custom ide driver causes "Badness in smp_call_function"

custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

[-- Attachment #1: Type: text/plain, Size: 1505 bytes --]

Hi all,

I have added a compact flash disk onto the local bus of an rm9224 mips
processor.  We are using an FPGA as an IDE host adaptor.  Our driver works
great when the kernel is compiled for a single processor.  But if SMP
support is enabled, the kernel dies with: 
            .
Uniform Multi-Platform E-IDE driver Revision: 7.00alpha2
Ide:  Assuming 50MHz system bus speed for PIO modes; override with
idebuss=xx
Badness in smp_call_function at arch/mips/kernel/smp.c:149
Call Trace:
 [<ffffffff8010aff8>] smp_call_function+0x1f8/0x200
 [<ffffffff8032adb0>] schedule+0x950/0xa08
 [<ffffffff8010b9f8>] local_rm9k_perfcounter_irq_startup+0x0/0x68
 [<ffffffff8010baa8>] rm9k_perfcounter_irq_startup+0x48/0x68
 [<ffffffff801658b8>] probe_irq_on+0x2b8/0x2e8
 [<ffffffff80146c00>] del_singleshot_timer_sync+0x38/0x50
 [<ffffffff8028e510>] try_to_identify+0x180/0x190
 [<ffffffff80147d88>] process_timeout+0x0/0x8
 [<ffffffff8028e66c>] do_probe+0x14c/0x338
 [<ffffffff8028e57c>] do_probe+0x5c/0x338
 [<ffffffff8028ec40>] wait_hwif_ready+0x178/0x180
 [<ffffffff8028f230>] probe_hwif+0x508/0x6e8
 [<ffffffff8028eec8>] probe_hwif+0x1a0/0x6e8
 [<ffffffff8028f42c>] probe_hwif_init_with_fixup+0x1c/0xe8
 [<ffffffff8029435c>] ide_3P_init+0xb4/0x118
 [<ffffffff80294324>] ide_3P_init+0x7c/0x118
 [<ffffffff8023d100>] idr_get_new+0x18/0x50
etc. 

Does anyone know what sort of bug could cause problems with SMP, but would
work fine otherwise?  I could supply my driver code if anyone is interested.
Thanks!

Bryan
 

[-- Attachment #2: Type: text/html, Size: 6244 bytes --]

custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

[-- Attachment #1: Type: text/plain, Size: 1505 bytes --]

Hi all,

I have added a compact flash disk onto the local bus of an rm9224 mips
processor.  We are using an FPGA as an IDE host adaptor.  Our driver works
great when the kernel is compiled for a single processor.  But if SMP
support is enabled, the kernel dies with: 
            .
Uniform Multi-Platform E-IDE driver Revision: 7.00alpha2
Ide:  Assuming 50MHz system bus speed for PIO modes; override with
idebuss=xx
Badness in smp_call_function at arch/mips/kernel/smp.c:149
Call Trace:
 [<ffffffff8010aff8>] smp_call_function+0x1f8/0x200
 [<ffffffff8032adb0>] schedule+0x950/0xa08
 [<ffffffff8010b9f8>] local_rm9k_perfcounter_irq_startup+0x0/0x68
 [<ffffffff8010baa8>] rm9k_perfcounter_irq_startup+0x48/0x68
 [<ffffffff801658b8>] probe_irq_on+0x2b8/0x2e8
 [<ffffffff80146c00>] del_singleshot_timer_sync+0x38/0x50
 [<ffffffff8028e510>] try_to_identify+0x180/0x190
 [<ffffffff80147d88>] process_timeout+0x0/0x8
 [<ffffffff8028e66c>] do_probe+0x14c/0x338
 [<ffffffff8028e57c>] do_probe+0x5c/0x338
 [<ffffffff8028ec40>] wait_hwif_ready+0x178/0x180
 [<ffffffff8028f230>] probe_hwif+0x508/0x6e8
 [<ffffffff8028eec8>] probe_hwif+0x1a0/0x6e8
 [<ffffffff8028f42c>] probe_hwif_init_with_fixup+0x1c/0xe8
 [<ffffffff8029435c>] ide_3P_init+0xb4/0x118
 [<ffffffff80294324>] ide_3P_init+0x7c/0x118
 [<ffffffff8023d100>] idr_get_new+0x18/0x50
etc. 

Does anyone know what sort of bug could cause problems with SMP, but would
work fine otherwise?  I could supply my driver code if anyone is interested.
Thanks!

Bryan
 

[-- Attachment #2: Type: text/html, Size: 6244 bytes --]

Re: custom ide driver causes "Badness in smp_call_function"

[Ralf Baechle]

On Tue, Aug 23, 2005 at 10:07:02AM -0400, Bryan Althouse wrote:

> I have added a compact flash disk onto the local bus of an rm9224 mips
> processor.  We are using an FPGA as an IDE host adaptor.  Our driver works
> great when the kernel is compiled for a single processor.  But if SMP
> support is enabled, the kernel dies with: 
>             .
> Uniform Multi-Platform E-IDE driver Revision: 7.00alpha2
> Ide:  Assuming 50MHz system bus speed for PIO modes; override with
> idebuss=xx
> Badness in smp_call_function at arch/mips/kernel/smp.c:149

That's just a rather drastically looking warning, btw.

> Call Trace:
>  [<ffffffff8010aff8>] smp_call_function+0x1f8/0x200
>  [<ffffffff8032adb0>] schedule+0x950/0xa08
>  [<ffffffff8010b9f8>] local_rm9k_perfcounter_irq_startup+0x0/0x68
>  [<ffffffff8010baa8>] rm9k_perfcounter_irq_startup+0x48/0x68
>  [<ffffffff801658b8>] probe_irq_on+0x2b8/0x2e8
>  [<ffffffff80146c00>] del_singleshot_timer_sync+0x38/0x50
>  [<ffffffff8028e510>] try_to_identify+0x180/0x190
>  [<ffffffff80147d88>] process_timeout+0x0/0x8
>  [<ffffffff8028e66c>] do_probe+0x14c/0x338
>  [<ffffffff8028e57c>] do_probe+0x5c/0x338
>  [<ffffffff8028ec40>] wait_hwif_ready+0x178/0x180
>  [<ffffffff8028f230>] probe_hwif+0x508/0x6e8
>  [<ffffffff8028eec8>] probe_hwif+0x1a0/0x6e8
>  [<ffffffff8028f42c>] probe_hwif_init_with_fixup+0x1c/0xe8
>  [<ffffffff8029435c>] ide_3P_init+0xb4/0x118
>  [<ffffffff80294324>] ide_3P_init+0x7c/0x118
>  [<ffffffff8023d100>] idr_get_new+0x18/0x50
> etc. 
> 
> Does anyone know what sort of bug could cause problems with SMP, but would
> work fine otherwise?  I could supply my driver code if anyone is interested.
> Thanks!

Your driver is probably fine.  The problem is that doing PIO may result in
cache aliases and that requires a cache flush.  Normally that's not
terribly hard to do - but in your case an SMP cacheflush is needed which
in turn requires smp_call_function to be used and that one again may
deadlock if called with interrupts disabled.  See also
include/asm-mips/mach-generic/ide.h.

So until we have a better implementation I suggest try to avoid the use
of ide_inb() etc. with interrupts disabled.

  Ralf

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Ralf,

Thank you for your help.  
I'm doing MMIO, not PIO, but it looks like your assessment is still valid.
I've been searching for places where ide MMIO is performed with interrupts
disabled.  I got excited when I found these lines in probe_hwif() of
ide-probe.c:

   irqd = hwif->irq;
   if (irqd)
        disable_irq(hwif->irq);

I was not initializing hwif->irq in my driver, so probably the interrupts
were being disabled here, and subsequent lines were causing the SMP badness.
I added the line "hwif->irq = 0" to my driver.  Interrupts are no longer
disabled here, but still I get the SMP badness.  I'll keep looking for other
places where the interrupts might be disabled.  

Does anyone know if the mips/swarm.c driver has this problem with SMP?
Thanks!
  
Bryan

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Ralf,

Thank you for your help.  
I'm doing MMIO, not PIO, but it looks like your assessment is still valid.
I've been searching for places where ide MMIO is performed with interrupts
disabled.  I got excited when I found these lines in probe_hwif() of
ide-probe.c:

   irqd = hwif->irq;
   if (irqd)
        disable_irq(hwif->irq);

I was not initializing hwif->irq in my driver, so probably the interrupts
were being disabled here, and subsequent lines were causing the SMP badness.
I added the line "hwif->irq = 0" to my driver.  Interrupts are no longer
disabled here, but still I get the SMP badness.  I'll keep looking for other
places where the interrupts might be disabled.  

Does anyone know if the mips/swarm.c driver has this problem with SMP?
Thanks!
  
Bryan

Re: custom ide driver causes "Badness in smp_call_function"

[Ralf Baechle]

On Thu, Aug 25, 2005 at 11:26:58AM -0400, Bryan Althouse wrote:
> From:	"Bryan Althouse" <bryan.althouse@3phoenix.com>
> To:	"'Ralf Baechle'" <ralf@linux-mips.org>
> Cc:	<linux-mips@linux-mips.org>
> Subject: RE: custom ide driver causes "Badness in smp_call_function"
> Date:	Thu, 25 Aug 2005 11:26:58 -0400
> Content-Type: text/plain;
> 	charset="us-ascii"
> 
> Ralf,
> 
> Thank you for your help.  
> I'm doing MMIO, not PIO, but it looks like your assessment is still valid.

... which still is programmed io ...

> I've been searching for places where ide MMIO is performed with interrupts
> disabled.  I got excited when I found these lines in probe_hwif() of
> ide-probe.c:
> 
>    irqd = hwif->irq;
>    if (irqd)
>         disable_irq(hwif->irq);
> 
> I was not initializing hwif->irq in my driver, so probably the interrupts
> were being disabled here, and subsequent lines were causing the SMP badness.
> I added the line "hwif->irq = 0" to my driver.  Interrupts are no longer
> disabled here, but still I get the SMP badness.  I'll keep looking for other
> places where the interrupts might be disabled.  
> 
> Does anyone know if the mips/swarm.c driver has this problem with SMP?
> Thanks!

No, SB1 has sane caches.

  Ralf

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Here are some observations...  

If I change the line "hwif->irq = 0" in my driver to "hwif->irq = 5", my SMP
kernel no longer experiences SMP badness.  Instead, I get many lines like
"hda: lost interrupt", and the drive is not usable.  If I compile the kernel
without SMP, the drive works properly as before.  I tried irq = 5 because I
noticed that /proc/interrupts indicated that ide0 was being probed at 5.
With the SMP kernel, /proc/interrupts shows a count of 0 for ide0.  My
non-SMP kernel shows a count that increments when the drive is being used
(as expected).

Bryan  

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Here are some observations...  

If I change the line "hwif->irq = 0" in my driver to "hwif->irq = 5", my SMP
kernel no longer experiences SMP badness.  Instead, I get many lines like
"hda: lost interrupt", and the drive is not usable.  If I compile the kernel
without SMP, the drive works properly as before.  I tried irq = 5 because I
noticed that /proc/interrupts indicated that ide0 was being probed at 5.
With the SMP kernel, /proc/interrupts shows a count of 0 for ide0.  My
non-SMP kernel shows a count that increments when the drive is being used
(as expected).

Bryan  

Re: custom ide driver causes "Badness in smp_call_function"

[Ralf Baechle]

On Thu, Aug 25, 2005 at 05:17:48PM -0400, Bryan Althouse wrote:

> If I change the line "hwif->irq = 0" in my driver to "hwif->irq = 5", my SMP
> kernel no longer experiences SMP badness.  Instead, I get many lines like
> "hda: lost interrupt", and the drive is not usable.  If I compile the kernel
> without SMP, the drive works properly as before.  I tried irq = 5 because I
> noticed that /proc/interrupts indicated that ide0 was being probed at 5.
> With the SMP kernel, /proc/interrupts shows a count of 0 for ide0.  My
> non-SMP kernel shows a count that increments when the drive is being used
> (as expected).

Try this patch below and let me know.  I would also like to ask those
people who used to suffer from aliases with IDE PIO to try this patch.

  Ralf

Index: include/asm-mips/mach-generic/ide.h
===================================================================
RCS file: /home/cvs/linux/include/asm-mips/mach-generic/ide.h,v
retrieving revision 1.9
diff -u -r1.9 ide.h
--- include/asm-mips/mach-generic/ide.h	19 Apr 2005 12:26:59 -0000	1.9
+++ include/asm-mips/mach-generic/ide.h	26 Aug 2005 14:04:38 -0000
@@ -19,6 +19,7 @@
 #include <linux/pci.h>
 #include <linux/stddef.h>
 #include <asm/processor.h>
+#include <asm/cacheflush.h>
 
 #ifndef MAX_HWIFS
 # ifdef CONFIG_BLK_DEV_IDEPCI
@@ -105,12 +106,14 @@
 
 /* MIPS port and memory-mapped I/O string operations.  */
 
-static inline void __ide_flush_dcache_range(unsigned long addr, unsigned long size)
+static inline void __ide_flush_dcache_range(unsigned long addr,
+	unsigned long size)
 {
-	if (cpu_has_dc_aliases) {
-		unsigned long end = addr + size;
-		for (; addr < end; addr += PAGE_SIZE)
-			flush_dcache_page(virt_to_page(addr));
+	unsigned long end = addr + size;
+
+	while (addr < end) {
+		SetPageDcacheDirty(virt_to_page(addr));
+		addr += PAGE_SIZE;
 	}
 }
 

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Ralf,

The patch doesn't seem to make any difference. :(

Bryan 

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Ralf,

The patch doesn't seem to make any difference. :(

Bryan 

RE: custom ide driver causes "Badness in smp_call_function"

[Alan Cox]

On Gwe, 2005-08-26 at 10:58 -0400, Bryan Althouse wrote: 
> Ralf,
> 
> The patch doesn't seem to make any difference. :(

Assuming your hardware is sane another approach might be to force
drive->unmask = 1. That will mean that PIO mode is running with
interrupts enabled which should avoid the problem.

Add a .fixup handler to your driver (assuming you are using a recent
2.6.x) and in the handler do something like this:

+void ide_unmask_interrupts(ide_hwif_t *hwif)
+{
+       int i;
+       for (i = 0; i < 2; i++) {
+               ide_drive_t *drive = &hwif->drives[i];
+               if(drive->present)
+                       drive->unmask = 1;
+       }
+}

hopefully that will be early enough.

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Alan,

Thanks for your suggestion.
I'm not sure how to write a .fixup handler.  I did some Googling, but got
nowhere.  I looked through drivers/ide to see what drive->unmask was doing.
I found this in ide-io.c:
     if (drive->unmask)
          local_irq_enable();
And this in ide-taskfile.c:
     if (!drive->unmask)
          local_irq_disable();
I modified both of these files so that execution would be as if unmask = 1.
This resulted in no change of behavior.

Bryan  

-----Original Message-----
From: Alan Cox [mailto:alan@lxorguk.ukuu.org.uk] 
Sent: Friday, August 26, 2005 11:47 AM
To: Bryan Althouse
Cc: linux-mips@linux-mips.org; 'Ralf Baechle'
Subject: RE: custom ide driver causes "Badness in smp_call_function"

On Gwe, 2005-08-26 at 10:58 -0400, Bryan Althouse wrote: 
> Ralf,
> 
> The patch doesn't seem to make any difference. :(

Assuming your hardware is sane another approach might be to force
drive->unmask = 1. That will mean that PIO mode is running with
interrupts enabled which should avoid the problem.

Add a .fixup handler to your driver (assuming you are using a recent
2.6.x) and in the handler do something like this:

+void ide_unmask_interrupts(ide_hwif_t *hwif)
+{
+       int i;
+       for (i = 0; i < 2; i++) {
+               ide_drive_t *drive = &hwif->drives[i];
+               if(drive->present)
+                       drive->unmask = 1;
+       }
+}

hopefully that will be early enough.

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Alan,

Thanks for your suggestion.
I'm not sure how to write a .fixup handler.  I did some Googling, but got
nowhere.  I looked through drivers/ide to see what drive->unmask was doing.
I found this in ide-io.c:
     if (drive->unmask)
          local_irq_enable();
And this in ide-taskfile.c:
     if (!drive->unmask)
          local_irq_disable();
I modified both of these files so that execution would be as if unmask = 1.
This resulted in no change of behavior.

Bryan  

-----Original Message-----
From: Alan Cox [mailto:alan@lxorguk.ukuu.org.uk] 
Sent: Friday, August 26, 2005 11:47 AM
To: Bryan Althouse
Cc: linux-mips@linux-mips.org; 'Ralf Baechle'
Subject: RE: custom ide driver causes "Badness in smp_call_function"

On Gwe, 2005-08-26 at 10:58 -0400, Bryan Althouse wrote: 
> Ralf,
> 
> The patch doesn't seem to make any difference. :(

Assuming your hardware is sane another approach might be to force
drive->unmask = 1. That will mean that PIO mode is running with
interrupts enabled which should avoid the problem.

Add a .fixup handler to your driver (assuming you are using a recent
2.6.x) and in the handler do something like this:

+void ide_unmask_interrupts(ide_hwif_t *hwif)
+{
+       int i;
+       for (i = 0; i < 2; i++) {
+               ide_drive_t *drive = &hwif->drives[i];
+               if(drive->present)
+                       drive->unmask = 1;
+       }
+}

hopefully that will be early enough.

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Alan,

Thanks for your suggestion.
I'm not sure how to write a .fixup handler.  I did some Googling, but got
nowhere.  I looked through drivers/ide to see what drive->unmask was doing.
I found this in ide-io.c:
     if (drive->unmask)
          local_irq_enable();
And this in ide-taskfile.c:
     if (!drive->unmask)
          local_irq_disable();
I modified both of these files so that execution would be as if unmask = 1.
This resulted in no change of behavior.

Bryan  

-----Original Message-----
From: Alan Cox [mailto:alan@lxorguk.ukuu.org.uk] 
Sent: Friday, August 26, 2005 11:47 AM
To: Bryan Althouse
Cc: linux-mips@linux-mips.org; 'Ralf Baechle'
Subject: RE: custom ide driver causes "Badness in smp_call_function"

On Gwe, 2005-08-26 at 10:58 -0400, Bryan Althouse wrote: 
> Ralf,
> 
> The patch doesn't seem to make any difference. :(

Assuming your hardware is sane another approach might be to force
drive->unmask = 1. That will mean that PIO mode is running with
interrupts enabled which should avoid the problem.

Add a .fixup handler to your driver (assuming you are using a recent
2.6.x) and in the handler do something like this:

+void ide_unmask_interrupts(ide_hwif_t *hwif)
+{
+       int i;
+       for (i = 0; i < 2; i++) {
+               ide_drive_t *drive = &hwif->drives[i];
+               if(drive->present)
+                       drive->unmask = 1;
+       }
+}

hopefully that will be early enough.

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Alan,

Thanks for your suggestion.
I'm not sure how to write a .fixup handler.  I did some Googling, but got
nowhere.  I looked through drivers/ide to see what drive->unmask was doing.
I found this in ide-io.c:
     if (drive->unmask)
          local_irq_enable();
And this in ide-taskfile.c:
     if (!drive->unmask)
          local_irq_disable();
I modified both of these files so that execution would be as if unmask = 1.
This resulted in no change of behavior.

Bryan  

-----Original Message-----
From: Alan Cox [mailto:alan@lxorguk.ukuu.org.uk] 
Sent: Friday, August 26, 2005 11:47 AM
To: Bryan Althouse
Cc: linux-mips@linux-mips.org; 'Ralf Baechle'
Subject: RE: custom ide driver causes "Badness in smp_call_function"

On Gwe, 2005-08-26 at 10:58 -0400, Bryan Althouse wrote: 
> Ralf,
> 
> The patch doesn't seem to make any difference. :(

Assuming your hardware is sane another approach might be to force
drive->unmask = 1. That will mean that PIO mode is running with
interrupts enabled which should avoid the problem.

Add a .fixup handler to your driver (assuming you are using a recent
2.6.x) and in the handler do something like this:

+void ide_unmask_interrupts(ide_hwif_t *hwif)
+{
+       int i;
+       for (i = 0; i < 2; i++) {
+               ide_drive_t *drive = &hwif->drives[i];
+               if(drive->present)
+                       drive->unmask = 1;
+       }
+}

hopefully that will be early enough.

Re: custom ide driver causes "Badness in smp_call_function"

[Ralf Baechle]

On Fri, Aug 26, 2005 at 10:58:40AM -0400, Bryan Althouse wrote:

> The patch doesn't seem to make any difference. :(

To clarify, the patch wasn't meant to resolve your interrupt problems but
the SMP cacheflush issues only.

  Ralf

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

I still get the SMP badness when hwif->irq is set to 0 in my driver.

Bryan

-----Original Message-----
From: Ralf Baechle [mailto:ralf@linux-mips.org] 
Sent: Friday, August 26, 2005 12:29 PM
To: Bryan Althouse
Cc: linux-mips@linux-mips.org
Subject: Re: custom ide driver causes "Badness in smp_call_function"

On Fri, Aug 26, 2005 at 10:58:40AM -0400, Bryan Althouse wrote:

> The patch doesn't seem to make any difference. :(

To clarify, the patch wasn't meant to resolve your interrupt problems but
the SMP cacheflush issues only.

  Ralf

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

I still get the SMP badness when hwif->irq is set to 0 in my driver.

Bryan

-----Original Message-----
From: Ralf Baechle [mailto:ralf@linux-mips.org] 
Sent: Friday, August 26, 2005 12:29 PM
To: Bryan Althouse
Cc: linux-mips@linux-mips.org
Subject: Re: custom ide driver causes "Badness in smp_call_function"

On Fri, Aug 26, 2005 at 10:58:40AM -0400, Bryan Althouse wrote:

> The patch doesn't seem to make any difference. :(

To clarify, the patch wasn't meant to resolve your interrupt problems but
the SMP cacheflush issues only.

  Ralf

Re: custom ide driver causes "Badness in smp_call_function"

[Ralf Baechle]

On Fri, Aug 26, 2005 at 12:36:26PM -0400, Bryan Althouse wrote:

> I still get the SMP badness when hwif->irq is set to 0 in my driver.

Interesting.  That implies smp_call_function is called through a different
path on your system that on mine here.  Doesn't invalidate the patch,
just means some extra debugging work for you ;-)

  Ralf

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Ralf,

I've done more testing with your patch.  When I use it with my non-SMP
kernel, disk access will cause a panic with a cache error message.  Is this
patch only intended for SMP, or is this a legitimate problem?

Bryan


Index: include/asm-mips/mach-generic/ide.h
===================================================================
RCS file: /home/cvs/linux/include/asm-mips/mach-generic/ide.h,v
retrieving revision 1.9
diff -u -r1.9 ide.h
--- include/asm-mips/mach-generic/ide.h	19 Apr 2005 12:26:59 -0000	1.9
+++ include/asm-mips/mach-generic/ide.h	26 Aug 2005 14:04:38 -0000
@@ -19,6 +19,7 @@
 #include <linux/pci.h>
 #include <linux/stddef.h>
 #include <asm/processor.h>
+#include <asm/cacheflush.h>
 
 #ifndef MAX_HWIFS
 # ifdef CONFIG_BLK_DEV_IDEPCI
@@ -105,12 +106,14 @@
 
 /* MIPS port and memory-mapped I/O string operations.  */
 
-static inline void __ide_flush_dcache_range(unsigned long addr, unsigned
long size)
+static inline void __ide_flush_dcache_range(unsigned long addr,
+	unsigned long size)
 {
-	if (cpu_has_dc_aliases) {
-		unsigned long end = addr + size;
-		for (; addr < end; addr += PAGE_SIZE)
-			flush_dcache_page(virt_to_page(addr));
+	unsigned long end = addr + size;
+
+	while (addr < end) {
+		SetPageDcacheDirty(virt_to_page(addr));
+		addr += PAGE_SIZE;
 	}
 }
 

RE: custom ide driver causes "Badness in smp_call_function"

[Bryan Althouse]

Ralf,

I've done more testing with your patch.  When I use it with my non-SMP
kernel, disk access will cause a panic with a cache error message.  Is this
patch only intended for SMP, or is this a legitimate problem?

Bryan


Index: include/asm-mips/mach-generic/ide.h
===================================================================
RCS file: /home/cvs/linux/include/asm-mips/mach-generic/ide.h,v
retrieving revision 1.9
diff -u -r1.9 ide.h
--- include/asm-mips/mach-generic/ide.h	19 Apr 2005 12:26:59 -0000	1.9
+++ include/asm-mips/mach-generic/ide.h	26 Aug 2005 14:04:38 -0000
@@ -19,6 +19,7 @@
 #include <linux/pci.h>
 #include <linux/stddef.h>
 #include <asm/processor.h>
+#include <asm/cacheflush.h>
 
 #ifndef MAX_HWIFS
 # ifdef CONFIG_BLK_DEV_IDEPCI
@@ -105,12 +106,14 @@
 
 /* MIPS port and memory-mapped I/O string operations.  */
 
-static inline void __ide_flush_dcache_range(unsigned long addr, unsigned
long size)
+static inline void __ide_flush_dcache_range(unsigned long addr,
+	unsigned long size)
 {
-	if (cpu_has_dc_aliases) {
-		unsigned long end = addr + size;
-		for (; addr < end; addr += PAGE_SIZE)
-			flush_dcache_page(virt_to_page(addr));
+	unsigned long end = addr + size;
+
+	while (addr < end) {
+		SetPageDcacheDirty(virt_to_page(addr));
+		addr += PAGE_SIZE;
 	}
 }
 

Re: custom ide driver causes "Badness in smp_call_function"

[Ralf Baechle DL5RB]

On Thu, Sep 01, 2005 at 12:15:31PM -0400, Bryan Althouse wrote:

> I've done more testing with your patch.  When I use it with my non-SMP
> kernel, disk access will cause a panic with a cache error message.  Is this
> patch only intended for SMP, or is this a legitimate problem?

There is a real problem but it only hits on SMP.

(The implementation is a hack though, it makes assumptions about the IDE
code's internal workings)

  Ralf

Re: custom ide driver causes "Badness in smp_call_function"

[Atsushi Nemoto]

>>>>> On Fri, 26 Aug 2005 15:10:47 +0100, Ralf Baechle <ralf@linux-mips.org> said:
ralf> Try this patch below and let me know.  I would also like to ask
ralf> those people who used to suffer from aliases with IDE PIO to try
ralf> this patch.

I'm using CPU which suffer from dcache aliasing.
I tried this patch for IDE PIO and it seems work fine too.

Maybe Cobalt user can try it.  Anyone?  :-)

---
Atsushi Nemoto