PROBLEM: 2.6.13 crash on sparc64 in sunsu_kbd_ms_interrupt
1. 2.6.13 crash on sparc64 in sunsu_kbd_ms_interrupt
2. I'm running 2.6.13/sparc64 on a diskless Sun Ultra 5 workstation. After an uptime between 5 and 20 minutes after start, the kernel crashes, system stalls.
3. sparc64 sun keyboard sunsu_kbd_ms_interrupt
4. vanilla 2.6.13
5. Last working kernel version is not known. 2.6.12.5 shows the same behaviour.
6. Kernel message is attached to this mail.
8.1. Gnu C 3.4.2, binutils 2.15
8.2.
sunny:~# cat /proc/cpuinfo cpu : TI UltraSparc IIi (Sabre) fpu : UltraSparc IIi integrated FPU promlib : Version 3 Revision 11 prom : 3.11.12 type : sun4u ncpus probed : 1 ncpus active : 1 Cpu0Bogo : 539.03 Cpu0ClkTck : 000000001017df80 MMU Type : Spitfire
8.3. no modules loaded
8.4.
sunny:~# cat /proc/ioports
1fe02000000-1fe02bfffff : SABRE0 PBMA
1fe02c00000-1fe02ffffff : SABRE0 PBMB
1fe02c00000-1fe02c00007 : 0000:01:03.0
1fe02c00008-1fe02c0000b : 0000:01:03.0
1fe02c00010-1fe02c00017 : 0000:01:03.0
1fe02c00018-1fe02c0001b : 0000:01:03.0
1fe02c00020-1fe02c0002f : 0000:01:03.0
1fe02c00400-1fe02c004ff : 0000:01:02.0
sunny:~# cat /proc/iomem
1ff00000000-1ffbfffffff : SABRE0 PBMA
1ff000a0000-1ff000bffff : Video RAM area
1ff000c0000-1ff000c7fff : Video ROM
1ff000f0000-1ff000fffff : System ROM
1ffe0000000-1ffffffffff : SABRE0 PBMB
1ffe0000000-1ffe0007fff : 0000:01:01.1
1ffe0000000-1ffe0007fff : sunhme
1ffe0008000-1ffe0008fff : 0000:01:02.0
1ffe0020000-1ffe003ffff : 0000:01:02.0
1ffe1000000-1ffe1ffffff : 0000:01:02.0
1ffe2000000-1ffe2ffffff : 0000:01:01.0
1ffe3000000-1ffe3ffffff : 0000:01:01.1
1fff0000000-1fff0ffffff : 0000:01:01.0
1fff0000000-1fff00fffff : flashprom
1fff1000000-1fff17fffff : 0000:01:01.0
1fff1000000-1fff1001fff : eeprom
1fff1200000-1fff12000ff : SUNW,CS4231
1fff130015c-1fff130015d : ecpp
1fff13023f0-1fff13023f7 : fdthree
1fff13043bc-1fff13043cb : ecpp
1fff13062f8-1fff13062ff : su
1fff13083f8-1fff13083ff : su
1fff1400000-1fff140007f : se
1fff1504000-1fff1504002 : SUNW,pll
1fff1700000-1fff170000f : ecpp
1fff1702000-1fff170200f : SUNW,CS4231
1fff1704000-1fff170400f : SUNW,CS4231
1fff1706000-1fff170600f : fdthree
1fff1720000-1fff1720003 : fdthree
1fff1722000-1fff1722003 : SUNW,CS4231
1fff1724000-1fff1724003 : power
1fff1726000-1fff1726003 : auxio
1fff1728000-1fff1728003 : auxio
1fff172a000-1fff172a003 : auxio
1fff172c000-1fff172c003 : auxio
1fff172f000-1fff172f003 : auxio8.5. lspci -vvv output attached
The kernel has been cross-compiled on a x86 Linux system, and converted from ELF to a.out format in order to boot it via network.
Regards,
Jim
Please turn off CONFIG_PREEMPT on sparc64, it really doesn't get very much testing on this platform.
From: "David S. Miller" <[email protected]>
Date: Mon, 03 Oct 2005 16:50:46 -0700 (PDT)
Quoted 2 lines
Please turn off
CONFIG_PREEMPTon sparc64, it really doesn't get very much testing on this platform.
Nevermind, there is a real bug there in the sunsu driver. We drop/retake the lock when it's not actually held.
The preempt debugging was good at catching this. :-)
Please try this fix, thanks.
diff-tree 62ba86fc7c72edc77a67e04b6fb593db5046cc92 (from baaf9033270c9cb10c1441195bc1a0a7665bee7e)
Author: David S. Miller <[email protected]> Date: Mon Oct 3 16:55:10 2005 -0700
[SUNSU]: Fix bogus locking in sunsu_change_mouse_baud()
The lock is not held when calling this function, so we shouldn't drop then reacquire it.
Based upon a report from Jim MacBaine.
drivers/serial/sunsu.c +0 -4
@@ -518,11 +518,7 @@ static void sunsu_change_mouse_baud(stru quot = up->port.uartclk / (16 * new_baud); -spin_unlock (&up->port.lock); -sunsu_change_speed (&up->port, up->cflag,0 , quot); - -spin_lock (&up->port.lock); }static void receive_kbd_ms_chars (struct uart_sunsu_port *up,struct pt_regs *regs,int is_break)