| Message ID | 1393429360-4344-1-git-send-email-clancy.kieran@gmail.com (mailing list archive) |
|---|---|
| State | Changes Requested, archived |
| Headers | show |
On Thursday, February 27, 2014 02:12:40 AM Kieran Clancy wrote: > A number of Samsung notebooks (530Uxx/535Uxx/540Uxx/550Pxx/900Xxx/etc) > continue to log events during sleep (lid open/close, AC plug/unplug, > battery level change), which accumulate in the EC until a buffer fills. > After the buffer is full (tests suggest it holds 8 events), GPEs stop > being triggered for new events. This state persists on wake or even on > power cycle, and prevents new events from being registered until the EC > is manually polled. > > This is the root cause of a number of bugs, including AC not being > detected properly, lid close not triggering suspend, and low ambient > light not triggering the keyboard backlight. The bug also seemed to be > responsible for performance issues on at least one user's machine. > > Juan Manuel Cabo found the cause of bug and the workaround of polling > the EC manually on wake. > > This patch: > - Adds a function acpi_ec_clear() which polls the EC, at most > ACPI_EC_CLEAR_MAX (currently 20) times. A warning is logged if this > limit is reached. > - Adds a flag EC_FLAGS_CLEAR_ON_RESUME which is set to 1 if the DMI > system vendor is Samsung. This check could be replaced by several more > specific DMI vendor/product pairs, but it's likely that the bug > affects more Samsung products than just the five series mentioned > above. Further, it should not be harmful to run acpi_ec_clear() on > systems without the bug; it will return immediately after finding no > data waiting. > - Runs acpi_ec_clear() on initialisation (boot), from acpi_ec_add() > - Runs acpi_ec_clear() on wake, from acpi_ec_unblock_transactions() > > References: https://bugzilla.kernel.org/show_bug.cgi?id=44161 > References: https://bugzilla.kernel.org/show_bug.cgi?id=45461 > References: https://bugzilla.kernel.org/show_bug.cgi?id=57271 > Signed-off-by: Kieran Clancy <clancy.kieran@gmail.com> > Signed-off-by: Lan Tianyu <tianyu.lan@intel.com> > Signed-off-by: Juan Manuel Cabo <juanmanuel.cabo@gmail.com> > Signed-off-by: Dennis Jansen <dennis.jansen@web.de> There are too many sign-offs under this patch. I suppose some of them should be Acked-by or Reviewed-by. Are you the author? > Tested-by: Maurizio D'Addona <mauritiusdadd@gmail.com> > Tested-by: San Zamoyski <san@plusnet.pl> > --- > drivers/acpi/ec.c | 61 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 61 insertions(+) > > diff --git a/drivers/acpi/ec.c b/drivers/acpi/ec.c > index 959d41a..f437d9a 100644 > --- a/drivers/acpi/ec.c > +++ b/drivers/acpi/ec.c > @@ -67,6 +67,8 @@ enum ec_command { > #define ACPI_EC_DELAY 500 /* Wait 500ms max. during EC ops */ > #define ACPI_EC_UDELAY_GLK 1000 /* Wait 1ms max. to get global lock */ > #define ACPI_EC_MSI_UDELAY 550 /* Wait 550us for MSI EC */ > +#define ACPI_EC_CLEAR_MAX 20 /* Maximum number of events to query > + * when trying to clear the EC */ > > 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 */ > > /* -------------------------------------------------------------------------- > 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 */ > + 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}, > {}, > }; > >
On Thu, Feb 27, 2014 at 10:15 AM, Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > On Thursday, February 27, 2014 02:12:40 AM Kieran Clancy wrote: >> Signed-off-by: Kieran Clancy <clancy.kieran@gmail.com> >> Signed-off-by: Lan Tianyu <tianyu.lan@intel.com> >> Signed-off-by: Juan Manuel Cabo <juanmanuel.cabo@gmail.com> >> Signed-off-by: Dennis Jansen <dennis.jansen@web.de> > > There are too many sign-offs under this patch. I suppose some of them > should be Acked-by or Reviewed-by. > > Are you the author? Sorry about that, I wasn't sure of the best way to acknowledge the people involved. I am the primary author, but I based the loop which calls acpi_ec_query_unlocked() on a patch by Lan Tianyu. Juan provided the initial idea (userspace workaround), and Dennis and Lan Tianyu reviewed and suggested some changes to the code. What would you usually do this kind of situation? I suppose I could put something such as: Signed-off-by: Kieran Clancy <clancy.kieran@gmail.com> Signed-off-by: Lan Tianyu <tianyu.lan@intel.com> Suggested-by: Juan Manuel Cabo <juanmanuel.cabo@gmail.com> Reviewed-by: Lan Tianyu <tianyu.lan@intel.com> Reviewed-by: Dennis Jansen <dennis.jansen@web.de> Would this be preferable? Thanks, Kieran Clancy -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Feb 27, 2014 at 11:15 AM, Kieran Clancy <clancy.kieran@gmail.com> wrote: > > I am the primary author, but I based the loop which calls > acpi_ec_query_unlocked() on a patch by Lan Tianyu. Juan provided the > initial idea (userspace workaround), and Dennis and Lan Tianyu > reviewed and suggested some changes to the code. Oh, and I should add that Juan and Dennis both helped to test the patch extensively. Kieran -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Feb 27, 2014 at 11:46 AM, Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > On Thursday, February 27, 2014 11:29:53 AM Kieran Clancy wrote: >> >> Oh, and I should add that Juan and Dennis both helped to test the >> patch extensively. > > You can use Tested-by or Reported-and-tested-by for that. Is it okay to have someone listed as both 'Reviewed-by' and 'Tested-by', or both 'Suggested-by' and 'Tested-by'? Thank you for your help. Once I fix up these things, I will send a patch v2. Kieran. -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thursday, February 27, 2014 11:15:05 AM Kieran Clancy wrote: > On Thu, Feb 27, 2014 at 10:15 AM, Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > > On Thursday, February 27, 2014 02:12:40 AM Kieran Clancy wrote: > >> Signed-off-by: Kieran Clancy <clancy.kieran@gmail.com> > >> Signed-off-by: Lan Tianyu <tianyu.lan@intel.com> > >> Signed-off-by: Juan Manuel Cabo <juanmanuel.cabo@gmail.com> > >> Signed-off-by: Dennis Jansen <dennis.jansen@web.de> > > > > There are too many sign-offs under this patch. I suppose some of them > > should be Acked-by or Reviewed-by. > > > > Are you the author? > > Sorry about that, I wasn't sure of the best way to acknowledge the > people involved. > > I am the primary author, but I based the loop which calls > acpi_ec_query_unlocked() on a patch by Lan Tianyu. Juan provided the > initial idea (userspace workaround), and Dennis and Lan Tianyu > reviewed and suggested some changes to the code. > > What would you usually do this kind of situation? I usually add acknowledgements to the changelog without using any tags, like "This patch is based on previous work by <somebody>." And then you can give a link to that work in References:. > I suppose I could put something such as: > > Signed-off-by: Kieran Clancy <clancy.kieran@gmail.com> > Signed-off-by: Lan Tianyu <tianyu.lan@intel.com> > Suggested-by: Juan Manuel Cabo <juanmanuel.cabo@gmail.com> > Reviewed-by: Lan Tianyu <tianyu.lan@intel.com> > Reviewed-by: Dennis Jansen <dennis.jansen@web.de> > > Would this be preferable? Yes, that'd be fine by me. Thanks!
On Thursday, February 27, 2014 11:29:53 AM Kieran Clancy wrote: > On Thu, Feb 27, 2014 at 11:15 AM, Kieran Clancy <clancy.kieran@gmail.com> wrote: > > > > I am the primary author, but I based the loop which calls > > acpi_ec_query_unlocked() on a patch by Lan Tianyu. Juan provided the > > initial idea (userspace workaround), and Dennis and Lan Tianyu > > reviewed and suggested some changes to the code. > > Oh, and I should add that Juan and Dennis both helped to test the > patch extensively. You can use Tested-by or Reported-and-tested-by for that.
Kieran Clancy wrote: > A number of Samsung notebooks (530Uxx/535Uxx/540Uxx/550Pxx/900Xxx/etc) > continue to log events during sleep (lid open/close, AC plug/unplug, > battery level change), which accumulate in the EC until a buffer fills. > After the buffer is full (tests suggest it holds 8 events), GPEs stop > being triggered for new events. This state persists on wake or even on > power cycle, and prevents new events from being registered until the EC > is manually polled. > > that's really nasty EC firmware! > This is the root cause of a number of bugs, including AC not being > detected properly, lid close not triggering suspend, and low ambient > light not triggering the keyboard backlight. The bug also seemed to be > responsible for performance issues on at least one user's machine. > > Juan Manuel Cabo found the cause of bug and the workaround of polling > the EC manually on wake. > > This patch: > - Adds a function acpi_ec_clear() which polls the EC, at most > ACPI_EC_CLEAR_MAX (currently 20) times. A warning is logged if this > limit is reached. > - Adds a flag EC_FLAGS_CLEAR_ON_RESUME which is set to 1 if the DMI > system vendor is Samsung. This check could be replaced by several more > specific DMI vendor/product pairs, but it's likely that the bug > affects more Samsung products than just the five series mentioned > above. Further, it should not be harmful to run acpi_ec_clear() on > systems without the bug; it will return immediately after finding no > data waiting. > - Runs acpi_ec_clear() on initialisation (boot), from acpi_ec_add() > - Runs acpi_ec_clear() on wake, from acpi_ec_unblock_transactions() > > References: https://bugzilla.kernel.org/show_bug.cgi?id=44161 > References: https://bugzilla.kernel.org/show_bug.cgi?id=45461 > References: https://bugzilla.kernel.org/show_bug.cgi?id=57271 > Signed-off-by: Kieran Clancy<clancy.kieran@gmail.com> > Signed-off-by: Lan Tianyu<tianyu.lan@intel.com> > Signed-off-by: Juan Manuel Cabo<juanmanuel.cabo@gmail.com> > Signed-off-by: Dennis Jansen<dennis.jansen@web.de> > Tested-by: Maurizio D'Addona<mauritiusdadd@gmail.com> > Tested-by: San Zamoyski<san@plusnet.pl> > --- > drivers/acpi/ec.c | 61 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 61 insertions(+) > > diff --git a/drivers/acpi/ec.c b/drivers/acpi/ec.c > index 959d41a..f437d9a 100644 > --- a/drivers/acpi/ec.c > +++ b/drivers/acpi/ec.c > @@ -67,6 +67,8 @@ enum ec_command { > #define ACPI_EC_DELAY 500 /* Wait 500ms max. during EC ops */ > #define ACPI_EC_UDELAY_GLK 1000 /* Wait 1ms max. to get global lock */ > #define ACPI_EC_MSI_UDELAY 550 /* Wait 550us for MSI EC */ > +#define ACPI_EC_CLEAR_MAX 20 /* Maximum number of events to query > + * when trying to clear the EC */ > 20 is enough? the query index is length of a byte. > > 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-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> 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. > 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). 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@gmail.com> >> >> 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-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
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@gmail.com> > > > > >>> 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@vger.kernel.org > 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-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 02/27/2014 12:14 AM, Li Guang wrote: > 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! > We tested each on our different samsung models (intel, amd), and it was 8 across. But you're right, there might be more in the future. I even saw a bug report in ubuntu's launchpad of an HP with a similar sounding problem, ( https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.20/+bug/89860 ) which I have no idea if it was caused by the same issue, but if in the future, the flag ec_clear_on_resume is used to match other DMI's, it might be a good idea to make the max iteration count bigger. The only reason that there is a max iteration count, was to prevent an unexpected case in which an unknown EC never returns 0 after queue emptied. So far it hasn't been the case. Can we count on it?. The loop currently does finish early when there are no more events. I guess changing it 255 or 1000 would be enough, right? Cheers!
Juan Manuel Cabo wrote: > On 02/27/2014 12:14 AM, Li Guang wrote: > >> 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! >> >> > We tested each on our different samsung models (intel, amd), and it > was 8 across. But you're right, there might be more in the future. > > I even saw a bug report in ubuntu's launchpad of an HP with a similar > sounding problem, ( https://bugs.launchpad.net/ubuntu/+source/linux-source-2.6.20/+bug/89860 ) > which I have no idea if it was caused by the same issue, but if in the future, > the flag ec_clear_on_resume is used to match other DMI's, it might > be a good idea to make the max iteration count bigger. > > The only reason that there is a max iteration count, was to prevent > an unexpected case in which an unknown EC never returns 0 after > queue emptied. So far it hasn't been the case. Can we count on it?. > The loop currently does finish early when there are no more events. > > I guess changing it 255 or 1000 would be enough, right? > > can't imagine 1K bytes be dissipated on Q event, EC's ram is usually expensive, I think 255 is really enough. :-) Thanks! -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Feb 27, 2014 at 12:29 PM, Li Guang <lig.fnst@cn.fujitsu.com> wrote: >> +#define ACPI_EC_CLEAR_MAX 20 /* Maximum number of events to >> query >> + * when trying to clear the EC */ >> > > > 20 is enough? > the query index is length of a byte. On my machine, 8 seems to be enough, so 20 seems to be a conservative maximum. Just reading your other email, maybe we should set this to 32? or 40? 100? If it's not enough, hopefully anyone seeing bugs will notice the warning "maximum of X stale EC events cleared". Here's what happens if I plug/replug the AC lots of times (more than 8) during suspend: [ 8807.019800] ACPI : EC: ---> status = 0x29 [ 8807.019804] ACPI : EC: ---> data = 0x66 [ 8807.020790] ACPI : EC: ---> status = 0x29 [ 8807.020793] ACPI : EC: ---> data = 0x66 [ 8807.021793] ACPI : EC: ---> status = 0x29 [ 8807.021798] ACPI : EC: ---> data = 0x66 [ 8807.022831] ACPI : EC: ---> status = 0x29 [ 8807.022834] ACPI : EC: ---> data = 0x66 [ 8807.023788] ACPI : EC: ---> status = 0x29 [ 8807.023792] ACPI : EC: ---> data = 0x66 [ 8807.024787] ACPI : EC: ---> status = 0x29 [ 8807.024791] ACPI : EC: ---> data = 0x66 [ 8807.025787] ACPI : EC: ---> status = 0x29 [ 8807.025790] ACPI : EC: ---> data = 0x66 [ 8807.026787] ACPI : EC: ---> status = 0x29 [ 8807.026790] ACPI : EC: ---> data = 0x66 [ 8807.027786] ACPI : EC: ---> status = 0x09 [ 8807.027790] ACPI : EC: ---> data = 0x00 [ 8807.027792] ACPI : EC: 8 stale EC events cleared Note that most of these have SCI_EVT set, but the OS is not notified according to ACPI specs (seemingly because these events happened during sleep). The _Q66 method in my DSDT, is: P8XH (Zero, 0x66) If (LEqual (B1EX, One)) { Notify (BAT1, 0x80) } So, basically, this is supposed to notify that the battery (BAT1 = PNP0C0A) has changed state, but they are stale events so we don't run the handlers. >> +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 :-) In my mind this is referring to the function name (acpi_ec_)clear. Perhaps we could just make the connection more explicit in the comment: /* needs acpi_ec_clear() on boot/resume */ Not sure if this is better? >> + /* Some hardware may need the EC to be cleared before use */ > > description is implicit, should specify what we clear is Q event, not EC. Are Q events the only thing we can get from the EC data port? I've read the relevant parts of the ACPI spec and I can't say I am 100% sure. Thank you for your advice, Kieran. -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Kieran Clancy wrote: > On Thu, Feb 27, 2014 at 12:29 PM, Li Guang<lig.fnst@cn.fujitsu.com> wrote: > >>> +#define ACPI_EC_CLEAR_MAX 20 /* Maximum number of events to >>> query >>> + * when trying to clear the EC */ >>> >>> >> >> 20 is enough? >> the query index is length of a byte. >> > On my machine, 8 seems to be enough, so 20 seems to be a conservative > maximum. Just reading your other email, maybe we should set this to > 32? or 40? 100? > > If it's not enough, hopefully anyone seeing bugs will notice the > warning "maximum of X stale EC events cleared". > > Here's what happens if I plug/replug the AC lots of times (more than > 8) during suspend: > > [ 8807.019800] ACPI : EC: ---> status = 0x29 > [ 8807.019804] ACPI : EC: ---> data = 0x66 > [ 8807.020790] ACPI : EC: ---> status = 0x29 > [ 8807.020793] ACPI : EC: ---> data = 0x66 > [ 8807.021793] ACPI : EC: ---> status = 0x29 > [ 8807.021798] ACPI : EC: ---> data = 0x66 > [ 8807.022831] ACPI : EC: ---> status = 0x29 > [ 8807.022834] ACPI : EC: ---> data = 0x66 > [ 8807.023788] ACPI : EC: ---> status = 0x29 > [ 8807.023792] ACPI : EC: ---> data = 0x66 > [ 8807.024787] ACPI : EC: ---> status = 0x29 > [ 8807.024791] ACPI : EC: ---> data = 0x66 > [ 8807.025787] ACPI : EC: ---> status = 0x29 > [ 8807.025790] ACPI : EC: ---> data = 0x66 > [ 8807.026787] ACPI : EC: ---> status = 0x29 > [ 8807.026790] ACPI : EC: ---> data = 0x66 > [ 8807.027786] ACPI : EC: ---> status = 0x09 > [ 8807.027790] ACPI : EC: ---> data = 0x00 > [ 8807.027792] ACPI : EC: 8 stale EC events cleared > > Note that most of these have SCI_EVT set, but the OS is not notified > according to ACPI specs (seemingly because these events happened > during sleep). > > The _Q66 method in my DSDT, is: > > P8XH (Zero, 0x66) > If (LEqual (B1EX, One)) > { > Notify (BAT1, 0x80) > } > > So, basically, this is supposed to notify that the battery (BAT1 = > PNP0C0A) has changed state, but they are stale events so we don't run > the handlers. > > >>> +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 :-) >> > In my mind this is referring to the function name (acpi_ec_)clear. > Perhaps we could just make the connection more explicit in the > comment: > > /* needs acpi_ec_clear() on boot/resume */ > > Not sure if this is better? > > >>> + /* Some hardware may need the EC to be cleared before use */ >>> >> description is implicit, should specify what we clear is Q event, not EC. >> > Are Q events the only thing we can get from the EC data port? I've > read the relevant parts of the ACPI spec and I can't say I am 100% > sure. > > I guess you want to clear Q events here, EC usually has ACPI space to be read by cmd 80. Thanks! -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hello Kieran!!
Dennis has an AMD laptop, NP535U3C-A04DE, and found it to be 9
(which I guess was an off by one in his modification
of the userspace workaround and was really 8):
https://bugzilla.kernel.org/show_bug.cgi?id=44161#c146
I found it to be 8 for me, both in the patch and in my new userspace workaround
(which doesn't ask for the same event twice now that it pauses for 1ms,
as the kernel does too):
https://bugzilla.kernel.org/show_bug.cgi?id=44161#c162
ANd I think I saw someone else with 8 too but don't remember where.
BUT, there might be more laptops existing or future, samsung or other,
that might queue more than 8. Li Guang (lig.fnst@cn.fujitsu.com)
said that he saw EC firmware's with 32, and that true maximum is 255.
So in my opinion it's best to put max. iters. at 255.
Cheers!
--jm
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thursday, February 27, 2014 11:39:42 AM Kieran Clancy wrote: > On Thu, Feb 27, 2014 at 11:46 AM, Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > > On Thursday, February 27, 2014 11:29:53 AM Kieran Clancy wrote: > >> > >> Oh, and I should add that Juan and Dennis both helped to test the > >> patch extensively. > > > > You can use Tested-by or Reported-and-tested-by for that. > > Is it okay to have someone listed as both 'Reviewed-by' and > 'Tested-by', or both 'Suggested-by' and 'Tested-by'? Yes, these are different things essentially and each of them requires time and effort on its own. > Thank you for your help. Once I fix up these things, I will send a patch v2. Thanks!
diff --git a/drivers/acpi/ec.c b/drivers/acpi/ec.c index 959d41a..f437d9a 100644 --- a/drivers/acpi/ec.c +++ b/drivers/acpi/ec.c @@ -67,6 +67,8 @@ enum ec_command { #define ACPI_EC_DELAY 500 /* Wait 500ms max. during EC ops */ #define ACPI_EC_UDELAY_GLK 1000 /* Wait 1ms max. to get global lock */ #define ACPI_EC_MSI_UDELAY 550 /* Wait 550us for MSI EC */ +#define ACPI_EC_CLEAR_MAX 20 /* Maximum number of events to query + * when trying to clear the EC */ 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 */ /* -------------------------------------------------------------------------- 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 */ + 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}, {}, };