usbip: fix off-by-one frame number calculation
Subsystems (2)
PATCH 1/2 usbip: fix off-by-one frame number calculation
vgadget_get_frame returns a frame number from 0 to 2046, which may require an expensive division operation to wrap at one lower than the usual number.
I can't see any reason for this, and all other drivers wrap at a power-of-two number. My best explanation is that it was a simple typo, so I'm changing the % modulo operator into a cheaper bitmask that the other drivers use, to make it wrap after 0x7ff rather than before it.
--- drivers/usb/usbip/vudc_dev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/usb/usbip/vudc_dev.c +1 -1
@@ -145,7 +145,7 @@ static int vgadget_get_frame(struct usb_gadget *_gadget)do_gettimeofday (&now);return ((now.tv_sec - udc->start_time.tv_sec) *1000 + (now.tv_usec - udc->start_time.tv_usec) /1000 ) - %0x7FF ; + &0x7FF ; }static int vgadget_set_selfpowered (struct usb_gadget *_gadget,int value)
Signature · 1 line
2.9.0
PATCH 2/2 usbip: use monotonic timestamps
This gets rid of the deprecated do_gettimeofday() function in usbip. The comment above vgadget_get_frame() mentions that it suffers from issues with the time jumps due to suspend and settimeofday, so I'm changing it to use ktime_get_ts64() to use monotonic times that don't have this problem.
I couldn't tell whether we should use CLOCK_MONOTONIC or CLOCK_MONOTONIC_RAW here, the difference being the exact rate when correcting for NTP. I picked monotonic time since it doesn't change the speed to the existing code and should be better synchronized with other machines we talk to.
--- drivers/usb/usbip/vudc.h | 2 +- drivers/usb/usbip/vudc_dev.c | 7 +++---- drivers/usb/usbip/vudc_sysfs.c | 2 +- 3 files changed, 5 insertions(+), 6 deletions(-)
drivers/usb/usbip/vudc.h +1 -1
@@ -105,7 +105,7 @@ struct vudc {struct usbip_device ud;struct transfer_timer tr_timer; -struct timeval start_time; +struct timespec64 start_time;struct list_head urb_queue;
drivers/usb/usbip/vudc_dev.c +3 -4
@@ -136,15 +136,14 @@ struct vep *vudc_find_endpoint(struct vudc *udc, u8 address)/* gadget ops */ -/* FIXME - this will probably misbehave when suspend/resume is added */ static int vgadget_get_frame (struct usb_gadget *_gadget) { -struct timeval now; +struct timespec64 now;struct vudc *udc =usb_gadget_to_vudc (_gadget); -do_gettimeofday (&now); +ktime_get_ts64 (&now);return ((now.tv_sec - udc->start_time.tv_sec) *1000 + - (now.tv_usec - udc->start_time.tv_usec) /1000 ) + (now.tv_nsec - udc->start_time.tv_nsec) /NSEC_PER_MSEC ) &0x7FF ; }
drivers/usb/usbip/vudc_sysfs.c +1 -1
@@ -162,7 +162,7 @@ static ssize_t store_sockfd(struct device *dev, struct device_attribute *attr, udc->ud.status =SDEV_ST_USED ;spin_unlock_irq (&udc->ud.lock); -do_gettimeofday (&udc->start_time); +ktime_get_ts64 (&udc->start_time);v_start_timer (udc); udc->connected =1 ; }else {
Signature · 1 line
2.9.0
Re: [PATCH 2/2] usbip: use monotonic timestamps
Quoted 6 lines · Arnd Bergmann
This gets rid of the deprecated
do_gettimeofday()function in usbip. The comment abovevgadget_get_frame()mentions that it suffers from issues with the time jumps due to suspend and settimeofday, so I'm changing it to usektime_get_ts64()to use monotonic times that don't have this problem.
Quoted 7 lines
I couldn't tell whether we should use
CLOCK_MONOTONICorCLOCK_MONOTONIC_RAWhere, the difference being the exact rate when correcting for NTP. I picked monotonic time since it doesn't change the speed to the existing code and should be better synchronized with other machines we talk to.Signed-off-by Arnd Bergmann <[email protected]>
Looks good to me:
Signature · 3 lines
Krzysztof Opasiak Samsung R&D Institute Poland Samsung Electronics
Re: [PATCH 2/2] usbip: use monotonic timestamps
Quoted 15 lines · Krzysztof Opasiak
Quoted 12 lines · Arnd Bergmann
This gets rid of the deprecated
do_gettimeofday()function in usbip. The comment abovevgadget_get_frame()mentions that it suffers from issues with the time jumps due to suspend and settimeofday, so I'm changing it to usektime_get_ts64()to use monotonic times that don't have this problem.I couldn't tell whether we should use
CLOCK_MONOTONICorCLOCK_MONOTONIC_RAWhere, the difference being the exact rate when correcting for NTP. I picked monotonic time since it doesn't change the speed to the existing code and should be better synchronized with other machines we talk to.
Quoted 5 lines
Quoted 1 line
Signed-off-by Arnd Bergmann <[email protected]>Looks good to me:
Reviewed-by Krzysztof Opasiak <[email protected]>
thanks,
-- Shuah
Quoted 11 lines · Arnd Bergmann
vgadget_get_framereturns a frame number from 0 to 2046, which may require an expensive division operation to wrap at one lower than the usual number.I can't see any reason for this, and all other drivers wrap at a power-of-two number. My best explanation is that it was a simple typo, so I'm changing the % modulo operator into a cheaper bitmask that the other drivers use, to make it wrap after
0x7ffrather than before it.Signed-off-by Arnd Bergmann <[email protected]>
Looks good to me:
Signature · 3 lines
Krzysztof Opasiak Samsung R&D Institute Poland Samsung Electronics
Quoted 13 lines · Krzysztof Opasiak
Quoted 10 lines · Arnd Bergmann
vgadget_get_framereturns a frame number from 0 to 2046, which may require an expensive division operation to wrap at one lower than the usual number.I can't see any reason for this, and all other drivers wrap at a power-of-two number. My best explanation is that it was a simple typo, so I'm changing the % modulo operator into a cheaper bitmask that the other drivers use, to make it wrap after
0x7ffrather than before it.
Quoted 5 lines
Quoted 1 line
Signed-off-by Arnd Bergmann <[email protected]>Looks good to me:
Reviewed-by Krzysztof Opasiak <[email protected]>
thanks,
-- Shuah