Skip to contentSkip to conversation

mmc: sh_mmcif: remove now superfluous sh_mmcif_host::data member

DISCUSSION Dan Carpenter 2 participants 3 messages Nov 6, 2013 – Dec 4, 2013 MMC, SD and SDIO

lore
Share & tools
lore
#0 lore raw

Hello Guennadi Liakhovetski,

This is a semi-automatic email about new static checker warnings.

The patch 699834045f1e: "mmc: sh_mmcif: remove now superfluous sh_mmcif_host::data member" from Dec 26, 2011, leads to the following Smatch complaint:

drivers/mmc/host/sh_mmcif.c:822 sh_mmcif_set_cmd()
error: we previously assumed 'data' could be null (see line 787)

drivers/mmc/host/sh_mmcif.c
786 /* WDAT / DATW */

   787		if (data) {
                    ^^^^

Check.

788 tmp |= CMD_SET_WDAT;
789 switch (host->bus_width) {
790 case MMC_BUS_WIDTH_1:
791 tmp |= CMD_SET_DATW_1;
792 break;
793 case MMC_BUS_WIDTH_4:
794 tmp |= CMD_SET_DATW_4;
795 break;
796 case MMC_BUS_WIDTH_8:
797 tmp |= CMD_SET_DATW_8;
798 break;
799 default:
800 dev_err(&host->pd->dev, "Unsupported bus width.\n");
801 break;
802 }
803 switch (host->timing) {
804 case MMC_TIMING_UHS_DDR50:
805 /*
806 * MMC core will only set this timing, if the host
807 * advertises the MMC_CAP_UHS_DDR50 capability. MMCIF
808 * implementations with this capability, e.g. sh73a0,
809 * will have to set it in their platform data.
810 */
811 tmp |= CMD_SET_DARS;
812 break;
813 }
814 }
815 /* DWEN */
816 if (opc == MMC_WRITE_BLOCK || opc == MMC_WRITE_MULTIPLE_BLOCK)
817 tmp |= CMD_SET_DWEN;
818 /* CMLTE/CMD12EN */
819 if (opc == MMC_READ_MULTIPLE_BLOCK || opc == MMC_WRITE_MULTIPLE_BLOCK) {
820 tmp |= CMD_SET_CMLTE | CMD_SET_CMD12EN;
821 sh_mmcif_bitset(host, MMCIF_CE_BLOCK_SET,

   822					data->blocks << 16);
                                        ^^^^^^^^^^^^

Dereference.

We used to assume that host->data might be NULL but that mrq->data was a valid pointer. But now they are unified into one pointer.

823 }
824 /* RIDXC[1:0] check bits */

regards,
dan carpenter

#1 lore raw

Re: *** GMX Spamverdacht *** re: mmc: sh_mmcif: remove now superfluous sh_mmcif_host::data member

Hi Dan,

Quoted 53 lines · Dan Carpenter

Hello Guennadi Liakhovetski,

This is a semi-automatic email about new static checker warnings.

The patch 699834045f1e: "mmc: sh_mmcif: remove now superfluous sh_mmcif_host::data member" from Dec 26, 2011, leads to the following Smatch complaint:

drivers/mmc/host/sh_mmcif.c:822 sh_mmcif_set_cmd()
error: we previously assumed 'data' could be null (see line 787)

drivers/mmc/host/sh_mmcif.c
786 /* WDAT / DATW */

   787		if (data) {
                    ^^^^

Check.

   788			tmp |= CMD_SET_WDAT;
   789			switch (host->bus_width) {
   790			case MMC_BUS_WIDTH_1:
   791				tmp |= CMD_SET_DATW_1;
   792				break;
   793			case MMC_BUS_WIDTH_4:
   794				tmp |= CMD_SET_DATW_4;
   795				break;
   796			case MMC_BUS_WIDTH_8:
   797				tmp |= CMD_SET_DATW_8;
   798				break;
   799			default:
   800				dev_err(&host->pd->dev, "Unsupported bus width.\n");
   801				break;
   802			}
   803			switch (host->timing) {
   804			case MMC_TIMING_UHS_DDR50:
   805				/*
   806				 * MMC core will only set this timing, if the host
   807				 * advertises the MMC_CAP_UHS_DDR50 capability. MMCIF
   808				 * implementations with this capability, e.g. sh73a0,
   809				 * will have to set it in their platform data.
   810				 */
   811				tmp |= CMD_SET_DARS;
   812				break;
   813			}
   814		}
   815		/* DWEN */
   816		if (opc == MMC_WRITE_BLOCK || opc == MMC_WRITE_MULTIPLE_BLOCK)
   817			tmp |= CMD_SET_DWEN;
   818		/* CMLTE/CMD12EN */
   819		if (opc == MMC_READ_MULTIPLE_BLOCK || opc == MMC_WRITE_MULTIPLE_BLOCK) {
   820			tmp |= CMD_SET_CMLTE | CMD_SET_CMD12EN;
   821			sh_mmcif_bitset(host, MMCIF_CE_BLOCK_SET,
   822					data->blocks << 16);
                                        ^^^^^^^^^^^^
Quoted 4 lines

Dereference.

We used to assume that host->data might be NULL but that mrq->data was a valid pointer. But now they are unified into one pointer.

Yes, formally this isn't correct, but in practice this is valid, because for those two opcodes, handled in this "if" data cannot be NULL.

Thanks
Guennadi

Quoted 7 lines
   823		}
   824		/* RIDXC[1:0] check bits */

regards,
dan carpenter

---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/

#2 lore raw

Re: *** GMX Spamverdacht *** re: mmc: sh_mmcif: remove now superfluous sh_mmcif_host::data member

Quoted 7 lines · Guennadi Liakhovetski
Quoted 3 lines

We used to assume that host->data might be NULL but that mrq->data was a valid pointer. But now they are unified into one pointer.

Yes, formally this isn't correct, but in practice this is valid, because for those two opcodes, handled in this "if" data cannot be NULL.

Fine fine. I suspected it might be something like that. We only send the warning messages one time so no worries.

regards,
dan carpenter