Skip to contentSkip to conversation

TTY improve n_gsm support

PATCH Superseded Gregory CLEMENT 5 participants 12 messages May 12, 2020 – May 18, 2020 TTY layer and serial drivers

lore
Share & tools
lore
#0 lore raw

v2 cover TTY improve n_gsm support

Hello,

This series is the second version of patch improving n_gms support especially with TELIT LE910. However the fix should benefit to any modem supporting cmux.

The first patch is just about improving debugging output.

The second one removes a tty optimization which make the LE910 hang.

The last one fixes an issue observed on the LE910 but should benefit to all the modem. We observed that pretty quickly the transfer done using the virtual tty were blocked. We found that it was due of a wakeup to the real tty. Without this fix, the real tty wait for indefinitely.

Thanks to Jiri Slaby for the review.

Changelog:
v1 -> v2:
- don't replace the pr_info by pr_debug
- remove the superfluous printk("\n");
- use --follow option with git log to find the original commit to fix
- use tty_port_tty_wakeup
- use 'for' loop instead of 'while'

Gregory

Gregory CLEMENT (3):
tty: n_gsm: Improve debug output
tty: n_gsm: Fix SOF skipping
tty: n_gsm: Fix waking up upper tty layer when room available

 drivers/tty/n_gsm.c | 48 +++++++++++++++++++++++----------------------
 1 file changed, 25 insertions(+), 23 deletions(-)
Signature · 1 line
2.26.2
#1 lore raw

PATCH v2 1/3 tty: n_gsm: Improve debug output

Use appropriate print helpers for debug messages.

---
 drivers/tty/n_gsm.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)
drivers/tty/n_gsm.c +2 -12
index d77ed82a4840..67c8f8173023 100644
@@ -504,18 +504,8 @@ static void gsm_print_packet(const char *hdr, int addr, int cr,
 	else
 		pr_cont("(F)");
 
-	if (dlen) {
-		int ct = 0;
-		while (dlen--) {
-			if (ct % 8 == 0) {
-				pr_cont("\n");
-				pr_debug("    ");
-			}
-			pr_cont("%02X ", *data++);
-			ct++;
-		}
-	}
-	pr_cont("\n");
+	if (dlen)
+		print_hex_dump_bytes("", DUMP_PREFIX_NONE, data, dlen);
 }
 
 
Signature · 1 line
2.26.2
#2 lore raw

Re: [PATCH v2 1/3] tty: n_gsm: Improve debug output

Quoted 18 lines · Gregory CLEMENT

Use appropriate print helpers for debug messages.

---
 drivers/tty/n_gsm.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)
drivers/tty/n_gsm.c +0 -3
index d77ed82a4840..67c8f8173023 100644
@@ -504,18 +504,8 @@ static void gsm_print_packet(const char *hdr, int addr, int cr,
 	else
 		pr_cont("(F)");
 
-	if (dlen) {
-		int ct = 0;
-		while (dlen--) {

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 507

Quoted 10 lines
drivers/tty/n_gsm.c +1 -9
-			if (ct % 8 == 0) {
-				pr_cont("\n");
-				pr_debug("    ");
-			}
-			pr_cont("%02X ", *data++);
-			ct++;
-		}
-	}
-	pr_cont("\n");
+	if (dlen)

This test is superfluous. print_hex_dump_* won't write anything when zero length is passed to it.

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 509

Quoted 2 lines
drivers/tty/n_gsm.c +1 -0
+		print_hex_dump_bytes("", DUMP_PREFIX_NONE, data, dlen);
 }

thanks,

Signature · 2 lines
js
suse labs
#3 lore raw

Re: [PATCH v2 1/3] tty: n_gsm: Improve debug output

Hello Jiri,

Quoted 16 lines · Gregory CLEMENT
Quoted 15 lines

Use appropriate print helpers for debug messages.

---
 drivers/tty/n_gsm.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)
drivers/tty/n_gsm.c +0 -0
index d77ed82a4840..67c8f8173023 100644
@@ -504,18 +504,8 @@ static void gsm_print_packet(const char *hdr, int addr, int cr,
 	else
 		pr_cont("(F)");
 
Quoted 16 lines

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 507

Quoted 13 lines
drivers/tty/n_gsm.c +1 -12
-	if (dlen) {
-		int ct = 0;
-		while (dlen--) {
-			if (ct % 8 == 0) {
-				pr_cont("\n");
-				pr_debug("    ");
-			}
-			pr_cont("%02X ", *data++);
-			ct++;
-		}
-	}
-	pr_cont("\n");
+	if (dlen)

This test is superfluous. print_hex_dump_* won't write anything when zero length is passed to it.

As I will send a v3 due to the issue found on the last patch, I am also going to fix this.

Gregory

Quoted 8 lines

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 509

Quoted 3 lines
drivers/tty/n_gsm.c +1 -0
+		print_hex_dump_bytes("", DUMP_PREFIX_NONE, data, dlen);
 }

thanks,

Signature · 2 lines
js
suse labs
Signature · 3 lines
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com
#4 lore raw

Re: [PATCH v2 1/3] tty: n_gsm: Improve debug output

Quoted 35 lines · Gregory CLEMENT

Hello Jiri,

Quoted 16 lines · Gregory CLEMENT
Quoted 15 lines

Use appropriate print helpers for debug messages.

---
 drivers/tty/n_gsm.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)
drivers/tty/n_gsm.c +0 -0
index d77ed82a4840..67c8f8173023 100644
@@ -504,18 +504,8 @@ static void gsm_print_packet(const char *hdr, int addr, int cr,
 	else
 		pr_cont("(F)");
 
Quoted 16 lines

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 507

Quoted 13 lines
drivers/tty/n_gsm.c +1 -12
-	if (dlen) {
-		int ct = 0;
-		while (dlen--) {
-			if (ct % 8 == 0) {
-				pr_cont("\n");
-				pr_debug("    ");
-			}
-			pr_cont("%02X ", *data++);
-			ct++;
-		}
-	}
-	pr_cont("\n");
+	if (dlen)

This test is superfluous. print_hex_dump_* won't write anything when zero length is passed to it.

Quoted 2 lines

As I will send a v3 due to the issue found on the last patch, I am also going to fix this.

Ugh, as I already applied this series, should I just revert them all, or are you going to send fix-ups on top of what I have applied instead?

thanks,

greg k-h

#5 lore raw

Re: [PATCH v2 1/3] tty: n_gsm: Improve debug output

Hello Greg,

Quoted 36 lines · Gregory CLEMENT
Quoted 35 lines

Hello Jiri,

Quoted 32 lines · Gregory CLEMENT
Quoted 18 lines

Use appropriate print helpers for debug messages.

---
 drivers/tty/n_gsm.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)
drivers/tty/n_gsm.c +0 -3
index d77ed82a4840..67c8f8173023 100644
@@ -504,18 +504,8 @@ static void gsm_print_packet(const char *hdr, int addr, int cr,
 	else
 		pr_cont("(F)");
 
-	if (dlen) {
-		int ct = 0;
-		while (dlen--) {

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 507

Quoted 10 lines
drivers/tty/n_gsm.c +1 -9
-			if (ct % 8 == 0) {
-				pr_cont("\n");
-				pr_debug("    ");
-			}
-			pr_cont("%02X ", *data++);
-			ct++;
-		}
-	}
-	pr_cont("\n");
+	if (dlen)

This test is superfluous. print_hex_dump_* won't write anything when zero length is passed to it.

Quoted 5 lines
Quoted 2 lines

As I will send a v3 due to the issue found on the last patch, I am also going to fix this.

Ugh, as I already applied this series, should I just revert them all, or are you going to send fix-ups on top of what I have applied instead?

I was about to send a new series, but I can just send fix-ups. It's up to you.

Gregory

Quoted 4 lines

thanks,

greg k-h

Signature · 3 lines
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com
#6 lore raw

Re: [PATCH v2 1/3] tty: n_gsm: Improve debug output

Quoted 38 lines · Gregory CLEMENT

Hello Greg,

Quoted 36 lines · Gregory CLEMENT
Quoted 35 lines

Hello Jiri,

Quoted 32 lines · Gregory CLEMENT
Quoted 18 lines

Use appropriate print helpers for debug messages.

---
 drivers/tty/n_gsm.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)
drivers/tty/n_gsm.c +0 -3
index d77ed82a4840..67c8f8173023 100644
@@ -504,18 +504,8 @@ static void gsm_print_packet(const char *hdr, int addr, int cr,
 	else
 		pr_cont("(F)");
 
-	if (dlen) {
-		int ct = 0;
-		while (dlen--) {

drivers/tty/n_gsm.c static void gsm_print_packet(const char *hdr, int addr, int cr, · line 507

Quoted 10 lines
drivers/tty/n_gsm.c +1 -9
-			if (ct % 8 == 0) {
-				pr_cont("\n");
-				pr_debug("    ");
-			}
-			pr_cont("%02X ", *data++);
-			ct++;
-		}
-	}
-	pr_cont("\n");
+	if (dlen)

This test is superfluous. print_hex_dump_* won't write anything when zero length is passed to it.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

As I will send a v3 due to the issue found on the last patch, I am also going to fix this.

Ugh, as I already applied this series, should I just revert them all, or are you going to send fix-ups on top of what I have applied instead?

I was about to send a new series, but I can just send fix-ups. It's up to you.

fix-ups are less work for me :)

#7 lore raw

PATCH 1/1 tty: n_gsm: gsm_print_packet: use scnprintf to avoid pr_cont

Use a temporary buffer to avoid multiple pr_cont uses.

---

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Ugh, as I already applied this series, should I just revert them all, or are you going to send fix-ups on top of what I have applied instead?

I was about to send a new series, but I can just send fix-ups. It's up to you.

fix-ups are less work for me :)

Perhaps use something like this instead?

It does increase object size a tiny bit.

 drivers/tty/n_gsm.c | 71 +++++++++++++++++++++++++++++++----------------------
 1 file changed, 42 insertions(+), 29 deletions(-)
drivers/tty/n_gsm.c +42 -29
index 69200bd411f7..7d7820aeb57b 100644
@@ -454,58 +454,71 @@ static u8 gsm_encode_modem(const struct gsm_dlci *dlci)
  */
 
 static void gsm_print_packet(const char *hdr, int addr, int cr,
-					u8 control, const u8 *data, int dlen)
+			     u8 control, const u8 *data, int dlen)
 {
+	char buf[100];
+	char *p;
+	char *type;
+
 	if (!(debug & 1))
 		return;
 
-	pr_info("%s %d) %c: ", hdr, addr, "RC"[cr]);
+	p = buf;
+	p += scnprintf(p, buf - p, "%s %d) %c: ", hdr, addr, "RC"[cr]);
 
 	switch (control & ~PF) {
 	case SABM:
-		pr_cont("SABM");
+		type = "SABM";
 		break;
 	case UA:
-		pr_cont("UA");
+		type = "UA";
 		break;
 	case DISC:
-		pr_cont("DISC");
+		type = "DISC";
 		break;
 	case DM:
-		pr_cont("DM");
+		type = "DM";
 		break;
 	case UI:
-		pr_cont("UI");
+		type = "UI";
 		break;
 	case UIH:
-		pr_cont("UIH");
+		type = "UIH";
 		break;
 	default:
-		if (!(control & 0x01)) {
-			pr_cont("I N(S)%d N(R)%d",
-				(control & 0x0E) >> 1, (control & 0xE0) >> 5);
-		} else switch (control & 0x0F) {
-			case RR:
-				pr_cont("RR(%d)", (control & 0xE0) >> 5);
-				break;
-			case RNR:
-				pr_cont("RNR(%d)", (control & 0xE0) >> 5);
-				break;
-			case REJ:
-				pr_cont("REJ(%d)", (control & 0xE0) >> 5);
-				break;
-			default:
-				pr_cont("[%02X]", control);
+		type = NULL;
+		break;
+	}
+
+	if (type) {
+		p += scnprintf(p, buf - p, "%s", type);
+	} else if (!(control & 0x01)) {
+		p += scnprintf(p, buf - p, "I N(S)%d N(R)%d",
+			       (control & 0x0E) >> 1, (control & 0xE0) >> 5);
+	} else {
+		switch (control & 0x0F) {
+		case RR:
+			p += scnprintf(p, buf - p, "RR(%d)",
+				       (control & 0xE0) >> 5);
+			break;
+		case RNR:
+			p += scnprintf(p, buf - p, "RNR(%d)",
+				       (control & 0xE0) >> 5);
+			break;
+		case REJ:
+			p += scnprintf(p, buf - p, "REJ(%d)",
+				       (control & 0xE0) >> 5);
+			break;
+		default:
+			p += scnprintf(p, buf - p, "[%02X]", control);
+			break;
 		}
 	}
 
-	if (control & PF)
-		pr_cont("(P)");
-	else
-		pr_cont("(F)");
+	p += scnprintf(p, buf - p, "(%c)", control & PF ? 'P' : 'F');
 
-	if (dlen)
-		print_hex_dump_bytes("", DUMP_PREFIX_NONE, data, dlen);
+	pr_info("%s\n", buf);
+	print_hex_dump_bytes("", DUMP_PREFIX_NONE, data, dlen);
 }
 
 
#8 lore raw

PATCH v2 2/3 tty: n_gsm: Fix SOF skipping

In reply to Gregory CLEMENT

For at least some modems like the TELIT LE910, skipping SOF makes transfers blocking indefinitely after a short amount of data transferred.

Given the small improvement provided by skipping the SOF (just one byte on about 100 bytes), it seems better to completely remove this "feature" than make it optional.

Fixes e1eaea46bb40 ("tty: n_gsm line discipline")
---
 drivers/tty/n_gsm.c | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)
drivers/tty/n_gsm.c +1 -7
index 67c8f8173023..d8d196645500 100644
@@ -667,7 +667,6 @@ static void gsm_data_kick(struct gsm_mux *gsm)
 {
 	struct gsm_msg *msg, *nmsg;
 	int len;
-	int skip_sof = 0;
 
 	list_for_each_entry_safe(msg, nmsg, &gsm->tx_list, list) {
 		if (gsm->constipated && msg->addr)
@@ -689,15 +688,10 @@ static void gsm_data_kick(struct gsm_mux *gsm)
 			print_hex_dump_bytes("gsm_data_kick: ",
 					     DUMP_PREFIX_OFFSET,
 					     gsm->txframe, len);
-
-		if (gsm->output(gsm, gsm->txframe + skip_sof,
-						len - skip_sof) < 0)
+		if (gsm->output(gsm, gsm->txframe, len) < 0)
 			break;
 		/* FIXME: Can eliminate one SOF in many more cases */
 		gsm->tx_bytes -= msg->len;
-		/* For a burst of frames skip the extra SOF within the
-		   burst */
-		skip_sof = 1;
 
 		list_del(&msg->list);
 		kfree(msg);
Signature · 1 line
2.26.2
#9 lore raw

PATCH v2 3/3 tty: n_gsm: Fix waking up upper tty layer when room available

In reply to Gregory CLEMENT

Warn the upper layer when n_gms is ready to receive data again. Without this the associated virtual tty remains blocked indefinitely.

Fixes e1eaea46bb40 ("tty: n_gsm line discipline")
---
 drivers/tty/n_gsm.c | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)
drivers/tty/n_gsm.c +22 -4
index d8d196645500..69200bd411f7 100644
@@ -663,7 +663,7 @@ static struct gsm_msg *gsm_data_alloc(struct gsm_mux *gsm, u8 addr, int len,
  *	FIXME: lock against link layer control transmissions
  */
 
-static void gsm_data_kick(struct gsm_mux *gsm)
+static void gsm_data_kick(struct gsm_mux *gsm, struct gsm_dlci *dlci)
 {
 	struct gsm_msg *msg, *nmsg;
 	int len;
@@ -695,6 +695,24 @@ static void gsm_data_kick(struct gsm_mux *gsm)
 
 		list_del(&msg->list);
 		kfree(msg);
+
+		if (dlci) {
+			tty_port_tty_wakeup(&dlci->port);
+		} else {
+			int i = 0;
+
+			for (i = 0; i < NUM_DLCI; i++) {
+				struct gsm_dlci *dlci;
+
+				dlci = gsm->dlci[i];
+				if (dlci == NULL) {
+					i++;
+					continue;
+				}
+
+				tty_port_tty_wakeup(&dlci->port);
+			}
+		}
 	}
 }
 
@@ -746,7 +764,7 @@ static void __gsm_data_queue(struct gsm_dlci *dlci, struct gsm_msg *msg)
 	/* Add to the actual output queue */
 	list_add_tail(&msg->list, &gsm->tx_list);
 	gsm->tx_bytes += msg->len;
-	gsm_data_kick(gsm);
+	gsm_data_kick(gsm, dlci);
 }
 
 /**
@@ -1207,7 +1225,7 @@ static void gsm_control_message(struct gsm_mux *gsm, unsigned int command,
 		gsm_control_reply(gsm, CMD_FCON, NULL, 0);
 		/* Kick the link in case it is idling */
 		spin_lock_irqsave(&gsm->tx_lock, flags);
-		gsm_data_kick(gsm);
+		gsm_data_kick(gsm, NULL);
 		spin_unlock_irqrestore(&gsm->tx_lock, flags);
 		break;
 	case CMD_FCOFF:
@@ -2529,7 +2547,7 @@ static void gsmld_write_wakeup(struct tty_struct *tty)
 	/* Queue poll */
 	clear_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);
 	spin_lock_irqsave(&gsm->tx_lock, flags);
-	gsm_data_kick(gsm);
+	gsm_data_kick(gsm, NULL);
 	if (gsm->tx_bytes < TX_THRESH_LO) {
 		gsm_dlci_data_sweep(gsm);
 	}
Signature · 1 line
2.26.2
#10 lore raw

Re: [PATCH v2 3/3] tty: n_gsm: Fix waking up upper tty layer when room available

Quoted 29 lines · Gregory CLEMENT

Warn the upper layer when n_gms is ready to receive data again. Without this the associated virtual tty remains blocked indefinitely.

Fixes e1eaea46bb40 ("tty: n_gsm line discipline")
---
 drivers/tty/n_gsm.c | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)
drivers/tty/n_gsm.c +3 -1
index d8d196645500..69200bd411f7 100644
@@ -663,7 +663,7 @@ static struct gsm_msg *gsm_data_alloc(struct gsm_mux *gsm, u8 addr, int len,
  *	FIXME: lock against link layer control transmissions
  */
 
-static void gsm_data_kick(struct gsm_mux *gsm)
+static void gsm_data_kick(struct gsm_mux *gsm, struct gsm_dlci *dlci)
 {
 	struct gsm_msg *msg, *nmsg;
 	int len;
@@ -695,6 +695,24 @@ static void gsm_data_kick(struct gsm_mux *gsm)
 
 		list_del(&msg->list);
 		kfree(msg);
+
+		if (dlci) {

drivers/tty/n_gsm.c static void gsm_data_kick(struct gsm_mux *gsm) · line 709

Quoted 10 lines
drivers/tty/n_gsm.c +10 -0
+			tty_port_tty_wakeup(&dlci->port);
+		} else {
+			int i = 0;
+
+			for (i = 0; i < NUM_DLCI; i++) {
+				struct gsm_dlci *dlci;
+
+				dlci = gsm->dlci[i];
+				if (dlci == NULL) {
+					i++;

This "i++" looks bogus here.

drivers/tty/n_gsm.c static void gsm_data_kick(struct gsm_mux *gsm) · line 713

Quoted 4 lines
drivers/tty/n_gsm.c +4 -0
+					continue;
+				}
+
+				tty_port_tty_wakeup(&dlci->port);

So simply:
for (i = 0; i < NUM_DLCI; i++) {
struct gsm_dlci *dlci = gsm->dlci[i];
if (dlci)

    tty_port_tty_wakeup(&dlci->port);
}

? Or even maybe directly:
for (i = 0; i < NUM_DLCI; i++)
if (gsm->dlci[i])

    tty_port_tty_wakeup(&gsm->dlci[i]->port);

thanks,

Signature · 2 lines
js
suse labs
#11 lore raw

Re: [PATCH v2 3/3] tty: n_gsm: Fix waking up upper tty layer when room available

Hi Jiri,

Quoted 26 lines · Gregory CLEMENT
Quoted 25 lines

Warn the upper layer when n_gms is ready to receive data again. Without this the associated virtual tty remains blocked indefinitely.

Fixes e1eaea46bb40 ("tty: n_gsm line discipline")
---
 drivers/tty/n_gsm.c | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)
drivers/tty/n_gsm.c +1 -1
index d8d196645500..69200bd411f7 100644
@@ -663,7 +663,7 @@ static struct gsm_msg *gsm_data_alloc(struct gsm_mux *gsm, u8 addr, int len,
  *	FIXME: lock against link layer control transmissions
  */
 
-static void gsm_data_kick(struct gsm_mux *gsm)
+static void gsm_data_kick(struct gsm_mux *gsm, struct gsm_dlci *dlci)
 {
 	struct gsm_msg *msg, *nmsg;
 	int len;
@@ -695,6 +695,24 @@ static void gsm_data_kick(struct gsm_mux *gsm)
 
Quoted 16 lines
Quoted 4 lines
drivers/tty/n_gsm.c +2 -0
 		list_del(&msg->list);
 		kfree(msg);
+
+		if (dlci) {

drivers/tty/n_gsm.c static void gsm_data_kick(struct gsm_mux *gsm) · line 709

Quoted 10 lines
drivers/tty/n_gsm.c +10 -0
+			tty_port_tty_wakeup(&dlci->port);
+		} else {
+			int i = 0;
+
+			for (i = 0; i < NUM_DLCI; i++) {
+				struct gsm_dlci *dlci;
+
+				dlci = gsm->dlci[i];
+				if (dlci == NULL) {
+					i++;

This "i++" looks bogus here.

You're right!
Sorry for this.

Quoted 11 lines

drivers/tty/n_gsm.c static void gsm_data_kick(struct gsm_mux *gsm) · line 713

Quoted 5 lines
drivers/tty/n_gsm.c +4 -0
+					continue;
+				}
+
+				tty_port_tty_wakeup(&dlci->port);

So simply:
for (i = 0; i < NUM_DLCI; i++) {
struct gsm_dlci *dlci = gsm->dlci[i];
if (dlci)

Quoted 7 lines
    tty_port_tty_wakeup(&dlci->port);
}

? Or even maybe directly:
for (i = 0; i < NUM_DLCI; i++)
if (gsm->dlci[i])

    tty_port_tty_wakeup(&gsm->dlci[i]->port);

I will do this, thanks,

Gregory

Quoted 5 lines

thanks,

Signature · 2 lines
js
suse labs
Signature · 3 lines
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com