omap24xx is broken in current head

Hi,

at least n810 is not booting with current l-o head

The problem is in clockdomain.c:omap2_clkdm_clk_enable/disable() functions.

Well, just for keep hacking I returned on top of the function and it's ok so far :-p

Below is the Unhandled fault exception from console.

<5>Kernel command line: root=1f03 rootfstype=jffs2 <6>Clocking rate (Crystal/DPLL/MPU): 19.2/658/329 MHz <1>Unhandled fault: alignment exception (0x001) at 0xc02c1b4e

Internal error: : 1 [#1]
Modules linked in:
CPU: 0    Not tainted  (2.6.26-omap1 #10)
PC is at omap2_clkdm_clk_enable+0x2c/0x74
LR is at omap2_clk_enable+0x54/0x9c
pc : [<c00348e0>]    lr : [<c0033118>]    psr: a00001d3

sp : c0307e90 ip : c0307ea8 fp : c0307ea4
r10: 800242c0 r9 : 4107b362 r8 : 00000013
r7 : c032b4c4 r6 : 00000002 r5 : 00000000 r4 : c02c1b42 r3 : c02c1b4e r2 : 00000000 r1 : c030b5d4 r0 : c02c1b42
Flags: NzCv IRQs off FIQs off Mode SVC_32 ISA ARM Segment kernel
Control: 00c5387f Table: 80004000 DAC: 00000017
Process swapper (pid: 0, stack limit = 0xc0306260)

Stack: (0xc0307e90 to 0xc0308000)

7e80: c030b5d4 c030b030 c0307ebc c0307ea8
7ea0: c0033118 c00348c0 c030b088 c030b030 c0307ed4 c0307ec0 c00330fc c00330d0 7ec0: c030b500 c030b030 c0307eec c0307ed8 c00330fc c00330d0 c030b6fc c030b030 7ee0: c0307f04 c0307ef0 c00330fc c00330d0 c030c09c c030b030 c0307f1c c0307f08 7f00: c00330fc c00330d0 c030dccc c030b030 c0307f34 c0307f20 c00330fc c00330d0 7f20: 800001d3 c030b030 c0307f4c c0307f38 c0043260 c00330d0 c030dccc c030b030 7f40: c0307f64 c0307f50 c0043340 c004322c 00000292 c030b030 c0307f94 c0307f68 7f60: c00114d0 c0043320 00000149 c0309ce0 c0307f94 00000000 c032a8c0 c0025f24 7f80: c0309ce0 8002435c c0307fac c0307f98 c00107e4 c001130c c0025f28 c032a8c0 7fa0: c0307fbc c0307fb0 c0011b70 c00107c8 c0307fcc c0307fc0 c000e438 c0011b68 7fc0: c0307ff4 c0307fd0 c0008c18 c000e40c c0008568 00000000 00000000 c0025f28 7fe0: 00c5387d c032b24c 00000000 c0307ff8 80008034 c0008ad4 00000000 00000000

Backtrace: 
[<c00348b4>] (omap2_clkdm_clk_enable+0x0/0x74) from [<c0033118>] (omap2_clk_enable+0x54/0x9c)
 r5:c030b030 r4:c030b5d4
[<c00330c4>] (omap2_clk_enable+0x0/0x9c) from [<c00330fc>] (omap2_clk_enable+0x38/0x9c)
 r5:c030b030 r4:c030b088
[<c00330c4>] (omap2_clk_enable+0x0/0x9c) from [<c00330fc>] (omap2_clk_enable+0x38/0x9c)
 r5:c030b030 r4:c030b500                                                            
[<c00330c4>] (omap2_clk_enable+0x0/0x9c) from [<c00330fc>] (omap2_clk_enable+0x38/0x9c)
 r5:c030b030 r4:c030b6fc                                                            
[<c00330c4>] (omap2_clk_enable+0x0/0x9c) from [<c00330fc>] (omap2_clk_enable+0x38/0x9c)
 r5:c030b030 r4:c030c09c
[<c00330c4>] (omap2_clk_enable+0x0/0x9c) from [<c00330fc>] (omap2_clk_enable+0x38/0x9c)
 r5:c030b030 r4:c030dccc
[<c00330c4>] (omap2_clk_enable+0x0/0x9c) from [<c0043260>] (clk_enable+0x40/0x54)
 r5:c030b030 r4:800001d3
[<c0043220>] (clk_enable+0x0/0x54) from [<c0043340>] (clk_enable_init_clocks+0x2c/0x4c)
 r5:c030b030 r4:c030dccc
[<c0043314>] (clk_enable_init_clocks+0x0/0x4c) from [<c00114d0>] (omap2_clk_init+0x1d0/0x23c)
 r5:c030b030 r4:00000292
[<c0011300>] (omap2_clk_init+0x0/0x23c) from [<c00107e4>] (omap2_init_common_hw+0x28/0x40)
 r8:8002435c r7:c0309ce0 r6:c0025f24 r5:c032a8c0 r4:00000000
[<c00107bc>] (omap2_init_common_hw+0x0/0x40) from [<c0011b70>] (nokia_n800_init_irq+0x14/0x5c)
 r5:c032a8c0 r4:c0025f28
[<c0011b5c>] (nokia_n800_init_irq+0x0/0x5c) from [<c000e438>] (init_IRQ+0x38/0x48)
[<c000e400>] (init_IRQ+0x0/0x48) from [<c0008c18>] (start_kernel+0x150/0x2a8)
[<c0008ac8>] (start_kernel+0x0/0x2a8) from [<80008034>] (0x80008034)
 r5:c032b24c r4:00c5387d
Code: 03a05001 03e00015 089da830 e284300c (e1931f9f) 

<4>---[ end trace 1b75b31a2719ed1c ]---

I'm gonna try to get more info on this later

balbi

Reverting commit 718fc6cd4db902aa2242a736cc3feb8744a4c71a, looks like gcc is not being able to align that union ??

I'm using codesourcery 2008q1-126 (gcc 4.2.3)

Did you guys have any problems with this version of CS ??

balbi

For OMAP3/Cortex-A8 at least some issues are known with 2008q1:

* Vectorization + NEON is broken
* building static binaries with cortex-a8 flag (or any ARMv7a core) is broken
* some armv6 compilations end in ICE

Therefore, for Beagle & friends we recommend 2007q3. This has a known issue with -Os option, though.

Dirk

Op 10 aug 2008, om 14:16 heeft Felipe Balbi het volgende geschreven:

"Broken as hell" if you want to do anything NEON related with it. I use 2007q3 with the -Os fix for compiling omap3 kernels and vanilla gcc 4.2.4 for all other arm kernels. I would rather uses vanilla gcc 4.3.1 + neon fix, but then I'd need to patch all kernels for the pld problem :/

regards,

Koen

yup
Very unlucky but very true.

Felipe: that's also what we should use internally, since it's the only one approved by our system performance team.

Cheers, Igor

---

Igor Stoppa
Maemo Software - Nokia Devices R&D - Helsinki

I'll revert this patch for now to avoid adding a compiler dependency as we already have a pile of other issues to solve.

Let's plan on re-applying this patch once the other issues are out of the way.

Tony

FYI, I just tried with 2007q3 and had the same issue :-p

I'll try later with vanilla gcc

balbi

Op 11 aug 2008, om 16:50 heeft Tony Lindgren het volgende geschreven:

Speaking of compiler dependencies, how about applying my patch that enable vanilla gcc to compile the kernel? Russel (or Erik) seems to be fond of rejecting my mails to the LAKML list, so I can't post it there. A copy of it can be found here:

http://ewi546.ewi.utwente.nl/git/?p=openembedded.git;a=blob;f=packages/linux/linux-omap2-git/beagleboard/no-harry-potter.diff;h=2bb20ab9c025312fcf0434de5bcb88aeceb4f87f;hb=HEAD

regards,

Koen

That seems like a good patch, you should see what the issue is with posting to LAKML. Maybe add it to Russell's patch system meanwhile? :)

Meanwhile, yeah I'd say if that works on more compilers, let's add it to l-o tree.

Tony

I still don't see this patch anywhere.. Koen, what's your mail problem with LAKML? Maybe Russell can help with this issue?

Anyways, I'll apply this to l-o tree. Seems like the dash is the right naming according to this:

http://sourceware.org/ml/binutils/2006-12/msg00149.html

Tony