mmc/dwmmc: remove recursive FIFO threshold setup
PATCH 1/1 mmc/dwmmc: remove recursive FIFO threshold setup
Removed code works properly only once after power-up because on every "first" invocation of "dwmci_init" existing value of "fifo_size" is used for calculation of itself. More over other bits in the same register (namely TX watermark and burst size) get corrupted (lost forever till the next power-toggle).
I understand that usually (in production systems) U-Boot starts only once and then passes control to Linux kernel etc. But when you do U-Boot debugging it might be loaded in memory via JTAG multiple times.
Also I wasn't able to find any instance of "host->fifoth_val" usage so I believe it's safe to just remove this code completely.
Another important thing - in the latest Linux kernel driver for DW MMC comtroller I don't see any utilization (configuration/modificatoin) of the register in question (SDMMC_FIFOTH). That's why I think we may want to keep it untouched as well so we don't force a situation when kernel driver gets DW MMC unexpectedly configured by U-Boot.
--- drivers/mmc/dw_mmc.c | 9 --------- include/dwmmc.h | 1 - 2 files changed, 10 deletions(-)
drivers/mmc/dw_mmc.c +0 -9
@@ -300,7 +300,6 @@ static void dwmci_set_ios(struct mmc *mmc)static int dwmci_init (struct mmc *mmc) {struct dwmci_host *host = (struct dwmci_host *)mmc->priv; -u32 fifo_size;if (host->quirks &DWMCI_QUIRK_DISABLE_SMU ) {dwmci_writel (host,EMMCP_MPSBEGIN0 ,0 );
@@ -330,14 +329,6 @@ static int dwmci_init(struct mmc *mmc)dwmci_writel (host,DWMCI_IDINTEN ,0 );dwmci_writel (host,DWMCI_BMOD ,1 ); -if (!host->fifoth_val) { - fifo_size =dwmci_readl (host,DWMCI_FIFOTH ); - fifo_size = ((fifo_size &RX_WMARK_MASK ) >>RX_WMARK_SHIFT ) +1 ; - host->fifoth_val =MSIZE (0x2 ) |RX_WMARK (fifo_size /2 -1 ) | -TX_WMARK (fifo_size /2 ); - } -dwmci_writel (host,DWMCI_FIFOTH , host->fifoth_val); -dwmci_writel (host,DWMCI_CLKENA ,0 );dwmci_writel (host,DWMCI_CLKSRC ,0 );
include/dwmmc.h +0 -1
@@ -137,7 +137,6 @@ struct dwmci_host {int dev_index;int buswidth;u32 clksel_val; -u32 fifoth_val;struct mmc *mmc;void (*clksel)(struct dwmci_host *host);
Signature · 1 line
1.8.4.2
PATCH 1/1 mmc/dwmmc: remove recursive FIFO threshold setup
Hi, Alexey,
I think good that use the initial fifoth value at register. Then we need not to change the value.
But according to my experiment, some SoC needs to change the fifoth value.
Quoted 6 lines · Alexey Brodkin
Removed code works properly only once after power-up because on every "first" invocation of "
dwmci_init" existing value of "fifo_size" is used for calculation of itself. More over other bits in the same register (namely TX watermark and burst size) get corrupted (lost forever till the next power-toggle).
Quoted 6 lines
I understand that usually (in production systems) U-Boot starts only once and then passes control to Linux kernel etc. But when you do U-Boot debugging it might be loaded in memory via JTAG multiple times.
Also I wasn't able to find any instance of "
host->fifoth_val" usage so I believe it's safe to just remove this code completely.
If developer is need to change the fifoth value, the use the host->fifoth_val. So it's existed.
Quoted 6 lines
Another important thing - in the latest Linux kernel driver for DW MMC comtroller I don't see any utilization (configuration/modificatoin) of the register in question (
SDMMC_FIFOTH). That's why I think we may want to keep it untouched as well so we don't force a situation when kernel driver gets DW MMC unexpectedly configured by U-Boot.
Right, if fifoth register is changed into U-boot,
then kernel driver use the value that fifoth register was set to value at U-Boot. I known that kernel driver can set to the value with dt file.
Best Regards,
Jaehoon Chung
Quoted 59 lines
Signed-off-by Alexey Brodkin <[email protected]>Cc Mischa Jonker <[email protected]>Cc Alim Akhtar <[email protected]>Cc Rajeshwari Shinde <[email protected]>Cc Jaehoon Chung <[email protected]>Cc Amar <[email protected]>Cc Kyungmin Park <[email protected]>Cc Hatim Ali <[email protected]>Cc Minkyu Kang <[email protected]>Cc Simon Glass <[email protected]>Cc Pantelis Antoniou <[email protected]>Cc Andy Fleming <[email protected]>--- drivers/mmc/dw_mmc.c | 9 --------- include/dwmmc.h | 1 - 2 files changed, 10 deletions(-)drivers/mmc/dw_mmc.c +0
-9@@ -300,7 +300,6 @@ static void dwmci_set_ios(struct mmc *mmc)static int dwmci_init (struct mmc *mmc) {struct dwmci_host *host = (struct dwmci_host *)mmc->priv; -u32 fifo_size;if (host->quirks &DWMCI_QUIRK_DISABLE_SMU ) {dwmci_writel (host,EMMCP_MPSBEGIN0 ,0 );@@ -330,14 +329,6 @@ static int dwmci_init(struct mmc *mmc)dwmci_writel (host,DWMCI_IDINTEN ,0 );dwmci_writel (host,DWMCI_BMOD ,1 ); -if (!host->fifoth_val) { - fifo_size =dwmci_readl (host,DWMCI_FIFOTH ); - fifo_size = ((fifo_size &RX_WMARK_MASK ) >>RX_WMARK_SHIFT ) +1 ; - host->fifoth_val =MSIZE (0x2 ) |RX_WMARK (fifo_size /2 -1 ) | -TX_WMARK (fifo_size /2 ); - } -dwmci_writel (host,DWMCI_FIFOTH , host->fifoth_val); -dwmci_writel (host,DWMCI_CLKENA ,0 );dwmci_writel (host,DWMCI_CLKSRC ,0 );include/dwmmc.h +0
-1@@ -137,7 +137,6 @@ struct dwmci_host {int dev_index;int buswidth;u32 clksel_val; -u32 fifoth_val;struct mmc *mmc;void (*clksel)(struct dwmci_host *host);
A message that is not in the archive (sent off-list or not yet received): 1385536307.2648.9.camel@abrodkin-8560l
PATCH 1/1 mmc/dwmmc: remove recursive FIFO threshold setup
Hi Alexey,
Quoted 10 lines · Alexey Brodkin
Hi Jaehoon,
Quoted 6 lines · Jaehoon Chung
Hi, Alexey,
I think good that use the initial fifoth value at register. Then we need not to change the value.
But according to my experiment, some SoC needs to change the fifoth value.
Quoted 6 lines
Ok, so let's leave possibility to set "fifoth" manually (if "
host->fifoth_val" != 0 then use provided value; else do nothing) on mmc controller instantiation.Will it work for you?
ATM if no host->fifoth_val value is provided the code will calculate one and write it. Otherwise it will write the one configured.
Quoted 5 lines
Still I have a couple of questions to you:
1. Why there's no example/instance of "host->fifoth_val" setup/usage in U-Boot source tree? If there was at least one I would leave "host->fifoth_val".
I would guess it's because everyone uses the default setting of 0 which results in a somewhat sane value.
Quoted 4 lines
2. How come there's no way to set/modify the register in question in Linux kernel? Don't people use Linux kernel on mentioned selected SoCs without U-Boot (i.e. "fifoth" gets set by U-Boot before kernel starts)?
That's a good question for the kernel driver devs.
Quoted 2 lines
Regards,
Alexey
Regards
-- Pantelis
A message that is not in the archive (sent off-list or not yet received): 1385538313.2648.24.camel@abrodkin-8560l
PATCH 1/1 mmc/dwmmc: remove recursive FIFO threshold setup
Hi Alexey,
Quoted 21 lines · Alexey Brodkin
Hi Pantelis,
Quoted 7 lines · Pantelis Antoniou
Hi Alexey,
On Nov 27, 2013, at 9:11 AM, Alexey Brodkin wrote:
ATM if no
host->fifoth_valvalue is provided the code will calculate one and write it. Otherwise it will write the one configured.Indeed. But what I faced - if you try to run U-Boot more than 1 time after powering up your board then this calculated value will be calculated improperly. And in my comment to the patch itself I described it.
So there're 2 options:
1. Re-implement calculation in such a way so it gets the same value on 2nd, 3rd and other U-Boot runs. Moreover it's not clear to me how this calculation is intended to work?
Quoted 7 lines
We take some initial value and then modify it - note it's not simply overrided with some value, but initial/default value is read from the register, then goes some math, and then we put new value back.
How do we know which value/configuration was by default so we do particular modifications later? What is a logic behind this?
Indeed this is bogus. What is the reset value of the fifoth register?
Instead of reading the register, and doing the weird calculations, just use the reset value of the register as if out of power on reset.
Quoted 5 lines
So why we need this functionality (unclear reconfiguration) if default register value works flawlessly in majority of cases and for some corner cases users may specify their custom value that will simply override default value.
Beats me. Let's fix this properly.
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines
Still I have a couple of questions to you:
1. Why there's no example/instance of "host->fifoth_val" setup/usage in U-Boot source tree? If there was at least one I would leave "host->fifoth_val".I would guess it's because everyone uses the default setting of 0 which results in a somewhat sane value.
Regards,
Alexey
Regards
-- Pantelis
PATCH 1/1 mmc/dwmmc: remove recursive FIFO threshold setup
Hi Pantelis,
Indeed this is bogus. What is the reset value of the fifoth register?
Quoted 1 line
Below is a description for the register in question with default
values specified.
----------------------------------------------
FIFO Threshold Watermark Register
----------------------------------------------
x - reserved
000 - Burst size of multiple transaction
yyyyyyyyyyy - Rx watermark = (FIFO_DEPTH - 1)
xxxx - reserved
000000000000 - Tx watermark
----------------------------------------------
Also databook says:
----------------------------------------------
During end of packet, request is generated regardless of threshold programming in order to complete any remaining data.
----------------------------------------------
So it's safe to leave watermark = 0.
Instead of reading the register, and doing the weird calculations, just
Quoted 2 lines
use the reset value of the register as if out of power on reset.
In other words both Rx and Tx watermarks should be sane for usage as they are on power-on.
Quoted 2 lines
So why we need this functionality (unclear reconfiguration) if default
Quoted 1 line
register value works flawlessly in majority of cases and for some
corner
Quoted 6 lines
Quoted 3 lines
cases users may specify their custom value that will simply override default value.
Beats me. Let's fix this properly.
So I'll submit v2 shortly.
-Alexey
PATCH 1/1 mmc/dwmmc: remove recursive FIFO threshold setup
In reply to a message not in the archiveHi, Alexey.
Quoted 8 lines · Alexey Brodkin
Hi Pantelis,
Quoted 4 lines · Pantelis Antoniou
Hi Alexey,
On Nov 27, 2013, at 9:11 AM, Alexey Brodkin wrote:
Quoted 7 lines
Quoted 2 lines
ATM if no
host->fifoth_valvalue is provided the code will calculate one and write it. Otherwise it will write the one configured.Indeed. But what I faced - if you try to run U-Boot more than 1 time after powering up your board then this calculated value will be calculated improperly. And in my comment to the patch itself I described it.
I understood your point. It's possible. If we set to calculated value, then it's set to value lower than previous value. right?
if host->fifoth_val is always used somewhere, this value should be fixed. But host->fifoth val was not used anywhere, so need to calculate the value. This is problem.
Best Regards,
Jaehoon Chung
Quoted 30 lines
So there're 2 options:
1. Re-implement calculation in such a way so it gets the same value on 2nd, 3rd and other U-Boot runs. Moreover it's not clear to me how this calculation is intended to work?We take some initial value and then modify it - note it's not simply overrided with some value, but initial/default value is read from the register, then goes some math, and then we put new value back.
How do we know which value/configuration was by default so we do particular modifications later? What is a logic behind this?
So why we need this functionality (unclear reconfiguration) if default register value works flawlessly in majority of cases and for some corner cases users may specify their custom value that will simply override default value.
Quoted 8 lines
Quoted 5 lines
Still I have a couple of questions to you:
1. Why there's no example/instance of "host->fifoth_val" setup/usage in U-Boot source tree? If there was at least one I would leave "host->fifoth_val".I would guess it's because everyone uses the default setting of 0 which results in a somewhat sane value.
Regards,
Alexey