Re: [PM][PATCH v3 2/4] OMAP3: Serial: Errata i202: fix for MDR1 access

[Nishanth Menon <nm@ti.com>]

Kattungal, Deepak had written, on 05/05/2010 07:19 PM, the following:
> Hi Kevin,
> 
> My Comments as below.
> 
> Regards
> Deepak
> 
> -----Original Message-----
> From: Kevin Hilman [mailto:khilman@deeprootsystems.com] 
> Sent: Wednesday, May 05, 2010 6:55 PM
> To: Menon, Nishanth
> Cc: linux-omap; Kattungal, Deepak; Raja, Govindraj; Tero Kristo
> Subject: Re: [PM][PATCH v3 2/4] OMAP3: Serial: Errata i202: fix for MDR1 access
> 
> Nishanth Menon <nm@ti.com> writes:
> 
>> From: Deepak K <deepak.k@ti.com>
>>
>> Original patch:
>> http://git.omapzoom.org/?p=kernel/omap.git;a=commitdiff;h=42d4a342c009bd9727c100abc8a4bc3063c22f0c
>>
>> Errata i202 (OMAP3430 - 1.12, OMAP3630 - 1.6):
> 
> This workaround is currently done for all OMAPs.  
> 
> Presumably, this errata will eventually be fixed.  So, as with other
> errata fixes, we need some sort of SoC-rev based flag and the errata
> workaround done based on that flag.
> 
> Deepak : This would be a good fix, thanks for the suggestion.
to confuse ;): I kinda disagree and agree on this..
Disagree:
The code in question is under CONFIG_ARCH_OMAP3 for context save and 
restore -> yes, this errata is valid for for all OMAP3s SOCs currently. 
OMAP4 and CONFIG_PM is not there yet in pm branch I suppose. at least in 
the OMAP4 errata does not seem to have uart erratas any that I see of, 
but given that the doc is pretty prelim, I will wait for final edition..

Agree: the patch mashes up errata code along with normal logic, making 
it a bit difficult to read thru..

ideally, I would like to introduce:
struct omap_uart_state {
	u16 errata;
}
set it and handle it accordingly.. 2cents: would have preferred to see 
serial_override also do something similar though..

> 
> Also, shouldn't there be a fix for this in the 8250 and omap-serial
> drivers too?
> 
> 
> Deepak : The 8250 is not using the MDR Register. This would be needed only by the omap-serial.
 > The 8250 being a general driver we may not require the access to MDR1 
register. Hence the fix not
 > required for 8250.
hmm.. surprised MDR reg is settings b/w ir and uart settings which 
should be ideally in serial.c(at the current location) and very omap 
specific 8250 wont hit it, but wondering how omap-serial hits it.. gotta 
dig at it sometime later..

any suggestions which path to take folks?

> 
> Kevin
> 
>> UART module MDR1 register access can cause a dummy underrun
>> condition which could result in a freeze in the case of IrDA
>> communication or if used as UART, corrupted data.
>>
>> Workaround is as follows for everytime MDR1 register is changed:
>> * setup all required UART registers
>> * setup MDR1.MODE_SELECT bit field
>> * Wait 5 L4 clk cycles + 5 UART functional clock cycles
>> * Clear the Tx and RX fifo using FCR register
>>
>> Note: The following step is not done as I am assuming it is not
>> needed due to reconfiguration being done and there is no halted
>> operation perse.
>> * Read if required, the RESUME register to resume halted operation
>>
>> Cc: Govindraj R <govindraj.raja@ti.com>
>> Cc: Kevin Hilman <khilman@deeprootsystems.com>
>> Cc: Tero Kristo <tero.kristo@nokia.com>
>>
>> Signed-off-by: Deepak K <deepak.k@ti.com>
>> Signed-off-by: Nishanth Menon <nm@ti.com>
>> ---
>> Changes from V2:
>>   * Use macros instead of hardcoded values
>>
>> Ref:
>> V2: http://marc.info/?t=127084514600010&r=1&w=2&n=2
>> v1: http://marc.info/?t=127074933100005&r=1&w=2
>>
>>  arch/arm/mach-omap2/serial.c |   41 ++++++++++++++++++++++++++++++++++++++---
>>  1 files changed, 38 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/arm/mach-omap2/serial.c b/arch/arm/mach-omap2/serial.c
>> index a7c45b5..6d13183 100644
>> --- a/arch/arm/mach-omap2/serial.c
>> +++ b/arch/arm/mach-omap2/serial.c
>> @@ -185,6 +185,42 @@ static inline void __init omap_uart_reset(struct omap_uart_state *uart)
>>  
>>  #if defined(CONFIG_PM) && defined(CONFIG_ARCH_OMAP3)
>>  
>> +/*
>> + * Work Around for Errata i202 (3430 - 1.12, 3630 - 1.6)
>> + * The access to uart register after MDR1 Access
>> + * causes UART to corrupt data.
>> + *
>> + * Need a delay =
>> + * 5 L4 clock cycles + 5 UART functional clock cycle (@48MHz = ~0.2uS)
>> + * give 10 times as much
>> + */
>> +static void omap_uart_mdr1_errataset(struct omap_uart_state *uart, u8 mdr1_val,
>> +		u8 fcr_val)
>> +{
>> +	struct plat_serial8250_port *p = uart->p;
>> +	u8 timeout = 255;
>> +
>> +	serial_write_reg(p, UART_OMAP_MDR1, mdr1_val);
>> +	udelay(2);
>> +	serial_write_reg(p, UART_FCR, fcr_val | UART_FCR_CLEAR_XMIT |
>> +			UART_FCR_CLEAR_RCVR);
>> +	/*
>> +	 * Wait for FIFO to empty: when empty, RX_FIFO_E bit is 0 and
>> +	 * TX_FIFO_E bit is 1.
>> +	 */
>> +	while (UART_LSR_THRE != (serial_read_reg(p, UART_LSR) &
>> +				(UART_LSR_THRE | UART_LSR_DR))) {
>> +		timeout--;
>> +		if (!timeout) {
>> +			/* Should *never* happen. we warn and carry on */
>> +			dev_crit(&uart->pdev.dev, "Errata i202: timedout %x\n",
>> +				serial_read_reg(p, UART_LSR));
>> +			break;
>> +		}
>> +		udelay(1);
>> +	}
>> +}
>> +
>>  static void omap_uart_save_context(struct omap_uart_state *uart)
>>  {
>>  	u16 lcr = 0;
>> @@ -222,7 +258,7 @@ static void omap_uart_restore_context(struct omap_uart_state *uart)
>>  
>>  	uart->context_valid = 0;
>>  
>> -	serial_write_reg(p, UART_OMAP_MDR1, 0x7);
>> +	omap_uart_mdr1_errataset(uart, 0x07, 0xA0);
>>  	serial_write_reg(p, UART_LCR, 0xBF); /* Config B mode */
>>  	efr = serial_read_reg(p, UART_EFR);
>>  	serial_write_reg(p, UART_EFR, UART_EFR_ECB);
>> @@ -235,14 +271,13 @@ static void omap_uart_restore_context(struct omap_uart_state *uart)
>>  	serial_write_reg(p, UART_IER, uart->ier);
>>  	serial_write_reg(p, UART_LCR, 0x80);
>>  	serial_write_reg(p, UART_MCR, uart->mcr);
>> -	serial_write_reg(p, UART_FCR, 0xA1);
>>  	serial_write_reg(p, UART_LCR, 0xBF); /* Config B mode */
>>  	serial_write_reg(p, UART_EFR, efr);
>>  	serial_write_reg(p, UART_LCR, UART_LCR_WLEN8);
>>  	serial_write_reg(p, UART_OMAP_SCR, uart->scr);
>>  	serial_write_reg(p, UART_OMAP_WER, uart->wer);
>>  	serial_write_reg(p, UART_OMAP_SYSC, uart->sysc);
>> -	serial_write_reg(p, UART_OMAP_MDR1, 0x00); /* UART 16x mode */
>> +	omap_uart_mdr1_errataset(uart, 0x00, 0xA1);
>>  }
>>  #else
>>  static inline void omap_uart_save_context(struct omap_uart_state *uart) {}
>> -- 
>> 1.6.3.3


-- 
Regards,
Nishanth Menon