Skip to contentSkip to conversation

endianness problems in fireworks/bebob_maudio

PATCH Applied Clemens Ladisch 3 participants 6 messages Dec 7, 2014 – Jan 7, 2015

Subsystems (2)
lore
Share & tools
lore

Clemens Ladisch

author maintainer
#0 lore raw

Hi,

sparse complains:

sound/firewire/fireworks/fireworks_transaction.c:127:18: warning: restricted __be32 degrades to integer

        t = (struct snd_efw_transaction *)data;
        length = min_t(size_t, t->length * sizeof(t->length), length);

't->length' is still a big-endian value. This means that the driver ends up always using 'length'.

sound/firewire/bebob/bebob_maudio.c:100:17: warning: incorrect type in initializer (different base types)
sound/firewire/bebob/bebob_maudio.c:100:17: expected restricted __be32
sound/firewire/bebob/bebob_maudio.c:100:17: got int

        __be32 cues[3] = {
                MAUDIO_BOOTLOADER_CUE1,
                MAUDIO_BOOTLOADER_CUE2,
                MAUDIO_BOOTLOADER_CUE3
        };

        rcode = fw_run_transaction(device->card, TCODE_WRITE_BLOCK_REQUEST,
                                   device->node_id, device->generation,
                                   device->max_speed, BEBOB_ADDR_REG_REQ,
                                   cues, sizeof(cues));

The three MAUDIO_BOOTLOADER_CUEx values will end up as a different byte sequence on big-endian machines. The simplest way to have these twelve bytes unchanged on the bus is to have a twelve-byte array in the driver.

Regards,
Clemens

maintainer
#1 lore raw

Hi Clemens,

Quoted 7 lines · Clemens Ladisch

sound/firewire/fireworks/fireworks_transaction.c:127:18: warning: restricted __be32 degrades to integer

        t = (struct snd_efw_transaction *)data;
        length = min_t(size_t, t->length * sizeof(t->length), length);

't->length' is still a big-endian value. This means that the driver ends up always using 'length'.

Exactly. I agree they should be fixed.

Quoted 15 lines

sound/firewire/bebob/bebob_maudio.c:100:17: warning: incorrect type in initializer (different base types)
sound/firewire/bebob/bebob_maudio.c:100:17: expected restricted __be32
sound/firewire/bebob/bebob_maudio.c:100:17: got int

        __be32 cues[3] = {
                MAUDIO_BOOTLOADER_CUE1,
                MAUDIO_BOOTLOADER_CUE2,
                MAUDIO_BOOTLOADER_CUE3
        };

        rcode = fw_run_transaction(device->card, TCODE_WRITE_BLOCK_REQUEST,
                                   device->node_id, device->generation,
                                   device->max_speed, BEBOB_ADDR_REG_REQ,
                                   cues, sizeof(cues));
Quoted 3 lines

The three MAUDIO_BOOTLOADER_CUEx values will end up as a different byte sequence on big-endian machines. The simplest way to have these twelve bytes unchanged on the bus is to have a twelve-byte array in the driver.

I'm too lazy for these codes. Yes, they should be fixed as long as we assume this driver is used in big-endian machine.

For my information, Would you tell me the way to generate these sparse report? I've been working with sparse for my patchset and never see such reports, furthermore never receive such reports from kernel build bot.

Thanks

Takashi Sakamoto
[email protected]

author maintainer
#2 lore raw
Quoted 2 lines · Takashi Sakamoto

For my information, Would you tell me the way to generate these sparse report?

make C=1 CF=-D__CHECK_ENDIAN__

Regards,
Clemens

maintainer
#3 lore raw
Quoted 5 lines · Clemens Ladisch
Quoted 2 lines · Takashi Sakamoto

For my information, Would you tell me the way to generate these sparse report?

make C=1 CF=-D__CHECK_ENDIAN__

I always check with 'make C=2' without the 'CF' flag. Firewire drivers handle big-endian data, so the flag is nice.

Thanks

Takashi Sakamoto
[email protected]

maintainer
#4 lore raw

PATCH 1/1 ALSA: fireworks: fix an endianness bug for transaction length

In reply to Clemens Ladisch

Although the 't->length' is a big-endian value, it's used without any conversion. This means that the driver always uses 'length' parameter.

Fixes 555e8a8f7f14("ALSA: fireworks: Add command/response functionality into hwdep interface")
---
 sound/firewire/fireworks/fireworks_transaction.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/firewire/fireworks/fireworks_transaction.c +1 -1
index 255dabc..2a85e42 100644
@@ -124,7 +124,7 @@ copy_resp_to_buf(struct snd_efw *efw, void *data, size_t length, int *rcode)
 	spin_lock_irq(&efw->lock);
 
 	t = (struct snd_efw_transaction *)data;
-	length = min_t(size_t, t->length * sizeof(t->length), length);
+	length = min_t(size_t, be32_to_cpu(t->length) * sizeof(u32), length);
 
 	if (efw->push_ptr < efw->pull_ptr)
 		capacity = (unsigned int)(efw->pull_ptr - efw->push_ptr);
Signature · 1 line
2.1.0
Applied
#5 lore raw

Re: [PATCH] ALSA: fireworks: fix an endianness bug for transaction length

At Thu, 8 Jan 2015 00:31:16 +0900,

Quoted 7 lines · Takashi Sakamoto

Although the 't->length' is a big-endian value, it's used without any conversion. This means that the driver always uses 'length' parameter.

Fixes 555e8a8f7f14("ALSA: fireworks: Add command/response functionality into hwdep interface")

Applied, thanks.

Takashi

Quoted 20 lines
---
 sound/firewire/fireworks/fireworks_transaction.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
sound/firewire/fireworks/fireworks_transaction.c +1 -1
index 255dabc..2a85e42 100644
@@ -124,7 +124,7 @@ copy_resp_to_buf(struct snd_efw *efw, void *data, size_t length, int *rcode)
 	spin_lock_irq(&efw->lock);
 
 	t = (struct snd_efw_transaction *)data;
-	length = min_t(size_t, t->length * sizeof(t->length), length);
+	length = min_t(size_t, be32_to_cpu(t->length) * sizeof(u32), length);
 
 	if (efw->push_ptr < efw->pull_ptr)
 		capacity = (unsigned int)(efw->pull_ptr - efw->push_ptr);
Signature · 1 line
2.1.0