Skip to contentSkip to conversation

spi: add support for microchip fpga spi controllers

bug report Dan Carpenter 2 participants 5 messages Jun 15, 2022 SPI

lore
Share & tools
lore
#0 lore raw

Hello Conor Dooley,

The patch 9ac8d17694b6: "spi: add support for microchip fpga spi controllers" from Jun 7, 2022, leads to the following Smatch static checker warning:

drivers/spi/spi-microchip-core.c:557 mchp_corespi_probe()
warn: passing zero to 'PTR_ERR'

drivers/spi/spi-microchip-core.c

    506 static int mchp_corespi_probe(struct platform_device *pdev)
    507 {
    508         struct spi_master *master;
    509         struct mchp_corespi *spi;
    510         struct resource *res;
    511         u32 num_cs;
    512         int ret = 0;
    513 
    514         master = spi_alloc_master(&pdev->dev, sizeof(*spi));
    515         if (!master)
    516                 return dev_err_probe(&pdev->dev, -ENOMEM,
    517                                      "unable to allocate master for SPI controller\n");
    518 
    519         platform_set_drvdata(pdev, master);
    520 
    521         if (of_property_read_u32(pdev->dev.of_node, "num-cs", &num_cs))
    522                 num_cs = MAX_CS;
    523 
    524         master->num_chipselect = num_cs;
    525         master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
    526         master->setup = mchp_corespi_setup;
    527         master->bits_per_word_mask = SPI_BPW_MASK(8);
    528         master->transfer_one = mchp_corespi_transfer_one;
    529         master->prepare_message = mchp_corespi_prepare_message;
    530         master->set_cs = mchp_corespi_set_cs;
    531         master->dev.of_node = pdev->dev.of_node;
    532 
    533         spi = spi_master_get_devdata(master);
    534 
    535         spi->regs = devm_platform_get_and_ioremap_resource(pdev, 0, &res);
    536         if (IS_ERR(spi->regs)) {
    537                 ret = PTR_ERR(spi->regs);
    538                 goto error_release_master;
    539         }
    540 
    541         spi->irq = platform_get_irq(pdev, 0);
    542         if (spi->irq <= 0) {
    543                 dev_err(&pdev->dev, "invalid IRQ %d for SPI controller\n", spi->irq);
    544                 ret = spi->irq;
    545                 goto error_release_master;
    546         }
    547 
    548         ret = devm_request_irq(&pdev->dev, spi->irq, mchp_corespi_interrupt,
    549                                IRQF_SHARED, dev_name(&pdev->dev), master);
    550         if (ret) {
    551                 dev_err(&pdev->dev, "could not request irq: %d\n", ret);
    552                 goto error_release_master;
    553         }
    554 
    555         spi->clk = devm_clk_get(&pdev->dev, NULL);
    556         if (!spi->clk || IS_ERR(spi->clk)) {
                     ^^^^^^^^

NULL

--> 557 ret = PTR_ERR(spi->clk);

ret is 0/success.

Normally when functions like this return NULL, you're supposed to just accept the NULL and add tests for it to avoid NULL related bugs. In this driver if spi->clk is NULL then it leads to spi_hz == 0 which leads to a divide by zero bug. So it's not clear which way to go on this? Fix the error code or add more checks for NULL?

    558                 dev_err(&pdev->dev, "could not get clk: %d\n", ret);
    559                 goto error_release_master;
    560         }
    561 
    562         ret = clk_prepare_enable(spi->clk);
    563         if (ret) {
    564                 dev_err(&pdev->dev, "failed to enable clock\n");
    565                 goto error_release_master;
    566         }
    567 
    568         mchp_corespi_init(master, spi);
    569 
    570         ret = devm_spi_register_master(&pdev->dev, master);
    571         if (ret) {
    572                 dev_err(&pdev->dev,
    573                         "unable to register master for SPI controller\n");
    574                 goto error_release_hardware;
    575         }
    576 
    577         dev_info(&pdev->dev, "Registered SPI controller %d\n", master->bus_num);
    578 
    579         return 0;
    580 
    581 error_release_hardware:
    582         mchp_corespi_disable(spi);
    583         clk_disable_unprepare(spi->clk);
    584 error_release_master:
    585         spi_master_put(master);
    586 
    587         return ret;
    588 }

regards,
dan carpenter

#1 lore raw
Quoted 68 lines · Dan Carpenter

EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

Hello Conor Dooley,

The patch 9ac8d17694b6: "spi: add support for microchip fpga spi controllers" from Jun 7, 2022, leads to the following Smatch static checker warning:

drivers/spi/spi-microchip-core.c:557 mchp_corespi_probe()
warn: passing zero to 'PTR_ERR'

drivers/spi/spi-microchip-core.c

     506 static int mchp_corespi_probe(struct platform_device *pdev)
     507 {
     508         struct spi_master *master;
     509         struct mchp_corespi *spi;
     510         struct resource *res;
     511         u32 num_cs;
     512         int ret = 0;
     513
     514         master = spi_alloc_master(&pdev->dev, sizeof(*spi));
     515         if (!master)
     516                 return dev_err_probe(&pdev->dev, -ENOMEM,
     517                                      "unable to allocate master for SPI controller\n");
     518
     519         platform_set_drvdata(pdev, master);
     520
     521         if (of_property_read_u32(pdev->dev.of_node, "num-cs", &num_cs))
     522                 num_cs = MAX_CS;
     523
     524         master->num_chipselect = num_cs;
     525         master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
     526         master->setup = mchp_corespi_setup;
     527         master->bits_per_word_mask = SPI_BPW_MASK(8);
     528         master->transfer_one = mchp_corespi_transfer_one;
     529         master->prepare_message = mchp_corespi_prepare_message;
     530         master->set_cs = mchp_corespi_set_cs;
     531         master->dev.of_node = pdev->dev.of_node;
     532
     533         spi = spi_master_get_devdata(master);
     534
     535         spi->regs = devm_platform_get_and_ioremap_resource(pdev, 0, &res);
     536         if (IS_ERR(spi->regs)) {
     537                 ret = PTR_ERR(spi->regs);
     538                 goto error_release_master;
     539         }
     540
     541         spi->irq = platform_get_irq(pdev, 0);
     542         if (spi->irq <= 0) {
     543                 dev_err(&pdev->dev, "invalid IRQ %d for SPI controller\n", spi->irq);
     544                 ret = spi->irq;
     545                 goto error_release_master;
     546         }
     547
     548         ret = devm_request_irq(&pdev->dev, spi->irq, mchp_corespi_interrupt,
     549                                IRQF_SHARED, dev_name(&pdev->dev), master);
     550         if (ret) {
     551                 dev_err(&pdev->dev, "could not request irq: %d\n", ret);
     552                 goto error_release_master;
     553         }
     554
     555         spi->clk = devm_clk_get(&pdev->dev, NULL);
     556         if (!spi->clk || IS_ERR(spi->clk)) {
                      ^^^^^^^^

NULL

--> 557 ret = PTR_ERR(spi->clk);

Quoted 7 lines

ret is 0/success.

Normally when functions like this return NULL, you're supposed to just accept the NULL and add tests for it to avoid NULL related bugs. In this driver if spi->clk is NULL then it leads to spi_hz == 0 which leads to a divide by zero bug. So it's not clear which way to go on this? Fix the error code or add more checks for NULL?

Am I being dumb here, or should the null check just be removed like every other driver? As in, devm_clk_get will only return a valid clk or an IS_ERR() condition.

The correct solution seems to me to be remove the !spi->clk check? Thanks for the report!
Conor.

Quoted 35 lines
     558                 dev_err(&pdev->dev, "could not get clk: %d\n", ret);
     559                 goto error_release_master;
     560         }
     561
     562         ret = clk_prepare_enable(spi->clk);
     563         if (ret) {
     564                 dev_err(&pdev->dev, "failed to enable clock\n");
     565                 goto error_release_master;
     566         }
     567
     568         mchp_corespi_init(master, spi);
     569
     570         ret = devm_spi_register_master(&pdev->dev, master);
     571         if (ret) {
     572                 dev_err(&pdev->dev,
     573                         "unable to register master for SPI controller\n");
     574                 goto error_release_hardware;
     575         }
     576
     577         dev_info(&pdev->dev, "Registered SPI controller %d\n", master->bus_num);
     578
     579         return 0;
     580
     581 error_release_hardware:
     582         mchp_corespi_disable(spi);
     583         clk_disable_unprepare(spi->clk);
     584 error_release_master:
     585         spi_master_put(master);
     586
     587         return ret;
     588 }

regards,
dan carpenter

#2 lore raw
Quoted 29 lines · On Wed, Jun 15, 2022 at 08:33:35AM +0000, [email protected] wrote:
Quoted 21 lines
     541         spi->irq = platform_get_irq(pdev, 0);
     542         if (spi->irq <= 0) {
     543                 dev_err(&pdev->dev, "invalid IRQ %d for SPI controller\n", spi->irq);
     544                 ret = spi->irq;
     545                 goto error_release_master;
     546         }
     547
     548         ret = devm_request_irq(&pdev->dev, spi->irq, mchp_corespi_interrupt,
     549                                IRQF_SHARED, dev_name(&pdev->dev), master);
     550         if (ret) {
     551                 dev_err(&pdev->dev, "could not request irq: %d\n", ret);
     552                 goto error_release_master;
     553         }
     554
     555         spi->clk = devm_clk_get(&pdev->dev, NULL);
     556         if (!spi->clk || IS_ERR(spi->clk)) {
                      ^^^^^^^^

NULL

--> 557 ret = PTR_ERR(spi->clk);

Quoted 7 lines

ret is 0/success.

Normally when functions like this return NULL, you're supposed to just accept the NULL and add tests for it to avoid NULL related bugs. In this driver if spi->clk is NULL then it leads to spi_hz == 0 which leads to a divide by zero bug. So it's not clear which way to go on this? Fix the error code or add more checks for NULL?

Quoted 3 lines

Am I being dumb here, or should the null check just be removed like every other driver? As in, devm_clk_get will only return a valid clk or an IS_ERR() condition.

It can return NULL if CONFIG_HAVE_CLK is disabled. I don't know the hardware or if that CONFIG_ is essential for booting.

Quoted 2 lines

The correct solution seems to me to be remove the !spi->clk check?

That's the normal solution, yes. But if you do that, then please add a check to prevent the divide by zero:
grep -w clk drivers/spi/spi-microchip-core.c

regards,
dan carpenter

#3 lore raw
Quoted 32 lines · Dan Carpenter

EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

Quoted 29 lines · On Wed, Jun 15, 2022 at 08:33:35AM +0000, [email protected] wrote:
Quoted 28 lines
      541         spi->irq = platform_get_irq(pdev, 0);
      542         if (spi->irq <= 0) {
      543                 dev_err(&pdev->dev, "invalid IRQ %d for SPI controller\n", spi->irq);
      544                 ret = spi->irq;
      545                 goto error_release_master;
      546         }
      547
      548         ret = devm_request_irq(&pdev->dev, spi->irq, mchp_corespi_interrupt,
      549                                IRQF_SHARED, dev_name(&pdev->dev), master);
      550         if (ret) {
      551                 dev_err(&pdev->dev, "could not request irq: %d\n", ret);
      552                 goto error_release_master;
      553         }
      554
      555         spi->clk = devm_clk_get(&pdev->dev, NULL);
      556         if (!spi->clk || IS_ERR(spi->clk)) {
                       ^^^^^^^^

NULL

--> 557 ret = PTR_ERR(spi->clk);

ret is 0/success.

Normally when functions like this return NULL, you're supposed to just accept the NULL and add tests for it to avoid NULL related bugs. In this driver if spi->clk is NULL then it leads to spi_hz == 0 which leads to a divide by zero bug. So it's not clear which way to go on this? Fix the error code or add more checks for NULL?

Quoted 6 lines
Quoted 3 lines

Am I being dumb here, or should the null check just be removed like every other driver? As in, devm_clk_get will only return a valid clk or an IS_ERR() condition.

It can return NULL if CONFIG_HAVE_CLK is disabled. I don't know the hardware or if that CONFIG_ is essential for booting.

Ehh I guess it is /possible/ that CONFIG_HAVE_CLK could be off if someone is accessing the FPGA from another device. In that case, neither option really particularly appeals to me. Just return -ENODEV I guess?

Quoted 11 lines
Quoted 3 lines

The correct solution seems to me to be remove the !spi->clk check?

That's the normal solution, yes. But if you do that, then please add a check to prevent the divide by zero:
grep -w clk drivers/spi/spi-microchip-core.c

regards,
dan carpenter

#4 lore raw
Quoted 37 lines · On Wed, Jun 15, 2022 at 08:58:53AM +0000, [email protected] wrote:
Quoted 36 lines · Dan Carpenter

EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

Quoted 32 lines · On Wed, Jun 15, 2022 at 08:33:35AM +0000, [email protected] wrote:
Quoted 21 lines
      541         spi->irq = platform_get_irq(pdev, 0);
      542         if (spi->irq <= 0) {
      543                 dev_err(&pdev->dev, "invalid IRQ %d for SPI controller\n", spi->irq);
      544                 ret = spi->irq;
      545                 goto error_release_master;
      546         }
      547
      548         ret = devm_request_irq(&pdev->dev, spi->irq, mchp_corespi_interrupt,
      549                                IRQF_SHARED, dev_name(&pdev->dev), master);
      550         if (ret) {
      551                 dev_err(&pdev->dev, "could not request irq: %d\n", ret);
      552                 goto error_release_master;
      553         }
      554
      555         spi->clk = devm_clk_get(&pdev->dev, NULL);
      556         if (!spi->clk || IS_ERR(spi->clk)) {
                       ^^^^^^^^

NULL

--> 557 ret = PTR_ERR(spi->clk);

Quoted 7 lines

ret is 0/success.

Normally when functions like this return NULL, you're supposed to just accept the NULL and add tests for it to avoid NULL related bugs. In this driver if spi->clk is NULL then it leads to spi_hz == 0 which leads to a divide by zero bug. So it's not clear which way to go on this? Fix the error code or add more checks for NULL?

Am I being dumb here, or should the null check just be removed like every other driver? As in, devm_clk_get will only return a valid clk or an IS_ERR() condition.

Quoted 8 lines
Quoted 2 lines

It can return NULL if CONFIG_HAVE_CLK is disabled. I don't know the hardware or if that CONFIG_ is essential for booting.

Ehh I guess it is /possible/ that CONFIG_HAVE_CLK could be off if someone is accessing the FPGA from another device. In that case, neither option really particularly appeals to me. Just return -ENODEV I guess?

To be honest, I always prefer just accepting the NULL check and adding the checks but also philosophical debates are kind of a waste of time.

Do whatever is easiest. :)

regards,
dan carpenter