Skip to contentSkip to conversation

x86/earlyprintk: Add a force option for pciserial device

PATCH Superseded Feng Tang 3 participants 14 messages Sep 28, 2018 – Oct 2, 2018 x86

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Feng Tang

author
#0 lore raw

PATCH v2 1/1 x86/earlyprintk: Add a force option for pciserial device

"pciserial" earlyprintk helps much on many modern x86 platforms, but unfortunately there are still some platforms whose PCI UART devices have wrong PCI class code, which will be blocked by current class code check.

Add a option "force" so that developer could still use a UART device even it has wrong class code, with format "",B:D.F,force,baud". And the original format ",B:D.F,baud" is kept unchanged.

---
 arch/x86/kernel/early_printk.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)
arch/x86/kernel/early_printk.c +20 -6
index 5e801c8..35d0f66 100644
@@ -213,8 +213,10 @@ static unsigned int mem32_serial_in(unsigned long addr, int offset)
  * early_pci_serial_init()
  *
  * This function is invoked when the early_printk param starts with "pciserial"
- * The rest of the param should be ",B:D.F,baud" where B, D & F describe the
- * location of a PCI device that must be a UART device.
+ * The rest of the param should be ",B:D.F,baud" or ",B:D.F,force,baud", where
+ * B, D & F describe the location of a PCI device that must be a UART device,
+ * "force" is optional and means insisting using a UART device with a wrong
+ * pci class code.
  */
 static __init void early_pci_serial_init(char *s)
 {
@@ -224,6 +226,7 @@ static __init void early_pci_serial_init(char *s)
 	u32 classcode, bar0;
 	u16 cmdreg;
 	char *e;
+	int force = 0;
 
 
 	/*
@@ -252,6 +255,15 @@ static __init void early_pci_serial_init(char *s)
 	if (*s == ',')
 		s++;
 
+	/* User may insist to use a UART device with wrong class code */
+	if (!strncmp(s, "force", 5)) {
+		force = 1;
+		s += 5;
+
+		if (*s == ',')
+			s++;
+	}
+
 	/*
 	 * Second, find the device from the BDF
 	 */
@@ -262,10 +274,12 @@ static __init void early_pci_serial_init(char *s)
 	/*
 	 * Verify it is a UART type device
 	 */
-	if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
-	     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
-	   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */
-		return;
+	if (!force) {
+		if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
+		     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
+		   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */
+			return;
+	}
 
 	/*
 	 * Determine if it is IO or memory mapped
Signature · 1 line
2.7.4
maintainer
#1 lore raw
Quoted 7 lines · Feng Tang

"pciserial" earlyprintk helps much on many modern x86 platforms, but unfortunately there are still some platforms whose PCI UART devices have wrong PCI class code, which will be blocked by current class code check.

Add a option "force" so that developer could still use a UART device even it has wrong class code, with format "",B:D.F,force,baud". And

This new parameter and its meaning needs to be documented where pciserial is documented.

Quoted 34 lines

the original format ",B:D.F,baud" is kept unchanged.

---
 arch/x86/kernel/early_printk.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)
arch/x86/kernel/early_printk.c +5 -2
index 5e801c8..35d0f66 100644
@@ -213,8 +213,10 @@ static unsigned int mem32_serial_in(unsigned long addr, int offset)
  * early_pci_serial_init()
  *
  * This function is invoked when the early_printk param starts with "pciserial"
- * The rest of the param should be ",B:D.F,baud" where B, D & F describe the
- * location of a PCI device that must be a UART device.
+ * The rest of the param should be ",B:D.F,baud" or ",B:D.F,force,baud", where
+ * B, D & F describe the location of a PCI device that must be a UART device,
+ * "force" is optional and means insisting using a UART device with a wrong
+ * pci class code.
  */
 static __init void early_pci_serial_init(char *s)
 {
@@ -224,6 +226,7 @@ static __init void early_pci_serial_init(char *s)
 	u32 classcode, bar0;
 	u16 cmdreg;
 	char *e;
+	int force = 0;
 
 
 	/*

arch/x86/kernel/early_printk.c static __init void early_pci_serial_init(char *s) · line 264

Quoted 10 lines
arch/x86/kernel/early_printk.c +7 -0
@@ -252,6 +255,15 @@ static __init void early_pci_serial_init(char *s)
 	if (*s == ',')
 		s++;
 
+	/* User may insist to use a UART device with wrong class code */
+	if (!strncmp(s, "force", 5)) {
+		force = 1;
+		s += 5;
+
+		if (*s == ',')
+			s++;

No, you need to force the presence of "," otherwise cmdlines like this:

earlyprintk=pciserial,0:XX.X,forcedoodoo

work too.

Quoted 8 lines
arch/x86/kernel/early_printk.c +2 -0
+	}
+
 	/*
 	 * Second, find the device from the BDF
 	 */
@@ -262,10 +274,12 @@ static __init void early_pci_serial_init(char *s)
 	/*
 	 * Verify it is a UART type device

arch/x86/kernel/early_printk.c static __init void early_pci_serial_init(char *s) · line 281

Quoted 10 lines
arch/x86/kernel/early_printk.c +5 -4
 	 */
-	if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
-	     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
-	   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */
-		return;
+	if (!force) {
+		if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
+		     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
+		   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */
+			return;

Move the force check in here ^

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.
author
#2 lore raw

Hi Borislav,

Quoted 11 lines · Borislav Petkov
Quoted 7 lines · Feng Tang

"pciserial" earlyprintk helps much on many modern x86 platforms, but unfortunately there are still some platforms whose PCI UART devices have wrong PCI class code, which will be blocked by current class code check.

Add a option "force" so that developer could still use a UART device even it has wrong class code, with format "",B:D.F,force,baud". And

This new parameter and its meaning needs to be documented where pciserial is documented.

Ok, will change the kernel-parameters.txt in next version.

Quoted 48 lines
Quoted 35 lines

the original format ",B:D.F,baud" is kept unchanged.

---
 arch/x86/kernel/early_printk.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)
arch/x86/kernel/early_printk.c +5 -2
index 5e801c8..35d0f66 100644
@@ -213,8 +213,10 @@ static unsigned int mem32_serial_in(unsigned long addr, int offset)
  * early_pci_serial_init()
  *
  * This function is invoked when the early_printk param starts with "pciserial"
- * The rest of the param should be ",B:D.F,baud" where B, D & F describe the
- * location of a PCI device that must be a UART device.
+ * The rest of the param should be ",B:D.F,baud" or ",B:D.F,force,baud", where
+ * B, D & F describe the location of a PCI device that must be a UART device,
+ * "force" is optional and means insisting using a UART device with a wrong
+ * pci class code.
  */
 static __init void early_pci_serial_init(char *s)
 {
@@ -224,6 +226,7 @@ static __init void early_pci_serial_init(char *s)
 	u32 classcode, bar0;
 	u16 cmdreg;
 	char *e;
+	int force = 0;
 
 
 	/*

arch/x86/kernel/early_printk.c static __init void early_pci_serial_init(char *s) · line 264

Quoted 10 lines
arch/x86/kernel/early_printk.c +7 -0
@@ -252,6 +255,15 @@ static __init void early_pci_serial_init(char *s)
 	if (*s == ',')
 		s++;
 
+	/* User may insist to use a UART device with wrong class code */
+	if (!strncmp(s, "force", 5)) {
+		force = 1;
+		s += 5;
+
+		if (*s == ',')
+			s++;

No, you need to force the presence of "," otherwise cmdlines like this:

Quoted 3 lines

earlyprintk=pciserial,0:XX.X,forcedoodoo

work too.

Got your point. As the following [,baudrate] is also optional, and this "force" could be the last option for pciserial. The check may be

	if (!strncmp(s, "force,", 6) || !strncmp(s, "force ", 6)) {
		force = 1;
		s += 6;
	}
Quoted 12 lines
Quoted 12 lines
arch/x86/kernel/early_printk.c +2 -3
+	}
+
 	/*
 	 * Second, find the device from the BDF
 	 */
@@ -262,10 +274,12 @@ static __init void early_pci_serial_init(char *s)
 	/*
 	 * Verify it is a UART type device
 	 */
-	if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
-	     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
-	   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */
Quoted 8 lines

arch/x86/kernel/early_printk.c static __init void early_pci_serial_init(char *s) · line 281

Quoted 6 lines
arch/x86/kernel/early_printk.c +5 -1
-		return;
+	if (!force) {
+		if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
+		     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
+		   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */
+			return;

Move the force check in here ^

Do you mean this?

	if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
	     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
	   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */ {
		if (!force)
			return;
	}

Thanks,
Feng

maintainer
#3 lore raw
Quoted 7 lines · Feng Tang

Got your point. As the following [,baudrate] is also optional, and this "force" could be the last option for pciserial. The check may be

	if (!strncmp(s, "force,", 6) || !strncmp(s, "force ", 6)) {
		force = 1;
		s += 6;
	}

You have !strncmp(s, "force,", 6) twice in there. I have no clue what you're trying to say.

Quoted 8 lines

Do you mean this?

	if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
	     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
	   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */ {
		if (!force)
			return;
	}

Yes.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.
author
#4 lore raw

Hi Boris,

Quoted 11 lines · Borislav Petkov
Quoted 7 lines · Feng Tang

Got your point. As the following [,baudrate] is also optional, and this "force" could be the last option for pciserial. The check may be

	if (!strncmp(s, "force,", 6) || !strncmp(s, "force ", 6)) {
		force = 1;
		s += 6;
	}

You have !strncmp(s, "force,", 6) twice in there. I have no clue what you're trying to say.

I was trying to address your last comments:

"No, you need to force the presence of "," otherwise cmdlines like this:

earlyprintk=pciserial,0:XX.X,forcedoodoo

work too."

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

Quoted 11 lines
Quoted 9 lines

Do you mean this?

	if (((classcode >> 16 != PCI_CLASS_COMMUNICATION_MODEM) &&
	     (classcode >> 16 != PCI_CLASS_COMMUNICATION_SERIAL)) ||
	   (((classcode >> 8) & 0xff) != 0x02)) /* 16550 I/F at BAR0 */ {
		if (!force)
			return;
	}

Yes.

Ok, thanks

Thanks,
Feng

maintainer
#5 lore raw
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

And what guarantees you have a space after the "force"?

	!strncmp(s, "force ", 6)
			  ^
Signature · 4 lines
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.
author
#6 lore raw

Hi Boris,

Quoted 9 lines · Borislav Petkov
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

And what guarantees you have a space after the "force"?

	!strncmp(s, "force ", 6)
			  ^

You are right, it can't be guranteed. Can we still use strncmp(s, "force", 5) and rely on developer to set it right? any suggestions? thanks,

- Feng

#7 lore raw
Quoted 8 lines · Feng Tang

Hi Boris,

Quoted 5 lines · Borislav Petkov
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

Quoted 7 lines
Quoted 4 lines

And what guarantees you have a space after the "force"?

	!strncmp(s, "force ", 6)
			  ^

You are right, it can't be guranteed. Can we still use strncmp(s, "force", 5) and rely on developer to set it right? any suggestions? thanks,

I don't know why you want strncmp() in the first place. "force" is null terminated already.

Thanks,

tglx

author
#8 lore raw

Hi Thomas,

Quoted 12 lines · Thomas Gleixner
Quoted 10 lines · Feng Tang

Hi Boris,

Quoted 7 lines · Borislav Petkov
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

And what guarantees you have a space after the "force"?

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
	!strncmp(s, "force ", 6)
			  ^

You are right, it can't be guranteed. Can we still use strncmp(s, "force", 5) and rely on developer to set it right? any suggestions? thanks,

I don't know why you want strncmp() in the first place. "force" is null terminated already.

Current code uses:
earlyprintk=pciserial,bus:device.function[,baudrate]
with the patch, it will be:
earlyprintk=pciserial,bus:device.function[,force][,baudrate]

So the force could be followed by ",baudrate".

Thanks,
Feng

#9 lore raw
Quoted 22 lines · Feng Tang
Quoted 12 lines · Thomas Gleixner
Quoted 10 lines · Feng Tang

Hi Boris,

Quoted 7 lines · Borislav Petkov
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

And what guarantees you have a space after the "force"?

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
	!strncmp(s, "force ", 6)
			  ^

You are right, it can't be guranteed. Can we still use strncmp(s, "force", 5) and rely on developer to set it right? any suggestions? thanks,

I don't know why you want strncmp() in the first place. "force" is null terminated already.

Quoted 6 lines

Current code uses:
earlyprintk=pciserial,bus:device.function[,baudrate]
with the patch, it will be:
earlyprintk=pciserial,bus:device.function[,force][,baudrate]

So the force could be followed by ",baudrate".

Sure, but that has nothing to do with strncmp(). The earlyprintk argument string is tokenized by commata. It really does not matter where the 'force' string is, but it matters that you check whether it is 'force' and not 'force$RANDOMCHARACTERS'. That's why strncmp() is the wrong thing to do, because it only compares the first 5 characters and ignores anything beyond, unless you make sure that the command line part is exactly 5 characters long.

Thanks,

tglx

author
#10 lore raw

Hi Thomas,

Quoted 30 lines · Thomas Gleixner
Quoted 22 lines · Feng Tang
Quoted 12 lines · Thomas Gleixner
Quoted 10 lines · Feng Tang

Hi Boris,

Quoted 7 lines · Borislav Petkov
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

And what guarantees you have a space after the "force"?

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
	!strncmp(s, "force ", 6)
			  ^

You are right, it can't be guranteed. Can we still use strncmp(s, "force", 5) and rely on developer to set it right? any suggestions? thanks,

I don't know why you want strncmp() in the first place. "force" is null terminated already.

Quoted 6 lines

Current code uses:
earlyprintk=pciserial,bus:device.function[,baudrate]
with the patch, it will be:
earlyprintk=pciserial,bus:device.function[,force][,baudrate]

So the force could be followed by ",baudrate".

Quoted 7 lines

Sure, but that has nothing to do with strncmp(). The earlyprintk argument string is tokenized by commata. It really does not matter where the 'force' string is, but it matters that you check whether it is 'force' and not 'force$RANDOMCHARACTERS'. That's why strncmp() is the wrong thing to do, because it only compares the first 5 characters and ignores anything beyond, unless you make sure that the command line part is exactly 5 characters long.

I don't know if I get it right due to my poor English, so for both cases of "xxx,force,baudrate" and "xxx,force", it could be covered by

	if ((!strcmp(s, "force") || !strncmp(s, "force," 6))
		force = 1;

Thanks,
Feng

author
#11 lore raw
Quoted 41 lines · Feng Tang

Hi Thomas,

Quoted 30 lines · Thomas Gleixner
Quoted 22 lines · Feng Tang
Quoted 12 lines · Thomas Gleixner
Quoted 10 lines · Feng Tang

Hi Boris,

Quoted 7 lines · Borislav Petkov
Quoted 3 lines · Feng Tang

As I rechecked, the baud rate for pciserial is optional, so there may be no ",baudrate" following the "force". So this 2 strncmp is to cover conditions for with and without baudrate.

And what guarantees you have a space after the "force"?

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
	!strncmp(s, "force ", 6)
			  ^

You are right, it can't be guranteed. Can we still use strncmp(s, "force", 5) and rely on developer to set it right? any suggestions? thanks,

I don't know why you want strncmp() in the first place. "force" is null terminated already.

Quoted 6 lines

Current code uses:
earlyprintk=pciserial,bus:device.function[,baudrate]
with the patch, it will be:
earlyprintk=pciserial,bus:device.function[,force][,baudrate]

So the force could be followed by ",baudrate".

Quoted 7 lines

Sure, but that has nothing to do with strncmp(). The earlyprintk argument string is tokenized by commata. It really does not matter where the 'force' string is, but it matters that you check whether it is 'force' and not 'force$RANDOMCHARACTERS'. That's why strncmp() is the wrong thing to do, because it only compares the first 5 characters and ignores anything beyond, unless you make sure that the command line part is exactly 5 characters long.

Quoted 5 lines

I don't know if I get it right due to my poor English, so for both cases of "xxx,force,baudrate" and "xxx,force", it could be covered by

	if ((!strcmp(s, "force") || !strncmp(s, "force," 6))
		force = 1;

One idea is to change the new format to
earlyprintk=pciserial[,force],bus:device.function[,baudrate]
Then we can use what Boris has suggested:

	!strncmp(s, "force,", 6)

Any thoughts?

Thanks,
Feng

#12 lore raw
Quoted 4 lines · Feng Tang

One idea is to change the new format to
earlyprintk=pciserial[,force],bus:device.function[,baudrate]
Then we can use what Boris has suggested:

	!strncmp(s, "force,", 6)

Works for me.

author
#13 lore raw
Quoted 7 lines · Thomas Gleixner
Quoted 4 lines · Feng Tang

One idea is to change the new format to
earlyprintk=pciserial[,force],bus:device.function[,baudrate]
Then we can use what Boris has suggested:

	!strncmp(s, "force,", 6)

Works for me.

Thanks, will send a v3.

- Feng