* [PATCH 1/3] firewire: core: allocate workqueue for AR/AT request/response contexts
2025-06-14 11:34 [PATCH 0/3] firewire: ohci: use regular workqueue to handle 1394 OHCI AT/AR context events Takashi Sakamoto
@ 2025-06-14 11:34 ` Takashi Sakamoto
2025-06-15 4:40 ` kernel test robot
2025-06-14 11:34 ` [PATCH 2/3] firewire: ohci: use workqueue to handle events of AR " Takashi Sakamoto
2025-06-14 11:34 ` [PATCH 3/3] firewire: ohci: use workqueue to handle events of AT " Takashi Sakamoto
2 siblings, 1 reply; 5+ messages in thread
From: Takashi Sakamoto @ 2025-06-14 11:34 UTC (permalink / raw)
To: linux1394-devel; +Cc: linux-kernel
Some tasklets (softIRQs) are still used as bottom-halves to handle
events for 1394 OHCI AR/AT contexts. However, using softIRQs for IRQ
bottom halves is generally discouraged today.
This commit adds a per-fw_card workqueue to accommodate the behaviour
specified by the 1394 OHCI specification.
According to the 1394 OHCI specification, system memory pages are
reserved for each asynchronous DMA context. This allows concurrent
operation across contexts. In the 1394 OHCI PCI driver implementation,
the hardware generates IRQs either upon receiving asynchronous packets
from other nodes (incoming) or after completing transmission to them
(outgoing). These independent events can occur in the same transmission
cycle, therefore the max_active parameter for the workqueue is set to the
total number of AR/AT contexts (=4). The WQ_UNBOUND flag is used to
allow the work to be scheduled on any available core, since there is
little CPU cache affinity benefit for the data.
Each DMA context uses a circular descriptor list in system memory,
allowing deferred data processing in software as long as buffer overrun
are avoided. Since the overall operation is sleepable except for small
atomic regions, WQ_BH is not used. As the descriptors contain
timestamps, WQ_HIGHPRI is specified to support semi-real-time
processing.
The asynchronous context is also used by the SCSI over IEEE 1394
protocol implementation (sbp2), which can be part of memory reclaim paths.
Therefore, WQ_MEM_RECLAIM is required.
To allow uses to adjust CPU affinity according to workload, WQ_SYSFS is
specified so that workqueue attributes are exposed to user space.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
---
drivers/firewire/core-card.c | 40 +++++++++++++++++++++++++++---------
include/linux/firewire.h | 1 +
2 files changed, 31 insertions(+), 10 deletions(-)
diff --git a/drivers/firewire/core-card.c b/drivers/firewire/core-card.c
index 2b6ad47b6d57..df0bb5b96ddc 100644
--- a/drivers/firewire/core-card.c
+++ b/drivers/firewire/core-card.c
@@ -574,7 +574,6 @@ EXPORT_SYMBOL(fw_card_initialize);
int fw_card_add(struct fw_card *card, u32 max_receive, u32 link_speed, u64 guid,
unsigned int supported_isoc_contexts)
{
- struct workqueue_struct *isoc_wq;
int ret;
// This workqueue should be:
@@ -589,12 +588,29 @@ int fw_card_add(struct fw_card *card, u32 max_receive, u32 link_speed, u64 guid,
// * == WQ_SYSFS Parameters are available via sysfs.
// * max_active == n_it + n_ir A hardIRQ could notify events for multiple isochronous
// contexts if they are scheduled to the same cycle.
- isoc_wq = alloc_workqueue("firewire-isoc-card%u",
- WQ_UNBOUND | WQ_FREEZABLE | WQ_HIGHPRI | WQ_SYSFS,
- supported_isoc_contexts, card->index);
- if (!isoc_wq)
+ card->isoc_wq = alloc_workqueue("firewire-isoc-card%u",
+ WQ_UNBOUND | WQ_FREEZABLE | WQ_HIGHPRI | WQ_SYSFS,
+ supported_isoc_contexts, card->index);
+ if (!card->isoc_wq)
return -ENOMEM;
+ // This workqueue should be:
+ // * != WQ_BH Sleepable.
+ // * == WQ_UNBOUND Any core can process data for asynchronous context.
+ // * == WQ_MEM_RECLAIM Used for any backend of block device.
+ // * == WQ_FREEZABLE The target device would not be available when being freezed.
+ // * == WQ_HIGHPRI High priority to process semi-realtime timestamped data.
+ // * == WQ_SYSFS Parameters are available via sysfs.
+ // * max_active == 4 A hardIRQ could notify events for a pair of requests and
+ // response AR/AT contexts.
+ card->async_wq = alloc_workqueue("firewire-async-card%u",
+ WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_HIGHPRI | WQ_SYSFS,
+ 4, card->index);
+ if (!card->async_wq) {
+ ret = -ENOMEM;
+ goto err_isoc;
+ }
+
card->max_receive = max_receive;
card->link_speed = link_speed;
card->guid = guid;
@@ -603,15 +619,17 @@ int fw_card_add(struct fw_card *card, u32 max_receive, u32 link_speed, u64 guid,
generate_config_rom(card, tmp_config_rom);
ret = card->driver->enable(card, tmp_config_rom, config_rom_length);
- if (ret < 0) {
- destroy_workqueue(isoc_wq);
- return ret;
- }
+ if (ret < 0)
+ goto err_async;
- card->isoc_wq = isoc_wq;
list_add_tail(&card->link, &card_list);
return 0;
+err_async:
+ destroy_workqueue(card->async_wq);
+err_isoc:
+ destroy_workqueue(card->isoc_wq);
+ return ret;
}
EXPORT_SYMBOL(fw_card_add);
@@ -744,6 +762,7 @@ void fw_core_remove_card(struct fw_card *card)
dummy_driver.stop_iso = card->driver->stop_iso;
card->driver = &dummy_driver;
drain_workqueue(card->isoc_wq);
+ drain_workqueue(card->async_wq);
scoped_guard(spinlock_irqsave, &card->lock)
fw_destroy_nodes(card);
@@ -753,6 +772,7 @@ void fw_core_remove_card(struct fw_card *card)
wait_for_completion(&card->done);
destroy_workqueue(card->isoc_wq);
+ destroy_workqueue(card->async_wq);
WARN_ON(!list_empty(&card->transaction_list));
}
diff --git a/include/linux/firewire.h b/include/linux/firewire.h
index b632eec3ab52..c55b8e30e700 100644
--- a/include/linux/firewire.h
+++ b/include/linux/firewire.h
@@ -136,6 +136,7 @@ struct fw_card {
__be32 maint_utility_register;
struct workqueue_struct *isoc_wq;
+ struct workqueue_struct *async_wq;
};
static inline struct fw_card *fw_card_get(struct fw_card *card)
--
2.48.1
^ permalink raw reply [flat|nested] 5+ messages in thread* [PATCH 2/3] firewire: ohci: use workqueue to handle events of AR request/response contexts
2025-06-14 11:34 [PATCH 0/3] firewire: ohci: use regular workqueue to handle 1394 OHCI AT/AR context events Takashi Sakamoto
2025-06-14 11:34 ` [PATCH 1/3] firewire: core: allocate workqueue for AR/AT request/response contexts Takashi Sakamoto
@ 2025-06-14 11:34 ` Takashi Sakamoto
2025-06-14 11:34 ` [PATCH 3/3] firewire: ohci: use workqueue to handle events of AT " Takashi Sakamoto
2 siblings, 0 replies; 5+ messages in thread
From: Takashi Sakamoto @ 2025-06-14 11:34 UTC (permalink / raw)
To: linux1394-devel; +Cc: linux-kernel
This commit adds a work item to handle events of 1394 OHCI AR
request/response contexts, and queues the item to the specific workqueue.
The call of struct fw_address_handler.address_callback() is done in the
workqueue when receiving any requests from the remove nodes. Additionally,
the call of struct fw_packet.callback() is done in the workqueue too when
receiving acknowledge to the asynchronous packet for the response
subaction of split transaction to the remote nodes.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
---
drivers/firewire/core-transaction.c | 7 ++++---
drivers/firewire/ohci.c | 27 +++++++++++----------------
2 files changed, 15 insertions(+), 19 deletions(-)
diff --git a/drivers/firewire/core-transaction.c b/drivers/firewire/core-transaction.c
index 2bd5deb9054e..d28477d84697 100644
--- a/drivers/firewire/core-transaction.c
+++ b/drivers/firewire/core-transaction.c
@@ -557,9 +557,10 @@ const struct fw_address_region fw_unit_space_region =
*
* region->start, ->end, and handler->length have to be quadlet-aligned.
*
- * When a request is received that falls within the specified address range,
- * the specified callback is invoked. The parameters passed to the callback
- * give the details of the particular request.
+ * When a request is received that falls within the specified address range, the specified callback
+ * is invoked. The parameters passed to the callback give the details of the particular request.
+ * The callback is invoked in the workqueue context in most cases. However, if the request is
+ * initiated by the local node, the callback is invoked in the initiator's context.
*
* To be called in process context.
* Return value: 0 on success, non-zero otherwise.
diff --git a/drivers/firewire/ohci.c b/drivers/firewire/ohci.c
index 40313a3ec63e..68317b5a64a7 100644
--- a/drivers/firewire/ohci.c
+++ b/drivers/firewire/ohci.c
@@ -101,7 +101,7 @@ struct ar_context {
void *pointer;
unsigned int last_buffer_index;
u32 regs;
- struct tasklet_struct tasklet;
+ struct work_struct work;
};
struct context;
@@ -1016,9 +1016,9 @@ static void ar_recycle_buffers(struct ar_context *ctx, unsigned int end_buffer)
}
}
-static void ar_context_tasklet(unsigned long data)
+static void ohci_ar_context_work(struct work_struct *work)
{
- struct ar_context *ctx = (struct ar_context *)data;
+ struct ar_context *ctx = from_work(ctx, work, work);
unsigned int end_buffer_index, end_buffer_offset;
void *p, *end;
@@ -1026,23 +1026,19 @@ static void ar_context_tasklet(unsigned long data)
if (!p)
return;
- end_buffer_index = ar_search_last_active_buffer(ctx,
- &end_buffer_offset);
+ end_buffer_index = ar_search_last_active_buffer(ctx, &end_buffer_offset);
ar_sync_buffers_for_cpu(ctx, end_buffer_index, end_buffer_offset);
end = ctx->buffer + end_buffer_index * PAGE_SIZE + end_buffer_offset;
if (end_buffer_index < ar_first_buffer_index(ctx)) {
- /*
- * The filled part of the overall buffer wraps around; handle
- * all packets up to the buffer end here. If the last packet
- * wraps around, its tail will be visible after the buffer end
- * because the buffer start pages are mapped there again.
- */
+ // The filled part of the overall buffer wraps around; handle all packets up to the
+ // buffer end here. If the last packet wraps around, its tail will be visible after
+ // the buffer end because the buffer start pages are mapped there again.
void *buffer_end = ctx->buffer + AR_BUFFERS * PAGE_SIZE;
p = handle_ar_packets(ctx, p, buffer_end);
if (p < buffer_end)
goto error;
- /* adjust p to point back into the actual buffer */
+ // adjust p to point back into the actual buffer
p -= AR_BUFFERS * PAGE_SIZE;
}
@@ -1057,7 +1053,6 @@ static void ar_context_tasklet(unsigned long data)
ar_recycle_buffers(ctx, end_buffer_index);
return;
-
error:
ctx->pointer = NULL;
}
@@ -1073,7 +1068,7 @@ static int ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci,
ctx->regs = regs;
ctx->ohci = ohci;
- tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
+ INIT_WORK(&ctx->work, ohci_ar_context_work);
for (i = 0; i < AR_BUFFERS; i++) {
ctx->pages[i] = dma_alloc_pages(dev, PAGE_SIZE, &dma_addr,
@@ -2238,10 +2233,10 @@ static irqreturn_t irq_handler(int irq, void *data)
}
if (event & OHCI1394_RQPkt)
- tasklet_schedule(&ohci->ar_request_ctx.tasklet);
+ queue_work(ohci->card.async_wq, &ohci->ar_request_ctx.work);
if (event & OHCI1394_RSPkt)
- tasklet_schedule(&ohci->ar_response_ctx.tasklet);
+ queue_work(ohci->card.async_wq, &ohci->ar_response_ctx.work);
if (event & OHCI1394_reqTxComplete)
tasklet_schedule(&ohci->at_request_ctx.tasklet);
--
2.48.1
^ permalink raw reply [flat|nested] 5+ messages in thread* [PATCH 3/3] firewire: ohci: use workqueue to handle events of AT request/response contexts
2025-06-14 11:34 [PATCH 0/3] firewire: ohci: use regular workqueue to handle 1394 OHCI AT/AR context events Takashi Sakamoto
2025-06-14 11:34 ` [PATCH 1/3] firewire: core: allocate workqueue for AR/AT request/response contexts Takashi Sakamoto
2025-06-14 11:34 ` [PATCH 2/3] firewire: ohci: use workqueue to handle events of AR " Takashi Sakamoto
@ 2025-06-14 11:34 ` Takashi Sakamoto
2 siblings, 0 replies; 5+ messages in thread
From: Takashi Sakamoto @ 2025-06-14 11:34 UTC (permalink / raw)
To: linux1394-devel; +Cc: linux-kernel
This commit adds a work item to handle events of 1394 OHCI AT
request/response contexts, and queues the item to the specific
workqueue. The call of struct fw_packet.callbaqck() is done in the
workqueue when receiving acknowledgement to the asynchronous packet
transferred to remote node.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
---
drivers/firewire/net.c | 4 ++--
drivers/firewire/ohci.c | 40 ++++++++++++++++++++++++----------------
include/linux/firewire.h | 11 +++++++++--
3 files changed, 35 insertions(+), 20 deletions(-)
diff --git a/drivers/firewire/net.c b/drivers/firewire/net.c
index 1bf0e15c1540..6d6446713539 100644
--- a/drivers/firewire/net.c
+++ b/drivers/firewire/net.c
@@ -1007,7 +1007,7 @@ static int fwnet_send_packet(struct fwnet_packet_task *ptask)
spin_lock_irqsave(&dev->lock, flags);
- /* If the AT tasklet already ran, we may be last user. */
+ /* If the AT work item already ran, we may be last user. */
free = (ptask->outstanding_pkts == 0 && !ptask->enqueued);
if (!free)
ptask->enqueued = true;
@@ -1026,7 +1026,7 @@ static int fwnet_send_packet(struct fwnet_packet_task *ptask)
spin_lock_irqsave(&dev->lock, flags);
- /* If the AT tasklet already ran, we may be last user. */
+ /* If the AT work item already ran, we may be last user. */
free = (ptask->outstanding_pkts == 0 && !ptask->enqueued);
if (!free)
ptask->enqueued = true;
diff --git a/drivers/firewire/ohci.c b/drivers/firewire/ohci.c
index 68317b5a64a7..a81a876819d0 100644
--- a/drivers/firewire/ohci.c
+++ b/drivers/firewire/ohci.c
@@ -158,7 +158,7 @@ struct context {
descriptor_callback_t callback;
- struct tasklet_struct tasklet;
+ struct work_struct work;
};
struct iso_context {
@@ -1176,9 +1176,9 @@ static void context_retire_descriptors(struct context *ctx)
}
}
-static void context_tasklet(unsigned long data)
+static void ohci_at_context_work(struct work_struct *work)
{
- struct context *ctx = (struct context *) data;
+ struct context *ctx = from_work(ctx, work, work);
context_retire_descriptors(ctx);
}
@@ -1243,7 +1243,6 @@ static int context_init(struct context *ctx, struct fw_ohci *ohci,
ctx->buffer_tail = list_entry(ctx->buffer_list.next,
struct descriptor_buffer, list);
- tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
ctx->callback = callback;
/*
@@ -1524,13 +1523,17 @@ static int at_context_queue_packet(struct context *ctx,
static void at_context_flush(struct context *ctx)
{
- tasklet_disable(&ctx->tasklet);
+ // Avoid dead lock due to programming mistake.
+ if (WARN_ON_ONCE(current_work() == &ctx->work))
+ return;
+
+ disable_work_sync(&ctx->work);
- ctx->flushing = true;
- context_tasklet((unsigned long)ctx);
- ctx->flushing = false;
+ WRITE_ONCE(ctx->flushing, true);
+ ohci_at_context_work(&ctx->work);
+ WRITE_ONCE(ctx->flushing, false);
- tasklet_enable(&ctx->tasklet);
+ enable_work(&ctx->work);
}
static int handle_at_packet(struct context *context,
@@ -1542,7 +1545,7 @@ static int handle_at_packet(struct context *context,
struct fw_ohci *ohci = context->ohci;
int evt;
- if (last->transfer_status == 0 && !context->flushing)
+ if (last->transfer_status == 0 && !READ_ONCE(context->flushing))
/* This descriptor isn't done yet, stop iteration. */
return 0;
@@ -1576,7 +1579,7 @@ static int handle_at_packet(struct context *context,
break;
case OHCI1394_evt_missing_ack:
- if (context->flushing)
+ if (READ_ONCE(context->flushing))
packet->ack = RCODE_GENERATION;
else {
/*
@@ -1598,7 +1601,7 @@ static int handle_at_packet(struct context *context,
break;
case OHCI1394_evt_no_status:
- if (context->flushing) {
+ if (READ_ONCE(context->flushing)) {
packet->ack = RCODE_GENERATION;
break;
}
@@ -2239,10 +2242,10 @@ static irqreturn_t irq_handler(int irq, void *data)
queue_work(ohci->card.async_wq, &ohci->ar_response_ctx.work);
if (event & OHCI1394_reqTxComplete)
- tasklet_schedule(&ohci->at_request_ctx.tasklet);
+ queue_work(ohci->card.async_wq, &ohci->at_request_ctx.work);
if (event & OHCI1394_respTxComplete)
- tasklet_schedule(&ohci->at_response_ctx.tasklet);
+ queue_work(ohci->card.async_wq, &ohci->at_response_ctx.work);
if (event & OHCI1394_isochRx) {
iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
@@ -2684,7 +2687,10 @@ static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
struct driver_data *driver_data = packet->driver_data;
int ret = -ENOENT;
- tasklet_disable_in_atomic(&ctx->tasklet);
+ // Avoid dead lock due to programming mistake.
+ if (WARN_ON_ONCE(current_work() == &ctx->work))
+ return 0;
+ disable_work_sync(&ctx->work);
if (packet->ack != 0)
goto out;
@@ -2703,7 +2709,7 @@ static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
packet->callback(packet, &ohci->card, packet->ack);
ret = 0;
out:
- tasklet_enable(&ctx->tasklet);
+ enable_work(&ctx->work);
return ret;
}
@@ -3765,11 +3771,13 @@ static int pci_probe(struct pci_dev *dev,
OHCI1394_AsReqTrContextControlSet, handle_at_packet);
if (err < 0)
return err;
+ INIT_WORK(&ohci->at_request_ctx.work, ohci_at_context_work);
err = context_init(&ohci->at_response_ctx, ohci,
OHCI1394_AsRspTrContextControlSet, handle_at_packet);
if (err < 0)
return err;
+ INIT_WORK(&ohci->at_response_ctx.work, ohci_at_context_work);
reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
ohci->ir_context_channels = ~0ULL;
diff --git a/include/linux/firewire.h b/include/linux/firewire.h
index c55b8e30e700..cceb70415ed2 100644
--- a/include/linux/firewire.h
+++ b/include/linux/firewire.h
@@ -308,8 +308,7 @@ struct fw_packet {
* For successful transmission, the status code is the ack received
* from the destination. Otherwise it is one of the juju-specific
* rcodes: RCODE_SEND_ERROR, _CANCELLED, _BUSY, _GENERATION, _NO_ACK.
- * The callback can be called from tasklet context and thus
- * must never block.
+ * The callback can be called from workqueue and thus must never block.
*/
fw_packet_callback_t callback;
int ack;
@@ -382,6 +381,10 @@ void __fw_send_request(struct fw_card *card, struct fw_transaction *t, int tcode
*
* A variation of __fw_send_request() to generate callback for response subaction without time
* stamp.
+ *
+ * The callback is invoked in the workqueue context in most cases. However, if an error is detected
+ * before queueing or the destination address refers to the local node, it is invoked in the
+ * current context instead.
*/
static inline void fw_send_request(struct fw_card *card, struct fw_transaction *t, int tcode,
int destination_id, int generation, int speed,
@@ -411,6 +414,10 @@ static inline void fw_send_request(struct fw_card *card, struct fw_transaction *
* @callback_data: data to be passed to the transaction completion callback
*
* A variation of __fw_send_request() to generate callback for response subaction with time stamp.
+ *
+ * The callback is invoked in the workqueue context in most cases. However, if an error is detected
+ * before queueing or the destination address refers to the local node, it is invoked in the current
+ * context instead.
*/
static inline void fw_send_request_with_tstamp(struct fw_card *card, struct fw_transaction *t,
int tcode, int destination_id, int generation, int speed, unsigned long long offset,
--
2.48.1
^ permalink raw reply [flat|nested] 5+ messages in thread