linux-kernel - [PATCH] Introduce callchains for guests

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 31 Aug 2023 16:09:02 +0800
From:   Tianyi Liu <i.pear@...look.com>
To:     peterz@...radead.org, acme@...nel.org,
        linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Cc:     mingo@...hat.com, mark.rutland@....com,
        alexander.shishkin@...ux.intel.com, jolsa@...nel.org,
        namhyung@...nel.org, irogers@...gle.com, adrian.hunter@...el.com,
        Tianyi Liu <i.pear@...look.com>
Subject: [PATCH] Introduce callchains for guests

This patch serves as the foundation for enabling callchains for guests,
(used by `perf kvm`). This functionality is useful for me, and I
noticed it holds the top spot on the perf wiki TODO list [1], so I'd like
to implement it. This patch introduces a `perf_callchain_guest` interface,
which different architectures can implement according to their own
specifications.

This is part of a series of patches. Since these patches are spread across
various modules like perf, kvm, arch/*, I plan to first submit some
foundational patches and gradually submit the complete implementation.
The full implementation can be found at [2], and it has been tested on
an x86_64 machine.

Since perf tools currently do not support callchain types with
PERF_CONTEXT_GUEST*, I temporarily used PERF_CONTEXT_KERNEL for ease of
testing and validation. This will be rectified once the perf tools patches
are applied.

Furthermore, do you think it's necessary to introduce a new guest filter
in the parameters of `get_perf_callchain`? Alternatively, should we add
two separate filters, `guest_user` and `guest_kernel`, to make these
filters entirely independent?

This is my first time submitting a patch for a new feature in the kernel,
and I would greatly appreciate any suggestions.

[1] https://perf.wiki.kernel.org/index.php/Todo
[2] https://github.com/i-Pear/linux/pull/2/files

Signed-off-by: Tianyi Liu <i.pear@...look.com>
---
 kernel/events/callchain.c | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
index 1273be84392c..eacb6ccf7034 100644
--- a/kernel/events/callchain.c
+++ b/kernel/events/callchain.c
@@ -45,6 +45,10 @@ __weak void perf_callchain_user(struct perf_callchain_entry_ctx *entry,
 {
 }

+__weak void perf_callchain_guest(struct perf_callchain_entry_ctx *entry)
+{
+}
+
 static void release_callchain_buffers_rcu(struct rcu_head *head)
 {
 	struct callchain_cpus_entries *entries;
@@ -183,6 +187,7 @@ get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
 	struct perf_callchain_entry *entry;
 	struct perf_callchain_entry_ctx ctx;
 	int rctx;
+	unsigned int guest_state;

 	entry = get_callchain_entry(&rctx);
 	if (!entry)
@@ -194,6 +199,26 @@ get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
 	ctx.contexts       = 0;
 	ctx.contexts_maxed = false;

+	guest_state = perf_guest_state();
+	if (guest_state) {
+		if (add_mark) {
+			if (guest_state & PERF_GUEST_USER)
+				/*
+				 * TODO: Change this to PERF_CONTEXT_GUEST_USER,
+				 * which is currently not recognized by perf utils.
+				 */
+				perf_callchain_store_context(&ctx, PERF_CONTEXT_KERNEL);
+			else
+				/*
+				 * TODO: Change this to PERF_CONTEXT_GUEST_KERNEL,
+				 * which is currently not recognized by perf utils.
+				 */
+				perf_callchain_store_context(&ctx, PERF_CONTEXT_KERNEL);
+		}
+		perf_callchain_guest(&ctx);
+		goto exit_put;
+	}
+
 	if (kernel && !user_mode(regs)) {
 		if (add_mark)
 			perf_callchain_store_context(&ctx, PERF_CONTEXT_KERNEL);
-- 
2.34.1