[alsa-devel] [PATCH 10/10] wss_lib: use wss detection code instead of ad1848 one

Krzysztof Helt krzysztof.h1 at poczta.fm
Thu Jul 24 20:47:34 CEST 2008


On Thu, 24 Jul 2008 19:19:15 +0200
Rene Herman <rene.herman at keyaccess.nl> wrote:


> *** 0004-wss-use-stuct-snd_wss-instead-of-snd_ad1848.patch
> 
> > +++ b/include/sound/ad1848.h
> > 
> > [ ... ]
> > 
> > +#include "wss.h"
> 
> Bad (not at top and "") but given that it's going anyway...
> 

During changes like renaming I got into "dummy" mode so I changed without deep thinking.

> 
> *** 0005-wss_lib-use-wss-constants-instead-of-ad1848-ones.patch
> 
> > --- a/sound/isa/ad1848/ad1848_lib.c
> > +++ b/sound/isa/ad1848/ad1848_lib.c
> > @@ -283,14 +283,12 @@ static int snd_ad1848_trigger(struct snd_wss *chip, unsigned char what,
> >                         return 0;
> >                 }
> >                 snd_ad1848_out(chip, AD1848_IFACE_CTRL, chip->image[AD1848_IFACE_CTRL] |= what);
> > -               chip->mode |= AD1848_MODE_RUNNING;
> 
> Is this now done in generic code? Not necessary anymore? Was no comment 
> in the changelog.
> 

The MODE_RUNNING is not used at all in the cs4231 code. I wonder what the purpose of it.

> >  static const char *snd_ad1848_chip_id(struct snd_wss *chip)
> >  {
> >         switch (chip->hardware) {
> > -       case AD1848_HW_AD1847:  return "AD1847";
> > -       case AD1848_HW_AD1848:  return "AD1848";
> > -       case AD1848_HW_CS4248:  return "CS4248";
> > -       case AD1848_HW_CMI8330: return "CMI8330/C3D";
> > -       default:                return "???";
> > +       case WSS_HW_AD1847:
> > +               return "AD1847";
> > +       case WSS_HW_AD1848:
> > +               return "AD1848";
> > +       case WSS_HW_CS4248:
> > +               return "CS4248";
> > +       case WSS_HW_CMI8330:
> > +               return "CMI8330/C3D";
> > +       default:
> > +               return "???";
> >         }
> >  }
> 
> These kinds of switches are just about the only time you get to 
> knowingly ignore coding style and you forego the opportunity? Tsssk...
> 

If switches like these generates more than one warnings 
by the checkpatch script I change them to be on safer side
with "checkpatch fundamentalists". Ok?

> > --- a/sound/isa/opti9xx/opti92x-ad1848.c
> > +++ b/sound/isa/opti9xx/opti92x-ad1848.c
> > @@ -775,7 +775,7 @@ static int __devinit snd_opti9xx_probe(struct snd_card *card)
> >  #else
> >         if ((error = snd_ad1848_create(card, chip->wss_base + 4,
> >                                        chip->irq, chip->dma1,
> > -                                      AD1848_HW_DETECT, &codec)) < 0)
> > +                                      WSS_HW_DETECT, &codec)) < 0)
> >                 return error;
> >         if ((error = snd_ad1848_pcm(codec, 0, &pcm)) < 0)
> >                 return error;
> 
> and
> 
> > --- a/sound/isa/sgalaxy.c
> > +++ b/sound/isa/sgalaxy.c
> > @@ -265,7 +265,7 @@ static int __devinit snd_sgalaxy_probe(struct device *devptr, unsigned int dev)
> >  
> >         if ((err = snd_ad1848_create(card, wssport[dev] + 4,
> >                                      xirq, xdma1,
> > -                                    AD1848_HW_DETECT, &chip)) < 0)
> > +                                    WSS_HW_DETECT, &chip)) < 0)
> >                 goto _err;
> >         card->private_data = chip;
> 
> For those that come after -- those error ifs are split when they are 
> turned into snd_wss_create() later.
> 

I tried to do as few as possible changes not related to what the patch did.
So I left ifs like these for now.

> 
> *** 0006-wss_lib-replace-ad1848-mixer-element-macros-with-ws.patch
> 
> >  #define AD1848_SINGLE_TLV(xname, xindex, reg, shift, mask, invert, xtlv) \
> > -{ .name = xname, \
> > -  .index = xindex, \
> > -  .type = AD1848_MIX_SINGLE, \
> > -  .private_value = AD1848_MIXVAL_SINGLE(reg, shift, mask, invert), \
> > -  .tlv = xtlv }
> > +{ .iface = SNDRV_CTL_ELEM_IFACE_MIXER, \
> > +  .access = SNDRV_CTL_ELEM_ACCESS_READWRITE | SNDRV_CTL_ELEM_ACCESS_TLV_READ, \
> > +  .name = xname, .index = xindex, \
> > +  .info = snd_ad1848_info_single, \
> > +  .get = snd_ad1848_get_single, .put = snd_ad1848_put_single, \
> > +  .private_value = reg | (shift << 8) | (mask << 16) | (invert << 24), \
> > +  .tlv = { .p = (xtlv) } }
> 
> Please just one per line (aligned is nice...)

Ok.

> 
> >  #define AD1848_DOUBLE_TLV(xname, xindex, left_reg, right_reg, shift_left, shift_right, mask, invert, xtlv) \
> > -{ .name = xname, \
> > -  .index = xindex, \
> > -  .type = AD1848_MIX_DOUBLE, \
> > -  .private_value = AD1848_MIXVAL_DOUBLE(left_reg, right_reg, shift_left, shift_right, mask, invert), \
> > -  .tlv = xtlv }
> > -
> > -static struct ad1848_mix_elem snd_ad1848_controls[] = {
> > -AD1848_DOUBLE("PCM Playback Switch", 0, AD1848_LEFT_OUTPUT, AD1848_RIGHT_OUTPUT, 7, 7, 1, 1),
> > -AD1848_DOUBLE_TLV("PCM Playback Volume", 0, AD1848_LEFT_OUTPUT, AD1848_RIGHT_OUTPUT, 0, 0, 63, 1,
> > +{ .iface = SNDRV_CTL_ELEM_IFACE_MIXER, \
> > +  .access = SNDRV_CTL_ELEM_ACCESS_READWRITE | SNDRV_CTL_ELEM_ACCESS_TLV_READ, \
> > +  .name = xname, .index = xindex, \
> > +  .info = snd_ad1848_info_double, \
> > +  .get = snd_ad1848_get_double, .put = snd_ad1848_put_double, \
> > +  .private_value = left_reg | (right_reg << 8) | (shift_left << 16) | \
> > +                  (shift_right << 19) | (mask << 24) | (invert << 22), \
> > +  .tlv = { .p = (xtlv) } }
> 
> Same.
> 
> Throughout this patch there's also still the comment about ignoring the 
> 80-column thing with these macros. The cmi8330.c ones are a wonderful 
> example of how much worse it gets. It's horrible.
> 

I wrote that changing it all into nice checkpatch code was probably not
sane. I left long macros in few places they were already.

> 
> *** 0007-wss_lib-use-CS4231P-instead-of-AD1848P-kill-the-AD.patch
> 
> > static void snd_ad1848_debug(struct snd_wss *chip)
> 
> Same. Otherwise

It is done to have fewer checkpatch warning. As the file was going
to be killed completely it does not matter.

> 
> Reviewed-by: Rene Herman <rene.herman at gmail.com>
> 
> One other comment:
> 
> > diff --git a/sound/isa/wss/wss_lib.c b/sound/isa/wss/wss_lib.c
> > index 08997d0..f695c85 100644
> > --- a/sound/isa/wss/wss_lib.c
> > +++ b/sound/isa/wss/wss_lib.c
> > @@ -164,7 +164,6 @@ static inline void wss_outb(struct snd_wss *chip, u8 offset, u8 val)
> >  {
> >         outb(val, chip->port + offset);
> >  }
> > -EXPORT_SYMBOL(snd_wss_out);
> >  
> >  static inline u8 wss_inb(struct snd_wss *chip, u8 offset)
> >  {
> > @@ -228,6 +227,7 @@ void snd_wss_out(struct snd_wss *chip, unsigned char reg, unsigned char value)
> >         snd_printdd("codec out - reg 0x%x = 0x%x\n",
> >                         chip->mce_bit | reg, value);
> >  }
> > +EXPORT_SYMBOL(snd_wss_out);
> 
> Ah. This fixes up one of my comments from yesterday already. If this 
> finds its way into 3, it should ofcourse be gone here.
> 

This is a proof that code can become alive and does thing programer
is not aware of ;-)

> 
> *** 0008-wss_lib-use-wss-mixer-code-instead-of-ad1848-one.patch
> 
> > +#define WSS_SINGLE_TLV(xname, xindex, reg, shift, mask, invert, xtlv) \
> > +{ .iface = SNDRV_CTL_ELEM_IFACE_MIXER, \
> > +  .access = SNDRV_CTL_ELEM_ACCESS_READWRITE | SNDRV_CTL_ELEM_ACCESS_TLV_READ, \
> > +  .name = xname, .index = xindex, \
> > +  .info = snd_wss_info_single, \
> > +  .get = snd_wss_get_single, .put = snd_wss_put_single, \
> > +  .private_value = reg | (shift << 8) | (mask << 16) | (invert << 24), \
> > +  .tlv = { .p = (xtlv) } }
> > +
> > +#define WSS_DOUBLE_TLV(xname, xindex, left_reg, right_reg, \
> > +                       shift_left, shift_right, mask, invert, xtlv) \
> > +{ .iface = SNDRV_CTL_ELEM_IFACE_MIXER, \
> > +  .access = SNDRV_CTL_ELEM_ACCESS_READWRITE | SNDRV_CTL_ELEM_ACCESS_TLV_READ, \
> > +  .name = xname, .index = xindex, \
> > +  .info = snd_wss_info_double, \
> > +  .get = snd_wss_get_double, .put = snd_wss_put_double, \
> > +  .private_value = left_reg | (right_reg << 8) | (shift_left << 16) | \
> > +                  (shift_right << 19) | (mask << 24) | (invert << 22), \
> > +  .tlv = { .p = (xtlv) } }
> 
> Ofcourse same with the please one member per line.
> 

Ok.

> 
> *** 0009-wss_lib-use-wss-pcm-code-instead-of-ad1848-one.patch
> 
> > --- a/sound/isa/wss/wss_lib.c
> > +++ b/sound/isa/wss/wss_lib.c
> > @@ -363,7 +363,7 @@ static void snd_wss_busy_wait(struct snd_wss *chip)
> >         for (timeout = 5; timeout > 0; timeout--)
> >                 wss_inb(chip, CS4231P(REGSEL));
> >         /* end of cleanup sequence */
> > -       for (timeout = 250;
> > +       for (timeout = 2500;
> >              timeout > 0 && (wss_inb(chip, CS4231P(REGSEL)) & CS4231_INIT);
> >              timeout--)
> >                 udelay(10);
> 
> Was this intentional? You comment on this in 10/10...

Yes. It was. The original ad1848 loop is 12000 x 100us. The cs4231 loop
was 250 x 10us. The code I tested had 2500 x 100us but I lowered the udelay()
argument for cs4231chips which were much faster (only few if any 10us 
during tests). I forgot to retest it again on the original ad1848 so the patch
went in with the error.

If this kind of change should be in a separate patch it must be a patch
before this patch otherwise we end up with perfectly bisectable error
the author was aware of while sending patches (so there is no point
in applying only one and no bisecting then).

> 
> > @@ -1360,6 +1381,11 @@ static int snd_wss_capture_open(struct s
> >  
> >         runtime->hw = snd_wss_capture;
> >  
> > +       /* hardware limitation of older chipsets */
> > +       if (chip->hardware == WSS_HW_INTERWAVE && chip->dma1 > 3)
> > +               runtime->hw.formats &= ~(SNDRV_PCM_FMTBIT_IMA_ADPCM |
> > +                                        SNDRV_PCM_FMTBIT_S16_BE);
> > +
> 
> If you'll be posting a V2 series, might as well fold 11/10 in here.
> 

I'll do fold the fix, but new limits for opti cards I would like to keep
as a separate one (missing it is not a regression).

> > const char *snd_wss_chip_id(struct snd_wss *chip)
> 
> Comment about the switch getting uglier again.
> 

The checkpatch likes the new switch more.

> 
> *** 0010-wss_lib-use-wss-detection-code-instead-of-ad1848-on.patch
> 
> > +++ b/sound/isa/wss/wss_lib.c
> > @@ -363,7 +363,7 @@ static void snd_wss_busy_wait(struct snd_wss *chip)
> >         for (timeout = 5; timeout > 0; timeout--)
> >                 wss_inb(chip, CS4231P(REGSEL));
> >         /* end of cleanup sequence */
> > -       for (timeout = 2500;
> > +       for (timeout = 25000;
> >              timeout > 0 && (wss_inb(chip, CS4231P(REGSEL)) & CS4231_INIT);
> >              timeout--)
> >                 udelay(10);
> 
> So now it's 100x total. Is this as planned?
> 

Yes. I have described it above.

> > -static int snd_wss_probe(struct snd_wss *chip)
> > +static int snd_ad1848_probe(struct snd_wss *chip)
> >  {
> >         unsigned long flags;
> > -       int i, id, rev;
> > -       unsigned char *ptr;
> > -       unsigned int hw;
> > +       int i, id, rev, ad1847;
> >  
> > -#if 0
> > -       snd_wss_debug(chip);
> > -#endif
> >         id = 0;
> > -       for (i = 0; i < 50; i++) {
> > +       ad1847 = 0;
> > +       for (i = 0; i < 1000; i++) {
> >                 mb();
> > -               if (wss_inb(chip, CS4231P(REGSEL)) & CS4231_INIT)
> > -                       udelay(2000);
> > +               if (inb(chip->port + CS4231P(REGSEL)) & CS4231_INIT)
> > +                       msleep(1);
> 
> Mmm. This was changed from a udelay(500) in the ad1848 case. It would be 
> better to keep these kinds of really functional changes seperated out.
> 

Again, without the change it may provide two patches but working
only if both applied.

> [ ... ]
> 
> > +       id = 0;
> > +       if (chip->hardware == WSS_HW_DETECT)
> > +               id = ad1847 ? WSS_HW_AD1847 : WSS_HW_AD1848;
> > +
> > +       spin_lock_irqsave(&chip->reg_lock, flags);
> > +       inb(chip->port + CS4231P(STATUS));      /* clear any pendings IRQ */
> > +       outb(0, chip->port + CS4231P(STATUS));
> 
> The original snd_ad1848_probe() had this below the below tests (which 
> were also outside the spinlock)
> 
> > +       mb();
> > +       if (id == WSS_HW_AD1848) {
> > +               /* check if there are more than 16 registers */
> > +               rev = snd_wss_in(chip, CS4231_MISC_INFO);
> > +               snd_wss_out(chip, CS4231_MISC_INFO, 0x40);
> > +               for (i = 0; i < 16; ++i) {
> > +                       if (snd_wss_in(chip, i) != snd_wss_in(chip, i + 16)) {
> > +                               id = WSS_HW_CMI8330;
> > +                               break;
> > +                       }
> > +               }
> > +               snd_wss_out(chip, CS4231_MISC_INFO, 0x00);
> > +               if (id != WSS_HW_CMI8330 && (rev & 0x80))
> > +                       id = WSS_HW_CS4248;
> 
> This one could also wind up different. Originally, snd_ad1848_probe() 
> would conclude CS4248 immediately upon (rev & 0x80) without doing that 
> register test. 

It did not work. It hit the Galaxy Waverider as it misdetected
the cs4231 as the cs4248. A smarter way was needed if ad1848 and
cs4231 families were present.

> It's probably all fine, but snd_ad1848 function changed 
> very significantly in the move and I'd rather it not do that. A patch
> 12 alone is much easier reviewable and any possible difficulty much 
> easier bisectable. Could you do that?
> 

It can be done but the patch which merges the code will incorrectly
detect chips (tested that it does). So both must be applied together.

> 
> *** 0011-wss-fix-capture-formats-limitations.patch
> 
> Can be folded, but otherwise no comments...
> 
> Will also put a next series through some tests here locally. I'l test at 
> least a few OPTi 92x cards since I don't see these in your report.

You may try merging opti92x-ad1848 and opti92x-cs4231 drivers now.
There are only few lines of difference (e.g. timer creation). I haven't
done this as I do not have these chips.

> Any 
> specific hardware you'd want tested more than others? I might have it.
> 

Yes. Try if older codecs (not yet tested) are correctly detected. Especially, 
these with special code paths in the detection function:
ad1847,
cmi8330, 
cs4248,

Also try if they are working as they may require even longer delays.

The cs4231 chips are much simpler as you have the revision register
so I do not have problem here. I would like to have tested the 
integrated or compatible cs4231 codecs:
yamaha opl3sa2 
interwave

If you have any of these please post results of your tests.

I did tests on cs4232 chip inside the Dell Laptitue CP250 and it works.
It is found by the PNP bios and configured automatically (the anti-crash
patch I posted is needed).
The chip is detected as CS4236, however. The cs4231 revision is 0x03 which
code detects as CS4236B. I cannot open the laptop (it is not mine)
so I assume the Dell replaced older chip with newer one but left the same
BIOS.

Thank you for your effort,
Regards,
Krzysztof

----------------------------------------------------------------------
Tanie rozmowy!
Sprawdz >>>  http://link.interia.pl/f1e91



More information about the Alsa-devel mailing list