On 27/05/2022 09:52, Petri Latvala wrote:
On Fri, May 27, 2022 at 08:53:04AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
On kernels without support for the feature we should skip showing the
clients header to avoid confusing users.
Simply briefly open a render node to the selected device during init and
look if the relevant fields are present in the fdinfo data.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
Issue: https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/issues/120
---
tools/intel_gpu_top.c | 22 ++++++++++++++++++++--
1 file changed, 20 insertions(+), 2 deletions(-)
diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 1984c10dca29..26986a822bb7 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -2389,6 +2389,23 @@ static void process_stdin(unsigned int timeout_us)
process_normal_stdin();
}
+static bool has_drm_fdinfo(const struct igt_device_card *card)
+{
+ struct drm_client_fdinfo info;
+ unsigned int cnt;
+ int fd;
+
+ fd = open(card->render, O_RDWR);
+ if (fd < 0)
+ return false;
+
+ cnt = igt_parse_drm_fdinfo(fd, &info);
+
+ close(fd);
+
+ return cnt > 0;
+}
+
static void show_help_screen(void)
{
printf(
@@ -2545,8 +2562,9 @@ int main(int argc, char **argv)
ret = EXIT_SUCCESS;
- clients = init_clients(card.pci_slot_name[0] ?
- card.pci_slot_name : IGPU_PCI);
+ if (has_drm_fdinfo(&card))
+ clients = init_clients(card.pci_slot_name[0] ?
+ card.pci_slot_name : IGPU_PCI);
Checked all usage of 'clients' below this, and everything handles NULL
properly.
That said, nothing seems to free() it, am I reading that correctly?
Process exit does ;) (or every loop iteration in pid aggregation mode) -
but okay, it is inconsistent that I free some bits but not all.
Anyway, that can be left for another patch, this change is
Reviewed-by: Petri Latvala <petri.latvala@xxxxxxxxx>
Thanks!
Regards,
Tvrtko
init_engine_classes(engines);
if (clients) {
clients->num_classes = engines->num_classes;
--
2.32.0
