Translation Fault (L1) while using pmap-v6-new

[Michal Meloun]

Dne 22.12.2015 v 16:23 Bartosz Szczepanek napsal(a):
> Hello,
> 
> currently I'm working on support for Armada38x on FreeBSD-CURRENT
> (patchset was submitted to Phabricator -
> https://reviews.freebsd.org/D4210). After switching to ARM_NEW_PMAP
> problems related with PCIe subsystem emerged, even though that worked
> fine on FreeBSD-10.2.
> 
> My setup consists of Marvell Armada38x GP development board equipped
> with Cortex-A9, PCIe controller serviced by arm/mv/mv_pci.c driver and
> RealTek GE PCI card (re driver). Enabling ARM_NEW_PMAP leads to
> 'Translation Fault (L1)' on write:
> 
>> Starting Network: lo0 re0.
>> lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> metric 0 mtu 16384
>> options=600003<RXCSUM,TXCSUM,RXCSUM_IPV6,TXCSUM_IPV6>
>> inet6 ::1 prefixlen 128
>> inet6 fe80::1%lo0 prefixlen 64 scopeid 0x2
>> inet 127.0.0.1 netmask 0xff000000
>> nd6 options=21<PERFORMNUD,AUTO_LINKLOCAL>
>> re0: flags=8802<FataBROADCAST,SIMPLEl keX,MULTICAST> metrnelric 0 mtu 1500
>> mod options=8209b<Re daXCSUM,TXCSUM,VLAta > aN_MTU,VLAN_HWTAGbortGING,VLAN_HWCSUM: 'T,WOL_MAGIC,LINKSransTATE>
>> on F ether 64lati:70:02:10:f7:20
>>    ault (L1)' on write
>> trapframe: 0xefe78b40
>> FSR=00000805, FAR=80000060, spsr=60000013
>> r0 =c0e796c0, r1 =80000000, r2 =00000004, r3 =00010000
>> r4 =c57f1000, r5 =c57f1000, r6 =00000000, r7 =00000001
>> r8 =c0e47e8c, r9 =c5be3780, r10=c57efb00, r11=efe78be0
>> r12=efe78d43, ssp=efe78bd0, slr=c0971ef4, pc =c0971f64
>> [ thread pid 241 tid 100068 ]
>> Stopped at      re_gmii_readreg+0x50:   str     r3, [r1, #0x060]
>> db>
> 
> (re_gmii_readreg is function in re driver, I made it non-static so it
> is visible in debugger)
> Address it crashes on lies in the PCI devices' memory range, and it
> was accessed successfully several times during boot proccess before
> crash (I put printfs in the exact function where fault occurs). So it
> seems just like the memory mapping has disappeared at some point. I
> put kdb_enter in re attach function (long before translation fault),
> from that point I see that 0x80000000 mapping exists:
> 
>> pcib0: <Marvell Integrated PCI/PCI-E Controller> mem 0xf1080000-0xf1081fff irq 1 on ofwbus0
>> [ thread pid 0 tid 100000 ]
>> Stopped at      kdb_enter+0x58: ldrb    r15, [r15, r15, ror r15]!
>> db> show pmap
>> pmap: 0xC0EAC544
>> PT2MAP: 0xBFC00000
>> pt2tab: 0xC0F04000
>> 0x80000000: Section 0x8001041A, s:1 g:1
>> 0x80100000: Section 0x8011041A, s:1 g:1
>> 0x80200000: Section 0x8021041A, s:1 g:1
>> 0x80300000: Section 0x8031041A, s:1 g:1
>> ...
> 
> Doing 'show pmap' after crash gives me long, long log without
> 0x80000000 occuring. On the other hand, adding vtophys(0x80000060)
> line before affected write operation translates address correctly. It
> is visible in log attached.
> 
> I've also tried to track various functions removing mapping in
> pmap-v6-new, but with no luck. However, problem seems to lie there, as
> system boots fine without ARM_NEW_PMAP option. I would be grateful for
> you advice - what else can I do to investigate the issue?
> 
> Best regards,
> Bartosz Szczepanek
> 
> 
> 
> _______________________________________________
> freebsd-arm at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-arm
> To unsubscribe, send any mail to "freebsd-arm-unsubscribe at freebsd.org"
> 

Isn't 0x80000060 little strange address for KVA?
Michal