+
+static void __exit qedr_exit_module(void)
+{
+ qede_roce_unregister_driver(&qedr_drv);
+}
+
+module_init(qedr_init_module);
+module_exit(qedr_exit_module);
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
new file mode 100644
index 0000000..5f85903
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -0,0 +1,64 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QEDR_H__
+#define __QEDR_H__
+
+#include <linux/pci.h>
+#include <rdma/ib_addr.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qede_roce.h>
+
+#define QEDR_MODULE_VERSION "8.10.10.0"
+#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
+#define DP_NAME(dev) ((dev)->ibdev.name)
+
+#define DP_DEBUG(dev, module, fmt, ...) \
+do { \
+ pr_debug("(%s) " module ": " fmt, \
+ DP_NAME(dev) ? DP_NAME(dev) : "", ## __VA_ARGS__); \
+} while (0)
+
+#define QEDR_MSG_INIT "INIT"
+
+
+struct qedr_dev {
+ struct ib_device ibdev;
+ struct qed_dev *cdev;
+ struct pci_dev *pdev;
+ struct net_device *ndev;
+
+ enum ib_atomic_cap atomic_cap;
+
+ u32 dp_module;
+ u8 dp_level;
+};
+#endif
diff --git a/drivers/net/ethernet/qlogic/qede/Makefile b/drivers/net/ethernet/qlogic/qede/Makefile
index 74a4985..28dc589 100644
--- a/drivers/net/ethernet/qlogic/qede/Makefile
+++ b/drivers/net/ethernet/qlogic/qede/Makefile
@@ -2,3 +2,4 @@ obj-$(CONFIG_QEDE) := qede.o
qede-y := qede_main.o qede_ethtool.o
qede-$(CONFIG_DCB) += qede_dcbnl.o
+qede-$(CONFIG_INFINIBAND_QEDR) += qede_roce.o
diff --git a/drivers/net/ethernet/qlogic/qede/qede.h b/drivers/net/ethernet/qlogic/qede/qede.h
index e01adce..28c0e9f 100644
--- a/drivers/net/ethernet/qlogic/qede/qede.h
+++ b/drivers/net/ethernet/qlogic/qede/qede.h
@@ -106,6 +106,13 @@ struct qede_vlan {
bool configured;
};
+struct qede_rdma_dev {
+ struct qedr_dev *qedr_dev;
+ struct list_head entry;
+ struct list_head roce_event_list;
+ struct workqueue_struct *roce_wq;
+};
+
struct qede_dev {
struct qed_dev *cdev;
struct net_device *ndev;
@@ -185,6 +192,8 @@ struct qede_dev {
unsigned long sp_flags;
u16 vxlan_dst_port;
u16 geneve_dst_port;
+
+ struct qede_rdma_dev rdma_info;
};
enum QEDE_STATE {
diff --git a/drivers/net/ethernet/qlogic/qede/qede_main.c b/drivers/net/ethernet/qlogic/qede/qede_main.c
index cd23a29..37f3a77 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_main.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_main.c
@@ -36,7 +36,7 @@
#include <linux/random.h>
#include <net/ip6_checksum.h>
#include <linux/bitops.h>
-
+#include <linux/qed/qede_roce.h>
#include "qede.h"
static char version[] =
@@ -189,8 +189,7 @@ static int qede_netdev_event(struct notifier_block *this, unsigned long event,
struct ethtool_drvinfo drvinfo;
struct qede_dev *edev;
- /* Currently only support name change */
- if (event != NETDEV_CHANGENAME)
+ if (event != NETDEV_CHANGENAME && event != NETDEV_CHANGEADDR)
goto done;
/* Check whether this is a qede device */
@@ -203,11 +202,18 @@ static int qede_netdev_event(struct notifier_block *this, unsigned long event,
goto done;
edev = netdev_priv(ndev);
- /* Notify qed of the name change */
- if (!edev->ops || !edev->ops->common)
- goto done;
- edev->ops->common->set_id(edev->cdev, edev->ndev->name,
- "qede");
+ switch (event) {
+ case NETDEV_CHANGENAME:
+ /* Notify qed of the name change */
+ if (!edev->ops || !edev->ops->common)
+ goto done;
+ edev->ops->common->set_id(edev->cdev, edev->ndev->name, "qede");
+ break;
+ case NETDEV_CHANGEADDR:
+ edev = netdev_priv(ndev);
+ qede_roce_event_changeaddr(edev);
+ break;
+ }
done:
return NOTIFY_DONE;
@@ -2541,10 +2547,14 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
qede_init_ndev(edev);
+ rc = qede_roce_dev_add(edev);
+ if (rc)
+ goto err3;
+
rc = register_netdev(edev->ndev);
if (rc) {
DP_NOTICE(edev, "Cannot register net-device\n");
- goto err3;
+ goto err4;
}
edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
@@ -2564,6 +2574,8 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
return 0;
+err4:
+ qede_roce_dev_remove(edev);
err3:
free_netdev(edev->ndev);
err2:
@@ -2610,8 +2622,11 @@ static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
DP_INFO(edev, "Starting qede_remove\n");
cancel_delayed_work_sync(&edev->sp_task);
+
unregister_netdev(ndev);
+ qede_roce_dev_remove(edev);
+
edev->ops->common->set_power_state(cdev, PCI_D0);
pci_set_drvdata(pdev, NULL);
@@ -3508,6 +3523,7 @@ static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
DP_INFO(edev, "Starting qede unload\n");
+ qede_roce_dev_event_close(edev);
mutex_lock(&edev->qede_lock);
edev->state = QEDE_STATE_CLOSED;
@@ -3608,6 +3624,7 @@ static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
/* Query whether link is already-up */
memset(&link_output, 0, sizeof(link_output));
edev->ops->common->get_link(edev->cdev, &link_output);
+ qede_roce_dev_event_open(edev);
qede_link_update(edev, &link_output);
DP_INFO(edev, "Ending successfully qede load\n");
diff --git a/drivers/net/ethernet/qlogic/qede/qede_roce.c b/drivers/net/ethernet/qlogic/qede/qede_roce.c
new file mode 100644
index 0000000..3d12a44
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qede/qede_roce.c
@@ -0,0 +1,309 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/qed/qede_roce.h>
+#include "qede.h"
+
+static struct qedr_driver *qedr_drv;
+static LIST_HEAD(qedr_dev_list);
+static DEFINE_MUTEX(qedr_dev_list_lock);
+
+bool qede_roce_supported(struct qede_dev *dev)
+{
+ return dev->dev_info.common.rdma_supported;
+}
+
+static void _qede_roce_dev_add(struct qede_dev *edev)
+{
+ if (!qedr_drv)
+ return;
+
+ edev->rdma_info.qedr_dev = qedr_drv->add(edev->cdev, edev->pdev,
+ edev->ndev);
+}
+
+static int qede_roce_create_wq(struct qede_dev *edev)
+{
+ INIT_LIST_HEAD(&edev->rdma_info.roce_event_list);
+ edev->rdma_info.roce_wq = create_singlethread_workqueue("roce_wq");
+ if (!edev->rdma_info.roce_wq) {
+ pr_err("qedr: Could not create workqueue\n");
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+static void qede_roce_cleanup_event(struct qede_dev *edev)
+{
+ struct list_head *head = &edev->rdma_info.roce_event_list;
+ struct qede_roce_event_work *event_node;
+
+ flush_workqueue(edev->rdma_info.roce_wq);
+ while (!list_empty(head)) {
+ event_node = list_entry(head->next, struct qede_roce_event_work,
+ list);
+ cancel_work_sync(&event_node->work);
+ list_del(&event_node->list);
+ kfree(event_node);
+ }
+}
+
+static void qede_roce_destroy_wq(struct qede_dev *edev)
+{
+ qede_roce_cleanup_event(edev);
+ destroy_workqueue(edev->rdma_info.roce_wq);
+}
+
+int qede_roce_dev_add(struct qede_dev *edev)
+{
+ int rc = 0;
+
+ if (qede_roce_supported(edev)) {
+ rc = qede_roce_create_wq(edev);
+ if (rc)
+ return rc;
+
+ INIT_LIST_HEAD(&edev->rdma_info.entry);
+ mutex_lock(&qedr_dev_list_lock);
+ list_add_tail(&edev->rdma_info.entry, &qedr_dev_list);
+ _qede_roce_dev_add(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+ }
+
+ return rc;
+}
+
+static void _qede_roce_dev_remove(struct qede_dev *edev)
+{
+ if (qedr_drv && qedr_drv->remove && edev->rdma_info.qedr_dev)
+ qedr_drv->remove(edev->rdma_info.qedr_dev);
+ edev->rdma_info.qedr_dev = NULL;
+}
+
+void qede_roce_dev_remove(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ qede_roce_destroy_wq(edev);
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_remove(edev);
+ list_del(&edev->rdma_info.entry);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+static void _qede_roce_dev_open(struct qede_dev *edev)
+{
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_UP);
+}
+
+void qede_roce_dev_open(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_open(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+static void _qede_roce_dev_close(struct qede_dev *edev)
+{
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_DOWN);
+}
+
+void qede_roce_dev_close(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_close(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+void qede_roce_dev_shutdown(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_CLOSE);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+int qede_roce_register_driver(struct qedr_driver *drv)
+{
+ struct qede_dev *edev;
+ u8 qedr_counter = 0;
+
+ mutex_lock(&qedr_dev_list_lock);
+ if (qedr_drv) {
+ mutex_unlock(&qedr_dev_list_lock);
+ return -EINVAL;
+ }
+ qedr_drv = drv;
+
+ list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
+ struct net_device *netdev;
+
+ qedr_counter++;
+ _qede_roce_dev_add(edev);
+ netdev = edev->ndev;
+ if (netif_running(netdev) && netif_oper_up(netdev))
+ _qede_roce_dev_open(edev);
+ }
+ mutex_unlock(&qedr_dev_list_lock);
+
+ pr_err("qedr: discovered and registered %d RoCE funcs\n", qedr_counter);
+
+ return 0;
+}
+EXPORT_SYMBOL(qede_roce_register_driver);
+
+void qede_roce_unregister_driver(struct qedr_driver *drv)
+{
+ struct qede_dev *edev;
+
+ mutex_lock(&qedr_dev_list_lock);
+ list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
+ if (edev->rdma_info.qedr_dev)
+ _qede_roce_dev_remove(edev);
+ }
+ qedr_drv = NULL;
+ mutex_unlock(&qedr_dev_list_lock);
+}
+EXPORT_SYMBOL(qede_roce_unregister_driver);
+
+void qede_roce_changeaddr(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_CHANGE_ADDR);
+}
+
+struct qede_roce_event_work *qede_roce_get_free_event_node(struct qede_dev
+ *edev)
+{
+ struct qede_roce_event_work *event_node = NULL;
+ struct list_head *list_node = NULL;
+ bool found = false;
+
+ list_for_each(list_node, &edev->rdma_info.roce_event_list) {
+ event_node = list_entry(list_node, struct qede_roce_event_work,
+ list);
+ if (!work_pending(&event_node->work)) {
+ found = true;
+ break;
+ }
+ }
+
+ if (!found) {
+ event_node = kzalloc(sizeof(*event_node), GFP_KERNEL);
+ if (!event_node) {
+ pr_err("qedr: Could not allocate memory for roce work\n");
+ return NULL;
+ }
+ list_add_tail(&event_node->list,
+ &edev->rdma_info.roce_event_list);
+ }
+
+ return event_node;
+}
+
+static void qede_roce_handle_event(struct work_struct *work)
+{
+ struct qede_roce_event_work *event_node;
+ enum qede_roce_event event;
+ struct qede_dev *edev;
+
+ event_node = container_of(work, struct qede_roce_event_work, work);
+ event = event_node->event;
+ edev = event_node->ptr;
+
+ switch (event) {
+ case QEDE_UP:
+ qede_roce_dev_open(edev);
+ break;
+ case QEDE_DOWN:
+ qede_roce_dev_close(edev);
+ break;
+ case QEDE_CLOSE:
+ qede_roce_dev_shutdown(edev);
+ break;
+ case QEDE_CHANGE_ADDR:
+ qede_roce_changeaddr(edev);
+ break;
+ default:
+ pr_err("qede: Invalid roce event %d", event);
+ }
+}
+
+static void qede_roce_add_event(struct qede_dev *edev,
+ enum qede_roce_event event)
+{
+ struct qede_roce_event_work *event_node;
+
+ event_node = qede_roce_get_free_event_node(edev);
+ if (!event_node)
+ return;
+
+ event_node->event = event;
+ event_node->ptr = edev;
+
+ INIT_WORK(&event_node->work, qede_roce_handle_event);
+ queue_work(edev->rdma_info.roce_wq, &event_node->work);
+}
+
+void qede_roce_dev_event_open(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_UP);
+}
+
+void qede_roce_dev_event_close(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_DOWN);
+}
+
+void qede_roce_event_changeaddr(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_CHANGE_ADDR);
+}
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index c2d74e8..e313742 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -260,11 +260,10 @@ struct qed_dev_info {
/* MFW version */
u32 mfw_rev;
- bool rdma_supported;
-
u32 flash_size;
u8 mf_mode;
bool tx_switching;
+ bool rdma_supported;
};
enum qed_sb_type {
diff --git a/include/linux/qed/qede_roce.h b/include/linux/qed/qede_roce.h
new file mode 100644
index 0000000..fd7acc1
--- /dev/null
+++ b/include/linux/qed/qede_roce.h
@@ -0,0 +1,88 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef QEDE_ROCE_H
+#define QEDE_ROCE_H
+
+struct qedr_dev;
+struct qed_dev;
+struct qede_dev;
+
+enum qede_roce_event {
+ QEDE_UP,
+ QEDE_DOWN,
+ QEDE_CHANGE_ADDR,
+ QEDE_CLOSE
+};
+
+struct qede_roce_event_work {
+ struct list_head list;
+ struct work_struct work;
+ void *ptr;
+ enum qede_roce_event event;
+};
+
+struct qedr_driver {
+ unsigned char name[32];
+
+ struct qedr_dev* (*add)(struct qed_dev *, struct pci_dev *,
+ struct net_device *);
+
+ void (*remove)(struct qedr_dev *);
+ void (*notify)(struct qedr_dev *, enum qede_roce_event);
+};
+
+/* APIs for RoCE driver to register callback handlers,
+ * which will be invoked when device is added, removed, ifup, ifdown
+ */
+int qede_roce_register_driver(struct qedr_driver *drv);
+void qede_roce_unregister_driver(struct qedr_driver *drv);
+
+bool qede_roce_supported(struct qede_dev *dev);
+
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+int qede_roce_dev_add(struct qede_dev *dev);
+void qede_roce_dev_event_open(struct qede_dev *dev);
+void qede_roce_dev_event_close(struct qede_dev *dev);
+void qede_roce_dev_remove(struct qede_dev *dev);
+void qede_roce_event_changeaddr(struct qede_dev *qedr);
+#else
+static int qede_roce_dev_add(struct qede_dev *dev)
+{
+ return 0;
+}
+
+static void qede_roce_dev_event_open(struct qede_dev *dev) {}
+static void qede_roce_dev_event_close(struct qede_dev *dev) {}
+static void qede_roce_dev_remove(struct qede_dev *dev) {}
+static void qede_roce_event_changeaddr(struct qede_dev *qedr) {}
+#endif
+#endif
diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
index 4040951..3973000 100644
--- a/include/uapi/linux/pci_regs.h
+++ b/include/uapi/linux/pci_regs.h
@@ -612,6 +612,8 @@
*/
#define PCI_EXP_DEVCAP2 36 /* Device Capabilities 2 */
#define PCI_EXP_DEVCAP2_ARI 0x00000020 /* Alternative Routing-ID */
+#define PCI_EXP_DEVCAP2_ATOMIC_ROUTE 0x00000040 /* Atomic Op routing */
+#define PCI_EXP_DEVCAP2_ATOMIC_COMP64 0x00000100 /* Atomic 64-bit comparison */
#define PCI_EXP_DEVCAP2_LTR 0x00000800 /* Latency tolerance reporting */
#define PCI_EXP_DEVCAP2_OBFF_MASK 0x000c0000 /* OBFF support mechanism */
#define PCI_EXP_DEVCAP2_OBFF_MSG 0x00040000 /* New message signaling */
@@ -619,6 +621,7 @@
#define PCI_EXP_DEVCTL2 40 /* Device Control 2 */
#define PCI_EXP_DEVCTL2_COMP_TIMEOUT 0x000f /* Completion Timeout Value */
#define PCI_EXP_DEVCTL2_ARI 0x0020 /* Alternative Routing-ID */
+#define PCI_EXP_DEVCTL2_ATOMIC_REQ 0x0040 /* Set Atomic requests */
#define PCI_EXP_DEVCTL2_IDO_REQ_EN 0x0100 /* Allow IDO for requests */
#define PCI_EXP_DEVCTL2_IDO_CMP_EN 0x0200 /* Allow IDO for completions */
#define PCI_EXP_DEVCTL2_LTR_EN 0x0400 /* Enable LTR mechanism */
--
1.8.3.1
^ permalink raw reply related
* [PATCH 10/11] qedr: Add GSI support
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem-fT/PcQaiUtIeIZ0/mPfg9Q, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: Ariel.Elior-YGCgFSpz5w/QT0dZR+AlfA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA,
Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA,
rajesh.borundia-YGCgFSpz5w/QT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Add support for GSI over light L2.
Signed-off-by: Rajesh Borundia <rajesh.borundia-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ram Amrani <Ram.Amrani-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/qedr/Makefile | 2 +-
drivers/infiniband/hw/qedr/main.c | 3 +
drivers/infiniband/hw/qedr/qedr.h | 15 +
drivers/infiniband/hw/qedr/qedr_cm.c | 636 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/qedr_cm.h | 21 +
drivers/infiniband/hw/qedr/verbs.c | 39 ++
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 95 ++++-
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 22 +-
drivers/net/ethernet/qlogic/qed/qed_roce.c | 66 +++
9 files changed, 888 insertions(+), 11 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.c
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
index b10f2b1..ba7067c 100644
--- a/drivers/infiniband/hw/qedr/Makefile
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
-qedr-y := main.o verbs.o
+qedr-y := main.o verbs.o qedr_cm.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 8ed1503..735bc70 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -138,6 +138,9 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.query_pkey = qedr_query_pkey;
+ dev->ibdev.create_ah = qedr_create_ah;
+ dev->ibdev.destroy_ah = qedr_destroy_ah;
+
dev->ibdev.get_dma_mr = qedr_get_dma_mr;
dev->ibdev.dereg_mr = qedr_dereg_mr;
dev->ibdev.reg_user_mr = qedr_reg_user_mr;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index bdfc599..5e4081a 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -57,6 +57,7 @@ do { \
#define QEDR_MSG_RQ " RQ"
#define QEDR_MSG_SQ " SQ"
#define QEDR_MSG_QP " QP"
+#define QEDR_MSG_GSI " GSI"
#define QEDR_CQ_MAGIC_NUMBER (0x11223344)
@@ -150,6 +151,10 @@ struct qedr_dev {
u8 num_hwfns;
uint wq_multiplier;
u8 gsi_ll2_mac_address[ETH_ALEN];
+ int gsi_qp_created;
+ struct qedr_cq *gsi_sqcq;
+ struct qedr_cq *gsi_rqcq;
+ struct qedr_qp *gsi_qp;
};
#define QEDR_MAX_SQ_PBL (0x8000)
@@ -248,6 +253,9 @@ struct qedr_cq {
u16 icid;
+ /* Lock to protect completion handler */
+ spinlock_t comp_handler_lock;
+
/* Lock to protect multiplem CQ's */
spinlock_t cq_lock;
u8 arm_flags;
@@ -294,6 +302,7 @@ struct qedr_qp_hwq_info {
u16 prod;
u16 cons;
u16 wqe_cons;
+ u16 gsi_cons;
u16 max_wr;
/* DB */
@@ -368,6 +377,7 @@ struct qedr_qp {
struct ib_sge sg_list[RDMA_MAX_SGE_PER_RQ_WQE];
u8 wqe_size;
+ u8 smac[ETH_ALEN];
u16 vlan_id;
int rc;
} *rqe_wr_id;
@@ -476,6 +486,11 @@ static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp)
return container_of(ibqp, struct qedr_qp, ibqp);
}
+static inline struct qedr_ah *get_qedr_ah(struct ib_ah *ibah)
+{
+ return container_of(ibah, struct qedr_ah, ibah);
+}
+
static inline struct qedr_mr *get_qedr_mr(struct ib_mr *ibmr)
{
return container_of(ibmr, struct qedr_mr, ibmr);
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.c b/drivers/infiniband/hw/qedr/qedr_cm.c
new file mode 100644
index 0000000..fe4874c
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_cm.c
@@ -0,0 +1,636 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/crc32.h>
+#include <linux/iommu.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/udp.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include "qedr_hsi.h"
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qedr.h"
+#include "qedr_hsi.h"
+#include "verbs.h"
+#include <rdma/qedr-abi.h>
+#include "qedr_hsi.h"
+#include "qedr_cm.h"
+
+void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info)
+{
+ info->gsi_cons = (info->gsi_cons + 1) % info->max_wr;
+}
+
+void qedr_store_gsi_qp_cq(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ dev->gsi_qp_created = 1;
+ dev->gsi_sqcq = get_qedr_cq(attrs->send_cq);
+ dev->gsi_rqcq = get_qedr_cq(attrs->recv_cq);
+ dev->gsi_qp = qp;
+}
+
+void qedr_ll2_tx_cb(void *_qdev, struct qed_roce_ll2_packet *pkt)
+{
+ struct qedr_dev *dev = (struct qedr_dev *)_qdev;
+ struct qedr_cq *cq = dev->gsi_sqcq;
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+
+ DP_DEBUG(dev, QEDR_MSG_GSI,
+ "LL2 TX CB: gsi_sqcq=%p, gsi_rqcq=%p, gsi_cons=%d, ibcq_comp=%s\n",
+ dev->gsi_sqcq, dev->gsi_rqcq, qp->sq.gsi_cons,
+ cq->ibcq.comp_handler ? "Yes" : "No");
+
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len, pkt->header.vaddr,
+ pkt->header.baddr);
+ kfree(pkt);
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+ qedr_inc_sw_gsi_cons(&qp->sq);
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (cq->ibcq.comp_handler) {
+ spin_lock_irqsave(&cq->comp_handler_lock, flags);
+ (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+ spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+ }
+}
+
+static inline void qedr_print_hex_dump(const char *str, void *buf, size_t size)
+{
+ print_hex_dump(KERN_INFO, str, DUMP_PREFIX_OFFSET, 32, 1, buf,
+ size, false);
+}
+
+void qedr_ll2_rx_cb(void *_dev, struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_rx_params *params)
+{
+ struct qedr_dev *dev = (struct qedr_dev *)_dev;
+ struct qedr_cq *cq = dev->gsi_rqcq;
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ qp->rqe_wr_id[qp->rq.gsi_cons].rc = params->rc;
+ qp->rqe_wr_id[qp->rq.gsi_cons].vlan_id = params->vlan_id;
+ qp->rqe_wr_id[qp->rq.gsi_cons].sg_list[0].length = pkt->payload[0].len;
+ ether_addr_copy(qp->rqe_wr_id[qp->rq.gsi_cons].smac, params->smac);
+
+ qedr_inc_sw_gsi_cons(&qp->rq);
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (cq->ibcq.comp_handler) {
+ spin_lock_irqsave(&cq->comp_handler_lock, flags);
+ (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+ spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+ }
+}
+
+static void qedr_destroy_gsi_cq(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ struct qed_rdma_destroy_cq_in_params iparams;
+ struct qed_rdma_destroy_cq_out_params oparams;
+ struct qedr_cq *cq;
+
+ cq = get_qedr_cq(attrs->send_cq);
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+
+ cq = get_qedr_cq(attrs->recv_cq);
+ /* if a dedicated recv_cq was used, delete it too */
+ if (iparams.icid != cq->icid) {
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+ }
+}
+
+static inline int qedr_check_gsi_qp_attrs(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ if (attrs->cap.max_recv_sge > QEDR_GSI_MAX_RECV_SGE) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_recv_sge is larger the max %d>%d\n",
+ attrs->cap.max_recv_sge, QEDR_GSI_MAX_RECV_SGE);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_recv_wr > QEDR_GSI_MAX_RECV_WR) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_recv_wr is too large %d>%d\n",
+ attrs->cap.max_recv_wr, QEDR_GSI_MAX_RECV_WR);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_wr > QEDR_GSI_MAX_SEND_WR) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_send_wr is too large %d>%d\n",
+ attrs->cap.max_send_wr, QEDR_GSI_MAX_SEND_WR);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs,
+ struct qedr_qp *qp)
+{
+ struct qed_roce_ll2_params ll2_params;
+ int rc;
+
+ rc = qedr_check_gsi_qp_attrs(dev, attrs);
+ if (rc)
+ return ERR_PTR(rc);
+
+ /* configure and start LL2 */
+ memset(&ll2_params, 0, sizeof(ll2_params));
+ ll2_params.max_tx_buffers = attrs->cap.max_send_wr;
+ ll2_params.max_rx_buffers = attrs->cap.max_recv_wr;
+ ll2_params.cbs.tx_cb = qedr_ll2_tx_cb;
+ ll2_params.cbs.rx_cb = qedr_ll2_rx_cb;
+ ll2_params.cb_cookie = (void *)dev;
+ ll2_params.mtu = dev->ndev->mtu;
+ ether_addr_copy(ll2_params.mac_address, dev->ndev->dev_addr);
+ rc = dev->ops->roce_ll2_start(dev->cdev, &ll2_params);
+ if (rc) {
+ DP_ERR(dev, "create gsi qp: failed on ll2 start. rc=%d\n", rc);
+ return ERR_PTR(rc);
+ }
+
+ /* create QP */
+ qp->ibqp.qp_num = 1;
+ qp->rq.max_wr = attrs->cap.max_recv_wr;
+ qp->sq.max_wr = attrs->cap.max_send_wr;
+
+ qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->rqe_wr_id) {
+ DP_ERR(dev, "create gsi qp: failed on rqe_wr_id allocation\n");
+ goto err;
+ }
+ qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->wqe_wr_id) {
+ DP_ERR(dev, "create gsi qp: failed on wqe_wr_id allocation\n");
+ goto err;
+ }
+
+ qedr_store_gsi_qp_cq(dev, qp, attrs);
+ ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
+
+ /* the GSI CQ is handled by the driver so remove it from the FW */
+ qedr_destroy_gsi_cq(dev, attrs);
+ dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI;
+ dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI;
+
+ DP_DEBUG(dev, QEDR_MSG_GSI, "created GSI QP %p\n", qp);
+
+ return &qp->ibqp;
+
+err:
+ kfree(qp->rqe_wr_id);
+
+ rc = dev->ops->roce_ll2_stop(dev->cdev);
+ if (rc)
+ DP_ERR(dev, "create gsi qp: failed destroy on create\n");
+
+ return ERR_PTR(-ENOMEM);
+}
+
+int qedr_destroy_gsi_qp(struct qedr_dev *dev)
+{
+ int rc;
+
+ rc = dev->ops->roce_ll2_stop(dev->cdev);
+ if (rc)
+ DP_ERR(dev, "destroy gsi qp: failed (rc=%d)\n", rc);
+ else
+ DP_DEBUG(dev, QEDR_MSG_GSI, "destroy gsi qp: success\n");
+
+ return rc;
+}
+
+#define QEDR_MAX_UD_HEADER_SIZE (100)
+#define QEDR_GSI_QPN (1)
+static inline int qedr_gsi_build_header(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_send_wr *swr,
+ struct ib_ud_header *udh,
+ int *roce_mode)
+{
+ bool has_vlan = false, has_grh_ipv6 = true;
+ struct ib_ah_attr *ah_attr = &get_qedr_ah(ud_wr(swr)->ah)->attr;
+ struct ib_global_route *grh = &ah_attr->grh;
+ union ib_gid sgid;
+ int send_size = 0;
+ u16 vlan_id = 0;
+ u16 ether_type;
+ struct ib_gid_attr sgid_attr;
+ int rc;
+ int ip_ver = 0;
+
+ bool has_udp = false;
+ int i;
+
+ send_size = 0;
+ for (i = 0; i < swr->num_sge; ++i)
+ send_size += swr->sg_list[i].length;
+
+ rc = ib_get_cached_gid(qp->ibqp.device, ah_attr->port_num,
+ grh->sgid_index, &sgid, &sgid_attr);
+ if (rc) {
+ DP_ERR(dev,
+ "gsi post send: failed to get cached GID (port=%d, ix=%d)\n",
+ ah_attr->port_num, grh->sgid_index);
+ return rc;
+ }
+
+ vlan_id = rdma_vlan_dev_vlan_id(sgid_attr.ndev);
+ if (vlan_id < VLAN_CFI_MASK)
+ has_vlan = true;
+ if (sgid_attr.ndev)
+ dev_put(sgid_attr.ndev);
+
+ if (!memcmp(&sgid, &zgid, sizeof(sgid))) {
+ DP_ERR(dev, "gsi post send: GID not found GID index %d\n",
+ ah_attr->grh.sgid_index);
+ return -ENOENT;
+ }
+
+ has_udp = (sgid_attr.gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP);
+ if (!has_udp) {
+ /* RoCE v1 */
+ ether_type = ETH_P_ROCE;
+ *roce_mode = ROCE_V1;
+ } else if (ipv6_addr_v4mapped((struct in6_addr *)&sgid)) {
+ /* RoCE v2 IPv4 */
+ ip_ver = 4;
+ ether_type = ETH_P_IP;
+ has_grh_ipv6 = false;
+ *roce_mode = ROCE_V2_IPV4;
+ } else {
+ /* RoCE v2 IPv6 */
+ ip_ver = 6;
+ ether_type = ETH_P_IPV6;
+ *roce_mode = ROCE_V2_IPV6;
+ }
+
+ rc = ib_ud_header_init(send_size, false, true, has_vlan,
+ has_grh_ipv6, ip_ver, has_udp, 0, udh);
+ if (rc) {
+ DP_ERR(dev, "gsi post send: failed to init header\n");
+ return rc;
+ }
+
+ /* ENET + VLAN headers */
+ ether_addr_copy(udh->eth.dmac_h, ah_attr->dmac);
+ ether_addr_copy(udh->eth.smac_h, dev->ndev->dev_addr);
+ if (has_vlan) {
+ udh->eth.type = htons(ETH_P_8021Q);
+ udh->vlan.tag = htons(vlan_id);
+ udh->vlan.type = htons(ether_type);
+ } else {
+ udh->eth.type = htons(ether_type);
+ }
+
+ /* BTH */
+ udh->bth.solicited_event = !!(swr->send_flags & IB_SEND_SOLICITED);
+ udh->bth.pkey = QEDR_ROCE_PKEY_DEFAULT;
+ udh->bth.destination_qpn = htonl(ud_wr(swr)->remote_qpn);
+ udh->bth.psn = htonl((qp->sq_psn++) & ((1 << 24) - 1));
+ udh->bth.opcode = IB_OPCODE_UD_SEND_ONLY;
+
+ /* DETH */
+ udh->deth.qkey = htonl(0x80010000);
+ udh->deth.source_qpn = htonl(QEDR_GSI_QPN);
+
+ if (has_grh_ipv6) {
+ /* GRH / IPv6 header */
+ udh->grh.traffic_class = grh->traffic_class;
+ udh->grh.flow_label = grh->flow_label;
+ udh->grh.hop_limit = grh->hop_limit;
+ udh->grh.destination_gid = grh->dgid;
+ memcpy(&udh->grh.source_gid.raw, &sgid.raw,
+ sizeof(udh->grh.source_gid.raw));
+ } else {
+ /* IPv4 header */
+ u32 ipv4_addr;
+
+ udh->ip4.protocol = IPPROTO_UDP;
+ udh->ip4.tos = htonl(ah_attr->grh.flow_label);
+ udh->ip4.frag_off = htons(IP_DF);
+ udh->ip4.ttl = ah_attr->grh.hop_limit;
+
+ ipv4_addr = qedr_get_ipv4_from_gid(sgid.raw);
+ udh->ip4.saddr = ipv4_addr;
+ ipv4_addr = qedr_get_ipv4_from_gid(ah_attr->grh.dgid.raw);
+ udh->ip4.daddr = ipv4_addr;
+ /* note: checksum is calculated by the device */
+ }
+
+ /* UDP */
+ if (has_udp) {
+ udh->udp.sport = htons(QEDR_ROCE_V2_UDP_SPORT);
+ udh->udp.dport = htons(ROCE_V2_UDP_DPORT);
+ udh->udp.csum = 0;
+ /* UDP length is untouched hence is zero */
+ }
+ return 0;
+}
+
+static inline int qedr_gsi_build_packet(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_send_wr *swr,
+ struct qed_roce_ll2_packet **p_packet)
+{
+ u8 ud_header_buffer[QEDR_MAX_UD_HEADER_SIZE];
+ struct qed_roce_ll2_packet *packet;
+ struct pci_dev *pdev = dev->pdev;
+ int roce_mode, header_size;
+ struct ib_ud_header udh;
+ int i, rc;
+
+ *p_packet = NULL;
+
+ rc = qedr_gsi_build_header(dev, qp, swr, &udh, &roce_mode);
+ if (rc)
+ return rc;
+
+ header_size = ib_ud_header_pack(&udh, &ud_header_buffer);
+
+ packet = kzalloc(sizeof(*packet), GFP_ATOMIC);
+ if (!packet) {
+ DP_ERR(dev, "gsi post send: failed to allocate packet\n");
+ return -ENOMEM;
+ }
+
+ packet->header.vaddr = dma_alloc_coherent(&pdev->dev, header_size,
+ &packet->header.baddr,
+ GFP_ATOMIC);
+ if (!packet->header.vaddr) {
+ DP_ERR(dev,
+ "cm header build: failed to allocate buffer for header\n");
+ kfree(packet);
+ return -ENOMEM;
+ }
+
+ if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h))
+ packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW;
+ else
+ packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB;
+
+ packet->roce_mode = roce_mode;
+ memcpy(packet->header.vaddr, ud_header_buffer, header_size);
+ packet->header.len = header_size;
+ packet->n_seg = swr->num_sge;
+ for (i = 0; i < packet->n_seg; i++) {
+ packet->payload[i].baddr = swr->sg_list[i].addr;
+ packet->payload[i].len = swr->sg_list[i].length;
+ }
+
+ *p_packet = packet;
+
+ return 0;
+}
+
+int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qed_roce_ll2_packet *pkt = NULL;
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_roce_ll2_tx_params params;
+ struct qedr_dev *dev = qp->dev;
+ unsigned long flags;
+ int rc;
+
+ if (qp->state != QED_ROCE_QP_STATE_RTS) {
+ *bad_wr = wr;
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTS\n",
+ qp->state);
+ return -EINVAL;
+ }
+
+ if (wr->num_sge > RDMA_MAX_SGE_PER_SQ_WQE) {
+ DP_ERR(dev, "gsi post send: num_sge is too large (%d>%d)\n",
+ wr->num_sge, RDMA_MAX_SGE_PER_SQ_WQE);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ if (wr->opcode != IB_WR_SEND) {
+ DP_ERR(dev,
+ "gsi post send: failed due to unsupported opcode %d\n",
+ wr->opcode);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ memset(¶ms, 0, sizeof(params));
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ rc = qedr_gsi_build_packet(dev, qp, wr, &pkt);
+ if (rc) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ goto err;
+ }
+
+ rc = dev->ops->roce_ll2_tx(dev->cdev, pkt, ¶ms);
+ if (!rc) {
+ qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
+ qedr_inc_sw_prod(&qp->sq);
+ DP_DEBUG(qp->dev, QEDR_MSG_GSI,
+ "gsi post send: opcode=%d, in_irq=%ld, irqs_disabled=%d, wr_id=%llx\n",
+ wr->opcode, in_irq(), irqs_disabled(), wr->wr_id);
+ } else {
+ if (rc == QED_ROCE_TX_HEAD_FAILURE) {
+ /* TX failed while posting header - release resources */
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len,
+ pkt->header.vaddr, pkt->header.baddr);
+ kfree(pkt);
+ } else if (rc == QED_ROCE_TX_FRAG_FAILURE) {
+ /* NTD since TX failed while posting a fragment. We will
+ * release the resources on TX callback
+ */
+ }
+
+ DP_ERR(dev, "gsi post send: failed to transmit (rc=%d)\n", rc);
+ rc = -EAGAIN;
+ *bad_wr = wr;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (wr->next) {
+ DP_ERR(dev,
+ "gsi post send: failed second WR. Only one WR may be passed at a time\n");
+ *bad_wr = wr->next;
+ rc = -EINVAL;
+ }
+
+ return rc;
+
+err:
+ *bad_wr = wr;
+ return rc;
+}
+
+int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_roce_ll2_buffer buf;
+ unsigned long flags;
+ int status = 0;
+ int rc;
+
+ if ((qp->state != QED_ROCE_QP_STATE_RTR) &&
+ (qp->state != QED_ROCE_QP_STATE_RTS)) {
+ *bad_wr = wr;
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTR/S\n",
+ qp->state);
+ return -EINVAL;
+ }
+
+ memset(&buf, 0, sizeof(buf));
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ while (wr) {
+ if (wr->num_sge > QEDR_GSI_MAX_RECV_SGE) {
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. too many sges %d>%d\n",
+ wr->num_sge, QEDR_GSI_MAX_RECV_SGE);
+ goto err;
+ }
+
+ buf.baddr = wr->sg_list[0].addr;
+ buf.len = wr->sg_list[0].length;
+
+ rc = dev->ops->roce_ll2_post_rx_buffer(dev->cdev, &buf, 0, 1);
+ if (rc) {
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer (rc=%d)\n",
+ rc);
+ goto err;
+ }
+
+ memset(&qp->rqe_wr_id[qp->rq.prod], 0,
+ sizeof(qp->rqe_wr_id[qp->rq.prod]));
+ qp->rqe_wr_id[qp->rq.prod].sg_list[0] = wr->sg_list[0];
+ qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id;
+
+ qedr_inc_sw_prod(&qp->rq);
+
+ wr = wr->next;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return status;
+err:
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ return -ENOMEM;
+}
+
+int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+ int i = 0;
+
+ spin_lock_irqsave(&cq->cq_lock, flags);
+
+ while (i < num_entries && qp->rq.cons != qp->rq.gsi_cons) {
+ memset(&wc[i], 0, sizeof(*wc));
+
+ wc[i].qp = &qp->ibqp;
+ wc[i].wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+ wc[i].opcode = IB_WC_RECV;
+ wc[i].pkey_index = 0;
+ wc[i].status = (qp->rqe_wr_id[qp->rq.cons].rc) ?
+ IB_WC_GENERAL_ERR : IB_WC_SUCCESS;
+ /* 0 - currently only one recv sg is supported */
+ wc[i].byte_len = qp->rqe_wr_id[qp->rq.cons].sg_list[0].length;
+ wc[i].wc_flags |= IB_WC_GRH | IB_WC_IP_CSUM_OK;
+ ether_addr_copy(wc[i].smac, qp->rqe_wr_id[qp->rq.cons].smac);
+ wc[i].wc_flags |= IB_WC_WITH_SMAC;
+ if (qp->rqe_wr_id[qp->rq.cons].vlan_id) {
+ wc[i].wc_flags |= IB_WC_WITH_VLAN;
+ wc[i].vlan_id = qp->rqe_wr_id[qp->rq.cons].vlan_id;
+ }
+
+ qedr_inc_sw_cons(&qp->rq);
+ i++;
+ }
+
+ while (i < num_entries && qp->sq.cons != qp->sq.gsi_cons) {
+ memset(&wc[i], 0, sizeof(*wc));
+
+ wc[i].qp = &qp->ibqp;
+ wc[i].wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id;
+ wc[i].opcode = IB_WC_SEND;
+ wc[i].status = IB_WC_SUCCESS;
+
+ qedr_inc_sw_cons(&qp->sq);
+ i++;
+ }
+
+ spin_unlock_irqrestore(&cq->cq_lock, flags);
+
+ DP_DEBUG(dev, QEDR_MSG_GSI,
+ "gsi poll_cq: requested entries=%d, actual=%d, qp->rq.cons=%d, qp->rq.gsi_cons=%x, qp->sq.cons=%d, qp->sq.gsi_cons=%d, qp_num=%d\n",
+ num_entries, i, qp->rq.cons, qp->rq.gsi_cons, qp->sq.cons,
+ qp->sq.gsi_cons, qp->ibqp.qp_num);
+
+ return i;
+}
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.h b/drivers/infiniband/hw/qedr/qedr_cm.h
index b8a8b76..9ba6e15 100644
--- a/drivers/infiniband/hw/qedr/qedr_cm.h
+++ b/drivers/infiniband/hw/qedr/qedr_cm.h
@@ -32,9 +32,30 @@
#ifndef LINUX_QEDR_CM_H_
#define LINUX_QEDR_CM_H_
+#define QEDR_GSI_MAX_RECV_WR (4096)
+#define QEDR_GSI_MAX_SEND_WR (4096)
+
+#define QEDR_GSI_MAX_RECV_SGE (1) /* LL2 FW limitation */
+
+#define ETH_P_ROCE (0x8915)
+#define QEDR_ROCE_V2_UDP_SPORT (0000)
+
static inline u32 qedr_get_ipv4_from_gid(u8 *gid)
{
return *(u32 *)(void *)&gid[12];
}
+/* RDMA CM */
+int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr);
+int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr);
+struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs,
+ struct qedr_qp *qp);
+void qedr_store_gsi_qp_cq(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_qp_init_attr *attrs);
+int qedr_destroy_gsi_qp(struct qedr_dev *dev);
+void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info);
#endif
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 130f47d..9f0a72f 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -1510,6 +1510,15 @@ struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
qedr_set_qp_init_params(dev, qp, pd, attrs);
+ if (attrs->qp_type == IB_QPT_GSI) {
+ if (udata) {
+ DP_ERR(dev,
+ "create qp: unexpected udata when creating GSI QP\n");
+ goto err0;
+ }
+ return qedr_create_gsi_qp(dev, attrs, qp);
+ }
+
memset(&in_params, 0, sizeof(in_params));
if (udata) {
@@ -2075,6 +2084,8 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
if (rc)
return rc;
+ } else {
+ qedr_destroy_gsi_qp(dev);
}
if (ibqp->uobject && ibqp->uobject->context) {
@@ -2090,6 +2101,23 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
return rc;
}
+struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
+{
+ struct qedr_ah *ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
+
+ ah->attr = *attr;
+
+ return &ah->ibah;
+}
+
+int qedr_destroy_ah(struct ib_ah *ibah)
+{
+ struct qedr_ah *ah = get_qedr_ah(ibah);
+
+ kfree(ah);
+ return 0;
+}
+
static void free_mr_info(struct qedr_dev *dev, struct mr_info *info)
{
struct qedr_pbl *pbl, *tmp;
@@ -2941,6 +2969,10 @@ int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
*bad_wr = NULL;
+ if (qp->qp_type == IB_QPT_GSI) {
+ return qedr_gsi_post_send(ibqp, wr, bad_wr);
+ }
+
spin_lock_irqsave(&qp->q_lock, flags);
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
@@ -2997,6 +3029,10 @@ int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
unsigned long flags;
int status = 0;
+ if (qp->qp_type == IB_QPT_GSI) {
+ return qedr_gsi_post_recv(ibqp, wr, bad_wr);
+ }
+
spin_lock_irqsave(&qp->q_lock, flags);
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
@@ -3432,6 +3468,9 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
int update = 0;
int done = 0;
+ if (cq->cq_type == QEDR_CQ_TYPE_GSI)
+ return qedr_gsi_poll_cq(ibcq, num_entries, wc);
+
spin_lock_irqsave(&cq->cq_lock, flags);
old_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
while (num_entries && is_valid_cqe(cq, cqe)) {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index c90f1fd..a6db107 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -278,6 +278,7 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
struct qed_ll2_tx_packet *p_pkt = NULL;
struct qed_ll2_info *p_ll2_conn;
struct qed_ll2_tx_queue *p_tx;
+ dma_addr_t tx_frag;
p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
if (!p_ll2_conn)
@@ -297,11 +298,22 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
p_tx->cur_completing_packet = *p_pkt;
p_tx->cur_completing_bd_idx = 1;
b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+ tx_frag = p_pkt->bds_set[0].tx_frag;
+ if (p_ll2_conn->gsi_enable)
+ qed_ll2b_release_tx_gsi_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
+ else
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
- qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
- p_pkt->cookie,
- p_pkt->bds_set[0].tx_frag,
- b_last_frag, b_last_packet);
}
}
@@ -313,6 +325,7 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
struct qed_ll2_tx_packet *p_pkt;
bool b_last_frag = false;
unsigned long flags;
+ dma_addr_t tx_frag;
int rc = -EINVAL;
spin_lock_irqsave(&p_tx->lock, flags);
@@ -353,11 +366,19 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
spin_unlock_irqrestore(&p_tx->lock, flags);
- qed_ll2b_complete_tx_packet(p_hwfn,
- p_ll2_conn->my_id,
- p_pkt->cookie,
- p_pkt->bds_set[0].tx_frag,
- b_last_frag, !num_bds);
+ tx_frag = p_pkt->bds_set[0].tx_frag;
+ if (p_ll2_conn->gsi_enable)
+ qed_ll2b_complete_tx_gsi_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag, !num_bds);
+ else
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag, !num_bds);
spin_lock_irqsave(&p_tx->lock, flags);
}
@@ -368,6 +389,54 @@ out:
return rc;
}
+static int
+qed_ll2_rxq_completion_gsi(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info,
+ union core_rx_cqe_union *p_cqe,
+ unsigned long lock_flags, bool b_last_cqe)
+{
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_info->rx_queue;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+ u16 packet_length, parse_flags, vlan;
+ u32 src_mac_addrhi;
+ u16 src_mac_addrlo;
+
+ if (!list_empty(&p_rx->active_descq))
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt) {
+ DP_NOTICE(p_hwfn,
+ "GSI Rx completion but active_descq is empty\n");
+ return -EIO;
+ }
+
+ list_del(&p_pkt->list_entry);
+ parse_flags = le16_to_cpu(p_cqe->rx_cqe_gsi.parse_flags.flags);
+ packet_length = le16_to_cpu(p_cqe->rx_cqe_gsi.data_length);
+ vlan = le16_to_cpu(p_cqe->rx_cqe_gsi.vlan);
+ src_mac_addrhi = le32_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrhi);
+ src_mac_addrlo = le16_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrlo);
+ if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
+ DP_NOTICE(p_hwfn,
+ "Mismatch between active_descq and the LL2 Rx chain\n");
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ spin_unlock_irqrestore(&p_rx->lock, lock_flags);
+ qed_ll2b_complete_rx_gsi_packet(p_hwfn,
+ p_ll2_info->my_id,
+ p_pkt->cookie,
+ p_pkt->rx_buf_addr,
+ packet_length,
+ p_cqe->rx_cqe_gsi.data_length_error,
+ parse_flags,
+ vlan,
+ src_mac_addrhi,
+ src_mac_addrlo, b_last_cqe);
+ spin_lock_irqsave(&p_rx->lock, lock_flags);
+
+ return 0;
+}
+
static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn,
union core_rx_cqe_union *p_cqe,
@@ -429,6 +498,10 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n");
rc = -EINVAL;
break;
+ case CORE_RX_CQE_TYPE_GSI_OFFLOAD:
+ rc = qed_ll2_rxq_completion_gsi(p_hwfn, p_ll2_conn,
+ cqe, flags, b_last_cqe);
+ break;
case CORE_RX_CQE_TYPE_REGULAR:
rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
cqe, flags, b_last_cqe);
@@ -527,6 +600,7 @@ static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
}
p_ramrod->action_on_error.error_type = action_on_error;
+ p_ramrod->gsi_offload_flag = p_ll2_conn->gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -589,6 +663,7 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
}
+ p_ramrod->gsi_offload_flag = p_ll2_conn->gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -775,6 +850,7 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
p_ll2_info->tx_dest = p_params->tx_dest;
p_ll2_info->ai_err_packet_too_big = p_params->ai_err_packet_too_big;
p_ll2_info->ai_err_no_buf = p_params->ai_err_no_buf;
+ p_ll2_info->gsi_enable = p_params->gsi_enable;
rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
if (rc)
@@ -1492,6 +1568,7 @@ static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
ll2_info.rx_vlan_removal_en = params->rx_vlan_stripping;
ll2_info.tx_tc = 0;
ll2_info.tx_dest = CORE_TX_DEST_NW;
+ ll2_info.gsi_enable = 1;
rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_info,
QED_LL2_RX_SIZE, QED_LL2_TX_SIZE,
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index b3783fd..80a5dc2 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -125,6 +125,7 @@ struct qed_ll2_info {
u8 tx_stats_en;
struct qed_ll2_rx_queue rx_queue;
struct qed_ll2_tx_queue tx_queue;
+ u8 gsi_enable;
};
/**
@@ -292,5 +293,24 @@ void qed_ll2_setup(struct qed_hwfn *p_hwfn,
*/
void qed_ll2_free(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_connections);
-
+void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ u16 data_length,
+ u8 data_length_error,
+ u16 parse_flags,
+ u16 vlan,
+ u32 src_mac_addr_hi,
+ u16 src_mac_addr_lo, bool b_last_packet);
+void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
+void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index b4f851b..83d841d 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -2633,6 +2633,71 @@ void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
+void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
+{
+ struct qed_roce_ll2_packet *packet = cookie;
+ struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
+
+ roce_ll2->cbs.tx_cb(roce_ll2->cb_cookie, packet);
+}
+
+void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
+{
+ qed_ll2b_complete_tx_gsi_packet(p_hwfn, connection_handle,
+ cookie, first_frag_addr,
+ b_last_fragment, b_last_packet);
+}
+
+void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ u16 data_length,
+ u8 data_length_error,
+ u16 parse_flags,
+ u16 vlan,
+ u32 src_mac_addr_hi,
+ u16 src_mac_addr_lo, bool b_last_packet)
+{
+ struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
+ struct qed_roce_ll2_rx_params params;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ struct qed_roce_ll2_packet pkt;
+
+ DP_VERBOSE(cdev,
+ QED_MSG_LL2,
+ "roce ll2 rx complete: bus_addr=%p, len=%d, data_len_err=%d\n",
+ (void *)(uintptr_t)rx_buf_addr,
+ data_length, data_length_error);
+
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.n_seg = 1;
+ pkt.payload[0].baddr = rx_buf_addr;
+ pkt.payload[0].len = data_length;
+
+ memset(¶ms, 0, sizeof(params));
+ params.vlan_id = vlan;
+ *((u32 *)¶ms.smac[0]) = ntohl(src_mac_addr_hi);
+ *((u16 *)¶ms.smac[4]) = ntohs(src_mac_addr_lo);
+
+ if (data_length_error) {
+ DP_ERR(cdev,
+ "roce ll2 rx complete: data length error %d, length=%d\n",
+ data_length_error, data_length);
+ params.rc = -EINVAL;
+ }
+
+ roce_ll2->cbs.rx_cb(roce_ll2->cb_cookie, &pkt, ¶ms);
+}
+
static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
u8 *old_mac_address,
u8 *new_mac_address)
@@ -2717,6 +2782,7 @@ static int qed_roce_ll2_start(struct qed_dev *cdev,
ll2_params.tx_dest = CORE_TX_DEST_NW;
ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
+ ll2_params.gsi_enable = true;
rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
params->max_rx_buffers,
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related
* [PATCH 09/11] qedr: Add LL2 RoCE interface
From: Ram Amrani @ 2016-09-26 14:16 UTC (permalink / raw)
To: davem-fT/PcQaiUtIeIZ0/mPfg9Q, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: Ariel.Elior-YGCgFSpz5w/QT0dZR+AlfA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA,
Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA,
rajesh.borundia-YGCgFSpz5w/QT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
Ram Amrani
In-Reply-To: <1474899407-12635-1-git-send-email-Ram.Amrani-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Add light L2 interface for RoCE.
Signed-off-by: Rajesh Borundia <rajesh.borundia-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ram Amrani <Ram.Amrani-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/qedr/main.c | 34 +++-
drivers/infiniband/hw/qedr/qedr.h | 2 +-
drivers/net/ethernet/qlogic/qed/qed.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_hsi.h | 3 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 20 ++-
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 241 +++++++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 1 +
include/linux/qed/qed_roce_if.h | 79 ++++++++++
9 files changed, 384 insertions(+), 4 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index b52cd8f..8ed1503 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -698,6 +698,38 @@ static void qedr_shutdown(struct qedr_dev *dev)
qedr_remove(dev);
}
+static void qedr_mac_address_change(struct qedr_dev *dev)
+{
+ union ib_gid *sgid = &dev->sgid_tbl[0];
+ u8 guid[8], mac_addr[6];
+ int rc;
+
+ /* Update SGID */
+ ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
+ guid[0] = mac_addr[0] ^ 2;
+ guid[1] = mac_addr[1];
+ guid[2] = mac_addr[2];
+ guid[3] = 0xff;
+ guid[4] = 0xfe;
+ guid[5] = mac_addr[3];
+ guid[6] = mac_addr[4];
+ guid[7] = mac_addr[5];
+ sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
+ memcpy(&sgid->raw[8], guid, sizeof(guid));
+
+ /* Update LL2 */
+ rc = dev->ops->roce_ll2_set_mac_filter(dev->cdev,
+ dev->gsi_ll2_mac_address,
+ dev->ndev->dev_addr);
+
+ ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
+
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+
+ if (rc)
+ DP_ERR(dev, "Error updating mac filter\n");
+}
+
/* event handling via NIC driver ensures that all the NIC specific
* initialization done before RoCE driver notifies
* event to stack.
@@ -715,7 +747,7 @@ static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event)
qedr_shutdown(dev);
break;
case QEDE_CHANGE_ADDR:
- qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+ qedr_mac_address_change(dev);
break;
default:
pr_err("Event not supported\n");
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index ba380d6..bdfc599 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -149,7 +149,7 @@ struct qedr_dev {
u8 dp_level;
u8 num_hwfns;
uint wq_multiplier;
-
+ u8 gsi_ll2_mac_address[ETH_ALEN];
};
#define QEDR_MAX_SQ_PBL (0x8000)
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index a92aa80..3387f47 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -429,6 +429,7 @@ struct qed_hwfn {
#endif
struct z_stream_s *stream;
+ struct qed_roce_ll2_info *ll2;
};
struct pci_params {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_hsi.h b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
index 2777d5b..72eee29 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_hsi.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
@@ -727,6 +727,9 @@ struct core_tx_bd_flags {
#define CORE_TX_BD_FLAGS_L4_PROTOCOL_SHIFT 6
#define CORE_TX_BD_FLAGS_L4_PSEUDO_CSUM_MODE_MASK 0x1
#define CORE_TX_BD_FLAGS_L4_PSEUDO_CSUM_MODE_SHIFT 7
+#define CORE_TX_BD_FLAGS_ROCE_FLAV_MASK 0x1
+#define CORE_TX_BD_FLAGS_ROCE_FLAV_SHIFT 12
+
};
struct core_tx_bd {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index e0ec8ed..c90f1fd 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -1026,6 +1026,7 @@ static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum core_roce_flavor_type type,
dma_addr_t first_frag,
u16 first_frag_len)
{
@@ -1046,6 +1047,9 @@ static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
DMA_REGPAIR_LE(start_bd->addr, first_frag);
start_bd->nbytes = cpu_to_le16(first_frag_len);
+ SET_FIELD(start_bd->bd_flags.as_bitfield, CORE_TX_BD_FLAGS_ROCE_FLAV,
+ type);
+
DP_VERBOSE(p_hwfn,
(NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
"LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
@@ -1137,11 +1141,13 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum qed_ll2_roce_flavor_type qed_roce_flavor,
dma_addr_t first_frag,
u16 first_frag_len, void *cookie, u8 notify_fw)
{
struct qed_ll2_tx_packet *p_curp = NULL;
struct qed_ll2_info *p_ll2_conn = NULL;
+ enum core_roce_flavor_type roce_flavor;
struct qed_ll2_tx_queue *p_tx;
struct qed_chain *p_tx_chain;
unsigned long flags;
@@ -1174,6 +1180,15 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
goto out;
}
+ if (qed_roce_flavor == QED_LL2_ROCE) {
+ roce_flavor = CORE_ROCE;
+ } else if (qed_roce_flavor == QED_LL2_RROCE) {
+ roce_flavor = CORE_RROCE;
+ } else {
+ rc = -EINVAL;
+ goto out;
+ }
+
/* Prepare packet and BD, and perhaps send a doorbell to FW */
qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
num_of_bds, first_frag,
@@ -1181,6 +1196,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
num_of_bds, CORE_TX_DEST_NW,
vlan, bd_flags, l4_hdr_offset_w,
+ roce_flavor,
first_frag, first_frag_len);
qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
@@ -1625,8 +1641,8 @@ static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
cdev->ll2->handle,
1 + skb_shinfo(skb)->nr_frags,
- vlan, flags, 0, mapping,
- skb->len, skb, 1);
+ vlan, flags, 0, 0 /* RoCE FLAVOR */,
+ mapping, skb->len, skb, 1);
if (rc)
goto err;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index a037c48..b3783fd 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -24,6 +24,12 @@
#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
+enum qed_ll2_roce_flavor_type {
+ QED_LL2_ROCE,
+ QED_LL2_RROCE,
+ MAX_QED_LL2_ROCE_FLAVOR_TYPE
+};
+
enum qed_ll2_conn_type {
QED_LL2_TYPE_RESERVED,
QED_LL2_TYPE_ISCSI,
@@ -199,6 +205,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum qed_ll2_roce_flavor_type qed_roce_flavor,
dma_addr_t first_frag,
u16 first_frag_len, void *cookie, u8 notify_fw);
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 04e33ee..b4f851b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -64,6 +64,7 @@
#include "qed_reg_addr.h"
#include "qed_sp.h"
#include "qed_roce.h"
+#include "qed_ll2.h"
void qed_async_roce_event(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe)
@@ -2632,6 +2633,240 @@ void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
+static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
+ u8 *old_mac_address,
+ u8 *new_mac_address)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_ptt *p_ptt;
+ int rc = 0;
+
+ if (!hwfn->ll2 || hwfn->ll2->handle == QED_LL2_UNUSED_HANDLE) {
+ DP_ERR(cdev,
+ "qed roce mac filter failed - roce_info/ll2 NULL\n");
+ return -EINVAL;
+ }
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_ERR(cdev,
+ "qed roce ll2 mac filter set: failed to acquire PTT\n");
+ return -EINVAL;
+ }
+
+ mutex_lock(&hwfn->ll2->lock);
+ if (old_mac_address)
+ qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ old_mac_address);
+ if (new_mac_address)
+ rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ new_mac_address);
+ mutex_unlock(&hwfn->ll2->lock);
+
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+
+ if (rc)
+ DP_ERR(cdev,
+ "qed roce ll2 mac filter set: failed to add mac filter\n");
+
+ return rc;
+}
+
+static int qed_roce_ll2_start(struct qed_dev *cdev,
+ struct qed_roce_ll2_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2;
+ struct qed_ll2_info ll2_params;
+ int rc;
+
+ if (!params) {
+ DP_ERR(cdev, "qed roce ll2 start: failed due to NULL params\n");
+ return -EINVAL;
+ }
+ if (!params->cbs.tx_cb || !params->cbs.rx_cb) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed due to NULL tx/rx. tx_cb=%p, rx_cb=%p\n",
+ params->cbs.tx_cb, params->cbs.rx_cb);
+ return -EINVAL;
+ }
+ if (!is_valid_ether_addr(params->mac_address)) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed due to invalid Ethernet address %pM\n",
+ params->mac_address);
+ return -EINVAL;
+ }
+
+ /* Initialize */
+ roce_ll2 = kzalloc(sizeof(*roce_ll2), GFP_ATOMIC);
+ if (!roce_ll2) {
+ DP_ERR(cdev, "qed roce ll2 start: failed memory allocation\n");
+ return -ENOMEM;
+ }
+ memset(roce_ll2, 0, sizeof(*roce_ll2));
+ roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
+ roce_ll2->cbs = params->cbs;
+ roce_ll2->cb_cookie = params->cb_cookie;
+ mutex_init(&roce_ll2->lock);
+
+ memset(&ll2_params, 0, sizeof(ll2_params));
+ ll2_params.conn_type = QED_LL2_TYPE_ROCE;
+ ll2_params.mtu = params->mtu;
+ ll2_params.rx_drop_ttl0_flg = true;
+ ll2_params.rx_vlan_removal_en = false;
+ ll2_params.tx_dest = CORE_TX_DEST_NW;
+ ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
+ ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
+
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
+ params->max_rx_buffers,
+ params->max_tx_buffers,
+ &roce_ll2->handle);
+ if (rc) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed to acquire LL2 connection (rc=%d)\n",
+ rc);
+ goto err;
+ }
+
+ rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle);
+ if (rc) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed to establish LL2 connection (rc=%d)\n",
+ rc);
+ goto err1;
+ }
+
+ hwfn->ll2 = roce_ll2;
+
+ rc = qed_roce_ll2_set_mac_filter(cdev, NULL, params->mac_address);
+ if (rc) {
+ hwfn->ll2 = NULL;
+ goto err2;
+ }
+ ether_addr_copy(roce_ll2->mac_address, params->mac_address);
+
+ return 0;
+
+err2:
+ qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+err1:
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+err:
+ kfree(roce_ll2);
+ return rc;
+}
+
+static int qed_roce_ll2_stop(struct qed_dev *cdev)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+ int rc;
+
+ if (!cdev) {
+ DP_ERR(cdev, "qed roce ll2 stop: invalid cdev\n");
+ return -EINVAL;
+ }
+
+ if (roce_ll2->handle == QED_LL2_UNUSED_HANDLE) {
+ DP_ERR(cdev, "qed roce ll2 stop: cannot stop an unused LL2\n");
+ return -EINVAL;
+ }
+
+ /* remove LL2 MAC address filter */
+ rc = qed_roce_ll2_set_mac_filter(cdev, roce_ll2->mac_address, NULL);
+ eth_zero_addr(roce_ll2->mac_address);
+
+ rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle);
+ if (rc)
+ DP_ERR(cdev,
+ "qed roce ll2 stop: failed to terminate LL2 connection (rc=%d)\n",
+ rc);
+
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+
+ roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
+
+ kfree(roce_ll2);
+
+ return rc;
+}
+
+static int qed_roce_ll2_tx(struct qed_dev *cdev,
+ struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_tx_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+ enum qed_ll2_roce_flavor_type qed_roce_flavor;
+ u8 flags = 0;
+ int rc;
+ int i;
+
+ if (!cdev || !pkt || !params) {
+ DP_ERR(cdev,
+ "roce ll2 tx: failed tx because one of the following is NULL - drv=%p, pkt=%p, params=%p\n",
+ cdev, pkt, params);
+ return -EINVAL;
+ }
+
+ qed_roce_flavor = (pkt->roce_mode == ROCE_V1) ? QED_LL2_ROCE
+ : QED_LL2_RROCE;
+
+ if (pkt->roce_mode == ROCE_V2_IPV4)
+ flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
+
+ /* Tx header */
+ rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev), roce_ll2->handle,
+ 1 + pkt->n_seg, 0, flags, 0,
+ qed_roce_flavor, pkt->header.baddr,
+ pkt->header.len, pkt, 1);
+ if (rc) {
+ DP_ERR(cdev, "roce ll2 tx: header failed (rc=%d)\n", rc);
+ return QED_ROCE_TX_HEAD_FAILURE;
+ }
+
+ /* Tx payload */
+ for (i = 0; i < pkt->n_seg; i++) {
+ rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle,
+ pkt->payload[i].baddr,
+ pkt->payload[i].len);
+ if (rc) {
+ /* If failed not much to do here, partial packet has
+ * been posted * we can't free memory, will need to wait
+ * for completion
+ */
+ DP_ERR(cdev,
+ "roce ll2 tx: payload failed (rc=%d)\n", rc);
+ return QED_ROCE_TX_FRAG_FAILURE;
+ }
+ }
+
+ return 0;
+}
+
+static int qed_roce_ll2_post_rx_buffer(struct qed_dev *cdev,
+ struct qed_roce_ll2_buffer *buf,
+ u64 cookie, u8 notify_fw)
+{
+ return qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
+ QED_LEADING_HWFN(cdev)->ll2->handle,
+ buf->baddr, buf->len,
+ (void *)(uintptr_t)cookie, notify_fw);
+}
+
+static int qed_roce_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+
+ return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle, stats);
+}
+
static const struct qed_rdma_ops qed_rdma_ops_pass = {
.common = &qed_common_ops_pass,
.fill_dev_info = &qed_fill_rdma_dev_info,
@@ -2659,6 +2894,12 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_free_tid = &qed_rdma_free_tid,
.rdma_register_tid = &qed_rdma_register_tid,
.rdma_deregister_tid = &qed_rdma_deregister_tid,
+ .roce_ll2_start = &qed_roce_ll2_start,
+ .roce_ll2_stop = &qed_roce_ll2_stop,
+ .roce_ll2_tx = &qed_roce_ll2_tx,
+ .roce_ll2_post_rx_buffer = &qed_roce_ll2_post_rx_buffer,
+ .roce_ll2_set_mac_filter = &qed_roce_ll2_set_mac_filter,
+ .roce_ll2_stats = &qed_roce_ll2_stats,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 71b37fc..f688a47 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -42,6 +42,7 @@
#include "qed.h"
#include "qed_dev_api.h"
#include "qed_hsi.h"
+#include "qed_ll2.h"
#define QED_RDMA_MAX_FMR (RDMA_MAX_TIDS)
#define QED_RDMA_MAX_P_KEY (1)
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 0b6df6e..53047d3 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -39,6 +39,16 @@
#include <linux/slab.h>
#include <linux/qed/qed_if.h>
#include <linux/qed/qed_ll2_if.h>
+#include <linux/qed/rdma_common.h>
+
+enum qed_roce_ll2_tx_dest {
+ /* Light L2 TX Destination to the Network */
+ QED_ROCE_LL2_TX_DEST_NW,
+
+ /* Light L2 TX Destination to the Loopback */
+ QED_ROCE_LL2_TX_DEST_LB,
+ QED_ROCE_LL2_TX_DEST_MAX
+};
#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
@@ -461,6 +471,61 @@ struct qed_rdma_counters_out_params {
#define QED_ROCE_TX_HEAD_FAILURE (1)
#define QED_ROCE_TX_FRAG_FAILURE (2)
+struct qed_roce_ll2_header {
+ void *vaddr;
+ dma_addr_t baddr;
+ size_t len;
+};
+
+struct qed_roce_ll2_buffer {
+ dma_addr_t baddr;
+ size_t len;
+};
+
+struct qed_roce_ll2_packet {
+ struct qed_roce_ll2_header header;
+ int n_seg;
+ struct qed_roce_ll2_buffer payload[RDMA_MAX_SGE_PER_SQ_WQE];
+ int roce_mode;
+ enum qed_roce_ll2_tx_dest tx_dest;
+};
+
+struct qed_roce_ll2_tx_params {
+ int reserved;
+};
+
+struct qed_roce_ll2_rx_params {
+ u16 vlan_id;
+ u8 smac[ETH_ALEN];
+ int rc;
+};
+
+struct qed_roce_ll2_cbs {
+ void (*tx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt);
+
+ void (*rx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_rx_params *params);
+};
+
+struct qed_roce_ll2_params {
+ u16 max_rx_buffers;
+ u16 max_tx_buffers;
+ u16 mtu;
+ u8 mac_address[ETH_ALEN];
+ struct qed_roce_ll2_cbs cbs;
+ void *cb_cookie;
+};
+
+struct qed_roce_ll2_info {
+ u8 handle;
+ struct qed_roce_ll2_cbs cbs;
+ u8 mac_address[ETH_ALEN];
+ void *cb_cookie;
+
+ /* Lock to protect ll2 */
+ struct mutex lock;
+};
+
enum qed_rdma_type {
QED_RDMA_TYPE_ROCE,
};
@@ -518,6 +583,20 @@ struct qed_rdma_ops {
int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
+ int (*roce_ll2_start)(struct qed_dev *cdev,
+ struct qed_roce_ll2_params *params);
+ int (*roce_ll2_stop)(struct qed_dev *cdev);
+ int (*roce_ll2_tx)(struct qed_dev *cdev,
+ struct qed_roce_ll2_packet *packet,
+ struct qed_roce_ll2_tx_params *params);
+ int (*roce_ll2_post_rx_buffer)(struct qed_dev *cdev,
+ struct qed_roce_ll2_buffer *buf,
+ u64 cookie, u8 notify_fw);
+ int (*roce_ll2_set_mac_filter)(struct qed_dev *cdev,
+ u8 *old_mac_address,
+ u8 *new_mac_address);
+ int (*roce_ll2_stats)(struct qed_dev *cdev,
+ struct qed_ll2_stats *stats);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox