Re: Problems with Kernels 3.17-rc1 and onwards on Acube Sam460 AMCC 460ex board

[Julian Margetson <runaway@candw.ms>]

On Wed, 2015-02-18 at 15:45 -0400, Julian Margetson wrote:
> On 2/15/2015 8:18 PM, Michael Ellerman wrote:
>
> > On Sun, 2015-02-15 at 08:16 -0400, Julian Margetson wrote:
> > > Hi
> > >
> > > I am unable to get any kernel beyond  the 3.16 branch working on an
> > > Acube Sam460ex
> > >  AMCC 460ex based motherboard. Kernel  up 3.16.7-ckt6 working.
> > Does reverting b0345bbc6d09 change anything?
> >
> > > [    6.364350] snd_hda_intel 0001:81:00.1: enabling device (0000 -> 0002)
> > > [    6.453794] snd_hda_intel 0001:81:00.1: ppc4xx_setup_msi_irqs: fail mapping irq
> > > [    6.487530] Unable to handle kernel paging request for data at address 0x0fa06c7c
> > > [    6.495055] Faulting instruction address: 0xc032202c
> > > [    6.500033] Vector: 300 (Data Access) at [efa31cf0]
> > > [    6.504922]     pc: c032202c: __reg_op+0xe8/0x100
> > > [    6.509697]     lr: c0014f88: msi_bitmap_free_hwirqs+0x50/0x94
> > > [    6.515600]     sp: efa31da0
> > > [    6.518491]    msr: 21000
> > > [    6.521112]    dar: fa06c7c
> > > [    6.523915]  dsisr: 0
> > > [    6.526190]   current = 0xef8bab00
> > > [    6.529603]     pid   = 115, comm = kworker/0:1
> > > [    6.534163] enter ? for help
> > > [    6.537054] [link register   ] c0014f88 msi_bitmap_free_hwirqs+0x50/0x94
> > > [    6.543811] [efa31da0] c0014f78 msi_bitmap_free_hwirqs+0x40/0x94 (unreliable)
> > > [    6.551001] [efa31dc0] c001aee8 ppc4xx_setup_msi_irqs+0xac/0xf4
> > > [    6.556973] [efa31e00] c03503a4 pci_enable_msi_range+0x1e0/0x280
> > > [    6.563032] [efa31e40] f92c2f74 azx_probe_work+0xe0/0x57c [snd_hda_intel]
> > > [    6.569906] [efa31e80] c0036344 process_one_work+0x1e8/0x2f0
> > > [    6.575627] [efa31eb0] c003677c worker_thread+0x2f4/0x438
> > > [    6.581079] [efa31ef0] c003a3e4 kthread+0xc8/0xcc
> > > [    6.585844] [efa31f40] c000aec4 ret_from_kernel_thread+0x5c/0x64
> > > [    6.591910] mon>  <no input ...>
>
> Managed to do a third git bisect  with the following results .

Great work.

> git bisect bad
> 9279d3286e10736766edcaf815ae10e00856e448 is the first bad commit
> commit 9279d3286e10736766edcaf815ae10e00856e448
> Author: Rasmus Villemoes <linux@rasmusvillemoes.dk>
> Date:   Wed Aug 6 16:10:16 2014 -0700
>
>     lib: bitmap: change parameter of bitmap_*_region to unsigned
>
>     Changing the pos parameter of __reg_op to unsigned allows the compiler
>     to generate slightly smaller and simpler code.  Also update its callers
>     bitmap_*_region to receive and pass unsigned int.  The return types of
>     bitmap_find_free_region and bitmap_allocate_region are still int to
>     allow a negative error code to be returned.  An int is certainly capable
>     of representing any realistic return value.

So that looks feasible as the culprit.

Looking at the 4xx MSI code, it just looks wrong:

static int ppc4xx_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
{
	...

	list_for_each_entry(entry, &dev->msi_list, list) {
		int_no = msi_bitmap_alloc_hwirqs(&msi_data->bitmap, 1);
		if (int_no >= 0)
			break;

That's backward, a *negative* return indicates an error.

		if (int_no < 0) {
			pr_debug("%s: fail allocating msi interrupt\n",
					__func__);
		}

This is the correct check, but it just prints a warning and then continues,
which is not going to work.

		virq = irq_of_parse_and_map(msi_data->msi_dev, int_no);

This will fail if int_no is negative.

		if (virq == NO_IRQ) {
			dev_err(&dev->dev, "%s: fail mapping irq\n", __func__);
			msi_bitmap_free_hwirqs(&msi_data->bitmap, int_no, 1);

And so here we can pass a negative int_no to the free routine, which 
then oopses.

			return -ENOSPC;
		}


So the bug is in the 4xx MSI code, and has always been there, in fact I 
don't
see how that code has *ever* worked. The commit you bisected to just 
caused the
existing bug to cause an oops.

Can you try this?

diff --git a/arch/powerpc/sysdev/ppc4xx_msi.c 
b/arch/powerpc/sysdev/ppc4xx_msi.c
index 6e2e6aa378bb..effb5b878a78 100644
--- a/arch/powerpc/sysdev/ppc4xx_msi.c
+++ b/arch/powerpc/sysdev/ppc4xx_msi.c
@@ -95,11 +95,9 @@ static int ppc4xx_setup_msi_irqs(struct pci_dev *dev, 
int nvec, int type)
   	list_for_each_entry(entry, &dev->msi_list, list) {
  		int_no = msi_bitmap_alloc_hwirqs(&msi_data->bitmap, 1);
-		if (int_no >= 0)
-			break;
  		if (int_no < 0) {
-			pr_debug("%s: fail allocating msi interrupt\n",
-					__func__);
+			pr_warn("%s: fail allocating msi interrupt\n", __func__);
+			return -ENOSPC;
  		}
  		virq = irq_of_parse_and_map(msi_data->msi_dev, int_no);
  		if (virq == NO_IRQ) {

cheers


Can also confirm patch working with kernel 3.18.7 .