From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tom Rini Date: Thu, 22 Feb 2018 10:43:31 -0500 Subject: [U-Boot] [RFC] ns16550: Add support for AUX regs usage on some ARC SoCs In-Reply-To: <20180221122605.8128-1-abrodkin@synopsys.com> References: <20180221122605.8128-1-abrodkin@synopsys.com> Message-ID: <20180222154331.GT4311@bill-the-cat> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de On Wed, Feb 21, 2018 at 03:26:05PM +0300, Alexey Brodkin wrote: > Synopsys Data Fusion subsystem (DFSS) is targeted to deeply built-in > use-cases and so to save some silicon area decision was made to > escape usage of any busses and use instead directly wired to CPU > peripherals. And one of those is DW APB UART. > > Later DFSS became a part of larger and more complicated SoCs with > some other peripherals connected via common buses but default UART > is still used via ARC core's auxulary registers which are not mapped > to "normal" address space and we use very special instructions to access > them, thus we cannot simply reuse whatever accessor we have in 16550 > UART as of now. > > Also we cannot just switch inb()/outb() to access ARC AUX regs always > for DFSS because other peripherals have normal memory-mapped control > registers and we need to use normal accessors for them. > > Frankly I don't like a lot what I did here but otherwise if I create > a special driver for this I'll need to reimplement > ns16550_serial_ops.putc()/getc() which will be pure copy-paseted from > ns16550.c because the only difference is only in > ns16550_{read|write}b(). > > As mentioned above we cannot remap those auxiliary registers to > normal memory address-space and thus we have to use very special > accessors write_aux_reg()/read_aux_reg() that directly use special > CPU intructions (namely LR/SR) for dealing with ARC AUX regs. > > Also note here I just use a check for a particular SoC being selected > (CONFIG_ARCH_DFSS will be introduced shortly) but that is done just for > simplicity, otherwise it might be a slecial Kconfig option for NS16550 > or anything else. > > I'd like to know what people think about possible colutions here. > And as always any comments are much appreciated! > > Signed-off-by: Alexey Brodkin > Cc: Simon Glass > Cc: Tom Rini > Cc: Stefan Roese > --- > drivers/serial/ns16550.c | 11 +++++++++-- > 1 file changed, 9 insertions(+), 2 deletions(-) > > diff --git a/drivers/serial/ns16550.c b/drivers/serial/ns16550.c > index 6f9ce689cfff..75f342f337f1 100644 > --- a/drivers/serial/ns16550.c > +++ b/drivers/serial/ns16550.c > @@ -13,6 +13,9 @@ > #include > #include > #include > +#ifdef CONFIG_ARCH_DFSS > +#include > +#endif > > DECLARE_GLOBAL_DATA_PTR; > > @@ -53,7 +56,9 @@ DECLARE_GLOBAL_DATA_PTR; > > static inline void serial_out_shift(void *addr, int shift, int value) > { > -#ifdef CONFIG_SYS_NS16550_PORT_MAPPED > +#ifdef CONFIG_ARCH_DFSS > + write_aux_reg((int)addr, value); > +#elif defined(CONFIG_SYS_NS16550_PORT_MAPPED) > outb(value, (ulong)addr); > #elif defined(CONFIG_SYS_NS16550_MEM32) && !defined(CONFIG_SYS_BIG_ENDIAN) > out_le32(addr, value); > @@ -70,7 +75,9 @@ static inline void serial_out_shift(void *addr, int shift, int value) > > static inline int serial_in_shift(void *addr, int shift) > { > -#ifdef CONFIG_SYS_NS16550_PORT_MAPPED > +#ifdef CONFIG_ARCH_DFSS > + return read_aux_reg((int)addr); > +#elif defined(CONFIG_SYS_NS16550_PORT_MAPPED) > return inb((ulong)addr); > #elif defined(CONFIG_SYS_NS16550_MEM32) && !defined(CONFIG_SYS_BIG_ENDIAN) > return in_le32(addr); As always, thanks for the detailed explanation. Yes, I think that of the options, putting the details in read/write_aux_reg (and please make sure read/write_aux_reg have a good function comment too) is the best choice. Thanks! -- Tom -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 819 bytes Desc: not available URL: