firewire: optimize iso queueing by setting wake only after the last packet

When queueing iso packets, the run time is dominated by the two MMIO accesses that set the DMA context's wake bit. Because most drivers submit packets in batches, we can save much time by removing all but the last wakeup. The internal kernel API is changed to require a call to fw_iso_context_queue_flush() after a batch of queued packets. The user space API does not change, so one call to FW_CDEV_IOC_QUEUE_ISO must specify multiple packets to take advantage of this optimization. In my measurements, this patch reduces the time needed to queue fifty skip packets from userspace to one sixth on a 2.5 GHz CPU, or to one third at 800 MHz. Signed-off-by: Clemens Ladisch <clemens@ladisch.de> Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>

firewire: optimize iso queueing by setting wake only after the last packet
When queueing iso packets, the run time is dominated by the two MMIO accesses that set the DMA context's wake bit. Because most drivers submit packets in batches, we can save much time by removing all but the last wakeup. The internal kernel API is changed to require a call to fw_iso_context_queue_flush() after a batch of queued packets. The user space API does not change, so one call to FW_CDEV_IOC_QUEUE_ISO must specify multiple packets to take advantage of this optimization. In my measurements, this patch reduces the time needed to queue fifty skip packets from userspace to one sixth on a 2.5 GHz CPU, or to one third at 800 MHz. Signed-off-by: Clemens Ladisch <clemens@ladisch.de> Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Clemens Ladisch · Stefan Richter
1 parent f30e6d3e41
Showing 9 changed files with 35 additions and 5 deletions Side-by-side Diff
drivers/firewire/core-card.c
drivers/firewire/core-cdev.c
drivers/firewire/core-iso.c
drivers/firewire/core.h
drivers/firewire/net.c
drivers/firewire/ohci.c
drivers/media/dvb/firewire/firedtv-fw.c
include/linux/firewire.h
sound/firewire/amdtp.c
@@ -630,6 +630,10 @@
 	return -ENODEV;
 }
  
+static void dummy_flush_queue_iso(struct fw_iso_context *ctx)
+{
+}
+
 static const struct fw_card_driver dummy_driver_template = {
 	.read_phy_reg		= dummy_read_phy_reg,
 	.update_phy_reg		= dummy_update_phy_reg,
@@ -641,6 +645,7 @@
 	.start_iso		= dummy_start_iso,
 	.set_iso_channels	= dummy_set_iso_channels,
 	.queue_iso		= dummy_queue_iso,
+	.flush_queue_iso	= dummy_flush_queue_iso,
 };
  
 void fw_card_release(struct kref *kref)
@@ -1107,6 +1107,7 @@
 		payload += u.packet.payload_length;
 		count++;
 	}
+	fw_iso_context_queue_flush(ctx);
  
 	a->size    -= uptr_to_u64(p) - a->packets;
 	a->packets  = uptr_to_u64(p);
@@ -185,6 +185,12 @@
 }
 EXPORT_SYMBOL(fw_iso_context_queue);
  
+void fw_iso_context_queue_flush(struct fw_iso_context *ctx)
+{
+	ctx->card->driver->flush_queue_iso(ctx);
+}
+EXPORT_SYMBOL(fw_iso_context_queue_flush);
+
 int fw_iso_context_stop(struct fw_iso_context *ctx)
 {
 	return ctx->card->driver->stop_iso(ctx);
@@ -97,6 +97,8 @@
 			 struct fw_iso_buffer *buffer,
 			 unsigned long payload);
  
+	void (*flush_queue_iso)(struct fw_iso_context *ctx);
+
 	int (*stop_iso)(struct fw_iso_context *ctx);
 };
  
@@ -881,7 +881,9 @@
  
 	spin_unlock_irqrestore(&dev->lock, flags);
  
-	if (retval < 0)
+	if (retval >= 0)
+		fw_iso_context_queue_flush(dev->broadcast_rcv_context);
+	else
 		fw_error("requeue failed\n");
 }
  
@@ -1192,9 +1192,6 @@
 	wmb(); /* finish init of new descriptors before branch_address update */
 	ctx->prev->branch_address = cpu_to_le32(d_bus | z);
 	ctx->prev = find_branch_descriptor(d, z);
-
-	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
-	flush_writes(ctx->ohci);
 }
  
 static void context_stop(struct context *ctx)
  
@@ -1348,8 +1345,12 @@
  
 	context_append(ctx, d, z, 4 - z);
  
-	if (!ctx->running)
+	if (ctx->running) {
+		reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
+		flush_writes(ohci);
+	} else {
 		context_run(ctx, 0);
+	}
  
 	return 0;
 }
@@ -3121,6 +3122,15 @@
 	return ret;
 }
  
+static void ohci_flush_queue_iso(struct fw_iso_context *base)
+{
+	struct context *ctx =
+			&container_of(base, struct iso_context, base)->context;
+
+	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
+	flush_writes(ctx->ohci);
+}
+
 static const struct fw_card_driver ohci_driver = {
 	.enable			= ohci_enable,
 	.read_phy_reg		= ohci_read_phy_reg,
@@ -3137,6 +3147,7 @@
 	.free_iso_context	= ohci_free_iso_context,
 	.set_iso_channels	= ohci_set_iso_channels,
 	.queue_iso		= ohci_queue_iso,
+	.flush_queue_iso	= ohci_flush_queue_iso,
 	.start_iso		= ohci_start_iso,
 	.stop_iso		= ohci_stop_iso,
 };
@@ -125,6 +125,7 @@
  
 		i = (i + 1) & (N_PACKETS - 1);
 	}
+	fw_iso_context_queue_flush(ctx->context);
 	ctx->current_packet = i;
 }
  
@@ -440,6 +440,7 @@
 			 struct fw_iso_packet *packet,
 			 struct fw_iso_buffer *buffer,
 			 unsigned long payload);
+void fw_iso_context_queue_flush(struct fw_iso_context *ctx);
 int fw_iso_context_start(struct fw_iso_context *ctx,
 			 int cycle, int sync, int tags);
 int fw_iso_context_stop(struct fw_iso_context *ctx);
@@ -396,6 +396,7 @@
  
 	for (i = 0; i < packets; ++i)
 		queue_out_packet(s, ++cycle);
+	fw_iso_context_queue_flush(s->context);
 }
  
 static int queue_initial_skip_packets(struct amdtp_out_stream *s)
...	...	@@ -630,6 +630,10 @@
630	630	return -ENODEV;
631	631	}
632	632
	633	+static void dummy_flush_queue_iso(struct fw_iso_context *ctx)
	634	+{
	635	+}
	636	+
633	637	static const struct fw_card_driver dummy_driver_template = {
634	638	.read_phy_reg = dummy_read_phy_reg,
635	639	.update_phy_reg = dummy_update_phy_reg,
...	...	@@ -641,6 +645,7 @@
641	645	.start_iso = dummy_start_iso,
642	646	.set_iso_channels = dummy_set_iso_channels,
643	647	.queue_iso = dummy_queue_iso,
	648	+ .flush_queue_iso = dummy_flush_queue_iso,
644	649	};
645	650
646	651	void fw_card_release(struct kref *kref)
...	...	@@ -1107,6 +1107,7 @@
1107	1107	payload += u.packet.payload_length;
1108	1108	count++;
1109	1109	}
	1110	+ fw_iso_context_queue_flush(ctx);
1110	1111
1111	1112	a->size -= uptr_to_u64(p) - a->packets;
1112	1113	a->packets = uptr_to_u64(p);
...	...	@@ -185,6 +185,12 @@
185	185	}
186	186	EXPORT_SYMBOL(fw_iso_context_queue);
187	187
	188	+void fw_iso_context_queue_flush(struct fw_iso_context *ctx)
	189	+{
	190	+ ctx->card->driver->flush_queue_iso(ctx);
	191	+}
	192	+EXPORT_SYMBOL(fw_iso_context_queue_flush);
	193	+
188	194	int fw_iso_context_stop(struct fw_iso_context *ctx)
189	195	{
190	196	return ctx->card->driver->stop_iso(ctx);
...	...	@@ -97,6 +97,8 @@
97	97	struct fw_iso_buffer *buffer,
98	98	unsigned long payload);
99	99
	100	+ void (flush_queue_iso)(struct fw_iso_context ctx);
	101	+
100	102	int (stop_iso)(struct fw_iso_context ctx);
101	103	};
102	104
...	...	@@ -881,7 +881,9 @@
881	881
882	882	spin_unlock_irqrestore(&dev->lock, flags);
883	883
884		- if (retval < 0)
	884	+ if (retval >= 0)
	885	+ fw_iso_context_queue_flush(dev->broadcast_rcv_context);
	886	+ else
885	887	fw_error("requeue failed\n");
886	888	}
887	889
...	...	@@ -1192,9 +1192,6 @@
1192	1192	wmb(); /* finish init of new descriptors before branch_address update */
1193	1193	ctx->prev->branch_address = cpu_to_le32(d_bus \| z);
1194	1194	ctx->prev = find_branch_descriptor(d, z);
1195		-
1196		- reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
1197		- flush_writes(ctx->ohci);
1198	1195	}
1199	1196
1200	1197	static void context_stop(struct context *ctx)
1201	1198
...	...	@@ -1348,8 +1345,12 @@
1348	1345
1349	1346	context_append(ctx, d, z, 4 - z);
1350	1347
1351		- if (!ctx->running)
	1348	+ if (ctx->running) {
	1349	+ reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
	1350	+ flush_writes(ohci);
	1351	+ } else {
1352	1352	context_run(ctx, 0);
	1353	+ }
1353	1354
1354	1355	return 0;
1355	1356	}
...	...	@@ -3121,6 +3122,15 @@
3121	3122	return ret;
3122	3123	}
3123	3124
	3125	+static void ohci_flush_queue_iso(struct fw_iso_context *base)
	3126	+{
	3127	+ struct context *ctx =
	3128	+ &container_of(base, struct iso_context, base)->context;
	3129	+
	3130	+ reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
	3131	+ flush_writes(ctx->ohci);
	3132	+}
	3133	+
3124	3134	static const struct fw_card_driver ohci_driver = {
3125	3135	.enable = ohci_enable,
3126	3136	.read_phy_reg = ohci_read_phy_reg,
...	...	@@ -3137,6 +3147,7 @@
3137	3147	.free_iso_context = ohci_free_iso_context,
3138	3148	.set_iso_channels = ohci_set_iso_channels,
3139	3149	.queue_iso = ohci_queue_iso,
	3150	+ .flush_queue_iso = ohci_flush_queue_iso,
3140	3151	.start_iso = ohci_start_iso,
3141	3152	.stop_iso = ohci_stop_iso,
3142	3153	};
...	...	@@ -125,6 +125,7 @@
125	125
126	126	i = (i + 1) & (N_PACKETS - 1);
127	127	}
	128	+ fw_iso_context_queue_flush(ctx->context);
128	129	ctx->current_packet = i;
129	130	}
130	131
...	...	@@ -440,6 +440,7 @@
440	440	struct fw_iso_packet *packet,
441	441	struct fw_iso_buffer *buffer,
442	442	unsigned long payload);
	443	+void fw_iso_context_queue_flush(struct fw_iso_context *ctx);
443	444	int fw_iso_context_start(struct fw_iso_context *ctx,
444	445	int cycle, int sync, int tags);
445	446	int fw_iso_context_stop(struct fw_iso_context *ctx);
...	...	@@ -396,6 +396,7 @@
396	396
397	397	for (i = 0; i < packets; ++i)
398	398	queue_out_packet(s, ++cycle);
	399	+ fw_iso_context_queue_flush(s->context);
399	400	}
400	401
401	402	static int queue_initial_skip_packets(struct amdtp_out_stream *s)