Re: [Y2038] [PATCH v2] Staging: fbtbt: Replace timespec with ktime_t

Hi,

On Wed, Sep 30, 2015 at 11:52 PM, Arnd Bergmann arnd@arndb.de wrote:

On Wednesday 30 September 2015 18:10:49 Ksenija Stanojevic wrote:

...

struct timespec will overflow in year 2038, so replace it with ktime_t. And replace functions that use struct timespec, timespec_sub with ktime_sub. Also use monotonic time instead of real time, by replacing getnstimeofday with ktime_get, to be more robust against leap seconds and settimeofday() calls. diff --git a/drivers/staging/fbtft/fbtft-core.c b/drivers/staging/fbtft/fbtft-core.c index 7f5fa3d..a1645e1 100644 --- a/drivers/staging/fbtft/fbtft-core.c +++ b/drivers/staging/fbtft/fbtft-core.c @@ -365,16 +365,15 @@ static void fbtft_update_display(struct fbtft_par *par, unsigned start_line, unsigned end_line) { size_t offset, len;

• struct timespec ts_start, ts_end, ts_fps, ts_duration;
  

• long fps_ms, fps_us, duration_ms, duration_us;
  

• long fps, throughput;
  

• ktime_t ts_start, ts_end, ts_fps;
  

• long  long fps, throughput;
  

Here you declare fps and throughput as 'long long', which causes problems later:

...

@@ -411,30 +410,22 @@ static void fbtft_update_display(struct fbtft_par *par, unsigned start_line, __func__);

  if (unlikely(timeit)) {

•         getnstimeofday(&ts_end);
  

•         if (par->update_time.tv_nsec == 0 && par->update_time.tv_sec == 0) {
  

•                 par->update_time.tv_sec = ts_start.tv_sec;
  

•                 par->update_time.tv_nsec = ts_start.tv_nsec;
  

•         }
  

•         ts_fps = timespec_sub(ts_start, par->update_time);
  

•         par->update_time.tv_sec = ts_start.tv_sec;
  

•         par->update_time.tv_nsec = ts_start.tv_nsec;
  

•         fps_ms = (ts_fps.tv_sec * 1000) + ((ts_fps.tv_nsec / 1000000) % 1000);
  

•         fps_us = (ts_fps.tv_nsec / 1000) % 1000;
  

•         fps = fps_ms * 1000 + fps_us;
  

•         ts_end = ktime_get();
  

•         if (par->update_time.tv64 == 0)
  

•                 par->update_time = ts_start;
  

It's better not to access the 'tv64' field of the ktime_t directly, this is supposed to be hidden. Just use ktime_to_ns() to do the same thing.

...

•         ts_fps = ktime_sub(ts_start, par->update_time);
  

•         par->update_time = ts_start;
  

•         fps = ktime_to_us(ts_fps);
  

This can be written slightly simpler using the ktime_us_delta() function, like you do below.

...

          fps = fps ? 1000000 / fps : 0;

•         ts_duration = timespec_sub(ts_end, ts_start);
  

•         duration_ms = (ts_duration.tv_sec * 1000) + ((ts_duration.tv_nsec / 1000000) % 1000);
  

•         duration_us = (ts_duration.tv_nsec / 1000) % 1000;
  

•         throughput = duration_ms * 1000 + duration_us;
  

•         throughput = ktime_us_delta(ts_end, ts_start);
          throughput = throughput ? (len * 1000) / throughput : 0;
          throughput = throughput * 1000 / 1024;
  

As mentioned above, throughput is a 64-bit 'long long', so the last line of the context will result in a 64-bit division, which is not allowed in 32-bit kernels.

This is a bit hard to detect, and the most reliable way to find issues like this is to compile the kernel for both a 32-bit target and a 64-bit target (on x86, just turn CONFIG_64BIT on/off) to see all the compile-time errors and warnings.

In my local repository I modified my .config file so that CONFIG_64BIT is not set, and after that I recompiled all directory, but I don't get any errors/warning at compile-time. Also I separetly compiled this specific file but still no warnings My .config looks something like this:

# # Automatically generated file; DO NOT EDIT. # Linux/x86 4.3.0-rc3 Kernel Configuration # # CONFIG_64BIT is not set CONFIG_X86_32=y CONFIG_X86=y CONFIG_INSTRUCTION_DECODER=y CONFIG_PERF_EVENTS_INTEL_UNCORE=y CONFIG_OUTPUT_FORMAT="elf32-i386" CONFIG_ARCH_DEFCONFIG="arch/x86/configs/i386_defconfig" . . .

Should I change my working kernel or .config file is just enough?

Thanks, Ksenija

    Arnd