Re: [PATCH v2 04/10] arm: socfpga: iossm: store size in bytes

[Sascha Hauer <s.hauer@pengutronix.de>]

On Fri, Apr 10, 2026 at 02:06:32PM +0200, Michael Tretter wrote:
> On Fri, 10 Apr 2026 10:37:09 +0200, Ahmad Fatoum wrote:
> > On 4/10/26 10:31 AM, Michael Tretter wrote:
> > > On Fri, 10 Apr 2026 10:13:41 +0200, Ahmad Fatoum wrote:
> > >> On 4/9/26 3:52 PM, Michael Tretter wrote:
> > >>> The mem_width_info is the memory size in gigabits. Convert it to bytes
> > >>> before storing it for each bank to have a more convenient format and
> > >>> simplify the conversion when reading the value.
> > >>>
> > >>> Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
> > >>> ---
> > >>> Changes in v2:
> > >>>
> > >>> - Change memory_size in io96b_mb_ctrl to phys_size_t to prevent overflow
> > >>
> > >> Thanks for addressing the concern. I looked it over and there's still a
> > >> chance for overflow.
> > >>
> > >>> diff --git a/arch/arm/mach-socfpga/iossm_mailbox.c b/arch/arm/mach-socfpga/iossm_mailbox.c
> > >>> index 9299fee71e0b..042ea4a99e5c 100644
> > >>> --- a/arch/arm/mach-socfpga/iossm_mailbox.c
> > >>> +++ b/arch/arm/mach-socfpga/iossm_mailbox.c
> > >>> @@ -9,6 +9,7 @@
> > >>>  #include <common.h>
> > >>>  #include <io.h>
> > >>>  #include <linux/bitfield.h>
> > >>> +#include <linux/sizes.h>
> > >>>  #include "iossm_mailbox.h"
> > >>>  #include <mach/socfpga/generic.h>
> > >>>  #include <mach/socfpga/soc64-regs.h>
> > >>> @@ -404,21 +405,25 @@ int io96b_get_mem_width_info(struct io96b_info *io96b_ctrl)
> > >>>  	struct io96b_mb_resp usr_resp;
> > >>>  	struct io96b_mb_ctrl *mb_ctrl;
> > >>>  	int i, j;
> > >>> -	u16 memory_size;
> > >>> -	u16 total_memory_size = 0;
> > >>> +	phys_size_t memory_size;
> > >>> +	u32 mem_width_info;
> > >>> +	phys_size_t total_memory_size = 0;
> > >>>  
> > >>>  	/* Get all memory interface(s) total memory size on all instance(s) */
> > >>>  	for (i = 0; i < io96b_ctrl->num_instance; i++) {
> > >>>  		mb_ctrl = &io96b_ctrl->io96b[i].mb_ctrl;
> > >>>  		memory_size = 0;
> > >>> +
> > >>>  		for (j = 0; j < mb_ctrl->num_mem_interface; j++) {
> > >>>  			io96b_mb_req_no_param(io96b_ctrl->io96b[i].io96b_csr_addr,
> > >>>  					      mb_ctrl->ip_type[j],
> > >>>  					      mb_ctrl->ip_instance_id[j],
> > >>>  					      CMD_GET_MEM_INFO, GET_MEM_WIDTH_INFO, &usr_resp);
> > >>> +			mem_width_info = usr_resp.cmd_resp_data[1] & GENMASK(7, 0);
> > >>
> > >> I assume this is the RAM memory width in bytes, e.g. 2 or 4 bytes, but
> > >> the mask  allows it to be up to 255.
> > > 
> > > mem_width_info is in Giga bits.
> > 
> > Uh, is memory width here not the number of bits transferred per clock
> > cycle..?
> 
> Maybe I'm confused, but I understand that it's not the bits transferred
> per clock cycle (the memory bus interface width), but information about
> the memory geometry under the assumption of a memory depth of 1 Giga
> bits. My understanding could be wrong, though.
> 
> > 
> > > 
> > >>
> > >>>  
> > >>> -			memory_size = memory_size +
> > >>> -				(usr_resp.cmd_resp_data[1] & GENMASK(7, 0));
> > >>> +			mb_ctrl->memory_size[j] = mem_width_info * (SZ_1G / SZ_8);
> > >>
> > >> SZ_4G / (SZ_1G / 8) = 32
> > >>
> > >> So the maximum value permitted in mem_width_info without overflowing is
> > >> 31, but the code accepts up to 255. As all types on the left hand side
> > >> are 32-bit only.
> > > 
> > > This driver is only available on Agilex5, which is arm64 and selects
> > > PHYS_ADDR_T_64BIT. Thus, phys_size_t has 64 bits and doesn't result in
> > > an overflow.
> > 
> > Sorry, left-right confusion. The right had size is all 32-bit integers,
> > so they may wrap around and are truncated before assignment to the
> > 64-bit left hand side.
> 
> Ah, indeed. I'll add the necessary cast to the right hand side.

Added a cast to phys_size_t to the right hand side while applying.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |