From mboxrd@z Thu Jan 1 00:00:00 1970 From: Boris Brezillon Subject: Re: [PATCH v8 2/4] dt-bindings: Add bindings for SPI NAND devices Date: Fri, 1 Jun 2018 23:07:20 +0200 Message-ID: <20180601230720.5619424e@bbrezillon> References: <20180601131400.17634-1-boris.brezillon@bootlin.com> <20180601131400.17634-3-boris.brezillon@bootlin.com> <20180601190902.17cf5f6f@bbrezillon> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: "linux-mtd" Errors-To: linux-mtd-bounces+gldm-linux-mtd-36=gmane.org@lists.infradead.org To: Geert Uytterhoeven Cc: Mark Rutland , "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" , Pawel Moll , Ian Campbell , Richard Weinberger , Kumar Gala , Rob Herring , linux-spi , Marek Vasut , Mark Brown , MTD Maling List , Miquel Raynal , Brian Norris , David Woodhouse List-Id: devicetree@vger.kernel.org On Fri, 1 Jun 2018 19:40:00 +0200 Geert Uytterhoeven wrote: > Hi Boris, > > On Fri, Jun 1, 2018 at 7:09 PM, Boris Brezillon > wrote: > > On Fri, 1 Jun 2018 16:42:26 +0200 > > Geert Uytterhoeven wrote: > >> On Fri, Jun 1, 2018 at 3:13 PM, Boris Brezillon > >> wrote: > >> > Add bindings for SPI NAND chips. > > >> > --- /dev/null > >> > +++ b/Documentation/devicetree/bindings/mtd/spi-nand.txt > >> > @@ -0,0 +1,27 @@ > >> > +SPI NAND flash > >> > + > >> > +Required properties: > >> > +- compatible: should be "spi-nand" > >> > +- reg: should encode the chip-select line used to access the NAND chip > >> > + > >> > +Optional properties > >> > +- spi-max-frequency: maximum frequency of the SPI bus the chip can operate at. > >> > + This should encode board limitations (i.e. max freq can't > >> > + be achieved due to crosstalk on IO lines). > >> > + When unspecified, the driver assumes the chip can run at > >> > + the max frequency defined in the spec (information > >> > + extracted chip detection time). > >> > >> This is a standard property according to > >> Documentation/devicetree/bindings/spi/spi-bus.txt. Can't you just refer > >> to that file, or just omit it, as it applies to all SPI slaves anyway? > > > > The thing is, the maximum frequency supported by a SPI NAND is directly > > encoded in the NAND device ID and can be auto-detected. Why should we > > define spi-max-frequency in the DT when we can automatically detect > > this information? The only reason one might want to override > > Because that's how we dealt with this traditionally. Or at least I thought so. > But include/linux/spi/spi.h says: > > * @max_speed_hz: Maximum clock rate to be used with this chip > * (on this board); may be changed by the device's driver. > * The spi_transfer.speed_hz can override this for each transfer. > > and many drivers seem to do so. And that's wrong, isn't it? I mean, if you have a constraint at the board level, why should we allow the SPI device driver to override this value? We should have: max(controller_max, device_max, board_max) and not (this is my understanding of how SPI max-freq selection works) if (board_max) max = board_max; else max = controller_max; if (device_max) max = device_max; > > > spi-max-frequency is when the board design impose such restrictions, > > hence the precision I give here. > > Which is exactly the meaning of the standard property, isn't it? Maybe it is how people expect the max-freq selection to work, but in practice it doesn't seem to work like that (see spi_setup() and how the value passed by the device driver overrides everything without even checking if the controller and board constraints impose a lower freq). > > So I think it can just be omitted here anyway. If it's present, the SPI > core will make sure it's taken into account. Yes, I should probably just refer to spi-bus.txt here. > > >> > +- spi-tx-bus-width: The bus width (number of data wires) that is used for MOSI. > >> > + Only encodes the board constraints (i.e. when not all IO > >> > + signals are routed on the board). Device constraints are > >> > + extracted when detecting the chip, and controller > >> > + constraints are exposed by the SPI mem controller. If this > >> > + property is missing that means no constraint at the board > >> > + level. > >> > +- spi-rx-bus-width: The bus width (number of data wires) that is used for MISO. > >> > + Only encodes the board constraints (i.e. when not all IO > >> > + signals are routed on the board). Device constraints are > >> > + extracted when detecting the chip, and controller > >> > + constraints are exposed by the SPI mem controller. If this > >> > + property is missing that means no constraint at the board > >> > + level. > >> > >> This does not match Documentation/devicetree/bindings/spi/spi-bus.txt, > >> which says the default is 1. > > > > Yes, I know. > > > >> As these properties are handled by the SPI core in of_spi_parse_dt, why > >> would you want to deviate? > > > > Because, again, this information can be extracted from the NAND ID, and > > the only reason we might want to override the information extracted > > from the NAND ID is when the board design adds extra restrictions > > (like, only 2 SIO lines wired on the 4 available). > > In struct spi_device, this is specified using the SPI_[RT]_X_{DUAL,QUAD} > bits in the mode field. Documentation says: > > * @mode: The spi mode defines how data is clocked out and in. > * This may be changed by the device's driver. > * The "active low" default for chipselect mode can be overridden > * (by specifying SPI_CS_HIGH) as can the "MSB first" default for > * each word in a transfer (by specifying SPI_LSB_FIRST). > > So this may also be specified by the device driver, but it seems no driver > does so for the dual/quad bits, except for drivers/mtd/spi-nor/nxp-spifi.c > (for rx only). Same problem as for spi-max-frequency, the driver can override the value that was specified in the DT, and the core does not seem to check if the board or controller constraint were stronger. > > But here we do have the generic SPI DT bindings saying the default is 1. > So personally, I wouldn't go for the option of the least surprise, and > don't deviate from the generic bindings. I do have a problem with this approach: we are in the process of converting existing spi-nor controller drivers to the SPI mem model in order to allow the same controller to indifferently control a SPI NAND or a SPI NOR. Such drivers had their own bindings which most of the time was close to the bindings described in spi-bus.txt with a few exceptions. One of these exception is that drivers tend to not explicitly describe the bus width but still use the max buswidth supported by the controller (QuadSPI) to interface with NOR chips. When we convert those drivers we have 2 choices: 1/ make existing setup work in a degraded mode (1-1-1) until they update their DT to explicitly specify the spi-{rx,tx}-bus-width 2/ consider that when spi-{rx,tx}-bus-width is missing the device should work in the fastest possible mode matching the device+controller constraints I went for option #2, but maybe #1 is fine. Note that spi-nand is not a problem here because this is a new binding, but I thought it would be better to clarify the meaning of these props before people start (ab)using them. > > >> Commenting to the question in the cover letter: what would be the > >> purpose of spi-max-bus-width? > > > > Defining how many IO lines are wired on the board design. If this > > property is missing we would assume all IO lines are wired and the > > restrictions would be negotiated between the controller and the device > > without requiring explicit description in the DT. > > Which is exactly the meaning of the standard property, except for the > default of all vs. 1. And that's a huge difference. Say you have a board design and several equivalence for the SPI NAND chip, one is supporting mode 1-1-4 (AKA QuadSPI) and the other one is supporting only mode 1-1-1 (Single SPI). With your solution we'd have to have 2 DTs or force all models to operate in 1-1-1 mode. > > I'll have to defer to Mark (broonie) for his opinion about deviating from > the way this is handled traditionally, and assuming different defaults... I'm not saying that we should blindly change the meaning of those existing props, I just wanted to start a discussion to see how the problem I'm mentioning here should be handled (addition of new props is something I'm perfectly fine with). ______________________________________________________ Linux MTD discussion mailing list http://lists.infradead.org/mailman/listinfo/linux-mtd/ From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mail.bootlin.com ([62.4.15.54]) by bombadil.infradead.org with esmtp (Exim 4.90_1 #2 (Red Hat Linux)) id 1fOrGw-0006lu-Eu for linux-mtd@lists.infradead.org; Fri, 01 Jun 2018 21:07:44 +0000 Date: Fri, 1 Jun 2018 23:07:20 +0200 From: Boris Brezillon To: Geert Uytterhoeven Cc: David Woodhouse , Brian Norris , Marek Vasut , Richard Weinberger , MTD Maling List , Miquel Raynal , Mark Brown , linux-spi , Rob Herring , Pawel Moll , Mark Rutland , Ian Campbell , Kumar Gala , "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" Subject: Re: [PATCH v8 2/4] dt-bindings: Add bindings for SPI NAND devices Message-ID: <20180601230720.5619424e@bbrezillon> In-Reply-To: References: <20180601131400.17634-1-boris.brezillon@bootlin.com> <20180601131400.17634-3-boris.brezillon@bootlin.com> <20180601190902.17cf5f6f@bbrezillon> MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit List-Id: Linux MTD discussion mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Fri, 1 Jun 2018 19:40:00 +0200 Geert Uytterhoeven wrote: > Hi Boris, > > On Fri, Jun 1, 2018 at 7:09 PM, Boris Brezillon > wrote: > > On Fri, 1 Jun 2018 16:42:26 +0200 > > Geert Uytterhoeven wrote: > >> On Fri, Jun 1, 2018 at 3:13 PM, Boris Brezillon > >> wrote: > >> > Add bindings for SPI NAND chips. > > >> > --- /dev/null > >> > +++ b/Documentation/devicetree/bindings/mtd/spi-nand.txt > >> > @@ -0,0 +1,27 @@ > >> > +SPI NAND flash > >> > + > >> > +Required properties: > >> > +- compatible: should be "spi-nand" > >> > +- reg: should encode the chip-select line used to access the NAND chip > >> > + > >> > +Optional properties > >> > +- spi-max-frequency: maximum frequency of the SPI bus the chip can operate at. > >> > + This should encode board limitations (i.e. max freq can't > >> > + be achieved due to crosstalk on IO lines). > >> > + When unspecified, the driver assumes the chip can run at > >> > + the max frequency defined in the spec (information > >> > + extracted chip detection time). > >> > >> This is a standard property according to > >> Documentation/devicetree/bindings/spi/spi-bus.txt. Can't you just refer > >> to that file, or just omit it, as it applies to all SPI slaves anyway? > > > > The thing is, the maximum frequency supported by a SPI NAND is directly > > encoded in the NAND device ID and can be auto-detected. Why should we > > define spi-max-frequency in the DT when we can automatically detect > > this information? The only reason one might want to override > > Because that's how we dealt with this traditionally. Or at least I thought so. > But include/linux/spi/spi.h says: > > * @max_speed_hz: Maximum clock rate to be used with this chip > * (on this board); may be changed by the device's driver. > * The spi_transfer.speed_hz can override this for each transfer. > > and many drivers seem to do so. And that's wrong, isn't it? I mean, if you have a constraint at the board level, why should we allow the SPI device driver to override this value? We should have: max(controller_max, device_max, board_max) and not (this is my understanding of how SPI max-freq selection works) if (board_max) max = board_max; else max = controller_max; if (device_max) max = device_max; > > > spi-max-frequency is when the board design impose such restrictions, > > hence the precision I give here. > > Which is exactly the meaning of the standard property, isn't it? Maybe it is how people expect the max-freq selection to work, but in practice it doesn't seem to work like that (see spi_setup() and how the value passed by the device driver overrides everything without even checking if the controller and board constraints impose a lower freq). > > So I think it can just be omitted here anyway. If it's present, the SPI > core will make sure it's taken into account. Yes, I should probably just refer to spi-bus.txt here. > > >> > +- spi-tx-bus-width: The bus width (number of data wires) that is used for MOSI. > >> > + Only encodes the board constraints (i.e. when not all IO > >> > + signals are routed on the board). Device constraints are > >> > + extracted when detecting the chip, and controller > >> > + constraints are exposed by the SPI mem controller. If this > >> > + property is missing that means no constraint at the board > >> > + level. > >> > +- spi-rx-bus-width: The bus width (number of data wires) that is used for MISO. > >> > + Only encodes the board constraints (i.e. when not all IO > >> > + signals are routed on the board). Device constraints are > >> > + extracted when detecting the chip, and controller > >> > + constraints are exposed by the SPI mem controller. If this > >> > + property is missing that means no constraint at the board > >> > + level. > >> > >> This does not match Documentation/devicetree/bindings/spi/spi-bus.txt, > >> which says the default is 1. > > > > Yes, I know. > > > >> As these properties are handled by the SPI core in of_spi_parse_dt, why > >> would you want to deviate? > > > > Because, again, this information can be extracted from the NAND ID, and > > the only reason we might want to override the information extracted > > from the NAND ID is when the board design adds extra restrictions > > (like, only 2 SIO lines wired on the 4 available). > > In struct spi_device, this is specified using the SPI_[RT]_X_{DUAL,QUAD} > bits in the mode field. Documentation says: > > * @mode: The spi mode defines how data is clocked out and in. > * This may be changed by the device's driver. > * The "active low" default for chipselect mode can be overridden > * (by specifying SPI_CS_HIGH) as can the "MSB first" default for > * each word in a transfer (by specifying SPI_LSB_FIRST). > > So this may also be specified by the device driver, but it seems no driver > does so for the dual/quad bits, except for drivers/mtd/spi-nor/nxp-spifi.c > (for rx only). Same problem as for spi-max-frequency, the driver can override the value that was specified in the DT, and the core does not seem to check if the board or controller constraint were stronger. > > But here we do have the generic SPI DT bindings saying the default is 1. > So personally, I wouldn't go for the option of the least surprise, and > don't deviate from the generic bindings. I do have a problem with this approach: we are in the process of converting existing spi-nor controller drivers to the SPI mem model in order to allow the same controller to indifferently control a SPI NAND or a SPI NOR. Such drivers had their own bindings which most of the time was close to the bindings described in spi-bus.txt with a few exceptions. One of these exception is that drivers tend to not explicitly describe the bus width but still use the max buswidth supported by the controller (QuadSPI) to interface with NOR chips. When we convert those drivers we have 2 choices: 1/ make existing setup work in a degraded mode (1-1-1) until they update their DT to explicitly specify the spi-{rx,tx}-bus-width 2/ consider that when spi-{rx,tx}-bus-width is missing the device should work in the fastest possible mode matching the device+controller constraints I went for option #2, but maybe #1 is fine. Note that spi-nand is not a problem here because this is a new binding, but I thought it would be better to clarify the meaning of these props before people start (ab)using them. > > >> Commenting to the question in the cover letter: what would be the > >> purpose of spi-max-bus-width? > > > > Defining how many IO lines are wired on the board design. If this > > property is missing we would assume all IO lines are wired and the > > restrictions would be negotiated between the controller and the device > > without requiring explicit description in the DT. > > Which is exactly the meaning of the standard property, except for the > default of all vs. 1. And that's a huge difference. Say you have a board design and several equivalence for the SPI NAND chip, one is supporting mode 1-1-4 (AKA QuadSPI) and the other one is supporting only mode 1-1-1 (Single SPI). With your solution we'd have to have 2 DTs or force all models to operate in 1-1-1 mode. > > I'll have to defer to Mark (broonie) for his opinion about deviating from > the way this is handled traditionally, and assuming different defaults... I'm not saying that we should blindly change the meaning of those existing props, I just wanted to start a discussion to see how the problem I'm mentioning here should be handled (addition of new props is something I'm perfectly fine with).