Skip to contentSkip to conversation

bsd-user/main: Allow setting tb-size

PATCH In review Ilya Leoshkevich 3 participants 7 messages Jul 31, 2024 – Aug 1, 2024

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bsd-user/main: Allow setting tb-size

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already handled.

---
 bsd-user/main.c | 9 +++++++++
 1 file changed, 9 insertions(+)
bsd-user/main.c +9 -0
index cc980e6f401..7c230b0c7a5 100644
@@ -60,6 +60,7 @@ uintptr_t qemu_host_page_size;
 intptr_t qemu_host_page_mask;
 
 static bool opt_one_insn_per_tb;
+static unsigned long opt_tb_size;
 uintptr_t guest_base;
 bool have_guest_base;
 /*
@@ -169,6 +170,7 @@ static void usage(void)
            "                  (use '-d help' for a list of log items)\n"
            "-D logfile        write logs to 'logfile' (default stderr)\n"
            "-one-insn-per-tb  run with one guest instruction per emulated TB\n"
+           "-tb-size size     TCG translation block cache size\n"
            "-strace           log system calls\n"
            "-trace            [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
            "                  specify tracing options\n"
@@ -387,6 +389,11 @@ int main(int argc, char **argv)
             seed_optarg = optarg;
         } else if (!strcmp(r, "one-insn-per-tb")) {
             opt_one_insn_per_tb = true;
+        } else if (!strcmp(r, "tb-size")) {
+            r = argv[optind++];
+            if (qemu_strtoul(r, NULL, 0, &opt_tb_size)) {
+                usage();
+            }
         } else if (!strcmp(r, "strace")) {
             do_strace = 1;
         } else if (!strcmp(r, "trace")) {
@@ -452,6 +459,8 @@ int main(int argc, char **argv)
         accel_init_interfaces(ac);
         object_property_set_bool(OBJECT(accel), "one-insn-per-tb",
                                  opt_one_insn_per_tb, &error_abort);
+        object_property_set_int(OBJECT(accel), "tb-size",
+                                opt_tb_size, &error_abort);
         ac->init_machine(NULL);
     }
 
Signature · 1 line
2.45.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 9 lines · Ilya Leoshkevich

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already handled.

---
  bsd-user/main.c | 9 +++++++++
  1 file changed, 9 insertions(+)
Reviewed-by on v1 1/1
#2 lore raw
In reply to Ilya Leoshkevich
Quoted 4 lines · Ilya Leoshkevich

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already handled.

Cool! Are you using bsd-user and need this for some reason? Or is this purely theoretical? Is there a larger context I can read about somewhere?

I'll merge it either way (so none of the above is a criticism, I'm genuinely curious) , but I don't get too many bsd-user fixes and this one is unusual.

Quoted 41 lines
---
 bsd-user/main.c | 9 +++++++++
 1 file changed, 9 insertions(+)
bsd-user/main.c +1 -0
index cc980e6f401..7c230b0c7a5 100644
@@ -60,6 +60,7 @@ uintptr_t qemu_host_page_size;
 intptr_t qemu_host_page_mask;
 
 static bool opt_one_insn_per_tb;
+static unsigned long opt_tb_size;
 uintptr_t guest_base;
 bool have_guest_base;
 /*
@@ -169,6 +170,7 @@ static void usage(void)
            "                  (use '-d help' for a list of log items)\n"
            "-D logfile        write logs to 'logfile' (default stderr)\n"
            "-one-insn-per-tb  run with one guest instruction per emulated

TB\n"
+ "-tb-size size TCG translation block cache size\n"
"-strace log system calls\n"
"-trace
[[enable=]<pattern>][,events=<file>][,file=<file>]\n"
" specify tracing options\n"

bsd-user/main.c +5 -0
@@ -387,6 +389,11 @@ int main(int argc, char **argv)
             seed_optarg = optarg;
         } else if (!strcmp(r, "one-insn-per-tb")) {
             opt_one_insn_per_tb = true;
+        } else if (!strcmp(r, "tb-size")) {
+            r = argv[optind++];
+            if (qemu_strtoul(r, NULL, 0, &opt_tb_size)) {
+                usage();
+            }
         } else if (!strcmp(r, "strace")) {
             do_strace = 1;
         } else if (!strcmp(r, "trace")) {

bsd-user/main.c int main(int argc, char **argv) · line 466

Quoted 8 lines
bsd-user/main.c +2 -0
@@ -452,6 +459,8 @@ int main(int argc, char **argv)
         accel_init_interfaces(ac);
         object_property_set_bool(OBJECT(accel), "one-insn-per-tb",
                                  opt_one_insn_per_tb, &error_abort);
+        object_property_set_int(OBJECT(accel), "tb-size",
+                                opt_tb_size, &error_abort);
         ac->init_machine(NULL);
     }
 

I'll queue this to my bsd-user-2024-q3-2 branch. I hope to land it, just after 9.1.0 release.

Warner

#3 lore raw
Quoted 9 lines · Warner Losh

On Wed, Jul 31, 2024 at 8:45 AM Ilya Leoshkevich <[email protected] <mailto:[email protected]>> wrote:

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already handled.

Cool! Are you using bsd-user and need this for some reason? Or is this purely theoretical? Is there a larger context I can read about somewhere?

Trying to keep user interface parity between linux/bsd.

Ideally this duplication should be unified in common-user/.

Quoted 10 lines

I'll merge it either way (so none of the above is a criticism, I'm genuinely curious) , but I don't get too many bsd-user fixes and this one is unusual.

<mailto:[email protected]>>

<mailto:[email protected]>>
---
 bsd-user/main.c | 9 +++++++++
 1 file changed, 9 insertions(+)

#4 lore raw
Quoted 9 lines · Philippe Mathieu-Daudé
Quoted 8 lines · Warner Losh

On Wed, Jul 31, 2024 at 8:45 AM Ilya Leoshkevich <[email protected] <mailto:[email protected]>> wrote:

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already handled.

Quoted 7 lines
Quoted 2 lines

Cool! Are you using bsd-user and need this for some reason? Or is this purely theoretical? Is there a larger context I can read about somewhere?

Trying to keep user interface parity between linux/bsd.

Ideally this duplication should be unified in common-user/.

I'd love that.

Anyway, both of these patches queued to my branch

Quoted 14 lines
Quoted 1 line

I'll merge it either way (so none of the above is a criticism, I'm

genuinely

Quoted 1 line

curious) , but I don't get too many bsd-user fixes and this one is

unusual.

Quoted 8 lines

<mailto:[email protected]>>

<mailto:[email protected]>>
---
bsd-user/main.c | 9 +++++++++
1 file changed, 9 insertions(+)

#5 lore raw
In reply to Warner Losh
Quoted 16 lines · Warner Losh

On Wed, Jul 31, 2024 at 8: 45 AM Ilya Leoshkevich
<iii@ linux. ibm. com> wrote: While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already

Quoted 5 lines · Ilya Leoshkevich

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how
one-insn-per-tb is already handled.

Quoted 4 lines

Cool! Are you using bsd-user and need this for some reason? Or is this
purely theoretical? Is there a larger context I can read about somewhere?

I needed this on Linux in order to debug an issue where I suspected full TB invalidation may be an issue.
It turned out to be something completely different, but I found it
useful: setting it to, e.g., 4096 makes full TB invalidation very rare, so if a problem is still reproducible, then the root causes is something else.
Philippe suggested to implement this for BSD as well in order to keep the interfaces in sync.

[...]

#6 lore raw
Quoted 18 lines · Ilya Leoshkevich

On Wed, 2024-07-31 at 15:21 -0600, Warner Losh wrote:

On Wed, Jul 31, 2024 at 8: 45 AM Ilya Leoshkevich <iii@ linux. ibm. com>
wrote: While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already
On Wed, Jul 31, 2024 at 8:45 AM Ilya Leoshkevich <[email protected]> wrote:

While qemu-system can set tb-size using -accel tcg,tb-size=n, there is no similar knob for qemu-bsd-user. Add one in a way similar to how one-insn-per-tb is already handled.

Cool! Are you using bsd-user and need this for some reason? Or is this purely theoretical? Is there a larger context I can read about somewhere?

Quoted 8 lines

I needed this on Linux in order to debug an issue where I suspected full TB invalidation may be an issue.
It turned out to be something completely different, but I found it useful:
setting it to, e.g., 4096 makes full TB invalidation very rare, so if a problem is still reproducible, then the root causes is something else. Philippe suggested to implement this for BSD as well in order to keep the interfaces in sync.

Excellent! Thank you for taking the time to do this! And the other bug fix. Both have been queued to my first post 9.1 pull branch.

Warner