Build failures on i386 / 32bit

[bzed]

Hi,

building on i386 (and most likely also on other 32bit architectures - see https://wiki.debian.org/ArchitectureSpecificsMemo) fails with

In file included from tcp_buffer_tuner.c:8:
1424
./tcp_buffer_tuner.skel.nobtf.h:35889:17: error: static assertion failed due to requirement 'sizeof (s->bss->bpftune_init_net) == 8': unexpected size of 'bpftune_init_net'
1425
 35889 |         _Static_assert(sizeof(s->bss->bpftune_init_net) == 8, "unexpected size of 'bpftune_init_net'");
1426
       |                        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1427
./tcp_buffer_tuner.skel.nobtf.h:35889:50: note: expression evaluates to '4 == 8'
1428
 35889 |         _Static_assert(sizeof(s->bss->bpftune_init_net) == 8, "unexpected size of 'bpftune_init_net'");
1429
       |                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~
1430
./tcp_buffer_tuner.skel.nobtf.h:35895:17: error: static assertion failed due to requirement 'sizeof (s->bss->nr_free_buffer_pages) == 8': unexpected size of 'nr_free_buffer_pages'
1431
 35895 |         _Static_assert(sizeof(s->bss->nr_free_buffer_pages) == 8, "unexpected size of 'nr_free_buffer_pages'");
1432
       |                        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1433
./tcp_buffer_tuner.skel.nobtf.h:35895:54: note: expression evaluates to '4 == 8'
1434
 35895 |         _Static_assert(sizeof(s->bss->nr_free_buffer_pages) == 8, "unexpected size of 'nr_free_buffer_pages'");
1435
       |                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~
1436
6 errors generated.

https://salsa.debian.org/debian/bpftune/-/jobs/6742374

A quick look at the code shows more usage of long/unsigned long, so I think this needs a more general fix, like using uint*_t or similar.

Or do you not want to support 32bit architectures?

[alan-maguire]

the problem with the above specific case is it's in auto-generated code (bpftool gen skeleton generates it). we could perhaps fix the type of the variable, but there may be problems lurking with kernel types that we use in BPF progs. plus my read (perhaps wrong?) of libbpf/libbpf-sys#27 suggests that 32-bit is not supported in libbpf. So I suspect the best approach would be not to support 32-bit. If you've got any thoughts on how best to specify/communicate that, do let me know. thanks!

[bzed]

Hmm, as libbpf is built in Debian for 32bit architectures and other packages use it, it can't be fully broken.

Completely untested (I also don't have an i386 vm/hardware rigt now), but bpftune compiles with this diff applied on i386:

diff --git a/include/bpftune/bpftune.bpf.h b/include/bpftune/bpftune.bpf.h
index 2d3a1b7..988bd2f 100644
--- a/include/bpftune/bpftune.bpf.h
+++ b/include/bpftune/bpftune.bpf.h
@@ -205,7 +205,7 @@ unsigned int tuner_id;
 unsigned int strategy_id;
 unsigned int bpftune_pid;
 /* init_net value used for older kernels since __ksym does not work */
-unsigned long bpftune_init_net;
+unsigned long long bpftune_init_net;

 /* TCP buffer tuning */
 #ifndef SO_SNDBUF
diff --git a/src/tcp_buffer_tuner.bpf.c b/src/tcp_buffer_tuner.bpf.c
index 2c0202f..62c535a 100644
--- a/src/tcp_buffer_tuner.bpf.c
+++ b/src/tcp_buffer_tuner.bpf.c
@@ -37,7 +37,7 @@ int kernel_page_size;
 int kernel_page_shift;
 int sk_mem_quantum;
 int sk_mem_quantum_shift;
-unsigned long nr_free_buffer_pages;
+unsigned long long nr_free_buffer_pages;

 #define tcp_tunable_corr(__id, __cookie, __newval, __tp, __field_type, __field)\
        {                                                               \
diff --git a/src/tcp_buffer_tuner.c b/src/tcp_buffer_tuner.c
index a1feb14..580f32d 100644
--- a/src/tcp_buffer_tuner.c
+++ b/src/tcp_buffer_tuner.c
@@ -89,7 +89,7 @@ int get_from_file(FILE *fp, const char *fmt, ...)
        return ret;
 }

-long nr_free_buffer_pages(bool initial)
+long long nr_free_buffer_pages(bool initial)
 {
        unsigned long nr_pages = 0;
        char *mzone = "Normal";
root@1cbc7162e97c:~/bpftune#

[alan-maguire]

I tested the above change and bpftune tests pass. would you like to submit it in a pull request? thanks!