[PATCH net-next] stmmac: fix signed 64-bit division

[PATCH net-next] stmmac: fix signed 64-bit division

[Arnd Bergmann]

I link error on 32-bit ARM points to yet another arithmetic bug:

drivers/net/ethernet/stmicro/stmmac/stmmac_tc.o: In function `tc_setup_cbs':
stmmac_tc.c:(.text+0x148): undefined reference to `__aeabi_uldivmod'
stmmac_tc.c:(.text+0x1fc): undefined reference to `__aeabi_uldivmod'
stmmac_tc.c:(.text+0x308): undefined reference to `__aeabi_uldivmod'
stmmac_tc.c:(.text+0x320): undefined reference to `__aeabi_uldivmod'
stmmac_tc.c:(.text+0x33c): undefined reference to `__aeabi_uldivmod'
drivers/net/ethernet/stmicro/stmmac/stmmac_tc.o:stmmac_tc.c:(.text+0x3a4): more undefined references to `__aeabi_uldivmod' follow

I observe that the last change to add the 'ul' prefix was incorrect,
as it did not turn the result of the multiplication into a 64-bit
expression on 32-bit architectures. Further, it seems that the
do_div() macro gets confused by the fact that we pass a signed
variable rather than unsigned into it.

This changes the code to instead use the div_s64() helper that is
meant for signed division, along with changing the constant suffix
to 'll' to actually make it a 64-bit argument everywhere, fixing
both of the issues I pointed out.

I'm not completely convinced that this makes the code correct, but
I'm fairly sure that we have two problems less than before.

Fixes: 1f705bc61aee ("net: stmmac: Add support for CBS QDISC")
Fixes: c18a9c096683 ("net: stmmac_tc: use 64-bit arithmetic instead of 32-bit")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c
index 8fedc288d138..1a96dd9c1091 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c
@@ -321,18 +321,16 @@ static int tc_setup_cbs(struct stmmac_priv *priv,
 	speed_div = (priv->speed == SPEED_100) ? 100000 : 1000000;
 
 	/* Final adjustments for HW */
-	value = qopt->idleslope * 1024UL * ptr;
-	do_div(value, speed_div);
+	value = div_s64(qopt->idleslope * 1024ll * ptr, speed_div);
 	priv->plat->tx_queues_cfg[queue].idle_slope = value & GENMASK(31, 0);
 
-	value = -qopt->sendslope * 1024UL * ptr;
-	do_div(value, speed_div);
+	value = div_s64(-qopt->sendslope * 1024ll * ptr, speed_div);
 	priv->plat->tx_queues_cfg[queue].send_slope = value & GENMASK(31, 0);
 
-	value = qopt->hicredit * 1024UL * 8;
+	value = qopt->hicredit * 1024ll * 8;
 	priv->plat->tx_queues_cfg[queue].high_credit = value & GENMASK(31, 0);
 
-	value = qopt->locredit * 1024UL * 8;
+	value = qopt->locredit * 1024ll * 8;
 	priv->plat->tx_queues_cfg[queue].low_credit = value & GENMASK(31, 0);
 
 	ret = stmmac_config_cbs(priv, priv->hw,
-- 
2.9.0

Re: [PATCH net-next] stmmac: fix signed 64-bit division

[Gustavo A. R. Silva]

Hi Arnd,

On 07/06/2018 08:36 AM, Arnd Bergmann wrote:
> I link error on 32-bit ARM points to yet another arithmetic bug:
> 
> drivers/net/ethernet/stmicro/stmmac/stmmac_tc.o: In function `tc_setup_cbs':
> stmmac_tc.c:(.text+0x148): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x1fc): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x308): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x320): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x33c): undefined reference to `__aeabi_uldivmod'
> drivers/net/ethernet/stmicro/stmmac/stmmac_tc.o:stmmac_tc.c:(.text+0x3a4): more undefined references to `__aeabi_uldivmod' follow
> 
> I observe that the last change to add the 'ul' prefix was incorrect,
> as it did not turn the result of the multiplication into a 64-bit
> expression on 32-bit architectures. Further, it seems that the
> do_div() macro gets confused by the fact that we pass a signed
> variable rather than unsigned into it.
> 

Thanks for this fix.

> This changes the code to instead use the div_s64() helper that is
> meant for signed division, along with changing the constant suffix
> to 'll' to actually make it a 64-bit argument everywhere, fixing
> both of the issues I pointed out.
> 
> I'm not completely convinced that this makes the code correct, but
> I'm fairly sure that we have two problems less than before.
> 
> Fixes: 1f705bc61aee ("net: stmmac: Add support for CBS QDISC")
> Fixes: c18a9c096683 ("net: stmmac_tc: use 64-bit arithmetic instead of 32-bit")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Acked-by: Gustavo A. R. Silva <gustavo@embeddedor.com>

> ---
>  drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c | 10 ++++------
>  1 file changed, 4 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c
> index 8fedc288d138..1a96dd9c1091 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c
> @@ -321,18 +321,16 @@ static int tc_setup_cbs(struct stmmac_priv *priv,
>  	speed_div = (priv->speed == SPEED_100) ? 100000 : 1000000;
>  
>  	/* Final adjustments for HW */
> -	value = qopt->idleslope * 1024UL * ptr;
> -	do_div(value, speed_div);
> +	value = div_s64(qopt->idleslope * 1024ll * ptr, speed_div);
>  	priv->plat->tx_queues_cfg[queue].idle_slope = value & GENMASK(31, 0);
>  
> -	value = -qopt->sendslope * 1024UL * ptr;
> -	do_div(value, speed_div);
> +	value = div_s64(-qopt->sendslope * 1024ll * ptr, speed_div);
>  	priv->plat->tx_queues_cfg[queue].send_slope = value & GENMASK(31, 0);
>  
> -	value = qopt->hicredit * 1024UL * 8;
> +	value = qopt->hicredit * 1024ll * 8;
>  	priv->plat->tx_queues_cfg[queue].high_credit = value & GENMASK(31, 0);
>  
> -	value = qopt->locredit * 1024UL * 8;
> +	value = qopt->locredit * 1024ll * 8;
>  	priv->plat->tx_queues_cfg[queue].low_credit = value & GENMASK(31, 0);
>  
>  	ret = stmmac_config_cbs(priv, priv->hw,
> 

Re: [PATCH net-next] stmmac: fix signed 64-bit division

[David Miller]

From: Arnd Bergmann <arnd@arndb.de>
Date: Fri,  6 Jul 2018 15:36:07 +0200

> I link error on 32-bit ARM points to yet another arithmetic bug:
> 
> drivers/net/ethernet/stmicro/stmmac/stmmac_tc.o: In function `tc_setup_cbs':
> stmmac_tc.c:(.text+0x148): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x1fc): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x308): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x320): undefined reference to `__aeabi_uldivmod'
> stmmac_tc.c:(.text+0x33c): undefined reference to `__aeabi_uldivmod'
> drivers/net/ethernet/stmicro/stmmac/stmmac_tc.o:stmmac_tc.c:(.text+0x3a4): more undefined references to `__aeabi_uldivmod' follow
> 
> I observe that the last change to add the 'ul' prefix was incorrect,
> as it did not turn the result of the multiplication into a 64-bit
> expression on 32-bit architectures. Further, it seems that the
> do_div() macro gets confused by the fact that we pass a signed
> variable rather than unsigned into it.
> 
> This changes the code to instead use the div_s64() helper that is
> meant for signed division, along with changing the constant suffix
> to 'll' to actually make it a 64-bit argument everywhere, fixing
> both of the issues I pointed out.
> 
> I'm not completely convinced that this makes the code correct, but
> I'm fairly sure that we have two problems less than before.
> 
> Fixes: 1f705bc61aee ("net: stmmac: Add support for CBS QDISC")
> Fixes: c18a9c096683 ("net: stmmac_tc: use 64-bit arithmetic instead of 32-bit")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Applied, thank you.