Hi Vidya,

-----Original Message-----
From: Purkait, Soham <soham.purkait@intel.com>
Sent: 06 April 2026 10:27
To: Srinivas, Vidya <vidya.srinivas@intel.com>; igt-dev@lists.freedesktop.org;
Konieczny, Kamil <kamil.konieczny@intel.com>
Subject: Re: tools/gputop: Fix zero output when stdout is not a terminal

Hi Vidya,

On 02-04-2026 19:15, Vidya Srinivas wrote:

When gputop output is redirected to a file or pipe, such as:

   gputop -n 5 -d 1 > results.txt

Isn’t only stdout (fd 1) redirected to the file here?

I guess some thing like "echo test | ./gputop" would be returning the default
value (-1 in this case) for con_h.

Hello,

Many thanks for the review.
I will update the patch like this. Should be okay?

if (ioctl(STDOUT_FILENO, TIOCGWINSZ, &ws) == -1 &&
    ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) == -1 &&
    ioctl(STDERR_FILENO, TIOCGWINSZ, &ws) == -1) {

I assume it's better to keep this : "if (ioctl(0, TIOCGWINSZ, &ws) == -1)" as is. As it won't make any  issue if the gputop output is redirected to any file or so (which only involve stdout).  The expression  will only be true when stdin is not a terminal, may be during running gputop in a ci.
And Its better to assign the default values to con_w and con_h in main. (please see below)

    ws.ws_col = 80;
    ws.ws_row = 50;
}

Regards
Vidya

   gputop -n 3 -d 2 | grep rcs

ioctl(0, TIOCGWINSZ) fails with -1 since stdin is not a terminal.
update_console_size() then returns without setting *w and *h, leaving
con_w and con_h at their initial value of -1.

The main display loop uses 'if (lines >= con_h) break' to limit output
to the terminal height. With con_h = -1, the condition (0 >= -1) is
immediately true on the very first line, causing all client output to
be silently suppressed. The result is that gputop produces only ANSI
clear- screen escape sequences and zero actual data.

This affects anyone using gputop in automation, CI pipelines, or any
non-interactive context on Linux or Android where output is redirected
or piped.

Fix this by falling back to a default console size of 80x50 when the
ioctl fails, consistent with the existing fallback for serial consoles
(where ws_col and ws_row are both 0).

Signed-off-by: Vidya Srinivas <vidya.srinivas@intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>
---
  tools/gputop.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/tools/gputop.c b/tools/gputop.c index
9b2e8cb6f..112ec5ddb 100644
--- a/tools/gputop.c
+++ b/tools/gputop.c
@@ -505,8 +505,11 @@ static void update_console_size(int *w, int *h)
  {
  	struct winsize ws = {};

-	if (ioctl(0, TIOCGWINSZ, &ws) == -1)
+	if (ioctl(0, TIOCGWINSZ, &ws) == -1) {
+		*w = 80;
+		*h = 50;

 Imho instead of setting the values (*w = 80;*h = 50) here , it's better to initialize these as default values in main as "int con_w = 80, con_h = 50;"

Thanks, Soham

How about assigning the values during initialization in the main function ?

eg : int con_w = 80, con_h = 50;

Thanks,
Soham

  		return;
+	}

  	*w = ws.ws_col;
  	*h = ws.ws_row;