linux-kernel - Re: [PATCH net-next v3 2/6] net: phy: microchip_t1s: replace read-modify-write code with phy_modify

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

Message-ID: <ZG9/E8Am2ICEHIbr@debian>
Date:   Thu, 25 May 2023 17:30:27 +0200
From:   Ramón Nordin Rodriguez 
        <ramon.nordin.rodriguez@...roamp.se>
To:     Parthiban Veerasooran <Parthiban.Veerasooran@...rochip.com>
Cc:     andrew@...n.ch, hkallweit1@...il.com, linux@...linux.org.uk,
        davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
        pabeni@...hat.com, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org, horatiu.vultur@...rochip.com,
        Woojung.Huh@...rochip.com, Nicolas.Ferre@...rochip.com,
        Thorsten.Kummermehr@...rochip.com
Subject: Re: [PATCH net-next v3 2/6] net: phy: microchip_t1s: replace
 read-modify-write code with phy_modify_mmd

On Wed, May 24, 2023 at 08:15:35PM +0530, Parthiban Veerasooran wrote:
> Replace read-modify-write code in the lan867x_config_init function to
> avoid handling data type mismatch and to simplify the code.
> 
> Signed-off-by: Parthiban Veerasooran <Parthiban.Veerasooran@...rochip.com>
> ---
>  drivers/net/phy/microchip_t1s.c | 23 +++++++++++------------
>  1 file changed, 11 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/net/phy/microchip_t1s.c b/drivers/net/phy/microchip_t1s.c
> index a42a6bb6e3bd..b5b5a95fa6e7 100644
> --- a/drivers/net/phy/microchip_t1s.c
> +++ b/drivers/net/phy/microchip_t1s.c
> @@ -31,19 +31,19 @@
>   * W   0x1F 0x0099 0x7F80 ------
>   */
>  
> -static const int lan867x_fixup_registers[12] = {
> +static const u32 lan867x_fixup_registers[12] = {
>  	0x00D0, 0x00D1, 0x0084, 0x0085,
>  	0x008A, 0x0087, 0x0088, 0x008B,
>  	0x0080, 0x00F1, 0x0096, 0x0099,
>  };
>  
> -static const int lan867x_fixup_values[12] = {
> +static const u16 lan867x_fixup_values[12] = {
>  	0x0002, 0x0000, 0x3380, 0x0006,
>  	0xC000, 0x801C, 0x033F, 0x0404,
>  	0x0600, 0x2400, 0x2000, 0x7F80,
>  };
>  
> -static const int lan867x_fixup_masks[12] = {
> +static const u16 lan867x_fixup_masks[12] = {
>  	0x0E03, 0x0300, 0xFFC0, 0x000F,
>  	0xF800, 0x801C, 0x1FFF, 0xFFFF,
>  	0x0600, 0x7F00, 0x2000, 0xFFFF,
> @@ -63,23 +63,22 @@ static int lan867x_config_init(struct phy_device *phydev)
>  	 * used, which might then write the same value back as read + modified.
>  	 */
>  
> -	int reg_value;
>  	int err;
> -	int reg;
>  
>  	/* Read-Modified Write Pseudocode (from AN1699)
>  	 * current_val = read_register(mmd, addr) // Read current register value
>  	 * new_val = current_val AND (NOT mask) // Clear bit fields to be written
>  	 * new_val = new_val OR value // Set bits
> -	 * write_register(mmd, addr, new_val) // Write back updated register value
> +	 * write_register(mmd, addr, new_val) // Write back updated register value.
> +	 * Although AN1699 says Read, Modify, Write, the write is not required if
> +	 * the register already has the required value.
>  	 */
>  	for (int i = 0; i < ARRAY_SIZE(lan867x_fixup_registers); i++) {
> -		reg = lan867x_fixup_registers[i];
> -		reg_value = phy_read_mmd(phydev, MDIO_MMD_VEND2, reg);
> -		reg_value &= ~lan867x_fixup_masks[i];
> -		reg_value |= lan867x_fixup_values[i];
> -		err = phy_write_mmd(phydev, MDIO_MMD_VEND2, reg, reg_value);
> -		if (err != 0)
> +		err = phy_modify_mmd(phydev, MDIO_MMD_VEND2,
> +				     lan867x_fixup_registers[i],
> +				     lan867x_fixup_masks[i],
> +				     lan867x_fixup_values[i]);

This change answers an uncertainty in the block comment in the top of
this func, pasted here for your convenience

	/* HW quirk: Microchip states in the application note (AN1699) for the phy
	 * that a set of read-modify-write (rmw) operations has to be performed
	 * on a set of seemingly magic registers.
	 * The result of these operations is just described as 'optimal performance'
	 * Microchip gives no explanation as to what these mmd regs do,
	 * in fact they are marked as reserved in the datasheet.
	 * It is unclear if phy_modify_mmd would be safe to use or if a write
	 * really has to happen to each register.
	 * In order to exactly conform to what is stated in the AN phy_write_mmd is
	 * used, which might then write the same value back as read + modified.
	 */

This change also invalidates most of the comment. I think this should be
reduced to something along the lines of:
	/* HW quirk: Microchip states in the application note (AN1699) for the phy
	 * that a set of read-modify-write (rmw) operations has to be performed
	 * on a set of seemingly magic registers.
	 * The result of these operations is just described as 'optimal performance'
	 * Microchip gives no explanation as to what these mmd regs do,
	 * in fact they are marked as reserved in the datasheet.*/

Additionally I don't mind it if you change the tone of the comment. This was brought
up in the sitdown we had, where it was explained from Microchip that
documenting what the reg operations actually does would expose to much
of the internal workings of the chip.

Possibly this comment should move down to where the fixup reg operations
are performed, and replace the comment about the 'read write modify'
stuff all togheter.
In my opinion I kind of loose context about what the func does when it
first explains the fixup stuff, then actually does something with the
STS2 regs, and finally actually does the fixup.
> +		if (err)
>  			return err;
>  	}
>  
> -- 
> 2.34.1
>