RE: [PATCH 3/3] edac: Supporting AST2400 and AST2600 edac driver

From: Troy Lee
Date: Tue Dec 01 2020 - 04:38:23 EST

Next message: Segher Boessenkool: "Re: [PATCH v2 3/4] Kbuild: make DWARF version a choice"
Previous message: David Hildenbrand: "Re: [PATCH v2] mm: Don't fault around userfaultfd-registered regions on reads"
Next in thread: Troy Lee: "RE: [PATCH 3/3] edac: Supporting AST2400 and AST2600 edac driver"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Andrew,

Great suggestion, responses inline below. We'll remove build-time SoC dependency driver code and submit an updated v2 patch.

Thanks,
Troy Lee

> -----Original Message-----
> From: Andrew Jeffery <andrew@xxxxxxxx>
> Sent: Tuesday, December 1, 2020 9:12 AM
> To: Troy Lee <troy_lee@xxxxxxxxxxxxxx>; Stefan M Schaeckeler
> <sschaeck@xxxxxxxxx>; Rob Herring <robh+dt@xxxxxxxxxx>; Joel Stanley
> <joel@xxxxxxxxx>; Borislav Petkov <bp@xxxxxxxxx>; Mauro Carvalho Chehab
> <mchehab@xxxxxxxxxx>; Tony Luck <tony.luck@xxxxxxxxx>; James Morse
> <james.morse@xxxxxxx>; Robert Richter <rrichter@xxxxxxxxxxx>; open
> list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS
> <devicetree@xxxxxxxxxxxxxxx>; moderated list:ARM/ASPEED MACHINE
> SUPPORT <linux-arm-kernel@xxxxxxxxxxxxxxxxxxx>; moderated
> list:ARM/ASPEED MACHINE SUPPORT <linux-aspeed@xxxxxxxxxxxxxxxx>; open
> list <linux-kernel@xxxxxxxxxxxxxxx>; open list:EDAC-CORE
> <linux-edac@xxxxxxxxxxxxxxx>
> Cc: leetroy@xxxxxxxxx; Ryan Chen <ryan_chen@xxxxxxxxxxxxxx>
> Subject: Re: [PATCH 3/3] edac: Supporting AST2400 and AST2600 edac driver
>
> Hi Troy,
>
> I like the idea of expanding the driver's support to cover the other SoC
> generations, but not at the cost of making the build of the driver SoC-specific.
> I've made some comments below in this regard.
>
> On Mon, 30 Nov 2020, at 19:03, Troy Lee wrote:
> > Adding AST2400 and AST2600 edac driver support.
> >
> > Signed-off-by: Troy Lee <troy_lee@xxxxxxxxxxxxxx>
> > ---
> > drivers/edac/Kconfig | 6 +-
> > drivers/edac/aspeed_edac.c | 114
> > +++++++++++++++++++++++++++++--------
> > 2 files changed, 94 insertions(+), 26 deletions(-)
> >
> > diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig index
> > fc30f2ef9782..8ea70746d0bf 100644
> > --- a/drivers/edac/Kconfig
> > +++ b/drivers/edac/Kconfig
> > @@ -508,10 +508,10 @@ config EDAC_QCOM
> > health, you should probably say 'Y' here.
> >
> > config EDAC_ASPEED
> > - tristate "Aspeed AST 2500 SoC"
> > - depends on MACH_ASPEED_G5
> > + tristate "Aspeed AST BMC SoC"
> > + depends on (MACH_ASPEED_G4 || MACH_ASPEED_G5 ||
> MACH_ASPEED_G6)
> > help
> > - Support for error detection and correction on the Aspeed AST 2500 SoC.
> > + Support for error detection and correction on the Aspeed AST BMC SoC.
> >
> > First, ECC must be configured in the bootloader. Then, this driver
> > will expose error counters via the EDAC kernel framework.
> > diff --git a/drivers/edac/aspeed_edac.c b/drivers/edac/aspeed_edac.c
> > index fbec28dc661d..03a3c12f6bf6 100644
> > --- a/drivers/edac/aspeed_edac.c
> > +++ b/drivers/edac/aspeed_edac.c
> > @@ -14,12 +14,11 @@
> > #include <linux/regmap.h>
> > #include "edac_module.h"
> >
> > -
> > #define DRV_NAME "aspeed-edac"
> >
> > -
> > #define ASPEED_MCR_PROT 0x00 /* protection key register */
> > #define ASPEED_MCR_CONF 0x04 /* configuration register */
> > +#define ASPEED_MCR_REQ 0x08 /* Graphics Memory Protection
> register */
> > #define ASPEED_MCR_INTR_CTRL 0x50 /* interrupt control/status
> register */
> > #define ASPEED_MCR_ADDR_UNREC 0x58 /* address of first
> un-recoverable error */
> > #define ASPEED_MCR_ADDR_REC 0x5c /* address of last recoverable
> error */
> > @@ -34,10 +33,8 @@
> > #define ASPEED_MCR_INTR_CTRL_CNT_UNREC GENMASK(15, 12)
> #define
> > ASPEED_MCR_INTR_CTRL_ENABLE (BIT(0) | BIT(1))
> >
> > -
> > static struct regmap *aspeed_regmap;
> >
> > -
> > static int regmap_reg_write(void *context, unsigned int reg, unsigned
> > int val) {
> > void __iomem *regs = (void __iomem *)context; @@ -53,7 +50,6 @@
> > static int regmap_reg_write(void *context, unsigned int reg, unsigned
> > int val)
> > return 0;
> > }
> >
> > -
> > static int regmap_reg_read(void *context, unsigned int reg, unsigned
> > int *val) {
> > void __iomem *regs = (void __iomem *)context; @@ -63,6 +59,76 @@
> > static int regmap_reg_read(void *context, unsigned int reg, unsigned
> > int *val)
> > return 0;
> > }
> >
> > +extern void aspeed_sdmc_disable_mem_protection(u8 req) {
> > + u32 req_val = 0;
> > +
> > + regmap_read(aspeed_regmap, ASPEED_MCR_REQ, &req_val);
> > +
> > + req_val &= ~BIT(req);
> > +
> > + regmap_write(aspeed_regmap, ASPEED_MCR_REQ, req_val); }
> > +EXPORT_SYMBOL(aspeed_sdmc_disable_mem_protection);
> > +
> > +static const u32 ast2400_dram_table[] = {
> > + 0x04000000, //64MB
> > + 0x08000000, //128MB
> > + 0x10000000, //256MB
> > + 0x20000000, //512MB
> > +};
>
> You could perhaps save us from the size comments here by using e.g.
>
> static const u32 ast2400_dram_table[] = {
> 64 << 20,
> 128 << 20,
> 256 << 20,
> 512 << 20,
> };
>
Fixed all in v2.

> > +
> > +static const u32 ast2500_dram_table[] = {
> > + 0x08000000, //128MB
> > + 0x10000000, //256MB
> > + 0x20000000, //512MB
> > + 0x40000000, //1024MB
> > +};
> > +
> > +static const u32 ast2600_dram_table[] = {
> > + 0x10000000, //256MB
> > + 0x20000000, //512MB
> > + 0x40000000, //1024MB
> > + 0x80000000, //2048MB
> > +};
> > +
> > +extern u32 aspeed_get_dram_size(void) {
> > + u32 reg04;
> > + u32 size;
> > +
> > + regmap_read(aspeed_regmap, ASPEED_MCR_CONF, &reg04);
> > +
> > +#if defined(CONFIG_MACH_ASPEED_G6)
> > + size = ast2600_dram_table[reg04 & 0x3]; #elif
> > +defined(CONFIG_MACH_ASPEED_G5)
> > + size = ast2500_dram_table[reg04 & 0x3]; #else
> > + size = ast2400_dram_table[reg04 & 0x3]; #endif
> > + return size;
> > +}
> > +EXPORT_SYMBOL(aspeed_get_dram_size);
>
> The driver must support running on any of the SoC generations without being
> recompiled. This requires the driver be compiled for a specific SoC generation.
>
> The right way to do this is to use the OF match table data. Please fix it.
>
Fixed in v2, driver can check the DRAM controller hardware version from MCR_CONF [31:28].

> > +
> > +static const u32 aspeed_vga_table[] = {
> > + 0x800000, //8MB
> > + 0x1000000, //16MB
> > + 0x2000000, //32MB
> > + 0x4000000, //64MB
> > +};
> > +
> > +extern u32 aspeed_get_vga_size(void)
> > +{
> > + u32 reg04;
> > + u32 size;
> > +
> > + regmap_read(aspeed_regmap, ASPEED_MCR_CONF, &reg04);
> > +
> > + size = aspeed_vga_table[((reg04 & 0xC) >> 2)];
> > + return size;
> > +}
> > +EXPORT_SYMBOL(aspeed_get_vga_size);
> > +
> > static bool regmap_is_volatile(struct device *dev, unsigned int reg)
> > {
> > switch (reg) {
> > @@ -209,8 +275,8 @@ static int config_irq(void *ctx, struct
> > platform_device *pdev)
> > /* register interrupt handler */
> > irq = platform_get_irq(pdev, 0);
> > dev_dbg(&pdev->dev, "got irq %d\n", irq);
> > - if (irq < 0)
> > - return irq;
> > + if (!irq)
> > + return -ENODEV;
>
> The documentation for platform_get_irq() says "Return: IRQ number on
> success, negative error number on failure."
>
> So this change doesn't look right.
>
Fixed in v2.

> >
> > rc = devm_request_irq(&pdev->dev, irq, mcr_isr, IRQF_TRIGGER_HIGH,
> > DRV_NAME, ctx);
> > @@ -239,7 +305,11 @@ static int init_csrows(struct mem_ctl_info *mci)
> > int rc;
> >
> > /* retrieve info about physical memory from device tree */
> > - np = of_find_node_by_path("/memory");
> > +#ifdef CONFIG_MACH_ASPEED_G4
> > + np = of_find_node_by_path("/memory@40000000");
> > +#else
> > + np = of_find_node_by_path("/memory@80000000");
> > +#endif
>
> Again, this makes the driver SoC-specific at compile time, and that's not going
> to fly. If we need it, this should be solved with a phandle to the appropriate
> memory node.
>
Fixed in v2 by using of_find_node_by_name(NULL, "memory") instead.

> > if (!np) {
> > dev_err(mci->pdev, "dt: missing /memory node\n");
> > return -ENODEV;
> > @@ -281,11 +351,19 @@ static int aspeed_probe(struct platform_device
> *pdev)
> > struct device *dev = &pdev->dev;
> > struct edac_mc_layer layers[2];
> > struct mem_ctl_info *mci;
> > + struct device_node *np;
> > + struct resource *res;
> > void __iomem *regs;
> > - u32 reg04;
> > int rc;
> >
> > - regs = devm_platform_ioremap_resource(pdev, 0);
> > + /* setup regmap */
> > + np = dev->of_node;
> > +
> > + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > + if (!res)
> > + return -ENOENT;
> > +
> > + regs = devm_ioremap_resource(dev, res);
> > if (IS_ERR(regs))
> > return PTR_ERR(regs);
>
> Why is this change necessary?
>
Not necessary, so I'll reverted this part of change in v2.

> >
> > @@ -294,13 +372,6 @@ static int aspeed_probe(struct platform_device
> *pdev)
> > if (IS_ERR(aspeed_regmap))
> > return PTR_ERR(aspeed_regmap);
> >
> > - /* bail out if ECC mode is not configured */
> > - regmap_read(aspeed_regmap, ASPEED_MCR_CONF, &reg04);
> > - if (!(reg04 & ASPEED_MCR_CONF_ECC)) {
> > - dev_err(&pdev->dev, "ECC mode is not configured in u-boot\n");
> > - return -EPERM;
> > - }
> > -
>
> Why shouldn't we do this check?
The driver can handle both ECC on and off, so this check might not be necessary.

> > edac_op_state = EDAC_OPSTATE_INT;
> >
> > /* allocate & init EDAC MC data structure */ @@ -373,13 +444,13 @@
> > static int aspeed_remove(struct platform_device *pdev)
> > return 0;
> > }
> >
> > -
> > static const struct of_device_id aspeed_of_match[] = {
> > + { .compatible = "aspeed,ast2400-sdram-edac" },
> > { .compatible = "aspeed,ast2500-sdram-edac" },
> > + { .compatible = "aspeed,ast2600-sdram-edac" },
>
> The .data member of struct of_device_id is where you need to attach your
> SoC-generation-specific data.
>
> Cheers,
>
> Andrew
>
> > {},
> > };
> >
> > -
> > static struct platform_driver aspeed_driver = {
> > .driver = {
> > .name = DRV_NAME,
> > @@ -395,18 +466,15 @@ static int __init aspeed_init(void)
> > return platform_driver_register(&aspeed_driver);
> > }
> >
> > -
> > static void __exit aspeed_exit(void)
> > {
> > platform_driver_unregister(&aspeed_driver);
> > }
> >
> > -
> > module_init(aspeed_init);
> > module_exit(aspeed_exit);
> >
> > -
> > MODULE_LICENSE("GPL");
> > MODULE_AUTHOR("Stefan Schaeckeler <sschaeck@xxxxxxxxx>");
> > -MODULE_DESCRIPTION("Aspeed AST2500 EDAC driver");
> > +MODULE_DESCRIPTION("Aspeed EDAC driver");
> > MODULE_VERSION("1.0");
> > --
> > 2.17.1
> >
> >

Next message: Segher Boessenkool: "Re: [PATCH v2 3/4] Kbuild: make DWARF version a choice"
Previous message: David Hildenbrand: "Re: [PATCH v2] mm: Don't fault around userfaultfd-registered regions on reads"
Next in thread: Troy Lee: "RE: [PATCH 3/3] edac: Supporting AST2400 and AST2600 edac driver"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]