serial: 8250_dw: Rework ->serial_out() & LCR retry logic
Subsystems (3)
Ilpo Järvinen
maintainercover serial: 8250_dw: Rework ->serial_out() & LCR retry logic
This series reworks DW UART's write logic such that the write in ->serial_out() is reused for LCR retries which allows removing those ugly duplicated writes in dw8250_check_lcr() (renamed to dw8250_verify_write() by this series).
I've used label+goto since the retry is really an exceptional thing. If somebody insists, I can convert those to do {} while (); but I feel it will give wrong impression that there's a "loop" there.
Ilpo Järvinen (4):
serial: 8250_dw: Use dw8250_serial_out() in dw8250_serial_out38x()
serial: 8250_dw: Rename offset to reg_offset
serial: 8250_dw: Move 16550 compatible & LCR checks to
dw8250_verify_write()
serial: 8250_dw: Rework ->serial_out() LCR write retry logic
drivers/tty/serial/8250/8250_dw.c | 90 ++++++++++++++++--------------- 1 file changed, 47 insertions(+), 43 deletions(-)
Signature · 1 line
2.30.2
Ilpo Järvinen
maintainerPATCH 1/4 serial: 8250_dw: Use dw8250_serial_out() in dw8250_serial_out38x()
Place dw8250_serial_out() before dw8250_serial_out38x() so that it can be called from dw8250_serial_out38x() to do the actual write.
--- drivers/tty/serial/8250/8250_dw.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-)
drivers/tty/serial/8250/8250_dw.c +6 -12
@@ -143,29 +143,23 @@ static void dw8250_tx_wait_empty(struct uart_port *p) } } -static void dw8250_serial_out38x (struct uart_port *p,int offset,int value) +static void dw8250_serial_out (struct uart_port *p,int offset,int value) {struct dw8250_data *d =to_dw8250_data (p->private_data); -/* Allow the TX to drain before we reconfigure */ -if (offset ==UART_LCR ) -dw8250_tx_wait_empty (p); -writeb (value, p->membase + (offset << p->regshift));if (offset ==UART_LCR && !d->uart_16550_compatible)dw8250_check_lcr (p, value); } - -static void dw8250_serial_out (struct uart_port *p,int offset,int value) +static void dw8250_serial_out38x (struct uart_port *p,int offset,int value) { -struct dw8250_data *d =to_dw8250_data (p->private_data); - -writeb (value, p->membase + (offset << p->regshift)); +/* Allow the TX to drain before we reconfigure */ +if (offset ==UART_LCR ) +dw8250_tx_wait_empty (p); -if (offset ==UART_LCR && !d->uart_16550_compatible) -dw8250_check_lcr (p, value); +dw8250_serial_out (p, offset, value); }static unsigned int dw8250_serial_in (struct uart_port *p,int offset)
Signature · 1 line
2.30.2
Ilpo Järvinen
maintainerPATCH 2/4 serial: 8250_dw: Rename offset to reg_offset
In reply to Ilpo JärvinenGet 'offset' variable out of the way of parameter named 'offset', rename it to 'reg_offset'. This is very short lived change as reg_offset is going to be soon removed.
--- drivers/tty/serial/8250/8250_dw.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/tty/serial/8250/8250_dw.c +5 -5
@@ -89,7 +89,7 @@ static void dw8250_force_idle(struct uart_port *p)static void dw8250_check_lcr (struct uart_port *p,int value) { -void __iomem *offset = p->membase + (UART_LCR << p->regshift); +void __iomem *reg_offset = p->membase + (UART_LCR << p->regshift);int tries =1000 ;/* Make sure LCR write wasn't ignored */
@@ -103,15 +103,15 @@ static void dw8250_check_lcr(struct uart_port *p, int value)#ifdef CONFIG_64BIT if (p->type ==PORT_OCTEON ) -__raw_writeq (value &0xff , offset); +__raw_writeq (value &0xff , reg_offset); else#endif if (p->iotype ==UPIO_MEM32 ) -writel (value, offset); +writel (value, reg_offset);else if (p->iotype ==UPIO_MEM32BE ) -iowrite32be (value, offset); +iowrite32be (value, reg_offset);else -writeb (value, offset); +writeb (value, reg_offset); }/* * FIXME: this deadlocks if port->lock is already held
Signature · 1 line
2.30.2
Andy Shevchenko
reviewerRe: [PATCH 2/4] serial: 8250_dw: Rename offset to reg_offset
Quoted 3 lines · Ilpo Järvinen
Get 'offset' variable out of the way of parameter named 'offset', rename it to '
reg_offset'. This is very short lived change asreg_offsetis going to be soon removed.
I'm not sure why this change then even needed...
Signature · 2 lines
With Best Regards, Andy Shevchenko
Ilpo Järvinen
maintainerRe: [PATCH 2/4] serial: 8250_dw: Rename offset to reg_offset
Quoted 6 lines · Andy Shevchenko
Quoted 3 lines · Ilpo Järvinen
Get 'offset' variable out of the way of parameter named 'offset', rename it to '
reg_offset'. This is very short lived change asreg_offsetis going to be soon removed.I'm not sure why this change then even needed...
I could either:
1) create one large patch doing many thing (2+3 or 2+3+4)
or
2) add the 'offset' parameter with some other name first and rename it to its final name after local var 'offset' is eliminated by patch 4
or
3) rename local var 'offset' first out of the way so that I can add 'offset' parameter in patch 3 (=this patch)
If I just drop patch 2 and only do 3, it won't build because 'offset' variable appears twice (as arg and local var).
Signature · 1 line
i.
Re: [PATCH 2/4] serial: 8250_dw: Rename offset to reg_offset
Quoted 6 lines · Ilpo Järvinen
Quoted 3 lines · Andy Shevchenko
Quoted 2 lines · Ilpo Järvinen
Get 'offset' variable out of the way of parameter named 'offset', rename it to '
reg_offset'. This is very short lived change as
a very
Quoted 13 lines
Quoted 3 lines
Quoted 1 line
reg_offsetis going to be soon removed.I'm not sure why this change then even needed...
I could either:
1) create one large patch doing many thing (2+3 or 2+3+4)
or
2) add the 'offset' parameter with some other name first and rename it to its final name after local var 'offset' is eliminated by patch 4
or
3) rename local var 'offset' first out of the way so that I can add 'offset' parameter in patch 3 (=this patch)
Quoted 2 lines
If I just drop patch 2 and only do 3, it won't build because 'offset' variable appears twice (as arg and local var).
Now I got it, thanks. See one remark above.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Ilpo Järvinen
maintainerPATCH 3/4 serial: 8250_dw: Move 16550 compatible & LCR checks to dw8250_verify_write()
In reply to Ilpo JärvinenRename dw8250_check_lcr() -> dw8250_verify_write() and add comment. Move LCR and 16550_compatible checks there. As offset is now passed and dw8250_verify_write() ensures it's UART_LCR, offset can use used instead of explicit UART_LCR.
--- drivers/tty/serial/8250/8250_dw.c | 35 +++++++++++++++---------------- 1 file changed, 17 insertions(+), 18 deletions(-)
drivers/tty/serial/8250/8250_dw.c +17 -18
@@ -87,14 +87,24 @@ static void dw8250_force_idle(struct uart_port *p) (void )p->serial_in (p,UART_RX ); } -static void dw8250_check_lcr (struct uart_port *p,int value) +/* +* DW UART can be configured to indicate BUSY in USR (with +* UART_16550_COMPATIBLE=NO or version prior to introducing that option). +* If BUSY is set while writing to LCR register, the write is ignored and +* needs to be retries. +*/ +static void dw8250_verify_write (struct uart_port *p,int offset,int value) { -void __iomem *reg_offset = p->membase + (UART_LCR << p->regshift); +void __iomem *reg_offset = p->membase + (offset << p->regshift); +struct dw8250_data *d =to_dw8250_data (p->private_data);int tries =1000 ; +if ((offset !=UART_LCR ) || !d->uart_16550_compatible) +return ; +/* Make sure LCR write wasn't ignored */ while (tries--) { -unsigned int lcr = p->serial_in (p,UART_LCR ); +unsigned int lcr = p->serial_in (p, offset);if ((value & ~UART_LCR_SPAR ) == (lcr & ~UART_LCR_SPAR ))return ;
@@ -145,12 +155,9 @@ static void dw8250_tx_wait_empty(struct uart_port *p)static void dw8250_serial_out (struct uart_port *p,int offset,int value) { -struct dw8250_data *d =to_dw8250_data (p->private_data); -writeb (value, p->membase + (offset << p->regshift)); -if (offset ==UART_LCR && !d->uart_16550_compatible) -dw8250_check_lcr (p, value); +dw8250_verify_write (p, offset, value); }static void dw8250_serial_out38x (struct uart_port *p,int offset,int value)
@@ -181,26 +188,20 @@ static unsigned int dw8250_serial_inq(struct uart_port *p, int offset)static void dw8250_serial_outq (struct uart_port *p,int offset,int value) { -struct dw8250_data *d =to_dw8250_data (p->private_data); - value &=0xff ;__raw_writeq (value, p->membase + (offset << p->regshift));/* Read back to ensure register write ordering. */ __raw_readq (p->membase + (UART_LCR << p->regshift)); -if (offset ==UART_LCR && !d->uart_16550_compatible) -dw8250_check_lcr (p, value); +dw8250_verify_write (p, offset, value); }#endif /* CONFIG_64BIT */ static void dw8250_serial_out32 (struct uart_port *p,int offset,int value) { -struct dw8250_data *d =to_dw8250_data (p->private_data); -writel (value, p->membase + (offset << p->regshift)); -if (offset ==UART_LCR && !d->uart_16550_compatible) -dw8250_check_lcr (p, value); +dw8250_verify_write (p, offset, value); }static unsigned int dw8250_serial_in32 (struct uart_port *p,int offset)
@@ -212,12 +213,10 @@ static unsigned int dw8250_serial_in32(struct uart_port *p, int offset)static void dw8250_serial_out32be (struct uart_port *p,int offset,int value) { -struct dw8250_data *d =to_dw8250_data (p->private_data);iowrite32be (value, p->membase + (offset << p->regshift)); -if (offset ==UART_LCR && !d->uart_16550_compatible) -dw8250_check_lcr (p, value); +dw8250_verify_write (p, offset, value); }static unsigned int dw8250_serial_in32be (struct uart_port *p,int offset)
Signature · 1 line
2.30.2
Re: [PATCH 3/4] serial: 8250_dw: Move 16550 compatible & LCR checks to dw8250_verify_write()
Quoted 5 lines · Ilpo Järvinen
Rename
dw8250_check_lcr()->dw8250_verify_write()and add comment. Move LCR and 16550_compatible checks there. As offset is now passed anddw8250_verify_write()ensures it'sUART_LCR, offset can use used instead of explicitUART_LCR.
...
Quoted 5 lines
+/*
+ * DW UART can be configured to indicate BUSY in USR (with
+ * UART_16550_COMPATIBLE=NO or version prior to introducing that option).
+ * If BUSY is set while writing to LCR register, the write is ignored and
+ * needs to be retries.
retried
Quoted 1 line
+ */
Signature · 2 lines
With Best Regards, Andy Shevchenko
Ilpo Järvinen
maintainerPATCH 4/4 serial: 8250_dw: Rework ->serial_out() LCR write retry logic
In reply to Ilpo JärvinenCurrently dw8250_verify_write() (was dw8250_check_lcr()) nullifies the benefit from differentiated ->serial_out() by having big if tree to select correct write type.
Rework the logic such that the LCR write can be retried within the relevant ->serial_out() handler:
1. Move retries counter on the caller level and pass as pointer to
dw8250_verify_write()
2. Make dw8250_verify_write() return bool
3. Retry the write on caller level (if needed)
--- drivers/tty/serial/8250/8250_dw.c | 59 ++++++++++++++++++------------- 1 file changed, 35 insertions(+), 24 deletions(-)
drivers/tty/serial/8250/8250_dw.c +35 -24
@@ -92,41 +92,36 @@ static void dw8250_force_idle(struct uart_port *p)* UART_16550_COMPATIBLE=NO or version prior to introducing that option). * If BUSY is set while writing to LCR register, the write is ignored and * needs to be retries. +* +* Returns: false if the caller should retry the write. */ -static void dw8250_verify_write (struct uart_port *p,int offset,int value) +static bool dw8250_verify_write (struct uart_port *p,int offset,int value, +unsigned int *retries) { -void __iomem *reg_offset = p->membase + (offset << p->regshift);struct dw8250_data *d =to_dw8250_data (p->private_data); -int tries =1000 ; +unsigned int lcr;if ((offset !=UART_LCR ) || !d->uart_16550_compatible) -return ; +return true;/* Make sure LCR write wasn't ignored */ -while (tries--) { -unsigned int lcr = p->serial_in (p, offset); + lcr = p->serial_in (p, offset); -if ((value & ~UART_LCR_SPAR ) == (lcr & ~UART_LCR_SPAR )) -return ; +if ((value & ~UART_LCR_SPAR ) == (lcr & ~UART_LCR_SPAR )) +return true; -dw8250_force_idle (p); +dw8250_force_idle (p); -#ifdef CONFIG_64BIT -if (p->type ==PORT_OCTEON ) -__raw_writeq (value &0xff , reg_offset); - else -#endif -if (p->iotype ==UPIO_MEM32 ) -writel (value, reg_offset); -else if (p->iotype ==UPIO_MEM32BE ) -iowrite32be (value, reg_offset); -else -writeb (value, reg_offset); +if (*retries) { + *retries -=1 ; +return false; } +/* * FIXME: this deadlocks if port->lock is already held * dev_err(p->dev, "Couldn't set LCR to %d\n", value); */ +return true; }/* Returns once the transmitter is empty or we run out of retries */
@@ -155,9 +150,13 @@ static void dw8250_tx_wait_empty(struct uart_port *p)static void dw8250_serial_out (struct uart_port *p,int offset,int value) { +unsigned int retries =1000 ; + +retry:writeb (value, p->membase + (offset << p->regshift)); -dw8250_verify_write (p, offset, value); +if (!dw8250_verify_write (p, offset, value, &retries)) + goto retry; }static void dw8250_serial_out38x (struct uart_port *p,int offset,int value)
@@ -188,20 +187,29 @@ static unsigned int dw8250_serial_inq(struct uart_port *p, int offset)static void dw8250_serial_outq (struct uart_port *p,int offset,int value) { +unsigned int retries =1000 ; + value &=0xff ; + +retry:__raw_writeq (value, p->membase + (offset << p->regshift));/* Read back to ensure register write ordering. */ __raw_readq (p->membase + (UART_LCR << p->regshift)); -dw8250_verify_write (p, offset, value); +if (!dw8250_verify_write (p, offset, value, &retries)) + goto retry; }#endif /* CONFIG_64BIT */ static void dw8250_serial_out32 (struct uart_port *p,int offset,int value) { +unsigned int retries =1000 ; + +retry:writel (value, p->membase + (offset << p->regshift)); -dw8250_verify_write (p, offset, value); +if (!dw8250_verify_write (p, offset, value, &retries)) + goto retry; }static unsigned int dw8250_serial_in32 (struct uart_port *p,int offset)
@@ -213,10 +221,13 @@ static unsigned int dw8250_serial_in32(struct uart_port *p, int offset)static void dw8250_serial_out32be (struct uart_port *p,int offset,int value) { +unsigned int retries =1000 ; +retry:iowrite32be (value, p->membase + (offset << p->regshift)); -dw8250_verify_write (p, offset, value); +if (!dw8250_verify_write (p, offset, value, &retries)) + goto retry; }static unsigned int dw8250_serial_in32be (struct uart_port *p,int offset)
Signature · 1 line
2.30.2
Re: [PATCH 4/4] serial: 8250_dw: Rework ->serial_out() LCR write retry logic
Quoted 11 lines · Ilpo Järvinen
Currently
dw8250_verify_write()(wasdw8250_check_lcr()) nullifies the benefit from differentiated ->serial_out() by having big if tree to select correct write type.Rework the logic such that the LCR write can be retried within the relevant ->serial_out() handler:
1. Move retries counter on the caller level and pass as pointer todw8250_verify_write()2. Make
dw8250_verify_write()return bool
3. Retry the write on caller level (if needed)
I'm wondering if it's possible to utilize one of iopoll.h macro here instead of copying retries and that not-so-obvious IO poll write.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Ilpo Järvinen
maintainerRe: [PATCH 4/4] serial: 8250_dw: Rework ->serial_out() LCR write retry logic
Quoted 14 lines · Andy Shevchenko
Quoted 11 lines · Ilpo Järvinen
Currently
dw8250_verify_write()(wasdw8250_check_lcr()) nullifies the benefit from differentiated ->serial_out() by having big if tree to select correct write type.Rework the logic such that the LCR write can be retried within the relevant ->serial_out() handler:
1. Move retries counter on the caller level and pass as pointer todw8250_verify_write()2. Make
dw8250_verify_write()return bool
3. Retry the write on caller level (if needed)
Quoted 2 lines
I'm wondering if it's possible to utilize one of
iopoll.hmacro here instead of copying retries and that not-so-obvious IO poll write.
Eh, are you suggesting I should do write as a side-effect inside one of the iopoll.h macros? Because those available seem to only read?
Or should I create another macro there which writes too?
Signature · 1 line
i.
Re: [PATCH 4/4] serial: 8250_dw: Rework ->serial_out() LCR write retry logic
Quoted 15 lines · Ilpo Järvinen
Quoted 14 lines · Andy Shevchenko
Quoted 11 lines · Ilpo Järvinen
Currently
dw8250_verify_write()(wasdw8250_check_lcr()) nullifies the benefit from differentiated ->serial_out() by having big if tree to select correct write type.Rework the logic such that the LCR write can be retried within the relevant ->serial_out() handler:
1. Move retries counter on the caller level and pass as pointer todw8250_verify_write()2. Make
dw8250_verify_write()return bool
3. Retry the write on caller level (if needed)
Quoted 7 lines
Quoted 2 lines
I'm wondering if it's possible to utilize one of
iopoll.hmacro here instead of copying retries and that not-so-obvious IO poll write.Eh, are you suggesting I should do write as a side-effect inside one of the
iopoll.hmacros? Because those available seem to only read?Or should I create another macro there which writes too?
It seems to me that it would be a macro on top of iopoll's one which will take an op read and op write arguments depending on the case. Note, for that special case you would need a custom write op instead of simple __raw_writeq().
Try and if it looks better, convert, otherwise it would be nice to hear why it won't fly in your opinion.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Ilpo Järvinen
maintainerRe: [PATCH 4/4] serial: 8250_dw: Rework ->serial_out() LCR write retry logic
Quoted 20 lines · Andy Shevchenko
Quoted 18 lines · Ilpo Järvinen
Quoted 16 lines · Andy Shevchenko
Quoted 11 lines · Ilpo Järvinen
Currently
dw8250_verify_write()(wasdw8250_check_lcr()) nullifies the benefit from differentiated ->serial_out() by having big if tree to select correct write type.Rework the logic such that the LCR write can be retried within the relevant ->serial_out() handler:
1. Move retries counter on the caller level and pass as pointer todw8250_verify_write()2. Make
dw8250_verify_write()return bool
3. Retry the write on caller level (if needed)I'm wondering if it's possible to utilize one of
iopoll.hmacro here instead of copying retries and that not-so-obvious IO poll write.
Quoted 7 lines
Quoted 4 lines
Eh, are you suggesting I should do write as a side-effect inside one of the
iopoll.hmacros? Because those available seem to only read?Or should I create another macro there which writes too?
It seems to me that it would be a macro on top of iopoll's one which will take an op read and op write arguments depending on the case.
The thing is those iopoll macros don't return until the timeout is exhausted so I don't think I can reuse them easily for this task ("on top of iopoll's one")? That is, w/o some major side-effect hack (which is IMHO a no-go).
Signature · 1 line
i.
Quoted 5 lines
Note, for that special case you would need a custom write op instead of simple
__raw_writeq().Try and if it looks better, convert, otherwise it would be nice to hear why it won't fly in your opinion.
Re: [PATCH 4/4] serial: 8250_dw: Rework ->serial_out() LCR write retry logic
Quoted 4 lines · Ilpo Järvinen
Quoted 3 lines · Andy Shevchenko
Quoted 1 line · Ilpo Järvinen
On Tue, 28 Jun 2022, Andy Shevchenko wrote:
...
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
Eh, are you suggesting I should do write as a side-effect inside one of the
iopoll.hmacros? Because those available seem to only read?Or should I create another macro there which writes too?
It seems to me that it would be a macro on top of iopoll's one which will take an op read and op write arguments depending on the case.
The thing is those iopoll macros don't return until the timeout is exhausted
It returns when the condition is true (in your case verify_lcr is OK).
Quoted 3 lines
so I don't think I can reuse them easily for this task ("on top of iopoll's one")? That is, w/o some major side-effect hack (which is IMHO a no-go).
Basically what we need is a write-read type of polling.
With your current approach I don't like that retries assignment is duplicated several times and decrement happens in the callee. What I'm trying to suggest is to research alternatives.
Signature · 2 lines
With Best Regards, Andy Shevchenko