CVS log for src/sys/dev/pci/if_rge.c

Up to [cvs.NetBSD.org] / src / sys / dev / pci

Request diff between arbitrary revisions

Keyword substitution: kv
Default branch: MAIN

Add MBUFTRACE

Pull up following revision(s) (requested by mrg in ticket #782):

	sys/dev/pci/if_rge.c: revision 1.29

- handle stuck transmitter (descriptor still owned)
- restart send queue after transmit
- count output packets
- use deferred start

Should fix PR 57694

if_stats(9): Add ifp argument to if_stat..._ref.

This will enable us to pass the ifp through to a dtrace probe inside.

No functional change intended in this change, but this is an API
change visible to modules so it shouldn't be pulled up.

PR kern/58377

rge(4): Print HW revision.

G/C sc_task and #include "opt_net_mpsafe.h"

- handle stuck transmitter (descriptor still owned)
- restart send queue after transmit
- count output packets
- use deferred start

Should fix PR 57694

Pull up following revision(s) (requested by mrg in ticket #434):

	sys/dev/pci/if_rge.c: revision 1.26
	sys/dev/pci/if_rge.c: revision 1.28

rge(4): check for all errors in rx buffer allocation
should fix a crash seen by by Chavdar Ivanov reported on current-users.
move the rx and tx list clean up into their own functions, and call the
rx clean up function from the init function if something fails.  this
should fix a potential leak in this case, and generally frees up memory
that won't be used without a successful init phase again.
slight application of 'static', much more could be done.

rge: properly handle mbuf allocation failures in rx interrupts
several changes that should fix crashes seen after an mbuf
allocation failure:
- create RX ring dma maps with BUS_DMA_ALLOCNOW, so that any
  future bus_dmamap_load*() call will succeed.  this avoids one
  error case in rge_newbuf(), that similar cases in eg wm(4)
  actually call panic for now (i think this idea can be copied
  into wm(4) as well.)
- extract the RX descriptor set into a common function that
  both rge_newbuf() and rge_rxeof() can both use.  it's almost
  identical to the old rge_discard_rxbuf() except it also sets
  the rge_addr (this is needed for the newbuf case.)
- move the bus_dmamap_unload() into rge_newbuf(), so that the
  existing mbuf will remain mapped until a new mbuf is allocated.
  (this part is what should fix crashes seen by wiz and Chavdar,
  as the unload follow by sync is what triggers the assert in
  x86 bus_dma.  without the assert, it will would have shortly
  triggered a page fault.)  remove the assignment to NULL for
  the rxq mbuf pointer, it is required for reload.
- add a couple of missing if_statinc() calls.

tested on amd64 and arm64.

rge: properly handle mbuf allocation failures in rx interrupts

several changes that should fix crashes seen after an mbuf
allocation failure:

- create RX ring dma maps with BUS_DMA_ALLOCNOW, so that any
  future bus_dmamap_load*() call will succeed.  this avoids one
  error case in rge_newbuf(), that similar cases in eg wm(4)
  actually call panic for now (i think this idea can be copied
  into wm(4) as well.)

- extract the RX descriptor set into a common function that
  both rge_newbuf() and rge_rxeof() can both use.  it's almost
  identical to the old rge_discard_rxbuf() except it also sets
  the rge_addr (this is needed for the newbuf case.)

- move the bus_dmamap_unload() into rge_newbuf(), so that the
  existing mbuf will remain mapped until a new mbuf is allocated.
  (this part is what should fix crashes seen by wiz and Chavdar,
  as the unload follow by sync is what triggers the assert in
  x86 bus_dma.  without the assert, it will would have shortly
  triggered a page fault.)  remove the assignment to NULL for
  the rxq mbuf pointer, it is required for reload.

- add a couple of missing if_statinc() calls.


tested on amd64 and arm64.

Pull up following revision(s) (requested by riastradh in ticket #414):

	sys/dev/pci/if_rge.c: revision 1.27

rge(4): callout_stop can't be right here; just do callout_halt.

rge(4): callout_stop can't be right here; just do callout_halt.

XXX pullup-10

rge(4): check for all errors in rx buffer allocation

should fix a crash seen by by Chavdar Ivanov reported on current-users.

move the rx and tx list clean up into their own functions, and call the
rx clean up function from the init function if something fails.  this
should fix a potential leak in this case, and generally frees up memory
that won't be used without a successful init phase again.

slight application of 'static', much more could be done.

Pull up following revision(s) (requested by nonaka in ticket #14):

	sys/dev/pci/if_rge.c: revision 1.25
	sys/dev/pci/if_rgereg.h: revision 1.7

Update the Rx descriptor based on the vendor driver for Linux.

This fixes a panic on RTL8125.

Patch from OpenBSD if_rge.c r1.20, if_rgereg.h r1.8.

Tested by msaitoh@n.o.

Update the Rx descriptor based on the vendor driver for Linux.

This fixes a panic on RTL8125.
Patch from OpenBSD if_rge.c r1.20, if_rgereg.h r1.8.

Tested by msaitoh@n.o.

Remove unnecessary include of <sys/malloc.h>.

Trailing whitespace

Avoid unaligned access in rge_get_macaddr(). Found by kUBSan.

Use unsigned to avoid undefined behavior. Found by kUBSan.

Add pmf_device_register().

Sync with HEAD.

Use pci_compatible_match().

Sync with HEAD.

Fix build failure for options NET_MPSAFE.

Update rge(4) from older OpenBSD, finish porting.

Should consider merging this all into re(4) and rgephy(4) someday.

Some cleanup tasks remain here.

if_rge.c: avoid aprint_error*() abuse

Sync with OpenBSD if_rge.c r1.4

"""
Change users of IFQ_SET_MAXLEN() and IFQ_IS_EMPTY() to use the "new" API.
"""

Use device_xname() to access dv_xname

Switch from printf to aprintf_error/normal

Initialise sc->sc_dev, otherwise unnamed interfaces show up and things go
downhill from there.

use device_private() instead of casting self as it doesn't work here.
Bump rcs tag which was missed in r1.9

Sync with r1.3 from OpenBSD

"Use correct bit mask to promptly exit rge_set_phy_power() and
rge_exit_oob() timeout loops.

Spotted by CID 1491296 and 1491309."

Merge changes from current as of 20200406

file if_rge.c was added on branch phil-wifi on 2020-04-08 14:08:09 +0000

Use the 64-bit PCI tag, if available.

Sync with head.

Cast to make up for the difference in type between our bus_dma_map(9) and
theirs. Where we expect kvap to be of void **, they expect caddr_t (typedef char *).

Add an implementation of MCLGETI.
Borrowed from sys/dev/pci/if_bwfm_pci.c

redundant include

Trailing whitespace

Adopt <net/if_stats.h>

Sync with head.

file if_rge.c was added on branch ad-namecache on 2020-01-17 21:47:31 +0000

Start to switch from OpenBSD timeout(9) API to callout(9) and workqueue(9)
Replace functions with NetBSD counterparts.

Start porting to NetBSD. (WIP, incomplete)

Import rge(4) from OpenBSD for Realtek's RTL8125 based 2.5GbE network cards.