Better handling of bad xfers/interrupt delays in psmouse
Hi,
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
Additionally, psmouse will also request resend if KBC indicates parity or timeout errors.
The patch should apply to 2.6.11-rc2+. A version of this patch for 2.6.10 is available here:
http://www.geocities.com/dt_or/input/2_6_10/
Signature · 119 lines
Dmitry =================================================================== [email protected], 2005-02-05 14:33:05-05:00, [email protected] Input: psmouse - better handle bad transfers and interrupt delays by requesting resend of the last packet so we don't need to guess if received byte is remainder of last packet or start of a new one. Signed-off-by: Dmitry Torokhov <[email protected]> psmouse-base.c | 48 +++++++++++++++++++++++++++++++++++++++++++----- psmouse.h | 2 ++ 2 files changed, 45 insertions(+), 5 deletions(-) =================================================================== diff -Nru a/drivers/input/mouse/psmouse-base.c b/drivers/input/mouse/psmouse-base.c --- a/drivers/input/mouse/psmouse-base.c 2005-02-05 14:37:51 -05:00 +++ b/drivers/input/mouse/psmouse-base.c 2005-02-05 14:37:51 -05:00 @@ -134,6 +134,42 @@ return PSMOUSE_FULL_PACKET; } +static void psmouse_handle_bad_xfer(struct psmouse *psmouse, unsigned int flags) +{ + if (psmouse->state == PSMOUSE_ACTIVATED) + printk(KERN_WARNING "psmouse.c: bad data from KBC -%s%s\n", + flags & SERIO_TIMEOUT ? " timeout" : "", + flags & SERIO_PARITY ? " bad parity" : ""); + + if (!psmouse->resend) { + /* + * This is first error. Try to request resend but not if we got + * timeout and we are in initialize phase - there most likely no + * mouse at all. + */ + if (psmouse->state > PSMOUSE_INITIALIZING || (~flags & SERIO_TIMEOUT)) { + if (serio_write(psmouse->ps2dev.serio, PSMOUSE_CMD_RESEND) == 0) { + psmouse->resend = 1; + psmouse->pktcnt = 0; + } + } + } else { + /* + * This is second error in a row. If mouse was itialized - attempt + * to rconnect, otherwise just signal failure. + */ + psmouse->resend = 0; + if (psmouse->state > PSMOUSE_INITIALIZING) { + psmouse->state = PSMOUSE_IGNORE; + printk(KERN_NOTICE "psmouse.c: issuing reconnect request\n"); + serio_reconnect(psmouse->ps2dev.serio); + } + } + + if (!psmouse->resend) + ps2_cmd_aborted(&psmouse->ps2dev); +} + /* * psmouse_interrupt() handles incoming characters, either gathering them into * packets or passing them to the command routine as command output. @@ -149,11 +185,7 @@ goto out; if (flags & (SERIO_PARITY|SERIO_TIMEOUT)) { - if (psmouse->state == PSMOUSE_ACTIVATED) - printk(KERN_WARNING "psmouse.c: bad data from KBC -%s%s\n", - flags & SERIO_TIMEOUT ? " timeout" : "", - flags & SERIO_PARITY ? " bad parity" : ""); - ps2_cmd_aborted(&psmouse->ps2dev); + psmouse_handle_bad_xfer(psmouse, flags); goto out; } @@ -168,11 +200,17 @@ if (psmouse->state == PSMOUSE_INITIALIZING) goto out; + psmouse->resend = 0; + if (psmouse->state == PSMOUSE_ACTIVATED && psmouse->pktcnt && time_after(jiffies, psmouse->last + HZ/2)) { printk(KERN_WARNING "psmouse.c: %s at %s lost synchronization, throwing %d bytes away.\n", psmouse->name, psmouse->phys, psmouse->pktcnt); psmouse->pktcnt = 0; + if (serio_write(serio, PSMOUSE_CMD_RESEND) == 0) { + psmouse->resend = 1; + goto out; + } } psmouse->last = jiffies; diff -Nru a/drivers/input/mouse/psmouse.h b/drivers/input/mouse/psmouse.h --- a/drivers/input/mouse/psmouse.h 2005-02-05 14:37:51 -05:00 +++ b/drivers/input/mouse/psmouse.h 2005-02-05 14:37:51 -05:00 @@ -13,6 +13,7 @@ #define PSMOUSE_CMD_ENABLE 0x00f4 #define PSMOUSE_CMD_DISABLE 0x00f5 #define PSMOUSE_CMD_RESET_DIS 0x00f6 +#define PSMOUSE_CMD_RESEND 0x00fe #define PSMOUSE_CMD_RESET_BAT 0x02ff #define PSMOUSE_RET_BAT 0xaa @@ -48,6 +49,7 @@ unsigned long last; unsigned long out_of_sync; enum psmouse_state state; + unsigned int resend; char devname[64]; char phys[32];
Quoted 6 lines · Dmitry Torokhov
Hi,
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
Have you actually tested the mouse is really sending the whole packet? I'd suspect it could just resend the last byte. :I Maybe using the GET_PACKET command would be more useful in this case.
Also, we could use GET_PACKET to check the mode of the mouse, when we get several successive lost bytes.
Quoted 131 lines
Additionally, psmouse will also request resend if KBC indicates parity or timeout errors.
The patch should apply to 2.6.11-rc2+. A version of this patch for 2.6.10 is available here:
http://www.geocities.com/dt_or/input/2_6_10/Signature · 25 lines
Dmitry =================================================================== [email protected], 2005-02-05 14:33:05-05:00, [email protected] Input: psmouse - better handle bad transfers and interrupt delays by requesting resend of the last packet so we don't need to guess if received byte is remainder of last packet or start of a new one. Signed-off-by: Dmitry Torokhov <[email protected]> psmouse-base.c | 48 +++++++++++++++++++++++++++++++++++++++++++----- psmouse.h | 2 ++ 2 files changed, 45 insertions(+), 5 deletions(-) =================================================================== diff -Nru a/drivers/input/mouse/psmouse-base.c b/drivers/input/mouse/psmouse-base.cdrivers/input/mouse/psmouse-base.c +43
-5@@ -134,6 +134,42 @@return PSMOUSE_FULL_PACKET ; } +static void psmouse_handle_bad_xfer (struct psmouse *psmouse,unsigned int flags) +{ +if (psmouse->state ==PSMOUSE_ACTIVATED ) +printk (KERN_WARNING "psmouse.c: bad data from KBC -%s%s\n" , + flags &SERIO_TIMEOUT ?" timeout" :"" , + flags &SERIO_PARITY ?" bad parity" :"" ); + +if (!psmouse->resend) { +/* +* This is first error. Try to request resend but not if we got +* timeout and we are in initialize phase - there most likely no +* mouse at all. +*/ +if (psmouse->state >PSMOUSE_INITIALIZING || (~flags &SERIO_TIMEOUT )) { +if (serio_write (psmouse->ps2dev.serio,PSMOUSE_CMD_RESEND ) ==0 ) { + psmouse->resend =1 ; + psmouse->pktcnt =0 ; + } + } + }else { +/* +* This is second error in a row. If mouse was itialized - attempt +* to rconnect, otherwise just signal failure. +*/ + psmouse->resend =0 ; +if (psmouse->state >PSMOUSE_INITIALIZING ) { + psmouse->state =PSMOUSE_IGNORE ; +printk (KERN_NOTICE "psmouse.c: issuing reconnect request\n" ); +serio_reconnect (psmouse->ps2dev.serio); + } + } + +if (!psmouse->resend) +ps2_cmd_aborted (&psmouse->ps2dev); +} +/* * psmouse_interrupt() handles incoming characters, either gathering them into * packets or passing them to the command routine as command output. @@ -149,11 +185,7 @@ goto out;if (flags & (SERIO_PARITY |SERIO_TIMEOUT )) { -if (psmouse->state ==PSMOUSE_ACTIVATED ) -printk (KERN_WARNING "psmouse.c: bad data from KBC -%s%s\n" , - flags &SERIO_TIMEOUT ?" timeout" :"" , - flags &SERIO_PARITY ?" bad parity" :"" ); -ps2_cmd_aborted (&psmouse->ps2dev); +psmouse_handle_bad_xfer (psmouse, flags); goto out; }@@ -168,11 +200,17 @@if (psmouse->state ==PSMOUSE_INITIALIZING ) goto out; + psmouse->resend =0 ; +if (psmouse->state ==PSMOUSE_ACTIVATED && psmouse->pktcnt &&time_after (jiffies, psmouse->last +HZ /2 )) {printk (KERN_WARNING "psmouse.c: %s at %s lost synchronization, throwing %d bytes away.\n" , psmouse->name, psmouse->phys, psmouse->pktcnt); psmouse->pktcnt =0 ; +if (serio_write (serio,PSMOUSE_CMD_RESEND ) ==0 ) { + psmouse->resend =1 ; + goto out; + } } psmouse->last = jiffies;diff -Nru
a/drivers/input/mouse/psmouse.hb/drivers/input/mouse/psmouse.hdrivers/input/mouse/psmouse.h +2
-0@@ -13,6 +13,7 @@#define PSMOUSE_CMD_ENABLE 0x00f4#define PSMOUSE_CMD_DISABLE 0x00f5#define PSMOUSE_CMD_RESET_DIS 0x00f6 +#define PSMOUSE_CMD_RESEND 0x00fe#define PSMOUSE_CMD_RESET_BAT 0x02ff#define PSMOUSE_RET_BAT 0xaa@@ -48,6 +49,7 @@unsigned long last;unsigned long out_of_sync;enum psmouse_state state; +unsigned int resend;char devname[64 ];char phys[32 ];
Signature · 2 lines
Vojtech Pavlik SuSE Labs, SuSE CR
Quoted 10 lines · Vojtech Pavlik
Quoted 6 lines · Dmitry Torokhov
Hi,
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
Have you actually tested the mouse is really sending the whole packet? I'd suspect it could just resend the last byte. :I Maybe using the
Well, I did test and my touchpad behaved properly. But then I tried 2 external mice and they are both sending ACK (and they should not) and then the last byte only. So I guess we'll have to scrap using 0xfe idea...
Quoted 2 lines
GET_PACKETcommand would be more useful in this case.
Are you talking about 0xeb? We could also try sending "set stream" mode as a sync marker...
Signature · 1 line
Dmitry
Quoted 16 lines · Dmitry Torokhov
Quoted 10 lines · Vojtech Pavlik
Quoted 6 lines · Dmitry Torokhov
Hi,
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
Have you actually tested the mouse is really sending the whole packet? I'd suspect it could just resend the last byte. :I Maybe using the
Well, I did test and my touchpad behaved properly. But then I tried 2 external mice and they are both sending ACK (and they should not) and then the last byte only. So I guess we'll have to scrap using
0xfeidea...
Quoted 6 lines
Quoted 2 lines
GET_PACKETcommand would be more useful in this case.Are you talking about
0xeb? We could also try sending "set stream" mode as a sync marker...
Ok, here is the patch using PSMOUSE_CMD_POLL. Seems to work fine with 2 external mice that I have and my touchpad in PS/2 compatibility mode.
Unfortunately POLL command kicks Synaptics out of absolute mode so I disabled all time-based sync checks for Synaptics altogether. This should be OK since Synaptics have pretty strict protocol rules and usually can resync on their own. I wonder what POLL does to ALPS?
Again, 2.6.10 version can be found here:
http://www.geocities.com/dt_or/input/2_6_10/
Comments/testing is appreciated.
Signature · 160 lines
Dmitry =================================================================== [email protected], 2005-02-06 01:58:47-05:00, [email protected] Input: psmouse - better handle bad transfers and interrupt delays by using POLL command to synchtonize on beginning of a packet so we don't need to guess whether received byte is remainder of the last packet or beginning of a new one. Signed-off-by: Dmitry Torokhov <[email protected]> psmouse-base.c | 83 +++++++++++++++++++++++++++++++++++++++++++++++++-------- psmouse.h | 1 2 files changed, 73 insertions(+), 11 deletions(-) =================================================================== diff -Nru a/drivers/input/mouse/psmouse-base.c b/drivers/input/mouse/psmouse-base.c --- a/drivers/input/mouse/psmouse-base.c 2005-02-06 02:15:18 -05:00 +++ b/drivers/input/mouse/psmouse-base.c 2005-02-06 02:15:18 -05:00 @@ -134,6 +134,49 @@ return PSMOUSE_FULL_PACKET; } +static void psmouse_schedule_reconnect(struct psmouse *psmouse) +{ + psmouse->state = PSMOUSE_IGNORE; + printk(KERN_NOTICE "psmouse.c: issuing reconnect request\n"); + serio_reconnect(psmouse->ps2dev.serio); +} + +static void psmouse_handle_bad_xfer(struct psmouse *psmouse, unsigned int flags) +{ + if (psmouse->state == PSMOUSE_ACTIVATED) + printk(KERN_WARNING "psmouse.c: bad data from KBC -%s%s\n", + flags & SERIO_TIMEOUT ? " timeout" : "", + flags & SERIO_PARITY ? " bad parity" : ""); + + if (!psmouse->resend) { + /* + * This is first error. Try to request resend but not if we + * got a timeout and we in initialize phase - there most + * likely no mouse at all. Also skip Synaptics as they don't + * like POLL command and pass-through ports since we can't + * use their serio_write in interrupt context. + */ + if ((psmouse->state > PSMOUSE_INITIALIZING || (~flags & SERIO_TIMEOUT)) && + psmouse->type != PSMOUSE_SYNAPTICS && !psmouse->ps2dev.serio->parent) { + if (serio_write(psmouse->ps2dev.serio, PSMOUSE_CMD_POLL & 0xff) == 0) { + psmouse->resend = 1; + psmouse->pktcnt = 0; + } + } + } else { + /* + * This is second error in a row. If mouse was itialized - attempt + * to rconnect, otherwise just signal failure. + */ + psmouse->resend = 0; + if (psmouse->state > PSMOUSE_INITIALIZING) + psmouse_schedule_reconnect(psmouse); + } + + if (!psmouse->resend) + ps2_cmd_aborted(&psmouse->ps2dev); +} + /* * psmouse_interrupt() handles incoming characters, either gathering them into * packets or passing them to the command routine as command output. @@ -149,11 +192,14 @@ goto out; if (flags & (SERIO_PARITY|SERIO_TIMEOUT)) { - if (psmouse->state == PSMOUSE_ACTIVATED) - printk(KERN_WARNING "psmouse.c: bad data from KBC -%s%s\n", - flags & SERIO_TIMEOUT ? " timeout" : "", - flags & SERIO_PARITY ? " bad parity" : ""); - ps2_cmd_aborted(&psmouse->ps2dev); + psmouse_handle_bad_xfer(psmouse, flags); + goto out; + } + + if (unlikely(psmouse->resend)) { + psmouse->resend = 0; + if (data != PSMOUSE_RET_ACK) + psmouse_schedule_reconnect(psmouse); goto out; } @@ -168,11 +214,28 @@ if (psmouse->state == PSMOUSE_INITIALIZING) goto out; + /* + * If there is a long delay between bytes in a packet we suspect + * that we lost synchronization and will try to restore it by + * requesting transmission of a full packet from th? mouse. + * Unfortunately POLL command kicks Synaptics touchpads out of + * absolute mode, so we skip them (but they have good protocol + * checks and usually can re-sync without problems). We also don't + * try to re-sync devices on pass-through ports as their serio_write + * can't be used in interrupt context and they usually in sync as + * long as their parent is. + */ if (psmouse->state == PSMOUSE_ACTIVATED && + psmouse->type != PSMOUSE_SYNAPTICS && + !serio->parent && psmouse->pktcnt && time_after(jiffies, psmouse->last + HZ/2)) { printk(KERN_WARNING "psmouse.c: %s at %s lost synchronization, throwing %d bytes away.\n", psmouse->name, psmouse->phys, psmouse->pktcnt); psmouse->pktcnt = 0; + if (serio_write(serio, PSMOUSE_CMD_POLL & 0xff) == 0) { + psmouse->resend = 1; + goto out; + } } psmouse->last = jiffies; @@ -206,11 +269,9 @@ psmouse->name, psmouse->phys, psmouse->pktcnt); psmouse->pktcnt = 0; - if (++psmouse->out_of_sync == psmouse->resetafter) { - psmouse->state = PSMOUSE_IGNORE; - printk(KERN_NOTICE "psmouse.c: issuing reconnect request\n"); - serio_reconnect(psmouse->ps2dev.serio); - } + if (++psmouse->out_of_sync == psmouse->resetafter) + psmouse_schedule_reconnect(psmouse); + break; case PSMOUSE_FULL_PACKET: @@ -594,7 +655,7 @@ { serio_pause_rx(psmouse->ps2dev.serio); psmouse->state = new_state; - psmouse->pktcnt = psmouse->out_of_sync = 0; + psmouse->pktcnt = psmouse->out_of_sync = psmouse->resend = 0; psmouse->ps2dev.flags = 0; serio_continue_rx(psmouse->ps2dev.serio); } diff -Nru a/drivers/input/mouse/psmouse.h b/drivers/input/mouse/psmouse.h --- a/drivers/input/mouse/psmouse.h 2005-02-06 02:15:18 -05:00 +++ b/drivers/input/mouse/psmouse.h 2005-02-06 02:15:18 -05:00 @@ -48,6 +48,7 @@ unsigned long last; unsigned long out_of_sync; enum psmouse_state state; + unsigned int resend; char devname[64]; char phys[32];
Quoted 12 lines · Dmitry Torokhov
Quoted 10 lines · Vojtech Pavlik
Quoted 6 lines · Dmitry Torokhov
Hi,
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
Have you actually tested the mouse is really sending the whole packet? I'd suspect it could just resend the last byte. :I Maybe using the
Quoted 8 lines
Well, I did test and my touchpad behaved properly. But then I tried 2 external mice and they are both sending ACK (and they should not) and then the last byte only. So I guess we'll have to scrap using
0xfeidea...Quoted 2 lines
GET_PACKETcommand would be more useful in this case.Are you talking about
0xeb?
Yes, sorry for the wrong name, it should've been "CMD_POLL".
Quoted 1 line
We could also try sending "set stream" mode as a sync marker...
That's an interesting idea, too. But I suspect we'll also have to turn the stream mode off first (CMD_DISABLE).
Signature · 2 lines
Vojtech Pavlik SuSE Labs, SuSE CR
Quoted 8 lines · Dmitry Torokhov
Ok, here is the patch using
PSMOUSE_CMD_POLL. Seems to work fine with 2 external mice that I have and my touchpad in PS/2 compatibility mode.Unfortunately POLL command kicks Synaptics out of absolute mode so I disabled all time-based sync checks for Synaptics altogether. This should be OK since Synaptics have pretty strict protocol rules and usually can resync on their own. I wonder what POLL does to ALPS?
Quoted 5 lines
Again, 2.6.10 version can be found here:
http://www.geocities.com/dt_or/input/2_6_10/Comments/testing is appreciated.
Did you check that issuing the POLL command in the middle of a mouse packet does indeed reset the counter of the streaming mode? I'd expect them to be separate, at least on some mice, if POLL is handled as a normal mouse command, and after the three bytes from POLL are sent, the mouse could just continue sending the packet like with any other command.
Signature · 2 lines
Vojtech Pavlik SuSE Labs, SuSE CR
Hi!
Quoted 4 lines
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
+ * This is second error in a row. If mouse was itialized - attempt
+ * to rconnect, otherwise just signal failure.
Tooo many typso in on coment?
Pavel
Signature · 2 lines
People were complaining that M$ turns users into beta-testers... ...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!
Quoted 14 lines · Vojtech Pavlik
Quoted 12 lines · Dmitry Torokhov
Ok, here is the patch using
PSMOUSE_CMD_POLL. Seems to work fine with 2 external mice that I have and my touchpad in PS/2 compatibility mode.Unfortunately POLL command kicks Synaptics out of absolute mode so I disabled all time-based sync checks for Synaptics altogether. This should be OK since Synaptics have pretty strict protocol rules and usually can resync on their own. I wonder what POLL does to ALPS?
Again, 2.6.10 version can be found here:
http://www.geocities.com/dt_or/input/2_6_10/
Quoted 9 lines
Quoted 1 line
Comments/testing is appreciated.
Did you check that issuing the POLL command in the middle of a mouse packet does indeed reset the counter of the streaming mode? I'd expect them to be separate, at least on some mice, if POLL is handled as a normal mouse command, and after the three bytes from POLL are sent, the mouse could just continue sending the packet like with any other command.
I have tested with Synaptics touchpad, a noname explorer mouse and a Belkin USB with USB-PS/2 converted, all of them resumed streaming with the full packet. I used the following patch to force polling:
===== drivers/input/mouse/psmouse-base.c 1.88 vs edited =====
drivers/input/mouse/psmouse-base.c +15 -0
@@ -185,6 +185,7 @@static irqreturn_t psmouse_interrupt (struct serio *serio,unsigned char data,unsigned int flags,struct pt_regs *regs) { +static int bcnt;struct psmouse *psmouse =serio_get_drvdata (serio);psmouse_ret_t rc;
@@ -237,6 +238,20 @@ goto out; } } + +if (psmouse->state ==PSMOUSE_ACTIVATED && + psmouse->type !=PSMOUSE_SYNAPTICS && + !serio->parent && + (++bcnt %502 ) ==0 ) { +printk (KERN_WARNING "psmouse.c: %s at %s - repolling, throwing %d bytes away.\n" , + psmouse->name, psmouse->phys, psmouse->pktcnt); + psmouse->pktcnt =0 ; +if (serio_write (serio,PSMOUSE_CMD_POLL &0xff ) ==0 ) { + psmouse->resend =1 ; + goto out; + } + } + psmouse->last = jiffies; psmouse->packet[psmouse->pktcnt++] = data;
We of course need more testing, I am particularly interested in ALPS reaction to poll command.
Signature · 1 line
Dmitry
Quoted 10 lines · Pavel Machek
Hi!
Quoted 4 lines
The patch below attempts to better handle situation when psmouse interrupt is delayed for more than 0.5 sec by requesting a resend. This will allow properly synchronize with the beginning of the packet as mouse is supposed to resend entire package.
(quoted diff) +2
-0+ * This is second error in a row. If mouse was itialized - attempt + * to rconnect, otherwise just signal failure.
Quoted 2 lines
Tooo many typso in on coment?
Pavel
Sorry about that. If the patch turns out to be working fine I'll fix the comments before resubmitting.
Hmm, wouldn't it be nice if they put spell checker in GCC? ;)
Signature · 1 line
Dmitry
Quoted 2 lines · Dmitry Torokhov
Hmm, wouldn't it be nice if they put spell checker in GCC? ;)
Isn't the resulting beast called "emacs"?
Signature · 2 lines
Ryan Anderson sometimes Pug Majere