There are 2 issues related to the enabling of GPEs:
1. Currently, our code clears GPE before enabling it. In case of edge
triggered GPEs, doing this risks GPE losses.
2. For edge-triggered GPEs, enabling it is not sufficiently to trigger an
already triggered GPE, we need to poll the GPE once it is enabled.
This patchset fixes these 2 problems.
Lv Zheng (2):
ACPICA: Events: Stop unconditionally clearing ACPI IRQs during
suspend/resume
ACPICA: Events: Dispatch GPEs after enabling for the first time
drivers/acpi/acpica/evgpe.c | 7 -------
drivers/acpi/acpica/evxfgpe.c | 22 ++++++++++++++++++++++
drivers/acpi/acpica/hwgpe.c | 1 -
drivers/acpi/acpica/hwsleep.c | 11 ++---------
drivers/acpi/acpica/hwxfsleep.c | 2 +-
5 files changed, 25 insertions(+), 18 deletions(-)
--
2.7.4
Unconditionally clearing ACPI IRQs during suspend/resume can lead to
unexpected IRQ losts. This patch fixes this issue by removing such IRQ
clearing code.
If this patch triggers regression, the regression should be in the GPE
handlers that cannot correctly determine some spurious triggered events as
no-ops. Please report any regression related to this commit to the ACPI
component on kernel bugzilla. Reported by Eric Bakula-Davis, fixed by Lv
Zheng.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=196249
Signed-off-by: Lv Zheng <[email protected]>
Reported-and-tested-by: Eric Bakula-Davis <[email protected]>
Tested-by: Mika Westerberg <[email protected]>
---
drivers/acpi/acpica/evgpe.c | 7 -------
drivers/acpi/acpica/hwgpe.c | 1 -
drivers/acpi/acpica/hwsleep.c | 11 ++---------
drivers/acpi/acpica/hwxfsleep.c | 2 +-
4 files changed, 3 insertions(+), 18 deletions(-)
diff --git a/drivers/acpi/acpica/evgpe.c b/drivers/acpi/acpica/evgpe.c
index 2293820..390a5ce 100644
--- a/drivers/acpi/acpica/evgpe.c
+++ b/drivers/acpi/acpica/evgpe.c
@@ -115,13 +115,6 @@ acpi_status acpi_ev_enable_gpe(struct acpi_gpe_event_info *gpe_event_info)
ACPI_FUNCTION_TRACE(ev_enable_gpe);
- /* Clear the GPE (of stale events) */
-
- status = acpi_hw_clear_gpe(gpe_event_info);
- if (ACPI_FAILURE(status)) {
- return_ACPI_STATUS(status);
- }
-
/* Enable the requested GPE */
status = acpi_hw_low_set_gpe(gpe_event_info, ACPI_GPE_ENABLE);
diff --git a/drivers/acpi/acpica/hwgpe.c b/drivers/acpi/acpica/hwgpe.c
index 5eb11b3..e004d5f 100644
--- a/drivers/acpi/acpica/hwgpe.c
+++ b/drivers/acpi/acpica/hwgpe.c
@@ -497,7 +497,6 @@ acpi_status acpi_hw_disable_all_gpes(void)
ACPI_FUNCTION_TRACE(hw_disable_all_gpes);
status = acpi_ev_walk_gpe_list(acpi_hw_disable_gpe_block, NULL);
- status = acpi_ev_walk_gpe_list(acpi_hw_clear_gpe_block, NULL);
return_ACPI_STATUS(status);
}
diff --git a/drivers/acpi/acpica/hwsleep.c b/drivers/acpi/acpica/hwsleep.c
index 1fe7387..8f54af6 100644
--- a/drivers/acpi/acpica/hwsleep.c
+++ b/drivers/acpi/acpica/hwsleep.c
@@ -85,15 +85,8 @@ acpi_status acpi_hw_legacy_sleep(u8 sleep_state)
return_ACPI_STATUS(status);
}
- /* Clear all fixed and general purpose status bits */
-
- status = acpi_hw_clear_acpi_status();
- if (ACPI_FAILURE(status)) {
- return_ACPI_STATUS(status);
- }
-
/*
- * 1) Disable/Clear all GPEs
+ * 1) Disable all GPEs
* 2) Enable all wakeup GPEs
*/
status = acpi_hw_disable_all_gpes();
@@ -299,7 +292,7 @@ acpi_status acpi_hw_legacy_wake(u8 sleep_state)
* might get fired there
*
* Restore the GPEs:
- * 1) Disable/Clear all GPEs
+ * 1) Disable all GPEs
* 2) Enable all runtime GPEs
*/
status = acpi_hw_disable_all_gpes();
diff --git a/drivers/acpi/acpica/hwxfsleep.c b/drivers/acpi/acpica/hwxfsleep.c
index 7ef1393..1888107 100644
--- a/drivers/acpi/acpica/hwxfsleep.c
+++ b/drivers/acpi/acpica/hwxfsleep.c
@@ -206,7 +206,7 @@ acpi_status acpi_enter_sleep_state_s4bios(void)
}
/*
- * 1) Disable/Clear all GPEs
+ * 1) Disable all GPEs
* 2) Enable all wakeup GPEs
*/
status = acpi_hw_disable_all_gpes();
--
2.7.4
After being enabled for the first time, the GPEs may have STS bits already
set. Setting EN bits is not sufficient to trigger the GPEs again, so this
patch polls GPEs after enabling them for the first time.
This is a simpler version on top of the "GPE clear" fix generated according
to Mika's report and Rafael's original Linux based fix. Based on Linux
commit originated from Rafael J. Wysocki, fixed by Lv Zheng, tested by Mika
Westerberg.
Original-by: Rafael J. Wysocki <[email protected]>
Signed-off-by: Lv Zheng <[email protected]>
Tested-by: Mika Westerberg <[email protected]>
---
drivers/acpi/acpica/evxfgpe.c | 22 ++++++++++++++++++++++
1 file changed, 22 insertions(+)
diff --git a/drivers/acpi/acpica/evxfgpe.c b/drivers/acpi/acpica/evxfgpe.c
index 57718a3..d8be21d 100644
--- a/drivers/acpi/acpica/evxfgpe.c
+++ b/drivers/acpi/acpica/evxfgpe.c
@@ -97,6 +97,14 @@ acpi_status acpi_update_all_gpes(void)
unlock_and_exit:
(void)acpi_ut_release_mutex(ACPI_MTX_EVENTS);
+ /*
+ * Poll GPEs to handle already triggered events.
+ * It is not sufficient to trigger edge-triggered GPE with specific
+ * GPE chips, software need to poll once after enabling.
+ */
+ if (acpi_gbl_all_gpes_initialized) {
+ acpi_ev_gpe_detect(acpi_gbl_gpe_xrupt_list_head);
+ }
return_ACPI_STATUS(status);
}
@@ -120,6 +128,7 @@ acpi_status acpi_enable_gpe(acpi_handle gpe_device, u32 gpe_number)
acpi_status status = AE_BAD_PARAMETER;
struct acpi_gpe_event_info *gpe_event_info;
acpi_cpu_flags flags;
+ u8 poll_gpes = FALSE;
ACPI_FUNCTION_TRACE(acpi_enable_gpe);
@@ -135,12 +144,25 @@ acpi_status acpi_enable_gpe(acpi_handle gpe_device, u32 gpe_number)
if (ACPI_GPE_DISPATCH_TYPE(gpe_event_info->flags) !=
ACPI_GPE_DISPATCH_NONE) {
status = acpi_ev_add_gpe_reference(gpe_event_info);
+ if (ACPI_SUCCESS(status) &&
+ gpe_event_info->runtime_count == 1) {
+ poll_gpes = TRUE;
+ }
} else {
status = AE_NO_HANDLER;
}
}
acpi_os_release_lock(acpi_gbl_gpe_lock, flags);
+
+ /*
+ * Poll GPEs to handle already triggered events.
+ * It is not sufficient to trigger edge-triggered GPE with specific
+ * GPE chips, software need to poll once after enabling.
+ */
+ if (poll_gpes && acpi_gbl_all_gpes_initialized) {
+ acpi_ev_gpe_detect(acpi_gbl_gpe_xrupt_list_head);
+ }
return_ACPI_STATUS(status);
}
ACPI_EXPORT_SYMBOL(acpi_enable_gpe)
--
2.7.4