Skip to contentSkip to conversation

fbdev: radeon: Remove 'struct timeval' usage

PATCH In review Tina Ruchandani 3 participants 4 messages May 25, 2015 – Aug 20, 2015

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fbdev: radeon: Remove 'struct timeval' usage

'struct timeval' uses a 32-bit representation for the seconds field which will overflow in the year 2038 and beyond. This patch replaces the usage of 'struct timeval' with ktime_t which uses a 64-bit time representation and does not suffer from the y2038 problem. This patch is part of a larger effort to remove all instances of 'struct timeval', 'struct timespec', time_t and other 32-bit timekeeping variables from the kernel.
The patch also replaces the use of real time (do_gettimeofday) with monotonic time (ktime_get).

---
 drivers/video/fbdev/aty/radeon_base.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)
drivers/video/fbdev/aty/radeon_base.c +14 -15
index 01237c8..9747e9e 100644
@@ -64,6 +64,7 @@
 #include <linux/slab.h>
 #include <linux/delay.h>
 #include <linux/time.h>
+#include <linux/ktime.h>
 #include <linux/fb.h>
 #include <linux/ioport.h>
 #include <linux/init.h>
@@ -461,8 +462,8 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 	int hTotal, vTotal, num, denom, m, n;
 	unsigned long long hz, vclk;
 	long xtal;
-	struct timeval start_tv, stop_tv;
-	long total_secs, total_usecs;
+	ktime_t start, stop;
+	s64 delta;
 	int i;
 
 	/* Ugh, we cut interrupts, bad bad bad, but we want some precision
@@ -478,7 +479,7 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 		if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) == 0)
 			break;
 
-	do_gettimeofday(&start_tv);
+	start = ktime_get();
 
 	for(i=0; i<1000000; i++)
 		if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) != 0)
@@ -487,20 +488,18 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 	for(i=0; i<1000000; i++)
 		if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) == 0)
 			break;
-	
-	do_gettimeofday(&stop_tv);
-	
+
+	stop = ktime_get();
+
 	local_irq_enable();
 
-	total_secs = stop_tv.tv_sec - start_tv.tv_sec;
-	if (total_secs > 10)
+	delta = ktime_us_delta(stop, start);
+
+	/* Return -1 if more than 10 seconds have elapsed */
+	if (delta > (10*1000000))
 		return -1;
-	total_usecs = stop_tv.tv_usec - start_tv.tv_usec;
-	total_usecs += total_secs * 1000000;
-	if (total_usecs < 0)
-		total_usecs = -total_usecs;
-	hz = 1000000/total_usecs;
- 
+	hz = 1000000/delta;
+
 	hTotal = ((INREG(CRTC_H_TOTAL_DISP) & 0x1ff) + 1) * 8;
 	vTotal = ((INREG(CRTC_V_TOTAL_DISP) & 0x3ff) + 1);
 	vclk = (long long)hTotal * (long long)vTotal * hz;
@@ -548,7 +547,7 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 		denom *= 3;
 		break;
 	case 6:
-		denom *= 6;   
+		denom *= 6;
 		break;
 	case 7:
 		denom *= 12;
Signature · 1 line
2.2.0.rc0.207.ga3a616c
#1 lore raw
Quoted 79 lines · Tina Ruchandani

'struct timeval' uses a 32-bit representation for the seconds field which will overflow in the year 2038 and beyond. This patch replaces the usage of 'struct timeval' with ktime_t which uses a 64-bit time representation and does not suffer from the y2038 problem. This patch is part of a larger effort to remove all instances of 'struct timeval', 'struct timespec', time_t and other 32-bit timekeeping variables from the kernel.
The patch also replaces the use of real time (do_gettimeofday) with monotonic time (ktime_get).

---
 drivers/video/fbdev/aty/radeon_base.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)
drivers/video/fbdev/aty/radeon_base.c +13 -14
index 01237c8..9747e9e 100644
@@ -64,6 +64,7 @@
 #include <linux/slab.h>
 #include <linux/delay.h>
 #include <linux/time.h>
+#include <linux/ktime.h>
 #include <linux/fb.h>
 #include <linux/ioport.h>
 #include <linux/init.h>
@@ -461,8 +462,8 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 	int hTotal, vTotal, num, denom, m, n;
 	unsigned long long hz, vclk;
 	long xtal;
-	struct timeval start_tv, stop_tv;
-	long total_secs, total_usecs;
+	ktime_t start, stop;
+	s64 delta;
 	int i;
 
 	/* Ugh, we cut interrupts, bad bad bad, but we want some precision
@@ -478,7 +479,7 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 		if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) == 0)
 			break;
 
-	do_gettimeofday(&start_tv);
+	start = ktime_get();
 
 	for(i=0; i<1000000; i++)
 		if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) != 0)
@@ -487,20 +488,18 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 	for(i=0; i<1000000; i++)
 		if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) == 0)
 			break;
-	
-	do_gettimeofday(&stop_tv);
-	
+
+	stop = ktime_get();
+
 	local_irq_enable();
 
-	total_secs = stop_tv.tv_sec - start_tv.tv_sec;
-	if (total_secs > 10)
+	delta = ktime_us_delta(stop, start);
+
+	/* Return -1 if more than 10 seconds have elapsed */
+	if (delta > (10*1000000))
 		return -1;
-	total_usecs = stop_tv.tv_usec - start_tv.tv_usec;
-	total_usecs += total_secs * 1000000;
-	if (total_usecs < 0)
-		total_usecs = -total_usecs;
-	hz = 1000000/total_usecs;
- 
+	hz = 1000000/delta;
+
 	hTotal = ((INREG(CRTC_H_TOTAL_DISP) & 0x1ff) + 1) * 8;
 	vTotal = ((INREG(CRTC_V_TOTAL_DISP) & 0x3ff) + 1);
 	vclk = (long long)hTotal * (long long)vTotal * hz;

drivers/video/fbdev/aty/radeon_base.c static int radeon_probe_pll_params(struct radeonfb_info *rinfo) · line 553

Quoted 8 lines
drivers/video/fbdev/aty/radeon_base.c +1 -1
@@ -548,7 +547,7 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
 		denom *= 3;
 		break;
 	case 6:
-		denom *= 6;   
+		denom *= 6;
 		break;
 	case 7:
 		denom *= 12;

The above change was extra. I removed it, and queued this for 4.2.

Tomi

#2 lore raw
Quoted 63 lines · Tomi Valkeinen
Quoted 60 lines · Tina Ruchandani

'struct timeval' uses a 32-bit representation for the seconds field which will overflow in the year 2038 and beyond. This patch replaces the usage of 'struct timeval' with ktime_t which uses a 64-bit time representation and does not suffer from the y2038 problem. This patch is part of a larger effort to remove all instances of 'struct timeval', 'struct timespec', time_t and other 32-bit timekeeping variables from the kernel.
The patch also replaces the use of real time (do_gettimeofday) with monotonic time (ktime_get).

---
 drivers/video/fbdev/aty/radeon_base.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)
drivers/video/fbdev/aty/radeon_base.c +7 -6
index 01237c8..9747e9e 100644
@@ -64,6 +64,7 @@
 #include <linux/slab.h>
 #include <linux/delay.h>
 #include <linux/time.h>
+#include <linux/ktime.h>
 #include <linux/fb.h>
 #include <linux/ioport.h>
 #include <linux/init.h>
@@ -461,8 +462,8 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
      int hTotal, vTotal, num, denom, m, n;
      unsigned long long hz, vclk;
      long xtal;
-     struct timeval start_tv, stop_tv;
-     long total_secs, total_usecs;
+     ktime_t start, stop;
+     s64 delta;
      int i;
 
      /* Ugh, we cut interrupts, bad bad bad, but we want some precision
@@ -478,7 +479,7 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
              if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) = 0)
                      break;
 
-     do_gettimeofday(&start_tv);
+     start = ktime_get();
 
      for(i=0; i<1000000; i++)
              if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) != 0)
@@ -487,20 +488,18 @@ static int radeon_probe_pll_params(struct radeonfb_info *rinfo)
      for(i=0; i<1000000; i++)
              if (((INREG(CRTC_VLINE_CRNT_VLINE) >> 16) & 0x3ff) = 0)
                      break;
-
-     do_gettimeofday(&stop_tv);
-
+
+     stop = ktime_get();
+
      local_irq_enable();
 
Quoted 14 lines
Quoted 4 lines
drivers/video/fbdev/aty/radeon_base.c +2 -2
-     total_secs = stop_tv.tv_sec - start_tv.tv_sec;
-     if (total_secs > 10)
+     delta = ktime_us_delta(stop, start);
+

drivers/video/fbdev/aty/radeon_base.c static int radeon_probe_pll_params(struct radeonfb_info *rinfo) · line 501

Quoted 10 lines
drivers/video/fbdev/aty/radeon_base.c +3 -6
+     /* Return -1 if more than 10 seconds have elapsed */
+     if (delta > (10*1000000))
              return -1;
-     total_usecs = stop_tv.tv_usec - start_tv.tv_usec;
-     total_usecs += total_secs * 1000000;
-     if (total_usecs < 0)
-             total_usecs = -total_usecs;
-     hz = 1000000/total_usecs;
-
+     hz = 1000000/delta;

This needs to be on of the do_div family.

Dave.

#3 lore raw
In reply to Tina Ruchandani

Hi Tina,

Quoted 11 lines · Tina Ruchandani

'struct timeval' uses a 32-bit representation for the seconds field which will overflow in the year 2038 and beyond. This patch replaces the usage of 'struct timeval' with ktime_t which uses a 64-bit time representation and does not suffer from the y2038 problem. This patch is part of a larger effort to remove all instances of 'struct timeval', 'struct timespec', time_t and other 32-bit timekeeping variables from the kernel.
The patch also replaces the use of real time (do_gettimeofday) with monotonic time (ktime_get).

Quoted 1 line

I had dropped this from 4.2 as there were issues reported, but forgot to write a mail about it...

Can you send a updated patch with the issues fixed?

Tomi