Re: [PATCH] ACPI / EC: Clear stale EC events on Samsung systems

From: Li Guang
Date: Wed Feb 26 2014 - 22:14:59 EST

Juan Manuel Cabo wrote:

that's really nasty EC firmware!
Yes!
And this bug has been unsolved for about two years.

20 is enough?
the query index is length of a byte.

According to our humble tests, 8 is the maximum number of
accumulated events. For instance, if the one plugs or unplugs the PSU
16 times, (or battery % changes 16 times) during S3 sleep, then the
EC returns no more than 8 events when polled by this patch or by
my userspace util.
And having reached 8 events, it won't produce its GPE until queried.

that surely indicts their EC firmware only queued 8 events,

the query index is length of a byte.
There is no query index, unless you refer to something else (I'm sorry if its
something that escapes me).

oh, sorry, I'm referring internal EC firmware code
for Q event queuing, not ACPI SPEC, ;-)
for machine you tested, 8 is the queue size,
but for some unknown also nasty EC firmwares(let's suppose it exists),
it may queue more Q events.
and I saw several firmwares queued 32 events by default,
then, let's say, they be used for some samsung products,
and also they also forgot to deal with sleep/resume state,
then, we'll also leave stale Q event there.

Thanks!

For us, a query is just: send 0x84 through EC CMD port, and read status
from CMD port and event type from EC DATA port. This is done with
the usual ec.c functions that would handle a query after a GPE interrupt,
but using them instead to poll (not GPE initiated) at awake. The EC would
then return status without 0x20 mask and 'event type'==0 when no more left.

--
Juan Manuel Cabo<juanmanuel.cabo@xxxxxxxxx>



enum {
EC_FLAGS_QUERY_PENDING, /* Query is pending */
@@ -116,6 +118,7 @@ EXPORT_SYMBOL(first_ec);
static int EC_FLAGS_MSI; /* Out-of-spec MSI controller */
static int EC_FLAGS_VALIDATE_ECDT; /* ASUStec ECDTs need to be validated */
static int EC_FLAGS_SKIP_DSDT_SCAN; /* Not all BIOS survive early DSDT scan */
+static int EC_FLAGS_CLEAR_ON_RESUME; /* EC should be polled on boot/resume */

seems name is implicit, what about EC_FLAGS_QEVENT_CLR_ON_RESUME?
seems too long :-)

/* --------------------------------------------------------------------------
Transaction Management
@@ -440,6 +443,26 @@ acpi_handle ec_get_handle(void)

EXPORT_SYMBOL(ec_get_handle);

+static int acpi_ec_query_unlocked(struct acpi_ec *ec, u8 *data);
+
+/* run with locked ec mutex */
+static void acpi_ec_clear(struct acpi_ec *ec)
+{
+ int i, status;
+ u8 value = 0;
+
+ for (i = 0; i< ACPI_EC_CLEAR_MAX; i++) {
+ status = acpi_ec_query_unlocked(ec,&value);
+ if (status || !value)
+ break;
+ }
+
+ if (i == ACPI_EC_CLEAR_MAX)
+ pr_warn("Warning: Maximum of %d stale EC events cleared\n", i);
+ else
+ pr_info("%d stale EC events cleared\n", i);
+}
+
void acpi_ec_block_transactions(void)
{
struct acpi_ec *ec = first_ec;
@@ -463,6 +486,10 @@ void acpi_ec_unblock_transactions(void)
mutex_lock(&ec->mutex);
/* Allow transactions to be carried out again */
clear_bit(EC_FLAGS_BLOCKED,&ec->flags);
+
+ if (EC_FLAGS_CLEAR_ON_RESUME)
+ acpi_ec_clear(ec);
+
mutex_unlock(&ec->mutex);
}

@@ -821,6 +848,13 @@ static int acpi_ec_add(struct acpi_device *device)

/* EC is fully operational, allow queries */
clear_bit(EC_FLAGS_QUERY_PENDING,&ec->flags);
+
+ /* Some hardware may need the EC to be cleared before use */

description is implicit, should specify what we clear is Q event, not EC.

Thanks!
Li Guang

+ if (EC_FLAGS_CLEAR_ON_RESUME) {
+ mutex_lock(&ec->mutex);
+ acpi_ec_clear(ec);
+ mutex_unlock(&ec->mutex);
+ }
return ret;
}

@@ -922,6 +956,30 @@ static int ec_enlarge_storm_threshold(const struct dmi_system_id *id)
return 0;
}

+/*
+ * On some hardware it is necessary to clear events accumulated by the EC during
+ * sleep. These ECs stop reporting GPEs until they are manually polled, if too
+ * many events are accumulated. (e.g. Samsung Series 5/9 notebooks)
+ *
+ * https://bugzilla.kernel.org/show_bug.cgi?id=44161
+ *
+ * Ideally, the EC should also be instructed not to accumulate events during
+ * sleep (which Windows seems to do somehow), but the interface to control this
+ * behaviour is not known at this time.
+ *
+ * Models known to be affected are Samsung 530Uxx/535Uxx/540Uxx/550Pxx/900Xxx,
+ * however it is very likely that other Samsung models are affected.
+ *
+ * On systems which don't accumulate EC events during sleep, this extra check
+ * should be harmless.
+ */
+static int ec_clear_on_resume(const struct dmi_system_id *id)
+{
+ pr_debug("Detected system needing EC poll on resume.\n");
+ EC_FLAGS_CLEAR_ON_RESUME = 1;
+ return 0;
+}
+
static struct dmi_system_id ec_dmi_table[] __initdata = {
{
ec_skip_dsdt_scan, "Compal JFL92", {
@@ -965,6 +1023,9 @@ static struct dmi_system_id ec_dmi_table[] __initdata = {
ec_validate_ecdt, "ASUS hardware", {
DMI_MATCH(DMI_SYS_VENDOR, "ASUSTek Computer Inc."),
DMI_MATCH(DMI_PRODUCT_NAME, "L4R"),}, NULL},
+ {
+ ec_clear_on_resume, "Samsung hardware", {
+ DMI_MATCH(DMI_SYS_VENDOR, "SAMSUNG ELECTRONICS CO., LTD.")}, NULL},
{},
};



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/