Skip to contentSkip to conversation

Emu10k1 SPDIF passthru doesn't work if CONFIG_NOHIGHMEM is not enabled

DISCUSSION German Gomez Garcia 6 participants 9 messages Feb 26, 2002 – Mar 2, 2002 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 1014586494.1518.23.camel@skyblade

German Gomez Garcia

author
#1 lore raw
Quoted 12 lines · José Carlos Monteiro

Hi!
After some more careful testing, I was able to identify the exact moment when the changes in the Linux kernel broke SPDIF passthru of Emu10k1 cards. I tested all the pre-patches between kernels 2.4.12 and 2.4.13 and I found that kernel 2.4.13-pre2 was the one that broke it. Up until 2.4.13-pre1, everything works fine. From 2.4.13-pre2 on, passthru sound is broken (if kernel option CONFIG_HIGHMEM4G or CONFIG_HIGHMEM64G is used).

According to the kernel Changelog, it appears that one of these changes was the responsible for it:

Quoted 8 lines

2.4.13-pre2:
- Alan Cox: more merging
- Ben Fennema: UDF module license
- Jeff Mahoney: reiserfs endian safeness
- Chris Mason: reiserfs O_SYNC/fsync performance improvements
- Jean Tourrilhes: wireless extension update
- Joerg Reuter: AX.25 updates
- David Miller: 64-bit DMA interfaces

Does the emu10k1 driver support the new DMA interface? I've
downloaded the patch-2.4.13-pre1-pre2.bz2 from

ftp://ftp.kernel.org/pub/linux/kernel/v2.4/testing/incr/

and checking it, it seems that the BIG change was the inclussion of David Miller 64-bit DMA. Does emu10k1 use DMA for SPDIF output? the main change affecting dma when HIGHMEM is enabled is in types.h, that define dma_addr_t as a u64 instead of u32, so does the hardware in the live or the driver (or whatever :-) support that?

Best regards,

- german

-------------------------------------------------------------------------
German Gomez Garcia | Send email with "SEND GPG KEY" as subject
<[email protected]> | to receive my GnuPG public key.

#2 lore raw
Quoted 23 lines · German Gomez Garcia
Quoted 12 lines · José Carlos Monteiro

Hi!
After some more careful testing, I was able to identify the exact moment when the changes in the Linux kernel broke SPDIF passthru of Emu10k1 cards. I tested all the pre-patches between kernels 2.4.12 and 2.4.13 and I found that kernel 2.4.13-pre2 was the one that broke it. Up until 2.4.13-pre1, everything works fine. From 2.4.13-pre2 on, passthru sound is broken (if kernel option CONFIG_HIGHMEM4G or CONFIG_HIGHMEM64G is used).

According to the kernel Changelog, it appears that one of these changes was the responsible for it:

Quoted 8 lines

2.4.13-pre2:
- Alan Cox: more merging
- Ben Fennema: UDF module license
- Jeff Mahoney: reiserfs endian safeness
- Chris Mason: reiserfs O_SYNC/fsync performance improvements
- Jean Tourrilhes: wireless extension update
- Joerg Reuter: AX.25 updates
- David Miller: 64-bit DMA interfaces

Quoted 2 lines

Does the emu10k1 driver support the new DMA interface? I've
downloaded the patch-2.4.13-pre1-pre2.bz2 from

We allocate memory with pci_alloc_consistent() which according to DMA-mapping.txt assures 32-bits PCI addresses. On top of that we set a dma_mask (with pci_set_dma_mask) of 29 bits (512Mib).

Quoted 4 lines

ftp://ftp.kernel.org/pub/linux/kernel/v2.4/testing/incr/

and checking it, it seems that the BIG change was the inclussion of David Miller 64-bit DMA. Does emu10k1 use DMA for SPDIF output?

It uses DMA for all sound input/output.

Quoted 8 lines

the
main change affecting dma when HIGHMEM is enabled is in types.h, that define dma_addr_t as a u64 instead of u32, so does the hardware in the live or the driver (or whatever :-) support that?

Best regards,

- german

The most bizzare is that in a machine with 192Mib of memory but with a kernel compiled with HIGHMEM support I see the same type of problems.

Rui Sousa

#3 lore raw

Re: [Emu10k1-devel] Re: Emu10k1 SPDIF passthru doesn't work if

Quoted 2 lines

The most bizzare is that in a machine with 192Mib of memory but with a=20 kernel compiled with HIGHMEM support I see the same type of problems.

Change of size in a structure or type ?

#4 lore raw

Re: [Emu10k1-devel] Re: Emu10k1 SPDIF passthru doesn't work if

On Wed, 27 Feb 2002, Alan Cox wrote:

It's true dma_addr_t does change from u32 to u64 and we do thinks like:

(32 bit pci register) = cpu_to_le32(dma_handle)

What is the correct way of doing this?

(32 bit pci register) = cpu_to_le32((u32)dma_handle) ?

Rui Sousa

Quoted 10 lines
Quoted 2 lines

The most bizzare is that in a machine with 192Mib of memory but with a=20 kernel compiled with HIGHMEM support I see the same type of problems.

Change of size in a structure or type ?

_______________________________________________
Emu10k1-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emu10k1-devel

#5 lore raw

Re: [Emu10k1-devel] Re: Emu10k1 SPDIF passthru doesn't work if

Quoted 10 lines · Rui Sousa

On Wed, 27 Feb 2002, Alan Cox wrote:

It's true dma_addr_t does change from u32 to u64 and we do thinks like:

(32 bit pci register) = cpu_to_le32(dma_handle)

What is the correct way of doing this?

(32 bit pci register) = cpu_to_le32((u32)dma_handle)

If you only have 32 bits, then I presume 64-bit DMA isn't supported.

So, (1) never pass more than 0xffffffff to pci_set_dma_mask, and (2) just truncate dma_addr_t (addr & 0xffffffff) so that you only read the low 32-bits, always.

Jeff

Signature · 3 lines
Jeff Garzik      | "UNIX enhancements aren't."
Building 1024    |           -- says /usr/games/fortune
MandrakeSoft     |
#6 lore raw

Re: [Emu10k1-devel] Re: Emu10k1 SPDIF passthru doesn't work if

In reply to Rui Sousa
Quoted 5 lines

It's true dma_addr_t does change from u32 to u64 and we do thinks like:
(32 bit pci register) = cpu_to_le32(dma_handle)

What is the correct way of doing this?
(32 bit pci register) = cpu_to_le32((u32)dma_handle)

The cast befor ethe cpu_to_ is safe if its 32bit I/O only. Maybe we should have cpu_to_le_dma_addr_t 8)

#7 lore raw

Re: [Emu10k1-devel] Re: Emu10k1 SPDIF passthru doesn't work if

From: Alan Cox <[email protected]>
Date: Fri, 1 Mar 2002 01:07:27 +0000 (GMT)

The cast befor ethe cpu_to_ is safe if its 32bit I/O only. Maybe we should have cpu_to_le_dma_addr_t 8)

Actually, the cast to 32-bit is safe if you've set your DMA mask properly :-)

#8 lore raw

Re: [Emu10k1-devel] Re: Emu10k1 SPDIF passthru doesn't work if

Hi,

Looks like the bug is from pushing the 64bit dma_handle on a 32bit va_arg list. Its fixed in CVS, and at least one person has reported success.

Quoted 9 lines · David S. Miller

From: Alan Cox <[email protected]>
Date: Fri, 1 Mar 2002 01:07:27 +0000 (GMT)

The cast befor ethe cpu_to_ is safe if its 32bit I/O only. Maybe we should have cpu_to_le_dma_addr_t 8)

Actually, the cast to 32-bit is safe if you've set your DMA mask properly :-)

Signature · 1 line
Daniel Bertrand
#9 lore raw

64bit dma_addr_t (was: Emu10k1 SPDIF ...)

In reply to David S. Miller

Hi,

Sorry to beat a dead horse, but wouldn't it be better if sizes didn't change within the same CPU arch? I don't see how binary-only modules can ever work reliably if the size of types change depending on people's .config file. (not that I use any binary-only modules :-).

Is this a common thing done in the kernel? For dma_addr_t, why not just have it always be 64bit?

Quoted 9 lines · David S. Miller

From: Alan Cox <[email protected]>
Date: Fri, 1 Mar 2002 01:07:27 +0000 (GMT)

The cast befor ethe cpu_to_ is safe if its 32bit I/O only. Maybe we should have cpu_to_le_dma_addr_t 8)

Actually, the cast to 32-bit is safe if you've set your DMA mask properly :-)

Signature · 1 line
Daniel Bertrand