linux-toradex.git/drivers/firewire/core-card.c, branch v6.11 Linux kernel for Apalis and Colibri modules firewire: core: record card index in tracepoinrts events derived from bus_reset_arrange_template 2024-06-15T05:59:17+00:00 Takashi Sakamoto o-takashi@sakamocchi.jp 2024-06-13T13:14:39+00:00 7507dbc46b780e9fa2ec3d4db7cd9ce07e722927 The asynchronous transmission of phy packet is initiated on one of 1394 OHCI controller, however the existing tracepoints events has the lack of data about it. This commit adds card_index member into event structure to store the index of host controller in use, and prints it. Link: https://lore.kernel.org/r/20240613131440.431766-8-o-takashi@sakamocchi.jp Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> 
The asynchronous transmission of phy packet is initiated on one of 1394
OHCI controller, however the existing tracepoints events has the lack of
data about it.

This commit adds card_index member into event structure to store the index
of host controller in use, and prints it.

Link: https://lore.kernel.org/r/20240613131440.431766-8-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Revert "firewire: core: option to log bus reset initiation" 2024-05-06T02:06:06+00:00 Takashi Sakamoto o-takashi@sakamocchi.jp 2024-05-01T07:32:37+00:00 01d860427f67fd4b797ae616a2adb57263baeee9 This reverts commit 6732491243045f5a7e1995b4be5f3c964b579ebd. The former commit adds some alternative tracepoints events to replace the reverted kernel log messages. Link: https://lore.kernel.org/r/20240501073238.72769-5-o-takashi@sakamocchi.jp Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> 
This reverts commit 6732491243045f5a7e1995b4be5f3c964b579ebd.

The former commit adds some alternative tracepoints events to replace the
reverted kernel log messages.

Link: https://lore.kernel.org/r/20240501073238.72769-5-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
firewire: core: add tracepoints events for initiating bus reset 2024-05-06T02:06:06+00:00 Takashi Sakamoto o-takashi@sakamocchi.jp 2024-05-01T07:32:36+00:00 08dd8602aab9fe9dbc3e41727caf7e416d2ccec4 At a commit 673249124304 ("firewire: core: option to log bus reset initiation"), some kernel log messages were added to trace initiation of bus reset. The kernel log messages are really helpful, while nowadays it is not preferable just for debugging purpose. For the purpose, Linux kernel tracepoints is more preferable. This commit adds some alternative tracepoints events. Link: https://lore.kernel.org/r/20240501073238.72769-4-o-takashi@sakamocchi.jp Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> 
At a commit 673249124304 ("firewire: core: option to log bus reset
initiation"), some kernel log messages were added to trace initiation of
bus reset. The kernel log messages are really helpful, while nowadays it
is not preferable just for debugging purpose. For the purpose, Linux
kernel tracepoints is more preferable.

This commit adds some alternative tracepoints events.

Link: https://lore.kernel.org/r/20240501073238.72769-4-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
firewire: core: option to log bus reset initiation 2024-05-06T02:06:03+00:00 Adam Goldman adamg@pobox.com 2024-03-24T22:37:25+00:00 f83a052f9a0b62318d4f6e9bdddf9fc0103ac908 Add a debug parameter to firewire-core, analogous to the one in firewire-ohci. When this is set to 1, log when we schedule, delay, or initiate a bus reset. Since FireWire bus resets can originate from any node on the bus, specific logging of the resets we initiate provides additional insight. Signed-off-by: Adam Goldman <adamg@pobox.com> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> 
Add a debug parameter to firewire-core, analogous to the one in
firewire-ohci. When this is set to 1, log when we schedule, delay, or
initiate a bus reset. Since FireWire bus resets can originate from any
node on the bus, specific logging of the resets we initiate provides
additional insight.

Signed-off-by: Adam Goldman <adamg@pobox.com>
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
firewire: core: use long bus reset on gap count error 2024-02-29T13:18:14+00:00 Takashi Sakamoto o-takashi@sakamocchi.jp 2024-02-29T13:17:37+00:00 d0b06dc48fb15902d7da09c5c0861e7f042a9381 When resetting the bus after a gap count error, use a long rather than short bus reset. IEEE 1394-1995 uses only long bus resets. IEEE 1394a adds the option of short bus resets. When video or audio transmission is in progress and a device is hot-plugged elsewhere on the bus, the resulting bus reset can cause video frame drops or audio dropouts. Short bus resets reduce or eliminate this problem. Accordingly, short bus resets are almost always preferred. However, on a mixed 1394/1394a bus, a short bus reset can trigger an immediate additional bus reset. This double bus reset can be interpreted differently by different nodes on the bus, resulting in an inconsistent gap count after the bus reset. An inconsistent gap count will cause another bus reset, leading to a neverending bus reset loop. This only happens for some bus topologies, not for all mixed 1394/1394a buses. By instead sending a long bus reset after a gap count inconsistency, we avoid the doubled bus reset, restoring the bus to normal operation. Signed-off-by: Adam Goldman <adamg@pobox.com> Link: https://sourceforge.net/p/linux1394/mailman/message/58741624/ Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> 
When resetting the bus after a gap count error, use a long rather than
short bus reset.

IEEE 1394-1995 uses only long bus resets. IEEE 1394a adds the option of
short bus resets. When video or audio transmission is in progress and a
device is hot-plugged elsewhere on the bus, the resulting bus reset can
cause video frame drops or audio dropouts. Short bus resets reduce or
eliminate this problem. Accordingly, short bus resets are almost always
preferred.

However, on a mixed 1394/1394a bus, a short bus reset can trigger an
immediate additional bus reset. This double bus reset can be interpreted
differently by different nodes on the bus, resulting in an inconsistent gap
count after the bus reset. An inconsistent gap count will cause another bus
reset, leading to a neverending bus reset loop. This only happens for some
bus topologies, not for all mixed 1394/1394a buses.

By instead sending a long bus reset after a gap count inconsistency, we
avoid the doubled bus reset, restoring the bus to normal operation.

Signed-off-by: Adam Goldman <adamg@pobox.com>
Link: https://sourceforge.net/p/linux1394/mailman/message/58741624/
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
firewire: core: send bus reset promptly on gap count error 2024-02-06T23:20:02+00:00 Takashi Sakamoto o-takashi@sakamocchi.jp 2024-02-06T23:01:17+00:00 7ed4380009e96d9e9c605e12822e987b35b05648 If we are bus manager and the bus has inconsistent gap counts, send a bus reset immediately instead of trying to read the root node's config ROM first. Otherwise, we could spend a lot of time trying to read the config ROM but never succeeding. This eliminates a 50+ second delay before the FireWire bus is usable after a newly connected device is powered on in certain circumstances. The delay occurs if a gap count inconsistency occurs, we are not the root node, and we become bus manager. One scenario that causes this is with a TI XIO2213B OHCI, the first time a Sony DSR-25 is powered on after being connected to the FireWire cable. In this configuration, the Linux box will not receive the initial PHY configuration packet sent by the DSR-25 as IRM, resulting in the DSR-25 having a gap count of 44 while the Linux box has a gap count of 63. FireWire devices have a gap count parameter, which is set to 63 on power-up and can be changed with a PHY configuration packet. This determines the duration of the subaction and arbitration gaps. For reliable communication, all nodes on a FireWire bus must have the same gap count. A node may have zero or more of the following roles: root node, bus manager (BM), isochronous resource manager (IRM), and cycle master. Unless a root node was forced with a PHY configuration packet, any node might become root node after a bus reset. Only the root node can become cycle master. If the root node is not cycle master capable, the BM or IRM should force a change of root node. After a bus reset, each node sends a self-ID packet, which contains its current gap count. A single bus reset does not change the gap count, but two bus resets in a row will set the gap count to 63. Because a consistent gap count is required for reliable communication, IEEE 1394a-2000 requires that the bus manager generate a bus reset if it detects that the gap count is inconsistent. When the gap count is inconsistent, build_tree() will notice this after the self identification process. It will set card->gap_count to the invalid value 0. If we become bus master, this will force bm_work() to send a bus reset when it performs gap count optimization. After a bus reset, there is no bus manager. We will almost always try to become bus manager. Once we become bus manager, we will first determine whether the root node is cycle master capable. Then, we will determine if the gap count should be changed. If either the root node or the gap count should be changed, we will generate a bus reset. To determine if the root node is cycle master capable, we read its configuration ROM. bm_work() will wait until we have finished trying to read the configuration ROM. However, an inconsistent gap count can make this take a long time. read_config_rom() will read the first few quadlets from the config ROM. Due to the gap count inconsistency, eventually one of the reads will time out. When read_config_rom() fails, fw_device_init() calls it again until MAX_RETRIES is reached. This takes 50+ seconds. Once we give up trying to read the configuration ROM, bm_work() will wake up, assume that the root node is not cycle master capable, and do a bus reset. Hopefully, this will resolve the gap count inconsistency. This change makes bm_work() check for an inconsistent gap count before waiting for the root node's configuration ROM. If the gap count is inconsistent, bm_work() will immediately do a bus reset. This eliminates the 50+ second delay and rapidly brings the bus to a working state. I considered that if the gap count is inconsistent, a PHY configuration packet might not be successful, so it could be desirable to skip the PHY configuration packet before the bus reset in this case. However, IEEE 1394a-2000 and IEEE 1394-2008 say that the bus manager may transmit a PHY configuration packet before a bus reset when correcting a gap count error. Since the standard endorses this, I decided it's safe to retain the PHY configuration packet transmission. Normally, after a topology change, we will reset the bus a maximum of 5 times to change the root node and perform gap count optimization. However, if there is a gap count inconsistency, we must always generate a bus reset. Otherwise the gap count inconsistency will persist and communication will be unreliable. For that reason, if there is a gap count inconstency, we generate a bus reset even if we already reached the 5 reset limit. Signed-off-by: Adam Goldman <adamg@pobox.com> Reference: https://sourceforge.net/p/linux1394/mailman/message/58727806/ Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> 
If we are bus manager and the bus has inconsistent gap counts, send a
bus reset immediately instead of trying to read the root node's config
ROM first. Otherwise, we could spend a lot of time trying to read the
config ROM but never succeeding.

This eliminates a 50+ second delay before the FireWire bus is usable after
a newly connected device is powered on in certain circumstances.

The delay occurs if a gap count inconsistency occurs, we are not the root
node, and we become bus manager. One scenario that causes this is with a TI
XIO2213B OHCI, the first time a Sony DSR-25 is powered on after being
connected to the FireWire cable. In this configuration, the Linux box will
not receive the initial PHY configuration packet sent by the DSR-25 as IRM,
resulting in the DSR-25 having a gap count of 44 while the Linux box has a
gap count of 63.

FireWire devices have a gap count parameter, which is set to 63 on power-up
and can be changed with a PHY configuration packet. This determines the
duration of the subaction and arbitration gaps. For reliable communication,
all nodes on a FireWire bus must have the same gap count.

A node may have zero or more of the following roles: root node, bus manager
(BM), isochronous resource manager (IRM), and cycle master. Unless a root
node was forced with a PHY configuration packet, any node might become root
node after a bus reset. Only the root node can become cycle master. If the
root node is not cycle master capable, the BM or IRM should force a change
of root node.

After a bus reset, each node sends a self-ID packet, which contains its
current gap count. A single bus reset does not change the gap count, but
two bus resets in a row will set the gap count to 63. Because a consistent
gap count is required for reliable communication, IEEE 1394a-2000 requires
that the bus manager generate a bus reset if it detects that the gap count
is inconsistent.

When the gap count is inconsistent, build_tree() will notice this after the
self identification process. It will set card->gap_count to the invalid
value 0. If we become bus master, this will force bm_work() to send a bus
reset when it performs gap count optimization.

After a bus reset, there is no bus manager. We will almost always try to
become bus manager. Once we become bus manager, we will first determine
whether the root node is cycle master capable. Then, we will determine if
the gap count should be changed. If either the root node or the gap count
should be changed, we will generate a bus reset.

To determine if the root node is cycle master capable, we read its
configuration ROM. bm_work() will wait until we have finished trying to
read the configuration ROM.

However, an inconsistent gap count can make this take a long time.
read_config_rom() will read the first few quadlets from the config ROM. Due
to the gap count inconsistency, eventually one of the reads will time out.
When read_config_rom() fails, fw_device_init() calls it again until
MAX_RETRIES is reached. This takes 50+ seconds.

Once we give up trying to read the configuration ROM, bm_work() will wake
up, assume that the root node is not cycle master capable, and do a bus
reset. Hopefully, this will resolve the gap count inconsistency.

This change makes bm_work() check for an inconsistent gap count before
waiting for the root node's configuration ROM. If the gap count is
inconsistent, bm_work() will immediately do a bus reset. This eliminates
the 50+ second delay and rapidly brings the bus to a working state.

I considered that if the gap count is inconsistent, a PHY configuration
packet might not be successful, so it could be desirable to skip the PHY
configuration packet before the bus reset in this case. However, IEEE
1394a-2000 and IEEE 1394-2008 say that the bus manager may transmit a PHY
configuration packet before a bus reset when correcting a gap count error.
Since the standard endorses this, I decided it's safe to retain the PHY
configuration packet transmission.

Normally, after a topology change, we will reset the bus a maximum of 5
times to change the root node and perform gap count optimization. However,
if there is a gap count inconsistency, we must always generate a bus reset.
Otherwise the gap count inconsistency will persist and communication will
be unreliable. For that reason, if there is a gap count inconstency, we
generate a bus reset even if we already reached the 5 reset limit.

Signed-off-by: Adam Goldman <adamg@pobox.com>
Reference: https://sourceforge.net/p/linux1394/mailman/message/58727806/
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Merge branch 'for-linus' into for-next 2022-05-23T05:48:27+00:00 Takashi Iwai tiwai@suse.de 2022-05-23T05:48:16+00:00 1212fa1b482ea09c19aed29001465799e85ea6c3 Merge for 5.18-rc1 Signed-off-by: Takashi Iwai <tiwai@suse.de> 
Merge for 5.18-rc1

Signed-off-by: Takashi Iwai <tiwai@suse.de>
firewire: core: extend card->lock in fw_core_handle_bus_reset 2022-04-25T06:01:09+00:00 Niels Dossche dossche.niels@gmail.com 2022-04-09T04:12:43+00:00 a7ecbe92b9243edbe94772f6f2c854e4142a3345 card->local_node and card->bm_retries are both always accessed under card->lock. fw_core_handle_bus_reset has a check whose condition depends on card->local_node and whose body writes to card->bm_retries. Both of these accesses are not under card->lock. Move the lock acquiring of card->lock to before this check such that these accesses do happen when card->lock is held. fw_destroy_nodes is called inside the check. Since fw_destroy_nodes already acquires card->lock inside its function body, move this out to the callsites of fw_destroy_nodes. Also add a comment to indicate which locking is necessary when calling fw_destroy_nodes. Cc: <stable@vger.kernel.org> Signed-off-by: Niels Dossche <dossche.niels@gmail.com> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20220409041243.603210-4-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de> 
card->local_node and card->bm_retries are both always accessed under
card->lock.
fw_core_handle_bus_reset has a check whose condition depends on
card->local_node and whose body writes to card->bm_retries.
Both of these accesses are not under card->lock. Move the lock acquiring
of card->lock to before this check such that these accesses do happen
when card->lock is held.
fw_destroy_nodes is called inside the check.
Since fw_destroy_nodes already acquires card->lock inside its function
body, move this out to the callsites of fw_destroy_nodes.
Also add a comment to indicate which locking is necessary when calling
fw_destroy_nodes.

Cc: <stable@vger.kernel.org>
Signed-off-by: Niels Dossche <dossche.niels@gmail.com>
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Link: https://lore.kernel.org/r/20220409041243.603210-4-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Iwai <tiwai@suse.de>
firewire: add kernel API to access CYCLE_TIME register 2022-04-05T16:23:04+00:00 Takashi Sakamoto o-takashi@sakamocchi.jp 2022-04-05T07:22:20+00:00 baa914cd81f51f4e4f3bae5bb59764b32ad8c353 1394 OHCI specification defined Isochronous Cycle Timer Register to get value of CYCLE_TIME register defined by IEEE 1394 for CSR architecture defined by ISO/IEC 13213. Unit driver can calculate packet time by compute with the value of CYCLE_TIME and timeStamp field in descriptor of each isochronous and asynchronous context. The resolution of CYCLE_TIME is 49.576 MHz, while the one of timeStamp is 8,000 Hz. Current implementation of Linux FireWire subsystem allows the driver to get the value of CYCLE_TIMER CSR register by transaction service. The transaction service has overhead in regard of access to MMIO register. This commit adds kernel API for unit driver to access the register directly. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20220405072221.226217-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de> 
1394 OHCI specification defined Isochronous Cycle Timer Register to get
value of CYCLE_TIME register defined by IEEE 1394 for CSR architecture
defined by ISO/IEC 13213. Unit driver can calculate packet time by
compute with the value of CYCLE_TIME and timeStamp field in descriptor
of each isochronous and asynchronous context. The resolution of CYCLE_TIME
is 49.576 MHz, while the one of timeStamp is 8,000 Hz.

Current implementation of Linux FireWire subsystem allows the driver to
get the value of CYCLE_TIMER CSR register by transaction service. The
transaction service has overhead in regard of access to MMIO register.

This commit adds kernel API for unit driver to access the register
directly.

Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Link: https://lore.kernel.org/r/20220405072221.226217-3-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Iwai <tiwai@suse.de>
firewire: Add dummy read_csr/write_csr functions 2022-04-05T16:23:03+00:00 Hector Martin marcan@marcan.st 2022-04-05T07:22:19+00:00 e70b6693cec229717dda683699ae1fd9ed160e3d (Hector Martin wrote) This fixes segfaults when a card gets yanked off of the PCIe bus while busy, e.g. with a userspace app trying to get the cycle time: [8638860.994310] Call Trace: [8638860.994313] ioctl_get_cycle_timer2+0x4f/0xd0 [firewire_core] [8638860.994323] fw_device_op_ioctl+0xae/0x150 [firewire_core] [8638860.994328] __x64_sys_ioctl+0x7d/0xb0 [8638860.994332] do_syscall_64+0x45/0x80 [8638860.994337] entry_SYSCALL_64_after_hwframe+0x44/0xae (Takashi Sakamoto wrote) As long as reading commit 20802224298c ("firewire: core: add forgotten dummy driver methods, remove unused ones"), three functions are not implemeted in dummy driver for reason; .read_csr, .write_csr, and .set_config_rom. In core of Linux FireWire subsystem, the callback of .set_config_rom is under acquisition of mutual exclusive for local list of card. The acquision is also done in process for removal of card, therefore it's safe for missing implementation of .set_config_rom. On the other hand, no lock primitive accompanies any call of .read_csr and .write_csr. For userspace client, check of node shutdown is done in the beginning of dispatch of ioctl request, while node shifts to shutdown state in workqueue context enough after card shifts to dummy driver. It's probable that these two functions are called for the dummy driver by the code of userspace client. In-kernel unit driver has similar situation. It's better to add implementation of the two functions for dummy driver. Signed-off-by: Hector Martin <marcan@marcan.st> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20220405072221.226217-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de> 
(Hector Martin wrote)
This fixes segfaults when a card gets yanked off of the PCIe bus while
busy, e.g. with a userspace app trying to get the cycle time:

[8638860.994310] Call Trace:
[8638860.994313]  ioctl_get_cycle_timer2+0x4f/0xd0 [firewire_core]
[8638860.994323]  fw_device_op_ioctl+0xae/0x150 [firewire_core]
[8638860.994328]  __x64_sys_ioctl+0x7d/0xb0
[8638860.994332]  do_syscall_64+0x45/0x80
[8638860.994337]  entry_SYSCALL_64_after_hwframe+0x44/0xae

(Takashi Sakamoto wrote)
As long as reading commit 20802224298c ("firewire: core: add forgotten
dummy driver methods, remove unused ones"), three functions are not
implemeted in dummy driver for reason; .read_csr, .write_csr, and
.set_config_rom.

In core of Linux FireWire subsystem, the callback of .set_config_rom is
under acquisition of mutual exclusive for local list of card. The
acquision is also done in process for removal of card, therefore it's
safe for missing implementation of .set_config_rom.

On the other hand, no lock primitive accompanies any call of .read_csr and
.write_csr. For userspace client, check of node shutdown is done in the
beginning of dispatch of ioctl request, while node shifts to shutdown
state in workqueue context enough after card shifts to dummy driver. It's
probable that these two functions are called for the dummy driver by the
code of userspace client. In-kernel unit driver has similar situation.
It's better to add implementation of the two functions for dummy driver.

Signed-off-by: Hector Martin <marcan@marcan.st>
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Link: https://lore.kernel.org/r/20220405072221.226217-2-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Iwai <tiwai@suse.de>