Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Support working under VFIO passthrough #9

Closed
wants to merge 5 commits into from
Closed

Support working under VFIO passthrough #9

wants to merge 5 commits into from

Conversation

mcbridematt
Copy link

This fixes some issues that were preventing the driver from working under VFIO / QEMU passthrough.
This has been tested with muvirt on Ten64 and MC firmware 10.20 (see our forum on how to set this up: https://forum.traverse.com.au/t/restool-in-muvirt/63/17?u=mcbridematt).

There are some changes to VFIO in more recent MC firmwares (>10.24) and QEMU versions so some adjustments may still be required.

Key changes:

  1. Extend ICID to 32-bits. The top 16 bits (marked as 'reserved' in NXP documentation) are used to differentiate host and guest interrupts in the (v)GIC [interrupt controller]. Without the full ICID the interrupts will not be routed in the kernel
dpaa2_rc0: Isolation context ID: 23 # Bare metal
dpaa2_rc0: Isolation context ID: 65536 # VM
  1. Treat an unreachable DPMAC (when we ask if it is PHY or FIXED) as a FIXED link.
    (This is because the DPMAC is usually sitting in the host / root DPRC and cannot be accessed by the child DPRC)
    Ideally, this would be avoided by having the DPMAC driver 'own' the PHY connection (signalling the MC on MII/MDIO events), and the DPNI receiving it's link state indications from the MC via interrupts. However, it's not worth the effort to implement this unless FreeBSD gains the ability to act as a host for VFIO passthrough.
  2. Support non-DPMAC partner types: DPNI, DPDMUX, DPSW.
    These function exactly the same as a DPMAC in FIXED mode, but their MAC addresses are set in the DPNI object only, so move the MAC query out of the DPMAC only section.

@mcbridematt
dpaa2: ni: assume a DPMAC not reachable (e.g outside our DPRC) is a F…
893702a 
…IXED link

We try to query the connected DPMAC to determine it's endpoint type (PHY or
FIXED), but this approach fails when the link partner is *outside*
our own DPRC.
For now just assume an unreachable DPMAC is a "fixed" link.
@mcbridematt
dpaa2: rc: extend ICID to 32-bits for VFIO support
dbfad2a 
While the ICID is only documented as a 16-bit value, 32-bits are
actually used (taking over the 'reserved' bits next to it).
The top 16 bits are then used to differentiate containers running
under a VM.
@mcbridematt
dpaa2: add DPDMUX and DPSW types
7848e48 
These do not have a driver implementation (yet), but a DPNI (or a DPMAC)
can be a link partner of these two objects, so we need to be
able to recognize them.
@mcbridematt
dpaa2: ni: make MAC address retreival common for all partner types
a0e0bb5 
Non-MAC partners like DPNI or DPDMUX have their MAC address stored in the
DPNI attributes which a ddpaa2_ni_set_mac_addr already handles.
@mcbridematt
dpaa2: ni: setup fixed link for DPDMUX and DPSW
350badb
@dsalychev dsalychev added bug Something isn't working enhancement New feature or request patch Patch is available labels Jun 2, 2022
@dsalychev dsalychev self-assigned this Jun 2, 2022
@dsalychev
Copy link

dsalychev commented Jun 2, 2022 •
edited
Loading

Patch applied starting from ef64c6e. Thanks!

@dsalychev dsalychev closed this Jun 2, 2022
dsalychev pushed a commit that referenced this pull request Apr 3, 2023
@rob-wing
FreeBSD: don't verify recycled vnode for zfs control directory
28251d8 
Under certain loads, the following panic is hit:

    panic: page fault
    KDB: stack backtrace:
    #0 0xffffffff805db025 at kdb_backtrace+0x65
    #1 0xffffffff8058e86f at vpanic+0x17f
    #2 0xffffffff8058e6e3 at panic+0x43
    #3 0xffffffff808adc15 at trap_fatal+0x385
    #4 0xffffffff808adc6f at trap_pfault+0x4f
    #5 0xffffffff80886da8 at calltrap+0x8
    #6 0xffffffff80669186 at vgonel+0x186
    #7 0xffffffff80669841 at vgone+0x31
    #8 0xffffffff8065806d at vfs_hash_insert+0x26d
    #9 0xffffffff81a39069 at sfs_vgetx+0x149
    #10 0xffffffff81a39c54 at zfsctl_snapdir_lookup+0x1e4
    #11 0xffffffff8065a28c at lookup+0x45c
    #12 0xffffffff806594b9 at namei+0x259
    #13 0xffffffff80676a33 at kern_statat+0xf3
    #14 0xffffffff8067712f at sys_fstatat+0x2f
    #15 0xffffffff808ae50c at amd64_syscall+0x10c
    #16 0xffffffff808876bb at fast_syscall_common+0xf8

The page fault occurs because vgonel() will call VOP_CLOSE() for active
vnodes. For this reason, define vop_close for zfsctl_ops_snapshot. While
here, define vop_open for consistency.

After adding the necessary vop, the bug progresses to the following
panic:

    panic: VERIFY3(vrecycle(vp) == 1) failed (0 == 1)
    cpuid = 17
    KDB: stack backtrace:
    #0 0xffffffff805e29c5 at kdb_backtrace+0x65
    #1 0xffffffff8059620f at vpanic+0x17f
    #2 0xffffffff81a27f4a at spl_panic+0x3a
    #3 0xffffffff81a3a4d0 at zfsctl_snapshot_inactive+0x40
    #4 0xffffffff8066fdee at vinactivef+0xde
    #5 0xffffffff80670b8a at vgonel+0x1ea
    #6 0xffffffff806711e1 at vgone+0x31
    #7 0xffffffff8065fa0d at vfs_hash_insert+0x26d
    #8 0xffffffff81a39069 at sfs_vgetx+0x149
    #9 0xffffffff81a39c54 at zfsctl_snapdir_lookup+0x1e4
    #10 0xffffffff80661c2c at lookup+0x45c
    #11 0xffffffff80660e59 at namei+0x259
    #12 0xffffffff8067e3d3 at kern_statat+0xf3
    #13 0xffffffff8067eacf at sys_fstatat+0x2f
    #14 0xffffffff808b5ecc at amd64_syscall+0x10c
    #15 0xffffffff8088f07b at fast_syscall_common+0xf8

This is caused by a race condition that can occur when allocating a new
vnode and adding that vnode to the vfs hash. If the newly created vnode
loses the race when being inserted into the vfs hash, it will not be
recycled as its usecount is greater than zero, hitting the above
assertion.

Fix this by dropping the assertion.

FreeBSD-issue: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=252700
Reviewed-by: Andriy Gapon <avg@FreeBSD.org>
Reviewed-by: Mateusz Guzik <mjguzik@gmail.com>
Reviewed-by: Alek Pinchuk <apinchuk@axcient.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Rob Wing <rob.wing@klarasystems.com>
Co-authored-by: Rob Wing <rob.wing@klarasystems.com>
Submitted-by: Klara, Inc.
Sponsored-by: rsync.net
Closes #14501
dsalychev pushed a commit that referenced this pull request Aug 3, 2023
@kprovost
mroute: avoid calling if_allmulti with the lock held
680ad06 
Avoid locking issues when if_allmulti() calls the driver's if_ioctl,
because that may acquire sleepable locks (while we hold a non-sleepable
rwlock).

Fortunately there's no pressing need to hold the mroute lock while we
do this, so we can postpone the call slightly, until after we've
released the lock.

This avoids the following WITNESS warning (with iflib drivers):

	lock order reversal: (sleepable after non-sleepable)
	 1st 0xffffffff82f64960 IPv4 multicast forwarding (IPv4 multicast forwarding, rw) @ /usr/src/sys/netinet/ip_mroute.c:1050
	 2nd 0xfffff8000480f180 iflib ctx lock (iflib ctx lock, sx) @ /usr/src/sys/net/iflib.c:4525
	lock order IPv4 multicast forwarding -> iflib ctx lock attempted at:
	#0 0xffffffff80bbd6ce at witness_checkorder+0xbbe
	#1 0xffffffff80b56d10 at _sx_xlock+0x60
	#2 0xffffffff80c9ce5c at iflib_if_ioctl+0x2dc
	#3 0xffffffff80c7c395 at if_setflag+0xe5
	#4 0xffffffff82f60a0e at del_vif_locked+0x9e
	#5 0xffffffff82f5f0d5 at X_ip_mrouter_set+0x265
	#6 0xffffffff80bfd402 at sosetopt+0xc2
	#7 0xffffffff80c02105 at kern_setsockopt+0xa5
	#8 0xffffffff80c02054 at sys_setsockopt+0x24
	#9 0xffffffff81046be8 at amd64_syscall+0x138
	#10 0xffffffff8101930b at fast_syscall_common+0xf8

See also:	https://redmine.pfsense.org/issues/12079
Reviewed by:	mjg
Sponsored by:	Rubicon Communications, LLC ("Netgate")
Differential Revision:	https://reviews.freebsd.org/D41209
dsalychev pushed a commit that referenced this pull request Sep 9, 2023
@np-2020
cxgbe(4): Avoid hang on kldunload on netlink enabled kernels.
3814249 
netlink(4) calls back into the driver during detach and it attempts to
start an internal synchronized op recursively, causing an interruptible
hang.  Fix it by failing the ioctl if the VI has been marked as DOOMED
by cxgbe_detach.

Here's the stack for the hang for reference.
 #6  begin_synchronized_op
 #7  cxgbe_media_status
 #8  ifmedia_ioctl
 #9  cxgbe_ioctl
 #10 if_ioctl
 #11 get_operstate_ether
 #12 get_operstate
 #13 dump_iface
 #14 rtnl_handle_ifevent
 #15 rtnl_handle_ifnet_event
 #16 rt_ifmsg
 #17 if_unroute
 #18 if_down
 #19 if_detach_internal
 #20 if_detach
 #21 ether_ifdetach
 #22 cxgbe_vi_detach
 #23 cxgbe_detach
 #24 DEVICE_DETACH

MFC after:	3 days
Sponsored by:	Chelsio Communications
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees

@dsalychev dsalychev

Labels
bug Something isn't working enhancement New feature or request patch Patch is available
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@mcbridematt @dsalychev