Message ID | 55369624.5040703@oracle.com |
---|---|
State | RFC |
Delegated to: | David Miller |
Headers | show |
From: chris hyser <chris.hyser@oracle.com> Date: Tue, 21 Apr 2015 14:25:40 -0400 > On 4/16/2015 3:41 PM, David Miller wrote: >> From: David Miller <davem@davemloft.net> >> Date: Thu, 09 Apr 2015 19:06:47 -0400 (EDT) >>> How will this work on T3, T2, and T1 which all neither have the >>> "socket" mdesc nodes nor level 3 caches? >> >> I'm still waiting for you to resolve this Chris. >> >> I don't think it's much effort to make this change back down >> to using the level=2 cache if no level=3 cache is found. Please >> implement that and resubmit. > > So that does appear to work. This is not the patch. I will send that > out shortly but I thought I'd give you a chance to provide feedback > while I'm getting that ready. > > Here is what I see on a T2. ... > Diff from the prior patch (I also see some cleanup here I will do): Yeah that should work just as well on T3 too. BTW, there are mdesc dumps in my prtconfs repo that you can use for reference when working on things that walk the mdesc graph. -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 4/21/2015 2:25 PM, chris hyser wrote: > mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets"); > if (mp == MDESC_NODE_NULL) > - return set_sock_ids_by_cache(hp, mp); > + if (!set_sock_ids_by_cache(hp, mp, 3)) > + set_sock_ids_by_cache(hp, mp, 2); > > return set_sock_ids_by_socket(hp, mp); So I missed the return here. Will fix that. Surprised that even worked. -chrish -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 4/21/2015 2:33 PM, David Miller wrote: > From: chris hyser <chris.hyser@oracle.com> > Date: Tue, 21 Apr 2015 14:25:40 -0400 > BTW, there are mdesc dumps in my prtconfs repo that you can use > for reference when working on things that walk the mdesc graph. Cool. I did not know that. Thanks. -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/arch/sparc/kernel/mdesc.c b/arch/sparc/kernel/mdesc.c index 78beb94..75e2890 100644 --- a/arch/sparc/kernel/mdesc.c +++ b/arch/sparc/kernel/mdesc.c @@ -694,23 +694,26 @@ static void set_core_ids(struct mdesc_handle *hp) } } -static void set_sock_ids_by_cache(struct mdesc_handle *hp, u64 mp) +static int set_sock_ids_by_cache(struct mdesc_handle *hp, u64 mp, int level) { int idx = 1; + int fnd = 0; /* Identify unique sockets by looking for cpus backpointed to by - * level 3 caches. + * shared level n caches. */ mdesc_for_each_node_by_name(hp, mp, "cache") { - const u64 *level; + const u64 *cur_lvl; - level = mdesc_get_property(hp, mp, "level", NULL); - if (*level != 3) + cur_lvl = mdesc_get_property(hp, mp, "level", NULL); + if (*cur_lvl != level) continue; mark_sock_ids(hp, mp, idx); idx++; + fnd = 1; } + return fnd; } static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp) @@ -742,11 +745,12 @@ static void set_sock_ids(struct mdesc_handle *hp) u64 mp; /* If machine description exposes sockets data use it. - * Otherwise fallback to use L3 cache + * Otherwise fallback to use shared L3 or L2 caches */ mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets"); if (mp == MDESC_NODE_NULL) - return set_sock_ids_by_cache(hp, mp); + if (!set_sock_ids_by_cache(hp, mp, 3)) + set_sock_ids_by_cache(hp, mp, 2); return set_sock_ids_by_socket(hp, mp); }