From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 63CB845823; Tue, 20 Aug 2024 16:38:54 +0200 (CEST) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 3A6FB4025F; Tue, 20 Aug 2024 16:38:54 +0200 (CEST) Received: from mail-pj1-f47.google.com (mail-pj1-f47.google.com [209.85.216.47]) by mails.dpdk.org (Postfix) with ESMTP id 73C43400D6 for ; Tue, 20 Aug 2024 16:38:52 +0200 (CEST) Received: by mail-pj1-f47.google.com with SMTP id 98e67ed59e1d1-2d3c9a67eaeso3656323a91.2 for ; Tue, 20 Aug 2024 07:38:52 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=iol.unh.edu; s=unh-iol; t=1724164731; x=1724769531; darn=dpdk.org; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:from:to:cc:subject:date :message-id:reply-to; bh=rrHwZ0Y/LVu9NtZIk+yXcFBod02J8ZfY66XbFZ3mrsU=; b=bs8CzvL8s41tqoj7BH3/5kX7PHbffDYLMnfsjJO5J5WH7shu3p8e6MAUKXEkWcV7nb xwJO0nUyeb4pRLnlUqjdqDNJNePJCV8Q8PSGcTqg51sFfaZCNW+8qGwGXc6mhJE9JOv1 72ka/Mnvn6UjJJEZW/IzfeLpDXW/MGOpVZfkg= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1724164731; x=1724769531; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=rrHwZ0Y/LVu9NtZIk+yXcFBod02J8ZfY66XbFZ3mrsU=; b=R2V9PshsIymD63OUa5EDWIc3fletSaxHkg2TnGvWKlSP4kRitkV8fyeUDalbn0D/od Ngx4VFt60cQJ6Bgr9wpNgjtQ5OpvKWyA+nyRD2Z//0d2mLZkcdmwnlhXrzXUkxILGvvi jlDY9xV8s/2cyQmOWJaJVl3WJGzwVukoGvXky53gZvOAnOI/84pnEvTSeWNXJkUtcdk7 dcEN2Wmd/DI88WDxUXEbRKX44djR7Vhc6G0r6s5nw8HYrBkdtFryaiRsptQ0fYBEi/dh iQzMtwQR5gvjMD4S9/YEVh+pfeNd9K30KPvJWISH7b6AF7mdLhHbAIyrvUCppJieew6r 5vZA== X-Forwarded-Encrypted: i=1; AJvYcCXuHdAAipkeNowGucI3wAg/i3qClIKlNaF5SwS9QHHys7x+vU3q5pp3QvSwfsEx5prTxyc3vVrZ0VSvlH0= X-Gm-Message-State: AOJu0YwiGxt+tERG0kQ4Qf3el0XHzeTG1cmb85dlACb4GhIo9wj3uq8N wVulnDd88sRrym24N6LngVupRXA4SWPla1IorBq+gGTMp69KqAPuWA1Mog9CHGtVS3138sF1ToM NTdXtC1f3LhYBcBL5e9OkV8FbdKb2IE+s4NR5iX4G+ZbfQZ3XLsrRzw== X-Google-Smtp-Source: AGHT+IGgNrCF2rTdDj7mRq6bpaj63T6bsY/SOkufLBzbO3SuBKhIH0Z3VwkMCiLq2DTwjR6kTur1Bjy/MWdb36NQrLg= X-Received: by 2002:a17:90b:10d:b0:2c3:2557:3de8 with SMTP id 98e67ed59e1d1-2d3e02e21c2mr12665111a91.33.1724164731338; Tue, 20 Aug 2024 07:38:51 -0700 (PDT) MIME-Version: 1.0 References: <20240514201436.2496-1-jspewock@iol.unh.edu> <20240709175341.183888-1-jspewock@iol.unh.edu> <20240709175341.183888-2-jspewock@iol.unh.edu> In-Reply-To: From: Jeremy Spewock Date: Tue, 20 Aug 2024 10:38:40 -0400 Message-ID: Subject: Re: [PATCH v7 1/2] dts: add methods for modifying MTU to testpmd shell To: =?UTF-8?Q?Juraj_Linke=C5=A1?= Cc: thomas@monjalon.net, Honnappa.Nagarahalli@arm.com, yoan.picchi@foss.arm.com, paul.szczepanek@arm.com, probb@iol.unh.edu, Luca.Vizzarro@arm.com, npratte@iol.unh.edu, wathsala.vithanage@arm.com, dev@dpdk.org Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org On Tue, Aug 20, 2024 at 9:05=E2=80=AFAM Juraj Linke=C5=A1 wrote: > > I'm trying to use this patch for the capabilities series. It works as I > need it to, so we just need to coordinate a bit to use this one patch > for both series. > > > diff --git a/dts/framework/remote_session/testpmd_shell.py b/dts/framew= ork/remote_session/testpmd_shell.py > > > @@ -82,12 +84,82 @@ class TestPmdForwardingModes(StrEnum): > > recycle_mbufs =3D auto() > > > > > > +T =3D TypeVarTuple("T") # type: ignore[misc] > > + > > + > > +class stop_then_start_port: > > Is there a particular reason why this is a class and not a function? We > can pass arguments even with a function (in that case we need two inner > wrapper functions). > There isn't really, I made it this way really just because I felt that it was easier to process at the time than the doubly nested functions. > In my capabilities patch, I've made a testpmd specific decorator a > static method to signify that the decorator is tied to testpmd methods. > This made sense to me, but maybe we don't want to do that. I actually also much prefer the static method approach, but at the time didn't think of it. Since then however I've seen it in other patches and agree that it makes the association more clear. > > > + """Decorator that stops a port, runs decorated function, then star= ts the port. > > + > > + The function being decorated must be a method defined in :class:`T= estPmdShell` that takes a > > + port ID (as an int) as its first parameter. The port ID will be pa= ssed into > > + :meth:`~TestPmdShell._stop_port` and :meth:`~TestPmdShell._start_p= ort` so that the correct port > > + is stopped/started. > > + > > + Note that, because this decorator is presented through a class to = allow for passing arguments > > + into the decorator, the class must be initialized when decorating = functions. This means that, > > + even when not modifying any arguments, the signature for decoratin= g with this class must be > > + "@stop_then_start_port()". > > + > > + Example usage on testpmd methods:: > > + > > + @stop_then_start_port() > > + def ex1(self, port_id, verify=3DTrue) > > + pass > > + > > + @stop_then_start_port(verify=3DFalse) > > + def ex2(self, port_id, verify=3DTrue) > > + pass > > + > > + Attributes: > > + verify: Whether to verify the stopping and starting of the por= t. > > + """ > > + > > + verify: bool > > + > > + def __init__(self, verify: bool =3D True) -> None: > > + """Store decorator options. > > + > > + Args: > > + verify: If :data:`True` the stopping/starting of ports wil= l be verified, otherwise they > > + will it won't. Defaults to :data:`True`. > > + """ > > + self.verify =3D verify > > + > > + def __call__( > > + self, func: Callable[["TestPmdShell", int, *T], None] # type:= ignore[valid-type] > > + ) -> Callable[["TestPmdShell", int, *T], None]: # type: ignore[va= lid-type] As a note, this typing monster that I made was also handled in a much more elegant way in Luca's patch (mentioned below) that I think even retains the variable names for added clarity, whereas this only shows you a tuple of what types it expects when calling the method and gives no hints regarding what they are. Definitely not super useful. > > + """Wrap decorated method. > > + > > + Args: > > + func: Decorated method to wrap. > > + > > + Returns: > > + Function that stops a port, runs the decorated method, the= n starts the port. > > + """ > > + > > + def wrapper(shell: "TestPmdShell", port_id: int, *args, **kwar= gs) -> None: > > + """Function that wraps the instance method of :class:`Test= PmdShell`. > > + > > + Args: > > + shell: Instance of the shell containing the method to = decorate. > > + port_id: ID of the port to stop/start. > > + """ > > + shell._stop_port(port_id, self.verify) > > + func(shell, port_id, *args, **kwargs) > > + shell._start_port(port_id, self.verify) > > Is it possible that the port will be stopped when the decorator is > called? In that case, we would start a port that's expected to be > stopped at the end. I think we should figure out what the port state is > and only start it if it started out as started. Luca has a patch that I think actually handles this problem [1]. He had the idea of making two decorators, one for a method that requires ports to be stopped, and another that signifies requiring ports to be started. This allows you to know the state of the port and only modify the state if needed. I mentioned on his patch that I actually like his approach more than this one, but the one aspect that it was missing compared to this was the verify parameter that we decided to make an argument to the decorator here. I guess the other main difference between these two patches is that this one tries to stop the specific port that needs modification whereas Luca's patch simply stops all ports. This might be a distinction that we are fine without honestly and it also cleans up the types a bit. Let me know what you think, but I personally think that these two patches should be combined into one based on which approach people prefer. As mentioned, I like Luca's approach more. [1] https://patchwork.dpdk.org/project/dpdk/patch/20240806124642.2580828-5-= luca.vizzarro@arm.com/ > > > + > > + return wrapper > > + > > + > > class TestPmdShell(InteractiveShell): > > """Testpmd interactive shell. > > > > The testpmd shell users should never use > > the :meth:`~.interactive_shell.InteractiveShell.send_command` met= hod directly, but rather > > - call specialized methods. If there isn't one that satisfies a need= , it should be added. > > + call specialized methods. If there isn't one that satisfies a need= , it should be added. Methods > > + of this class can be optionally decorated by :func:`~stop_then_sta= rt_port` if their first > > + parameter is the ID of a port in testpmd. This decorator will stop= the port before running the > > + method and then start it again once the method is finished. > > > > This explanation is more from the "this decorator exists and does this" > point of view, but I think a more fitting explanation would be how to > configure ports using the decorator, something like: > "In order to configure ports in TestPmd, the ports (may) need to be > stopped" and so on. This would be more of a "this how you implement > configuration in this class" explanation. This is a good thought, it probably would be more useful if it followed the second perspective. > > > Attributes: > > number_of_ports: The number of ports which were allowed on th= e command-line when testpmd > > @@ -227,6 +299,63 @@ def set_forward_mode(self, mode: TestPmdForwarding= Modes, verify: bool =3D True): > > f"Test pmd failed to set fwd mode to {mode.value}" > > ) > > > > + def _stop_port(self, port_id: int, verify: bool =3D True) -> None: > > + """Stop port with `port_id` in testpmd. > > + > > + Depending on the PMD, the port may need to be stopped before c= onfiguration can take place. > > What is this dependence? How do we determine which PMDs need this? I > guess we don't really need to concern ourselves with this as mentioned > in set_port_mtu(). I'm not sure if there is a way to consistently distinguish between PMDs that need it and ones that don't, but I know that vfio-pci, for example, can update the MTU of the port without stopping it but a bifurcated driver like mlx5_core needs the port to be stopped first. I think, in truth, the port always needs to be stopped for this configuration to happen but the more likely difference is that some PMDs will just stop the port for you automatically. > > I think we should actually remove this line. It doesn't really add much > (and the same thing is mentioned in set_port_mtu()) and the method could > actually used in other contexts. Ack. > > > + This method wraps the command needed to properly stop ports an= d take their link down. > > + > > + Raises: > > + InteractiveCommandExecutionError: If `verify` is :data:`Tr= ue` and the port did not > > + successfully stop. > > + """ > > + stop_port_output =3D self.send_command(f"port stop {port_id}") > > + if verify and ("Done" not in stop_port_output): > > + self._logger.debug(f"Failed to stop port {port_id}. Output= was:\n{stop_port_output}") > > + raise InteractiveCommandExecutionError(f"Test pmd failed t= o stop port {port_id}.") > > + > > + def _start_port(self, port_id: int, verify: bool =3D True) -> None= : > > + """Start port with `port_id` in testpmd. > > + > > + Because the port may need to be stopped to make some configura= tion changes, it naturally > > + follows that it will need to be started again once those chang= es have been made. > > The same reasoning applies here, we don't really need this sentence. > However, we could add the other sentence about the method wrapping the > command to unify the doctrings a bit. Ack.