svn commit: r199289 - in stable/7/sys: conf dev/bce
Stanislav Sedov
Date: Sun Nov 15 11:43:28 2009
New Revision: 199289
URL: http://svn.freebsd.org/changeset/base/199289
Log:
- MFC r198320:
Introduce new option BCE_JUMBO_HDRSPLIT that allows user to enable header
in bce(4) instead of (ab)using ZERO_COPY_SOCKETS that was not
into if_bce.c anyway. It is disabled by default.
> PR: If a GNATS PR is affected by the change.
> Submitted by: If someone else sent in the change.
> Reviewed by: If someone else reviewed your modification.
> Approved by: If you needed approval for this commit.
> Obtained from: If the change is from a third party.
> MFC after: N [day[s]|week[s]|month[s]]. Request a reminder email.
> Security: Vulnerability reference (one per line) or description.
> Empty fields above will be automatically removed.
_M 7/sys
M 7/sys/conf/NOTES
M 7/sys/conf/options
M 7/sys/dev/bce/if_bce.c
M 7/sys/dev/bce/if_bcereg.h
_M 7/sys/contrib/pf
Modified:
stable/7/sys/conf/NOTES
stable/7/sys/conf/options
stable/7/sys/dev/bce/if_bce.c
stable/7/sys/dev/bce/if_bcereg.h
Directory Properties:
stable/7/sys/ (props changed)
stable/7/sys/contrib/pf/ (props changed)
Modified: stable/7/sys/conf/NOTES
==============================================================================
--- stable/7/sys/conf/NOTES Sun Nov 15 11:30:59 2009 (r199288)
+++ stable/7/sys/conf/NOTES Sun Nov 15 11:43:28 2009 (r199289)
@@ -1929,6 +1929,12 @@ device lmc
# only works for Tigon II chips, and has no effect for Tigon I chips.
options TI_JUMBO_HDRSPLIT
+#
+# Use header splitting feature on bce(4) adapters.
+# This may help to reduce the amount of jumbo-sized memory buffers used.
+#
+options BCE_JUMBO_HDRSPLIT
+
# These two options allow manipulating the mbuf cluster size and mbuf size,
# respectively. Be very careful with NIC driver modules when changing
# these from their default values, because that can potentially cause a
Modified: stable/7/sys/conf/options
==============================================================================
--- stable/7/sys/conf/options Sun Nov 15 11:30:59 2009 (r199288)
+++ stable/7/sys/conf/options Sun Nov 15 11:43:28 2009 (r199289)
@@ -493,6 +493,7 @@ DRM_DEBUG opt_drm.h
ZERO_COPY_SOCKETS opt_zero.h
TI_PRIVATE_JUMBOS opt_ti.h
TI_JUMBO_HDRSPLIT opt_ti.h
+BCE_JUMBO_HDRSPLIT opt_bce.h
# XXX Conflict: # of devices vs network protocol (Native ATM).
# This makes "atm.h" unusable.
Modified: stable/7/sys/dev/bce/if_bce.c
==============================================================================
--- stable/7/sys/dev/bce/if_bce.c Sun Nov 15 11:30:59 2009 (r199288)
+++ stable/7/sys/dev/bce/if_bce.c Sun Nov 15 11:43:28 2009 (r199289)
@@ -293,12 +293,12 @@ static void bce_dump_enet (str
static void bce_dump_mbuf (struct bce_softc *, struct mbuf *);
static void bce_dump_tx_mbuf_chain (struct bce_softc *, u16, int);
static void bce_dump_rx_mbuf_chain (struct bce_softc *, u16, int);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
static void bce_dump_pg_mbuf_chain (struct bce_softc *, u16, int);
#endif
static void bce_dump_txbd (struct bce_softc *, int, struct tx_bd *);
static void bce_dump_rxbd (struct bce_softc *, int, struct rx_bd *);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
static void bce_dump_pgbd (struct bce_softc *, int, struct rx_bd *);
#endif
static void bce_dump_l2fhdr (struct bce_softc *, int, struct l2_fhdr *);
@@ -306,7 +306,7 @@ static void bce_dump_ctx (struct bce_s
static void bce_dump_ftqs (struct bce_softc *);
static void bce_dump_tx_chain (struct bce_softc *, u16, int);
static void bce_dump_rx_chain (struct bce_softc *, u16, int);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
static void bce_dump_pg_chain (struct bce_softc *, u16, int);
#endif
static void bce_dump_status_block (struct bce_softc *);
@@ -391,7 +391,7 @@ static int bce_init_rx_chain (struct b
static void bce_fill_rx_chain (struct bce_softc *);
static void bce_free_rx_chain (struct bce_softc *);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
static int bce_get_pg_buf (struct bce_softc *, struct mbuf *, u16 *, u16 *);
static int bce_init_pg_chain (struct bce_softc *);
static void bce_fill_pg_chain (struct bce_softc *);
@@ -597,7 +597,7 @@ bce_print_adapter_info(struct bce_softc
/* Firmware version and device features. */
printf("B/C (0x%08X); Flags( ", sc->bce_bc_ver);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
printf("SPLT ");
#endif
if (sc->bce_flags & BCE_MFW_ENABLE_FLAG)
@@ -1013,7 +1013,7 @@ bce_attach(device_t dev)
* This may change later if the MTU size is set to
* something other than 1500.
*/
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
sc->rx_bd_mbuf_alloc_size = MHLEN;
/* Make sure offset is 16 byte aligned for hardware. */
sc->rx_bd_mbuf_align_pad = roundup2((MSIZE - MHLEN), 16) -
@@ -2753,7 +2753,7 @@ bce_dma_free(struct bce_softc *sc)
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/* Free, unmap and destroy all page buffer descriptor chain pages. */
for (i = 0; i < PG_PAGES; i++ ) {
if (sc->pg_bd_chain[i] != NULL) {
@@ -2817,7 +2817,7 @@ bce_dma_free(struct bce_softc *sc)
sc->rx_mbuf_tag = NULL;
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/* Unload and destroy the page mbuf maps. */
for (i = 0; i < TOTAL_PG_BD; i++) {
if (sc->pg_mbuf_map[i] != NULL) {
@@ -3263,7 +3263,7 @@ bce_dma_alloc(device_t dev)
/*
* Create a DMA tag for RX mbufs.
*/
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
max_size = max_seg_size = ((sc->rx_bd_mbuf_alloc_size < MCLBYTES) ?
MCLBYTES : sc->rx_bd_mbuf_alloc_size);
#else
@@ -3304,7 +3304,7 @@ bce_dma_alloc(device_t dev)
}
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/*
* Create a DMA tag for the page buffer descriptor chain,
* allocate and clear the memory, and fetch the physical
@@ -4393,7 +4393,7 @@ bce_stop(struct bce_softc *sc)
bce_disable_intr(sc);
/* Free RX buffers. */
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
bce_free_pg_chain(sc);
#endif
bce_free_rx_chain(sc);
@@ -4831,7 +4831,7 @@ bce_get_rx_buf(struct bce_softc *sc, str
goto bce_get_rx_buf_exit);
/* This is a new mbuf allocation. */
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
MGETHDR(m_new, M_DONTWAIT, MT_DATA);
#else
if (sc->rx_bd_mbuf_alloc_size <= MCLBYTES)
@@ -4912,7 +4912,7 @@ bce_get_rx_buf_exit:
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/****************************************************************************/
/* Encapsulate an mbuf cluster into the page chain. */
/* */
@@ -5021,7 +5021,7 @@ bce_get_pg_buf_exit:
return(rc);
}
-#endif /* ZERO_COPY_SOCKETS */
+#endif /* BCE_JUMBO_HDRSPLIT */
/****************************************************************************/
/* Initialize the TX context memory. */
@@ -5377,7 +5377,7 @@ bce_free_rx_chain(struct bce_softc *sc)
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/****************************************************************************/
/* Allocate memory and initialize the page data structures. */
/* Assumes that bce_init_rx_chain() has not already been called. */
@@ -5541,7 +5541,7 @@ bce_free_pg_chain(struct bce_softc *sc)
DBEXIT(BCE_VERBOSE_RESET | BCE_VERBOSE_RECV | BCE_VERBOSE_UNLOAD);
}
-#endif /* ZERO_COPY_SOCKETS */
+#endif /* BCE_JUMBO_HDRSPLIT */
/****************************************************************************/
@@ -5714,7 +5714,7 @@ bce_rx_intr(struct bce_softc *sc)
unsigned int pkt_len;
u16 sw_rx_cons, sw_rx_cons_idx, hw_rx_cons;
u32 status;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
unsigned int rem_len;
u16 sw_pg_cons, sw_pg_cons_idx;
#endif
@@ -5730,7 +5730,7 @@ bce_rx_intr(struct bce_softc *sc)
bus_dmamap_sync(sc->rx_bd_chain_tag,
sc->rx_bd_chain_map[i], BUS_DMASYNC_POSTREAD);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/* Prepare the page chain pages to be accessed by the host CPU. */
for (int i = 0; i < PG_PAGES; i++)
bus_dmamap_sync(sc->pg_bd_chain_tag,
@@ -5742,7 +5742,7 @@ bce_rx_intr(struct bce_softc *sc)
/* Get working copies of the driver's view of the consumer indices. */
sw_rx_cons = sc->rx_cons;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
sw_pg_cons = sc->pg_cons;
#endif
@@ -5803,7 +5803,7 @@ bce_rx_intr(struct bce_softc *sc)
*/
m_adj(m0, sizeof(struct l2_fhdr) + ETHER_ALIGN);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/*
* Check whether the received frame fits in a single
* mbuf or not (i.e. packet data + FCS <=
@@ -5977,7 +5977,7 @@ bce_rx_int_next_rx:
if (m0) {
/* Make sure we don't lose our place when we release the lock. */
sc->rx_cons = sw_rx_cons;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
sc->pg_cons = sw_pg_cons;
#endif
@@ -5987,7 +5987,7 @@ bce_rx_int_next_rx:
/* Recover our place. */
sw_rx_cons = sc->rx_cons;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
sw_pg_cons = sc->pg_cons;
#endif
}
@@ -5998,7 +5998,7 @@ bce_rx_int_next_rx:
}
/* No new packets to process. Refill the RX and page chains and exit. */
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
sc->pg_cons = sw_pg_cons;
bce_fill_pg_chain(sc);
#endif
@@ -6011,7 +6011,7 @@ bce_rx_int_next_rx:
bus_dmamap_sync(sc->rx_bd_chain_tag,
sc->rx_bd_chain_map[i], BUS_DMASYNC_PREWRITE);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
for (int i = 0; i < PG_PAGES; i++)
bus_dmamap_sync(sc->pg_bd_chain_tag,
sc->pg_bd_chain_map[i], BUS_DMASYNC_PREWRITE);
@@ -6257,7 +6257,7 @@ bce_init_locked(struct bce_softc *sc)
* Calculate and program the hardware Ethernet MTU
* size. Be generous on the receive if we have room.
*/
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
if (ifp->if_mtu <= (sc->rx_bd_mbuf_data_len + sc->pg_bd_mbuf_alloc_size))
ether_mtu = sc->rx_bd_mbuf_data_len + sc->pg_bd_mbuf_alloc_size;
#else
@@ -6289,7 +6289,7 @@ bce_init_locked(struct bce_softc *sc)
/* Program appropriate promiscuous/multicast filtering. */
bce_set_rx_mode(sc);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
DBPRINT(sc, BCE_INFO_LOAD, "%s(): pg_bd_mbuf_alloc_size = %d\n",
__FUNCTION__, sc->pg_bd_mbuf_alloc_size);
@@ -6802,7 +6802,7 @@ bce_ioctl(struct ifnet *ifp, u_long comm
BCE_LOCK(sc);
ifp->if_mtu = ifr->ifr_mtu;
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/* No buffer allocation size changes are necessary. */
#else
/* Recalculate our buffer allocation sizes. */
@@ -7505,7 +7505,7 @@ bce_tick(void *xsc)
bce_stats_update(sc);
/* Top off the receive and page chains. */
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
bce_fill_pg_chain(sc);
#endif
bce_fill_rx_chain(sc);
@@ -7685,7 +7685,7 @@ bce_sysctl_dump_tx_chain(SYSCTL_HANDLER_
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/****************************************************************************/
/* Provides a sysctl interface to allow dumping the page chain. */
/* */
@@ -8313,7 +8313,7 @@ bce_add_sysctls(struct bce_softc *sc)
(void *)sc, 0,
bce_sysctl_dump_tx_chain, "I", "Dump tx_bd chain");
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
SYSCTL_ADD_PROC(ctx, children, OID_AUTO,
"dump_pg_chain", CTLTYPE_INT | CTLFLAG_RW,
(void *)sc, 0,
@@ -8608,7 +8608,7 @@ bce_dump_rx_mbuf_chain(struct bce_softc
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/****************************************************************************/
/* Prints out the mbufs in the mbuf page chain. */
/* */
@@ -8732,7 +8732,7 @@ bce_dump_rxbd(struct bce_softc *sc, int
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/****************************************************************************/
/* Prints out a rx_bd structure in the page chain. */
/* */
@@ -9219,7 +9219,7 @@ bce_dump_rx_chain(struct bce_softc *sc,
}
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/****************************************************************************/
/* Prints out the page chain. */
/* */
@@ -9700,7 +9700,7 @@ bce_dump_driver_state(struct bce_softc *
"0x%08X:%08X - (sc->rx_bd_chain) rx_bd chain virtual address\n",
val_hi, val_lo);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
val_hi = BCE_ADDR_HI(sc->pg_bd_chain);
val_lo = BCE_ADDR_LO(sc->pg_bd_chain);
BCE_PRINTF(
@@ -9720,7 +9720,7 @@ bce_dump_driver_state(struct bce_softc *
"0x%08X:%08X - (sc->rx_mbuf_ptr) rx mbuf chain virtual address\n",
val_hi, val_lo);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
val_hi = BCE_ADDR_HI(sc->pg_mbuf_ptr);
val_lo = BCE_ADDR_LO(sc->pg_mbuf_ptr);
BCE_PRINTF(
@@ -9773,7 +9773,7 @@ bce_dump_driver_state(struct bce_softc *
BCE_PRINTF(" 0x%08X - (sc->free_rx_bd) free rx_bd's\n",
sc->free_rx_bd);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
BCE_PRINTF(" 0x%04X(0x%04X) - (sc->pg_prod) page producer index\n",
sc->pg_prod, (u16) PG_CHAIN_IDX(sc->pg_prod));
@@ -10279,7 +10279,7 @@ bce_breakpoint(struct bce_softc *sc)
bce_dump_tpat_state(sc, 0);
bce_dump_cp_state(sc, 0);
bce_dump_com_state(sc, 0);
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
bce_dump_pgbd(sc, 0, NULL);
bce_dump_pg_mbuf_chain(sc, 0, USABLE_PG_BD);
bce_dump_pg_chain(sc, 0, USABLE_PG_BD);
Modified: stable/7/sys/dev/bce/if_bcereg.h
==============================================================================
--- stable/7/sys/dev/bce/if_bcereg.h Sun Nov 15 11:30:59 2009 (r199288)
+++ stable/7/sys/dev/bce/if_bcereg.h Sun Nov 15 11:43:28 2009 (r199289)
@@ -6205,7 +6205,7 @@ struct l2_fhdr {
#define RX_PAGE(x) (((x) & ~USABLE_RX_BD_PER_PAGE) >> (BCM_PAGE_BITS - 4))
#define RX_IDX(x) ((x) & USABLE_RX_BD_PER_PAGE)
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/*
* To accomodate jumbo frames, the page chain should
* be 4 times larger than the receive chain.
@@ -6227,7 +6227,7 @@ struct l2_fhdr {
#define PG_PAGE(x) (((x) & ~USABLE_PG_BD_PER_PAGE) >> (BCM_PAGE_BITS - 4))
#define PG_IDX(x) ((x) & USABLE_PG_BD_PER_PAGE)
-#endif /* ZERO_COPY_SOCKETS */
+#endif /* BCE_JUMBO_HDRSPLIT */
#define CTX_INIT_RETRY_COUNT 10
@@ -6505,7 +6505,7 @@ struct bce_softc
u16 tx_cons;
u32 tx_prod_bseq; /* Counts the bytes used. */
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
u16 pg_prod;
u16 pg_cons;
#endif
@@ -6522,7 +6522,7 @@ struct bce_softc
int rx_bd_mbuf_data_len;
int rx_bd_mbuf_align_pad;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
int pg_bd_mbuf_alloc_size;
#endif
@@ -6544,7 +6544,7 @@ struct bce_softc
struct rx_bd *rx_bd_chain[RX_PAGES];
bus_addr_t rx_bd_chain_paddr[RX_PAGES];
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/* H/W maintained page buffer descriptor chain structure. */
bus_dma_tag_t pg_bd_chain_tag;
bus_dmamap_t pg_bd_chain_map[PG_PAGES];
@@ -6581,7 +6581,7 @@ struct bce_softc
bus_dma_tag_t rx_mbuf_tag;
bus_dma_tag_t tx_mbuf_tag;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
bus_dma_tag_t pg_mbuf_tag;
#endif
@@ -6593,7 +6593,7 @@ struct bce_softc
bus_dmamap_t rx_mbuf_map[TOTAL_RX_BD];
struct mbuf *rx_mbuf_ptr[TOTAL_RX_BD];
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
/* S/W maintained mbuf page chain structure. */
bus_dmamap_t pg_mbuf_map[TOTAL_PG_BD];
struct mbuf *pg_mbuf_ptr[TOTAL_PG_BD];
@@ -6605,7 +6605,7 @@ struct bce_softc
u16 used_tx_bd;
u16 max_tx_bd;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
u16 free_pg_bd;
u16 max_pg_bd;
#endif
@@ -6693,7 +6693,7 @@ struct bce_softc
int debug_tx_mbuf_alloc;
int debug_rx_mbuf_alloc;
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
int debug_pg_mbuf_alloc;
#endif
@@ -6710,7 +6710,7 @@ struct bce_softc
u32 rx_low_watermark; /* Lowest number of rx_bd's free. */
u32 rx_empty_count; /* Number of times the RX chain was empty. */
-#ifdef ZERO_COPY_SOCKETS
+#ifdef BCE_JUMBO_HDRSPLIT
u32 pg_low_watermark; /* Lowest number of pages free. */
u32 pg_empty_count; /* Number of times the page chain was empty. */
#endif
_______________________________________________
svn-s...@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-all
To unsubscribe, send any mail to "svn-src-all...@freebsd.org"
Maxim Sobolev
should be changed to avoid any confusion.
-Maxim
Tom Judge
Hash: SHA1
Maxim Sobolev wrote:
> Is the option related to Jumbo Frames in any way? If not, IMHO the name
> should be changed to avoid any confusion.
The bug is easier to reproduce when jumbo frames are turned on however
we have seen it with normal frames also.
Tom
- --
TJU13-ARIN
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.13 (FreeBSD)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iQEcBAEBAgAGBQJLFCZVAAoJEMSwVS7lr0OdMLgIAKzeXBKtLHvx6BmAMn97ky+K
mc0qempP/6gnydQkzJCCosoL/78mLzelqId7ff6jAayOriNkh4iQkSl3cZUtewD2
pJXrN1vhWWR0TxNuXLg77g3VXTtBNwxZ/9y02KWaf2xOGVVMkAOILdfhwHmCPNiV
btUjtZpcr9Fe5N4IM0Mpgr8SNAPDbigAWs/58Nn8jpoo6Dy3qYf1Pl7P7kvQDZIj
XLC91Vun1gY6B1YSXfZQWeHaOm/6YJwLFjEJs8G4MHF4rjy9Yfb27HX9AzJr3DvS
Tm7tgZG+CU0PGrzgojxhqSDOND3MoDhPVSSqQ19oQHlTWSEOZQaq+qrtPvYfPRk=
=hEox
-----END PGP SIGNATURE-----
Maxim Sobolev
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Maxim Sobolev wrote:
>> Is the option related to Jumbo Frames in any way? If not, IMHO the name
>> should be changed to avoid any confusion.
>
> The bug is easier to reproduce when jumbo frames are turned on however
> we have seen it with normal frames also.
Can we change the name then? We've seen the similar issue without the
jumbo frames, which made me wonder. Also, perhaps we should make it a
default and/or tunable. Having user to recompile kernel just to save 20
bytes of text segment is not very practical these days.
-Maxim
Maxim Sobolev
> On Mon, 30 Nov 2009 13:25:03 -0800
> Maxim Sobolev <sob...@FreeBSD.org> mentioned:
>
>> Can we change the name then? We've seen the similar issue without the
>> jumbo frames, which made me wonder.
>
> so the issue I was experiencing should not be reproducable with jumbo frames
> disabled. Can you describe a bit more what kind of problem you're experiencing?
The traffic through interface abruptly halts randomly when under a load,
without any signs of a problem. What is interesting is that the
communication through unix domain sockets appears to be down too, since
some scripts that connect to database not able to complete as well, so
whatever this problem is it affect the whole network subsystem. We tried
to replace hardware and it did not help. Only after we replaced bce(4)
with IntelPro card (em) the problem has went away.
Tom Judge
Hash: SHA1
Maxim Sobolev wrote:
> Stanislav Sedov wrote:
>> On Mon, 30 Nov 2009 13:25:03 -0800
>> Maxim Sobolev <sob...@FreeBSD.org> mentioned:
>>
>>> Can we change the name then? We've seen the similar issue without the
>>> jumbo frames, which made me wonder.
>>
>> It's strange, because 'normal' frames are less than a page size in
>> length,
>> so the issue I was experiencing should not be reproducable with jumbo
>> frames
>> disabled. Can you describe a bit more what kind of problem you're
>> experiencing?
>
> The traffic through interface abruptly halts randomly when under a load,
> without any signs of a problem. What is interesting is that the
> communication through unix domain sockets appears to be down too, since
> some scripts that connect to database not able to complete as well, so
> whatever this problem is it affect the whole network subsystem. We tried
> to replace hardware and it did not help. Only after we replaced bce(4)
> with IntelPro card (em) the problem has went away.
>
> -Maxim
When you see this lock up do you also see denied requests for mbufs?
All the times I have seen the issue, the system was denying mbuf requests.
Tom
- --
TJU13-ARIN
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.13 (FreeBSD)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iQEcBAEBAgAGBQJLFRbkAAoJEMSwVS7lr0OdL84H/0rPW0l53qCqMv8HQ45KLd21
9PMEYRT9GkIMR9xQ489BCtcKQEILqA9o9WJoGOHPL/9I3ecYiB3phMp/cTvtcQiu
3I2UNdeDUg3YWqZyPwgXg4snUUod9flITkCzwn8+yx7EmD4cEz9BVi7XXyqwGxoW
GM9BoX2rBttO3DEP2h19JazG9U5BXPW1zZ1Jf47Or6WMMfYeC6dSMA0WMr1fxjgY
VsQVyIH2EfR19ddwZkJqkx32y5+TDgjsevlu0cw/0KjxLii2NqAtLtPQNLfDC4Wo
wKWpi0CXqdkCinwJsKcVISYj6TLI3wAz5U2q5l4bDNRkp/0YkgHAjdj+VSwKErg=
=o5DH
-----END PGP SIGNATURE-----
Maxim Sobolev
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Maxim Sobolev wrote:
>> Stanislav Sedov wrote:
>>> On Mon, 30 Nov 2009 13:25:03 -0800
>>> Maxim Sobolev <sob...@FreeBSD.org> mentioned:
>>>
>>>> Can we change the name then? We've seen the similar issue without the
>>>> jumbo frames, which made me wonder.
>>> It's strange, because 'normal' frames are less than a page size in
>>> length,
>>> so the issue I was experiencing should not be reproducable with jumbo
>>> frames
>>> disabled. Can you describe a bit more what kind of problem you're
>>> experiencing?
>> The traffic through interface abruptly halts randomly when under a load,
>> without any signs of a problem. What is interesting is that the
>> communication through unix domain sockets appears to be down too, since
>> some scripts that connect to database not able to complete as well, so
>> whatever this problem is it affect the whole network subsystem. We tried
>> to replace hardware and it did not help. Only after we replaced bce(4)
>> with IntelPro card (em) the problem has went away.
>
>
> All the times I have seen the issue, the system was denying mbuf requests.
Not sure about that, since we only have had network connection to that
box, no console. However, I think that the problem with the other
network subsystems that we see at the same time suggests that it might
be an issue with the mbufs indeed.
-Maxim
Attila Nagy
> Stanislav Sedov wrote:
>> On Mon, 30 Nov 2009 13:25:03 -0800
>> Maxim Sobolev <sob...@FreeBSD.org> mentioned:
>>
>>> Can we change the name then? We've seen the similar issue without
>>> the jumbo frames, which made me wonder.
>>
>> It's strange, because 'normal' frames are less than a page size in
>> length,
>> so the issue I was experiencing should not be reproducable with jumbo
>> frames
>> disabled. Can you describe a bit more what kind of problem you're
>> experiencing?
>
> The traffic through interface abruptly halts randomly when under a
> load, without any signs of a problem. What is interesting is that the
> communication through unix domain sockets appears to be down too,
> since some scripts that connect to database not able to complete as
> well, so whatever this problem is it affect the whole network
> subsystem. We tried to replace hardware and it did not help. Only
> after we replaced bce(4) with IntelPro card (em) the problem has went
> away.
I can see similar, although only (at least for now) on version 8
(running something around RC2), also with bce interfaces and "heavy" -in
terms of pps, this is a DNS server- network load.
Everything which involves network halts for some minutes, but the
machine is completely usable from the console. top shows some processes
(I don't have an output, but as far as I can remember they were only
that which did network IO) in the "keglim" state.
Then something happens and the machine starts to work again.
Can this be the same, or is it completely different?
Tom Judge
Hash: SHA1
Maxim Sobolev wrote:
> Tom Judge wrote:
>>
>>
>> When you see this lock up do you also see denied requests for mbufs?
>>
>> All the times I have seen the issue, the system was denying mbuf
>> requests.
>
> Not sure about that, since we only have had network connection to that
> box, no console. However, I think that the problem with the other
> network subsystems that we see at the same time suggests that it might
> be an issue with the mbufs indeed.
>
If you could put some simple cron job in that dumps the denied counters
to disk every minute or so that would help diagnose the issue.
Tom
- --
TJU13-ARIN
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.13 (FreeBSD)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iQEcBAEBAgAGBQJLFaNYAAoJEMSwVS7lr0OdIwkIAJdQXEU9a+nQ9tjq8YezkRHU
/SMEst2To8NxH+AktOLr0dkOmMi484VFJRWLUEyBBi1AAKELNfLucTwCisvhTLmX
6LgqDWghJdPnx014HWozHLbkakeeL/rmFyo5ZdbfWhkQgPxwcxyB5EkKHLcOz0OF
SnRT57vTdFQwHPOSMV5DcvO2XwTP2zHdqecT2edvAbY4LQeP8QTtLkOrwuPHnP1w
vknrpNxgd+A7Wt1ME0HMUD4UuZUvuHekw5DhM07YJGWSJXdu3HmO2DJf3jMznLAV
EC43I+Uouz6zo3fVCp5Jhvmf57fK6wAFxhhdvuVPn0niD6VDgnxj5gSoUKOyyNg=
=mWfX
-----END PGP SIGNATURE-----
Maxim Sobolev
It really looks like the same issue. Those boxes are heavy-duty VoIP
switches, so that they are handling high PPS as well in our case.
-Maxim
Attila Nagy
restarted, sadly this is a diskless machine):
last pid: 86028; load averages: 0.53, 0.35, 0.27 up 0+05:23:37
12:49:15
47 processes: 2 running, 43 sleeping, 1 zombie, 1 lock
CPU: 0.1% user, 0.0% nice, 0.1% system, 27.3% interrupt, 72.6% idle
Mem: 3200M Active, 21M Inact, 221M Wired, 228K Cache, 43M Buf, 4471M Free
Swap:
PID USERNAME THR PRI NICE SIZE RES STATE C TIME WCPU COMMAND
720 bind 1 64 0 3153M 3154M RUN 0 63:40 0.83% unbound
85972 root 1 60 0 37428K 10936K *udp 1 0:00 0.15% python
721 bind 3 44 0 70892K 24328K uwait 2 7:38 0.00%
python2.6
2575 root 1 76 0 8212K 2376K pause 1 0:06 0.00% csh
556 root 1 44 0 14972K 1848K select 3 0:05 0.00%
syslog-ng
I can't type on already established ssh connections, and nothing comes
out from the machine (for example via syslog).
This is on 8-STABLE from yesterday...
Thanks,