Linux on HP PA-RISC development discussion ()

Joel Soete | 1 Jul 2008 16:29

Picon

About iommu-helpers.h: where can I find detailed info on the scatter list programming?

Hello Grant,

I am looking for such kind of reference because I am totally ignorant on this
subject (as well as many others  ) and didn't find yet the good entry point?

Re-reading ccio-dma code for the n*1000 time, I noticed your comment:
/* Fast path single entry scatterlists */
if (nents == 1) {
[snip]

This comment make me thought: it's just a shortcut and the rest of the code
would just slowdown operation?

No?

In fact the 2 drivers (sba and ccio) without this hunk of code make panicing
the boot. 
For the ccio: some where in the initialization of the first scsi drive (no
more info)
For the sba: seems nearly at the same place but with addtional message
[snip]
sym0: <895a> rev 0x1 at pci 0000:00:0f.0 irq 20
sym0: PA-RISC Firmware, ID 7, Fast-40, LVD, parity checking
sym0: SCSI BUS has been reset.
scsi0 : sym-2.2.3
Kernel panic - not syncing:
/Extra/linux-current-trace/drivers/parisc/sba_iommu.c: I/O MMU  <at>  0000a000 is
out of mapping resources   

Is it normal?

(Continue reading)

Grant Grundler | 2 Jul 2008 06:28

Re: in ccio_io_pdir_entry(),BUG_ON() seems to break gcc-4.2 optimization?

On Mon, Jun 30, 2008 at 06:28:36PM +0000, Joel Soete wrote:
...
> Taken also into account of David remarks, here is the latest stuff tested 
> with success:
>
>         __asm__ __volatile__ (
>         "lci    %%r0(%%sr1, %2), %0\n"
>         "\textru        %0,19,12,%0\n"
>         "\tdepw         %0,15,12,%1\n"
>         : "=r" (ci), "=r" (pa)
>         : "r" (vba)
>         );

So can you remind me (and Kyle) why this is better than what is
there now?

> Tx to all for kind comments, they learn me a lot 

welcome! :)

grant
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Joel Soete | 2 Jul 2008 20:01

Picon

Re: in ccio_io_pdir_entry(),BUG_ON() seems to break gcc-4.2 optimization?


Grant Grundler wrote:
> On Mon, Jun 30, 2008 at 06:28:36PM +0000, Joel Soete wrote:
> ...
>> Taken also into account of David remarks, here is the latest stuff tested 
>> with success:
>>
>>         __asm__ __volatile__ (
>>         "lci    %%r0(%%sr1, %2), %0\n"
>>         "\textru        %0,19,12,%0\n"
>>         "\tdepw         %0,15,12,%1\n"
>>         : "=r" (ci), "=r" (pa)
>>         : "r" (vba)
>>         );
> 
> So can you remind me (and Kyle) why this is better than what is
> there now?
> 
mmm may be because gcc consider it now as only 1 'volatile' instruction:

in <ccio_map_sg>
-----------< Original code > --------------+------------< New stuff >-----------------
200: 06 80 53 13  lci r0(sr1,r20),r19      |  200: 08 1a 02 54  copy r26,r20
204: d2 73 1a 74  extrw,u r19,19,12,r19    |  204: 06 a0 53 14  lci r0(sr1,r21),r20
208: 08 1a 02 5c  copy r26,ret0            |  208: d2 94 1a 74  extrw,u r20,19,12,r20
20c: d7 93 0e 14  depw r19,15,12,ret0      |  20c: d7 94 0e 14  depw r20,15,12,ret0
210: 0e dc 12 80  stw ret0,0(r22)             210: 0e dc 12 80  stw ret0,0(r22)
214: 06 c0 12 80  fdc r0(r22)                 214: 06 c0 12 80  fdc r0(r22)
218: 00 00 04 00  sync                        218: 00 00 04 00  sync

(Continue reading)

Grant Grundler | 7 Jul 2008 17:28

Re: in ccio_io_pdir_entry(),BUG_ON() seems to break gcc-4.2 optimization?

On Wed, Jul 02, 2008 at 06:01:45PM +0000, Joel Soete wrote:
>> So can you remind me (and Kyle) why this is better than what is
>> there now?
> mmm may be because gcc consider it now as only 1 'volatile' instruction:
>
> in <ccio_map_sg>
> -----------< Original code > --------------+------------< New stuff 
> >-----------------
> 200: 06 80 53 13  lci r0(sr1,r20),r19      |  200: 08 1a 02 54  copy r26,r20
> 204: d2 73 1a 74  extrw,u r19,19,12,r19    |  204: 06 a0 53 14  lci r0(sr1,r21),r20
> 208: 08 1a 02 5c  copy r26,ret0            |  208: d2 94 1a 74  extrw,u r20,19,12,r20
> 20c: d7 93 0e 14  depw r19,15,12,ret0      |  20c: d7 94 0e 14  depw r20,15,12,ret0
> 210: 0e dc 12 80  stw ret0,0(r22)             210: 0e dc 12 80  stw ret0,0(r22)
> 214: 06 c0 12 80  fdc r0(r22)                 214: 06 c0 12 80  fdc r0(r22)
> 218: 00 00 04 00  sync                        218: 00 00 04 00  sync
>
> thought?

I'm not comfortable "New stuff" version is correct.

In "New Stuff", r20 gets clobbered and values copied from
r26 are lost. That's not the case in the original code.
"depw" only replaces 12 bits of "ret0" in the original code.

thanks,
grant
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

(Continue reading)

Joel Soete | 8 Jul 2008 11:04

Picon

Re: in ccio_io_pdir_entry(),BUG_ON() seems to break gcc-4.2 optimization?

[...]
> 
> In "New Stuff", r20 gets clobbered and values copied from
> r26 are lost. That's not the case in the original code.
> "depw" only replaces 12 bits of "ret0" in the original code.
> 
Good catch, in fact I would just have to save the read+write constraint on pa;
i.e this seems to make the drill:

        __asm__ __volatile__ (
        "lci    %%r0(%%sr1, %2), %0\n"
        "\textru        %0,19,12,%0\n"
        "\tdepw         %0,15,12,%1\n"
        : "=r" (ci), "+r" (pa)
        : "r" (vba)
        );

comparing produced objdump:
00000000 <ccio_io_pdir_entry>:           00000000 <ccio_io_pdir_entry>:
 0: 00 19 58 20  mtsp r25,sr1             0: 00 19 58 20  mtsp r25,sr1
 4: 23 80 0e 01  ldil L%-10000000,ret0    4: 23 80 0e 01  ldil L%-10000000,ret0
 8: 0b 98 0a 1c  add,l r24,ret0,ret0      8: 0b 98 0a 1c  add,l r24,ret0,ret0
 c: d7 97 0c 14  depw r23,31,12,ret0      c: d7 97 0c 14  depw r23,31,12,ret0
10: 0f 5c 12 88  stw ret0,4(r26)       | 10: 34 13 00 00  ldi 0,r19
14: 07 00 53 18  lci r0(sr1,r24),r24   | 14: 0f 5c 12 88  stw ret0,4(r26)
18: d3 18 1a 74  extrw,u r24,19,12,r24 | 18: 07 00 53 1c  lci r0(sr1,r24),ret0
1c: 34 1c 00 00  ldi 0,ret0            | 1c: d3 9c 1a 74  extrw,u ret0,19,12,ret0
20: d7 98 0e 14  depw r24,15,12,ret0   | 20: d6 7c 0e 14  depw ret0,15,12,r19
24: 0f 5c 12 80  stw ret0,0(r26)       | 24: 0f 53 12 80  stw r19,0(r26)
28: 07 40 12 80  fdc r0(r26)             28: 07 40 12 80  fdc r0(r26)

(Continue reading)

Helge Deller | 8 Jul 2008 22:06

X-Face

Picon

Picon

[PATCH] drop superfluous .align 16

This .align 16 is misplaced and should be in front of the ENTRY(lws_lock_start).
Works because of pure luck since we have a .align PAGE_SIZE before it instead.

Signed-off-by: Helge Deller <deller <at> gmx.de>

--- a/arch/parisc/kernel/syscall.S
+++ b/arch/parisc/kernel/syscall.S
 <at>  <at>  -640,7 +640,6  <at>  <at>  END(sys_call_table64)
 	.align	PAGE_SIZE
 ENTRY(lws_lock_start)
 	/* lws locks */
-	.align 16
 	.rept 16
 	/* Keep locks aligned at 16-bytes */
 	.word 1
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Carlos O'Donell | 9 Jul 2008 14:14

Re: [PATCH] drop superfluous .align 16

On Tue, Jul 8, 2008 at 4:06 PM, Helge Deller <deller <at> gmx.de> wrote:
> This .align 16 is misplaced and should be in front of the ENTRY(lws_lock_start).
> Works because of pure luck since we have a .align PAGE_SIZE before it instead.

This is not true, the .align 16 aligns almost any object in the
subsection including .word.

However, it is superfluous since we have a .align PAGE_SIZE, but I was
probably being safe in the event that someone put an object before
lws_lock_start.

> Signed-off-by: Helge Deller <deller <at> gmx.de>
>
> --- a/arch/parisc/kernel/syscall.S
> +++ b/arch/parisc/kernel/syscall.S
>  <at>  <at>  -640,7 +640,6  <at>  <at>  END(sys_call_table64)
>        .align  PAGE_SIZE
>  ENTRY(lws_lock_start)
>        /* lws locks */
> -       .align 16
>        .rept 16
>        /* Keep locks aligned at 16-bytes */
>        .word 1

Cheers,
Carlos.
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

(Continue reading)

Helge Deller | 9 Jul 2008 21:50

Picon

Picon

Re: [PATCH] drop superfluous .align 16

Carlos O'Donell wrote:
> On Tue, Jul 8, 2008 at 4:06 PM, Helge Deller <deller <at> gmx.de> wrote:
>> This .align 16 is misplaced and should be in front of the ENTRY(lws_lock_start).
>> Works because of pure luck since we have a .align PAGE_SIZE before it instead.
> 
> This is not true, the .align 16 aligns almost any object in the
> subsection including .word.

Sure, but it does not take care of lws_lock_start itself.

> However, it is superfluous since we have a .align PAGE_SIZE, but I was
> probably being safe in the event that someone put an object before
> lws_lock_start.

What I mean is, that the lws_lock_start label should start at 16byte 
boundary, right? (it is due to the .align PAGE_SIZE).
Now, if e.g. the .word555 (see below) would be in there, then
lws_lock_start would be at PAGE_SIZE+4, while the .word1 would start at 
PAGE_SIZE+16, which is wrong for accessing the lws_lock_start variables, 
where the first entry should be "1".

Example:
        .align  PAGE_SIZE
        .word 5555  <<<- think this would be here.
ENTRY(lws_lock_start)
       /* lws locks */
      .align 16
      .rept 16
      /* Keep locks aligned at 16-bytes */
      .word 1

(Continue reading)

Carlos O'Donell | 9 Jul 2008 22:22

Re: [PATCH] drop superfluous .align 16

On Wed, Jul 9, 2008 at 3:50 PM, Helge Deller <deller <at> gmx.de> wrote:
> So, I still think it would be better to remove the .align16, or
> alternatively to move it in front of ENTRY(lws_lock_start) [to be on the
> safe side].
>
> Am I really that wrong?

No, you are absolutely right. I forgot the locks are referenced via
lws_lock_start, and the symbol needs to be aligned. Sorry, I thought
there was another symbol elsewhere.

Signed-off-by: Carlos O'Donell <carlos <at> systemhalted.org>

Cheers,
Carlos.
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

John David Anglin | 9 Jul 2008 22:55

Picon

Re: [PATCH] drop superfluous .align 16

> On Wed, Jul 9, 2008 at 3:50 PM, Helge Deller <deller <at> gmx.de> wrote:
> > So, I still think it would be better to remove the .align16, or
> > alternatively to move it in front of ENTRY(lws_lock_start) [to be on the
> > safe side].
> >
> > Am I really that wrong?
> 
> No, you are absolutely right. I forgot the locks are referenced via
> lws_lock_start, and the symbol needs to be aligned. Sorry, I thought
> there was another symbol elsewhere.

This would be clearer if there was an ENTRY_ALIGNED macro (e.g.,
ENTRY_ALIGNED(lws_lock_start, 16)).  It's probably overkill though.

Dave
--

-- 
J. David Anglin                                  dave.anglin <at> nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Group

gmane.linux.ports.parisc.

Advertisement

Project Web Page

Linux on HP PA-RISC development discussion ()

Search Archive

Page

1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10

Articles in period: 98

July 2008

Language

Change language

Options

Current view: Threads only / Showing only 20 lines / Not hiding cited text.
Change to All messages, whole messages, or hide cited text.

Post a message
NNTP Newsgroup
Classic Gmane web interface
XML

XML

RSS Feed
List Information

About Gmane

Recent entries

[GIT PULL] parisc updates for v3.10 (18 Jun 2013)
[PATCH] parisc: delete __cpuinit usage from all users (17 Jun 2013)
[PATCH RESEND] fix WARNING: at kernel/cpu/idle.c:96 (14 Jun 2013)
[PATCH] parisc: fix LMMIO mismatch between PAT length and MASK registe (14 Jun 2013)
[PATCH] parisc: provide pci_mmap_page_range() for parisc (14 Jun 2013)
[PATCH] parisc: fix serial ports on C8000 workstation (9 Jun 2013)
PROCES A ODKUPU VAŠÍ VYHRÁL CENU £ 7.500,000LIBRA BRITSKÁ LIBRA { (6 Jun 2013)
[PATCH] parisc: fix kernel BUG at arch/parisc/include/asm/mmzone.h:50 (5 Jun 2013)
[PATCH] parisc: Use unshadowed index register for flush instructions i (2 Jun 2013)
[PATCH] parisc: Fix cache coherency for copy/clear_user_page operation (2 Jun 2013)
[GIT PULL] parisc updates for v3.10-rc4 (1 Jun 2013)
[PATCH] parisc: parport0: fix this legacy no-device port driver! (30 May 2013)
[PATCH] parport_pc: disable PARPORT_PC_SUPERIO on parisc architecture (30 May 2013)
[PATCH] parisc/PCI: lba: fix: convert to pci_create_root_bus() for cor (30 May 2013)
[PATCH] parisc: MAINTAINERS - add sound, video and documentation refer (30 May 2013)
[PATCH] arch: parisc: kernel: using strlcpy() instead of strcpy() (30 May 2013)
[PATCH] parisc: fix kernel BUG at arch/parisc/include/asm/mmzone.h:50 (28 May 2013)
[PATCH] parisc: fix race conditions flushing user cache pages (28 May 2013)
[PATCH] console/dummy: Move screen size selection from CPP to Kconfig (27 May 2013)
[PATCH] arch: parisc: kernel: memory overflow, 'name' length is too sh (27 May 2013)
linux-next: manual merge of the parisc-hd tree with Linus' tree (27 May 2013)

Archives

40	June 2013
97	May 2013
84	April 2013
27	March 2013
68	February 2013
29	January 2013
15	December 2012
20	November 2012
56	October 2012
23	September 2012
25	August 2012
9	July 2012
19	June 2012
150	May 2012
25	April 2012
19	March 2012
61	February 2012
39	January 2012
22	December 2011
12	November 2011
53	October 2011
26	September 2011
20	August 2011
69	July 2011
36	June 2011
54	May 2011
112	April 2011
42	March 2011
69	February 2011
37	January 2011
52	December 2010
55	November 2010
85	October 2010
23	September 2010
37	August 2010
17	July 2010
83	June 2010
62	May 2010
75	April 2010
57	March 2010
50	February 2010
62	January 2010
120	December 2009
75	November 2009
60	October 2009
124	September 2009
116	August 2009
130	July 2009
126	June 2009
112	May 2009
157	April 2009
92	March 2009
88	February 2009
174	January 2009
187	December 2008
66	November 2008
18	October 2008
111	September 2008
114	August 2008
98	July 2008
114	June 2008
72	May 2008
42	April 2008
106	March 2008
55	February 2008
30	January 2008
60	December 2007

Design

Zawodny | Right Menu | Left Menu

Your Own Design

Paste the URL of your CSS below. Download CSS template