Alan, Garrett,
I made bug and improvement list which includes corresponding
line numbers. It's big.
-masa
================================================
Bugs:
Original driver version was not included in sfe.c.
sfe.c
55
PCI parity check was disabled. It should be enabled.
sfe.c
562
579
616-617
sfereg.h
177
Tx fifo setting was incorrect. It caused tx timeouts.
It was exposed after the sis900 MII autonegotiation problem was fixed.
The root cause was a typo of valiable name. :-(
sfe.c
1062
Reading tx/rx descriptors were not atomic. It may cause
hangs or lost packets for sparc platforms.
1256-1257
1361-1362
NS DP83815 hung in transmitting under heavy network traffic.
sfe.c
1247
1267-1271
2160-2163
External PHY was not supported for NS chipset.
sfe.c
1675-1725
2219-2220
Procedures for accessing MII PHY were incorrect for sis900 that
made autonegotiation failed (CR#6655415)
sfe.c
1731
1791
1800
1805-1806
1809
1814
1820-1821
1826
1850
1852
1756-1759 in old sfe.c were removed
Incorrect comma at the end of statement.
sfe.c
2060
sfe hung in transmitting vlan packets in sparc platforms.
(I really didn't know why it happened only for vlan packets
under sparc)
sfe_util.c
971
1064
1066-1070
1123-1125
1149
1215
1114-1116 in old sfe_util.c were removed
1298 in old sfe_util.c was removed
1495-1499
1364-1373 in old sfe_util.c were removed
1535
1436-1441 in old sfe_util.c were removed
1467-1473 in old sfe_util.c were removed
1666
1675-1681
1698
1709-1710
1985-1086
1991-1992
1999-2001
2006-2007
2549-2550 in old sfe_util.c were removed
2703-2705
3098
3028 in old sfe_util.c
Incorrect tx statistics
sfe_util.c
1251
1327
1532
1558
Netperfs in test08 i.e. vlan tests of nicdrv caused timeouts or
were not scheduled fairly.
sfe_util.c
1480
1346-1350 in old sfe_util.c were removed
1486
3101
sfe_util.h
233
242
sfe didn't work with NWAM
sfe_util.c
2256-2267
2810-2821
2968
sfe_util.h
307
cap_autoneg in kstat may be incorrect.
sfe_util.c
4410
Vlan capabiliy was enabled for sis900 which didn't support vlan.
sfe_util.c
4664-4666
Forced mode was not setup correctly
sfe_util.c
4697-4704
4973
Corrupted warning messages
sfe_util.c
4737
4747
Packets may be corrupted in sparc platforms, because rx/tx buffers
may not be aligned for io cache.
sfe_util.c
4864-4866
Detach won't work correctly if mac_unregister will return errors.
sfe_util.c
5091-5094
4989-4990 in old sfe_util.c
-----------------------------------------------------------
Making the bug list above, I found another bug. I'd like to fix below too.
Incorrect tx timeout period. It's too short.
1675 if (dp->tx_blocked &&
1676 now - dp->tx_blocked > dp->gc.gc_tx_timeout_interval) {
should be
1675 if (dp->tx_blocked &&
1676 now - dp->tx_blocked > dp->gc.gc_tx_timeout) {
================================================================
Improvements:
Improving performance in making tx/rx descriptors
sfe.c
1146-1147
1177-1180
1216-1217
1234-1237
1243
1254
Use tx_ring_size and rx_ring_size in gem_conf structure rather
than the constants. In future, the ring size constant
should be seperated from hardware dependant methods.
sfe.c
1188
1194
1243
1254
1296
1437
1450
1457
1470
1498
1550
Make NS DP83815 PHY registers non-writable after modifying
PHY parameters.
sfe.c
1644
Patching PHY parameters for recent NS DP83815/83816 chipset
sfe.c
1654-1669
Improving performance inconverting value to boolean
sfe_util.c
117
A new function dumping tx packets was added for debugging purpose.
sfe_util.c
214-248
251-352
Removing unused vlan tag function
sfe_util.c
230-250 in old sfe_util.c
253
Changing dump funtions because of framework change.
sfe_util.c
437
LSO support. (it doesn't affect sfe)
sfe_util.c
964-966
1111-1114
sfe_util.h
463-466
531
Improving performance
sfe_util.c
1149
Unused lines for vlan tag processing were removed
sfe_util.c
1163-1184 in old sfe_util.c
Readablity of gem_send_common() was improved.
A new subroutine was seperated from gem_semd_common()
sfe_util.c
1374-1420
1301-1302, 1383 in old sfe_util.c were removed.
1507
1383 in old sfe_util.c was removed
1391-1410, 1412-1413 in old sfe_util were removed
1526-1530
1425-1432 in old sfe_util were removed
Debugging information were improved
sfe_util.c
1575-1580
1693
1709-1710
Use macro instead of constant, to specify parameter
in calling gem_restart_nic()
sfe_util.c
1696
ddi_dma_sync() performance improved
sfe_util.c
1777-1782
sync iocache for rx header in received packets. (it
doesn't affect sfe.)
sfe_util.c
1812
1835-1838
Unused vlan tag code were removed
1875
1769-1777 in old sfe_util.c were removed
Autonegotiations didn't start for some PHYs. (It will not
affect sis900.)
sfe_util.c
2797-2798
cstyle issues were fixed
sfe_util.c
2806
3749
receive buffers were not ready when the mac started. (It doesn't
affect sfe.
sfe_util.c
3132-3135
3016-3019 in old sfe_util.c were removed
Workarond for panics in getting statistics
It happened in GLDv2. It should be fixed for GLDv3 too, because
I'm not sure whether it will happen in GLDv3. The root cause is
calling methods in GLD with holding a private mutex lock, intrlock
in gem_dev structure. But I cannot fix the root cause for now
because it will need big change.
sfe_util.c
4233-4237
rx_pkt_delay may become too big. It doesn't affect sfe.
sfe_util.c
4599
mrf_normal_blank_time was incorrect. (It doesn't affect sfe)
gem controles packet count, not blank time.
sfe_util.c
4621
Unused code were removed
sfe_util.c
4712 in old sfe_util.c
4719-4726 in old sfe_util.c
performance improvement in gem polling mode. The default value
was changed. (It doesn't affect sfe)
sfe_util.c
4902
Fix for supportting multiport devices, like marvell nics. (It
doesn't affect sfe)
sfe_util.c
5040
5137
5013 in old sfe_util.c was removed.
sfe_util.h
559
An unused member in gem_dev was removed
sfe_util.h
336
----- Original Message -----
>Date: Thu, 12 Jun 2008 14:54:13 -0700 (PDT)
>From: Alan DuBoff <[EMAIL PROTECTED]>
>To: "Garrett D'Amore" <[EMAIL PROTECTED]>
>Cc: driver-discuss@opensolaris.org
>Subject: Re: [driver-discuss] Codereview for CR# 6655415 sfe auto-negotiate
>
>
>On Wed, 11 Jun 2008, Garrett D'Amore wrote:
>
>> So maybe instead of a gazillion different CRs, you could just do a single C
R
>> saying "incorporate changes from upstream supplier". I'd still like to have
>> some explanation in that CR for what all the changes are for, but you can
>> coordinate with Masa to get the details.
>>
>> (Of course, listing all the CRs individually is even better, but I do
>> recognize it is potentially a fair bit more effort.)
>
>Garrett,
>
>It looks like I can change a CR as fix-delivered.
>
>I think I should file another CR for all the bug fixes which Masa creates
>a list of, and then just mark this one fixed-delivered...and refer to the
>new CR in this old one. New list would include the auto-negotiate bug of
>course.
>
>Does that sound ok?
>
>--
>
>Alan DuBoff - Solaris x86 IHV/OEM Group
>_______________________________________________
>driver-discuss mailing list
>driver-discuss@opensolaris.org
>http://mail.opensolaris.org/mailman/listinfo/driver-discuss
_______________________________________________
driver-discuss mailing list
driver-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/driver-discuss
