linux_dsm_epyc7002/sound/firewire/fireworks/fireworks_transaction.c
Takashi Sakamoto 6b1ca4bcad ALSA: fireworks: accessing to user space outside spinlock
In hwdep interface of fireworks driver, accessing to user space is in a
critical section with disabled local interrupt. Depending on architecture,
accessing to user space can cause page fault exception. Then local
processor stores machine status and handles the synchronous event. A
handler corresponding to the event can call task scheduler to wait for
preparing pages. In a case of usage of single core processor, the state to
disable local interrupt is worse because it don't handle usual interrupts
from hardware.

This commit fixes this bug, performing the accessing outside spinlock. This
commit also gives up counting the number of queued response messages to
simplify ring-buffer management.

Reported-by: Vaishali Thakkar <vaishali.thakkar@oracle.com>
Cc: stable@vger.kernel.org
Fixes: 555e8a8f7f14('ALSA: fireworks: Add command/response functionality into hwdep interface')
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
2016-08-31 16:17:15 +02:00

326 lines
8.0 KiB
C

/*
* fireworks_transaction.c - a part of driver for Fireworks based devices
*
* Copyright (c) 2013-2014 Takashi Sakamoto
*
* Licensed under the terms of the GNU General Public License, version 2.
*/
/*
* Fireworks have its own transaction. The transaction can be delivered by AV/C
* Vendor Specific command frame or usual asynchronous transaction. At least,
* Windows driver and firmware version 5.5 or later don't use AV/C command.
*
* Transaction substance:
* At first, 6 data exist. Following to the data, parameters for each command
* exist. All of the parameters are 32 bit aligned to big endian.
* data[0]: Length of transaction substance
* data[1]: Transaction version
* data[2]: Sequence number. This is incremented by the device
* data[3]: Transaction category
* data[4]: Transaction command
* data[5]: Return value in response.
* data[6-]: Parameters
*
* Transaction address:
* command: 0xecc000000000
* response: 0xecc080000000 (default)
*
* I note that the address for response can be changed by command. But this
* module uses the default address.
*/
#include "./fireworks.h"
#define MEMORY_SPACE_EFW_COMMAND 0xecc000000000ULL
#define MEMORY_SPACE_EFW_RESPONSE 0xecc080000000ULL
#define ERROR_RETRIES 3
#define ERROR_DELAY_MS 5
#define EFC_TIMEOUT_MS 125
static DEFINE_SPINLOCK(instances_lock);
static struct snd_efw *instances[SNDRV_CARDS] = SNDRV_DEFAULT_PTR;
static DEFINE_SPINLOCK(transaction_queues_lock);
static LIST_HEAD(transaction_queues);
enum transaction_queue_state {
STATE_PENDING,
STATE_BUS_RESET,
STATE_COMPLETE
};
struct transaction_queue {
struct list_head list;
struct fw_unit *unit;
void *buf;
unsigned int size;
u32 seqnum;
enum transaction_queue_state state;
wait_queue_head_t wait;
};
int snd_efw_transaction_cmd(struct fw_unit *unit,
const void *cmd, unsigned int size)
{
return snd_fw_transaction(unit, TCODE_WRITE_BLOCK_REQUEST,
MEMORY_SPACE_EFW_COMMAND,
(void *)cmd, size, 0);
}
int snd_efw_transaction_run(struct fw_unit *unit,
const void *cmd, unsigned int cmd_size,
void *resp, unsigned int resp_size)
{
struct transaction_queue t;
unsigned int tries;
int ret;
t.unit = unit;
t.buf = resp;
t.size = resp_size;
t.seqnum = be32_to_cpu(((struct snd_efw_transaction *)cmd)->seqnum) + 1;
t.state = STATE_PENDING;
init_waitqueue_head(&t.wait);
spin_lock_irq(&transaction_queues_lock);
list_add_tail(&t.list, &transaction_queues);
spin_unlock_irq(&transaction_queues_lock);
tries = 0;
do {
ret = snd_efw_transaction_cmd(t.unit, (void *)cmd, cmd_size);
if (ret < 0)
break;
wait_event_timeout(t.wait, t.state != STATE_PENDING,
msecs_to_jiffies(EFC_TIMEOUT_MS));
if (t.state == STATE_COMPLETE) {
ret = t.size;
break;
} else if (t.state == STATE_BUS_RESET) {
msleep(ERROR_DELAY_MS);
} else if (++tries >= ERROR_RETRIES) {
dev_err(&t.unit->device, "EFW transaction timed out\n");
ret = -EIO;
break;
}
} while (1);
spin_lock_irq(&transaction_queues_lock);
list_del(&t.list);
spin_unlock_irq(&transaction_queues_lock);
return ret;
}
static void
copy_resp_to_buf(struct snd_efw *efw, void *data, size_t length, int *rcode)
{
size_t capacity, till_end;
struct snd_efw_transaction *t;
t = (struct snd_efw_transaction *)data;
length = min_t(size_t, be32_to_cpu(t->length) * sizeof(u32), length);
spin_lock_irq(&efw->lock);
if (efw->push_ptr < efw->pull_ptr)
capacity = (unsigned int)(efw->pull_ptr - efw->push_ptr);
else
capacity = snd_efw_resp_buf_size -
(unsigned int)(efw->push_ptr - efw->pull_ptr);
/* confirm enough space for this response */
if (capacity < length) {
*rcode = RCODE_CONFLICT_ERROR;
goto end;
}
/* copy to ring buffer */
while (length > 0) {
till_end = snd_efw_resp_buf_size -
(unsigned int)(efw->push_ptr - efw->resp_buf);
till_end = min_t(unsigned int, length, till_end);
memcpy(efw->push_ptr, data, till_end);
efw->push_ptr += till_end;
if (efw->push_ptr >= efw->resp_buf + snd_efw_resp_buf_size)
efw->push_ptr -= snd_efw_resp_buf_size;
length -= till_end;
data += till_end;
}
/* for hwdep */
wake_up(&efw->hwdep_wait);
*rcode = RCODE_COMPLETE;
end:
spin_unlock_irq(&efw->lock);
}
static void
handle_resp_for_user(struct fw_card *card, int generation, int source,
void *data, size_t length, int *rcode)
{
struct fw_device *device;
struct snd_efw *efw;
unsigned int i;
spin_lock_irq(&instances_lock);
for (i = 0; i < SNDRV_CARDS; i++) {
efw = instances[i];
if (efw == NULL)
continue;
device = fw_parent_device(efw->unit);
if ((device->card != card) ||
(device->generation != generation))
continue;
smp_rmb(); /* node id vs. generation */
if (device->node_id != source)
continue;
break;
}
if (i == SNDRV_CARDS)
goto end;
copy_resp_to_buf(efw, data, length, rcode);
end:
spin_unlock_irq(&instances_lock);
}
static void
handle_resp_for_kernel(struct fw_card *card, int generation, int source,
void *data, size_t length, int *rcode, u32 seqnum)
{
struct fw_device *device;
struct transaction_queue *t;
unsigned long flags;
spin_lock_irqsave(&transaction_queues_lock, flags);
list_for_each_entry(t, &transaction_queues, list) {
device = fw_parent_device(t->unit);
if ((device->card != card) ||
(device->generation != generation))
continue;
smp_rmb(); /* node_id vs. generation */
if (device->node_id != source)
continue;
if ((t->state == STATE_PENDING) && (t->seqnum == seqnum)) {
t->state = STATE_COMPLETE;
t->size = min_t(unsigned int, length, t->size);
memcpy(t->buf, data, t->size);
wake_up(&t->wait);
*rcode = RCODE_COMPLETE;
}
}
spin_unlock_irqrestore(&transaction_queues_lock, flags);
}
static void
efw_response(struct fw_card *card, struct fw_request *request,
int tcode, int destination, int source,
int generation, unsigned long long offset,
void *data, size_t length, void *callback_data)
{
int rcode, dummy;
u32 seqnum;
rcode = RCODE_TYPE_ERROR;
if (length < sizeof(struct snd_efw_transaction)) {
rcode = RCODE_DATA_ERROR;
goto end;
} else if (offset != MEMORY_SPACE_EFW_RESPONSE) {
rcode = RCODE_ADDRESS_ERROR;
goto end;
}
seqnum = be32_to_cpu(((struct snd_efw_transaction *)data)->seqnum);
if (seqnum > SND_EFW_TRANSACTION_USER_SEQNUM_MAX + 1) {
handle_resp_for_kernel(card, generation, source,
data, length, &rcode, seqnum);
if (snd_efw_resp_buf_debug)
handle_resp_for_user(card, generation, source,
data, length, &dummy);
} else {
handle_resp_for_user(card, generation, source,
data, length, &rcode);
}
end:
fw_send_response(card, request, rcode);
}
void snd_efw_transaction_add_instance(struct snd_efw *efw)
{
unsigned int i;
spin_lock_irq(&instances_lock);
for (i = 0; i < SNDRV_CARDS; i++) {
if (instances[i] != NULL)
continue;
instances[i] = efw;
break;
}
spin_unlock_irq(&instances_lock);
}
void snd_efw_transaction_remove_instance(struct snd_efw *efw)
{
unsigned int i;
spin_lock_irq(&instances_lock);
for (i = 0; i < SNDRV_CARDS; i++) {
if (instances[i] != efw)
continue;
instances[i] = NULL;
}
spin_unlock_irq(&instances_lock);
}
void snd_efw_transaction_bus_reset(struct fw_unit *unit)
{
struct transaction_queue *t;
spin_lock_irq(&transaction_queues_lock);
list_for_each_entry(t, &transaction_queues, list) {
if ((t->unit == unit) &&
(t->state == STATE_PENDING)) {
t->state = STATE_BUS_RESET;
wake_up(&t->wait);
}
}
spin_unlock_irq(&transaction_queues_lock);
}
static struct fw_address_handler resp_register_handler = {
.length = SND_EFW_RESPONSE_MAXIMUM_BYTES,
.address_callback = efw_response
};
int snd_efw_transaction_register(void)
{
static const struct fw_address_region resp_register_region = {
.start = MEMORY_SPACE_EFW_RESPONSE,
.end = MEMORY_SPACE_EFW_RESPONSE +
SND_EFW_RESPONSE_MAXIMUM_BYTES
};
return fw_core_add_address_handler(&resp_register_handler,
&resp_register_region);
}
void snd_efw_transaction_unregister(void)
{
WARN_ON(!list_empty(&transaction_queues));
fw_core_remove_address_handler(&resp_register_handler);
}