Received-SPF: pass (google.com: domain of linux-kernel+bounces-25814-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
From: Valentin Schneider <vschneid@redhat.com>
To: Yong-Xuan Wang <yongxuan.wang@sifive.com>, linux-riscv@lists.infradead.org
Cc: greentime.hu@sifive.com, vincent.chen@sifive.com, Yong-Xuan Wang
 <yongxuan.wang@sifive.com>, Paul Walmsley <paul.walmsley@sifive.com>,
 Palmer Dabbelt <palmer@dabbelt.com>, Albert Ou <aou@eecs.berkeley.edu>,
 Marc Zyngier <maz@kernel.org>, Anup Patel <apatel@ventanamicro.com>, Atish
 Patra <atishp@rivosinc.com>, Conor Dooley <conor.dooley@microchip.com>,
 Samuel Holland <samuel.holland@sifive.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/1] RISC-V: add IPI tracepoints
In-Reply-To: <20240115071246.32675-1-yongxuan.wang@sifive.com>
References: <20240115071246.32675-1-yongxuan.wang@sifive.com>
Date: Mon, 15 Jan 2024 10:41:44 +0100
Message-ID: <xhsmhh6jfhr1j.mognet@vschneid-thinkpadt14sgen2i.remote.csb>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain

On 15/01/24 07:12, Yong-Xuan Wang wrote:
> The strings used to list IPIs in /proc/interrupts are reused for tracing
> purposes. Also slightly clean up the code by removing send_ipi_single().
>

So this is adding trace events for sending & receiving IPIs.

Regarding reception, AFAICT arm/arm64 had that historically because IPIs
were handled directly by the irqchip (see rambling on the first part of the
cover letter at [1]), but now that this is no longer the case they are
obsolete: they show up in trace_irq_handler* events.

I don't know riscv IRQ handling, but riscv_ipi_set_virq_range() looks
awfully similar to arm64's set_smp_ipi_range() and does a "proper"
request_percpu_irq() for handle_IPI(). So if one wants a trace footprint of
IPI reception, there already is infrastructure for it.

Regarding sending IPIs, there now is trace_ipi_send_cpu{mask} which is in
the core kernel and thus benefits all architectures, though it
only covers IPI_RESCHEDULE, IPI_CALL_FUNC and IPI_IRQ_WORK.

Long story short, I know the ipi_raise and ipi_{entry_exit} TPs already
exist and we can't remove them, but IMO they're not that helpful because
they just give a string rather than an actual function pointer, *and* they
have to be manually added to the right place in each architecture. I'd
rather see trace_ipi_send_cpu{mask} extended to cover the missing cases.

[1]: https://lore.kernel.org/lkml/20230307143558.294354-1-vschneid@redhat.com/