| Message ID | 20200406113310.3041-2-nikita.shubin@maquefel.me (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | remoteproc: imx_rproc: add virtio support | expand |
Hi Nikita, On Mon, Apr 06, 2020 at 02:33:08PM +0300, nikita.shubin@maquefel.me wrote: > In case elf file interrupt vector is not supposed to be at OCRAM_S, > it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 > firmware. > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > The firmware must set stack poiner as first instruction: > > Reset_Handler: > ldr sp, = __stack /* set stack pointer */ > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> The address in the SoB has to match what is found in the "From:" field of the email header. Checkpatch is complaining about that, something I would have expected to be fixed before sending this set out. > --- > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > 1 file changed, 15 insertions(+), 1 deletion(-) > > diff --git a/drivers/remoteproc/imx_rproc.c b/drivers/remoteproc/imx_rproc.c > index 3e72b6f38d4b..bebc58d0f711 100644 > --- a/drivers/remoteproc/imx_rproc.c > +++ b/drivers/remoteproc/imx_rproc.c > @@ -45,6 +45,8 @@ > > #define IMX7D_RPROC_MEM_MAX 8 > > +#define IMX_BOOT_PC 0x4 > + > /** > * struct imx_rproc_mem - slim internal memory structure > * @cpu_addr: MPU virtual address of the memory region > @@ -85,6 +87,7 @@ struct imx_rproc { > const struct imx_rproc_dcfg *dcfg; > struct imx_rproc_mem mem[IMX7D_RPROC_MEM_MAX]; > struct clk *clk; > + void __iomem *bootreg; > }; > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc *rproc) > struct device *dev = priv->dev; > int ret; > > + /* write entry point to program counter */ > + writel(rproc->bootaddr, priv->bootreg); What happens on all the other IMX systems where this fix is not needed? Will they continue to work properly? > + > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > dcfg->src_mask, dcfg->src_start); > if (ret) > dev_err(dev, "Failed to enable M4!\n"); > > + dev_info(&rproc->dev, "Started from 0x%x\n", rproc->bootaddr); > + > return ret; > } > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) > if (ret) > dev_err(dev, "Failed to stop M4!\n"); > > + /* clear entry points */ > + writel(0, priv->bootreg); > + > return ret; > } > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc *rproc, u64 da, int len) > static const struct rproc_ops imx_rproc_ops = { > .start = imx_rproc_start, > .stop = imx_rproc_stop, > - .da_to_va = imx_rproc_da_to_va, > + .da_to_va = imx_rproc_da_to_va, > + .get_boot_addr = rproc_elf_get_boot_addr, How is this useful? Sure it will set rproc->bootaddr in rproc_fw_boot() but what good does that do when it is invariably set again in imx_rproc_start() ? > }; > > static int imx_rproc_addr_init(struct imx_rproc *priv, > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct platform_device *pdev) > goto err_put_rproc; > } > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, sizeof(u32)); > + > /* > * clk for M4 block including memory. Should be > * enabled before .start for FW transfer. > -- > 2.25.1 >
On Tue, 14 Apr 2020 10:45:19 -0600 Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > Hi Nikita, > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, nikita.shubin@maquefel.me > wrote: > > In case elf file interrupt vector is not supposed to be at OCRAM_S, > > it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 > > firmware. > > > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > > > The firmware must set stack poiner as first instruction: > > > > Reset_Handler: > > ldr sp, = __stack /* set stack pointer */ > > > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > The address in the SoB has to match what is found in the "From:" > field of the email header. Checkpatch is complaining about that, > something I would have expected to be fixed before sending this set > out. > > > --- > > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > > 1 file changed, 15 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/remoteproc/imx_rproc.c > > b/drivers/remoteproc/imx_rproc.c index 3e72b6f38d4b..bebc58d0f711 > > 100644 --- a/drivers/remoteproc/imx_rproc.c > > +++ b/drivers/remoteproc/imx_rproc.c > > @@ -45,6 +45,8 @@ > > > > #define IMX7D_RPROC_MEM_MAX 8 > > > > +#define IMX_BOOT_PC 0x4 > > + > > /** > > * struct imx_rproc_mem - slim internal memory structure > > * @cpu_addr: MPU virtual address of the memory region > > @@ -85,6 +87,7 @@ struct imx_rproc { > > const struct imx_rproc_dcfg *dcfg; > > struct imx_rproc_mem > > mem[IMX7D_RPROC_MEM_MAX]; struct clk *clk; > > + void __iomem *bootreg; > > }; > > > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc > > *rproc) struct device *dev = priv->dev; > > int ret; > > > > + /* write entry point to program counter */ > > + writel(rproc->bootaddr, priv->bootreg); > > What happens on all the other IMX systems where this fix is not > needed? Will they continue to work properly? Mathieu you are totally correct imx6/imx7 use different addresses they boot. For imx7: | On i.MX 7Dual/7Solo, the boot vector for the Cortex-M4 core is located | at the start of the OCRAM_S (On Chip RAM - Secure) whose address is | 0x0018_0000 from Cortex-A7. For imx6: | The Boot vector for the Cortex-M4 core is located at the start of the | TCM_L whose address is 0x007F_8000 from the Cortex-A9. This is a | different location than on the i.MX 7Dual/7Solo But on imx7 0x0 is translated to 0x0018_0000 by imx_rproc_da_to_va, and on imx7 0x0 is translated to 0x007F_8000, using imx_rproc_att_imx7d and imx_rproc_att_imx6sx respectively. I have no information about IMX8 (i have found none available publicity), but should be the same as Cortex-M boots from 0x0. > > > + > > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > > dcfg->src_mask, dcfg->src_start); > > if (ret) > > dev_err(dev, "Failed to enable M4!\n"); > > > > + dev_info(&rproc->dev, "Started from 0x%x\n", > > rproc->bootaddr); + > > return ret; > > } > > > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) > > if (ret) > > dev_err(dev, "Failed to stop M4!\n"); > > > > + /* clear entry points */ > > + writel(0, priv->bootreg); > > + > > return ret; > > } > > > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc > > *rproc, u64 da, int len) static const struct rproc_ops > > imx_rproc_ops = { .start = imx_rproc_start, > > .stop = imx_rproc_stop, > > - .da_to_va = imx_rproc_da_to_va, > > + .da_to_va = imx_rproc_da_to_va, > > + .get_boot_addr = rproc_elf_get_boot_addr, > > How is this useful? Sure it will set rproc->bootaddr in > rproc_fw_boot() but what good does that do when it is invariably set > again in imx_rproc_start() ? > > > }; > > > > static int imx_rproc_addr_init(struct imx_rproc *priv, > > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct > > platform_device *pdev) goto err_put_rproc; > > } > > > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, > > sizeof(u32)); + > > /* > > * clk for M4 block including memory. Should be > > * enabled before .start for FW transfer. > > -- > > 2.25.1 > >
On Fri, 17 Apr 2020 at 06:12, Nikita Shubin <nikita.shubin@maquefel.me> wrote: > > On Tue, 14 Apr 2020 10:45:19 -0600 > Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > > > Hi Nikita, > > > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, nikita.shubin@maquefel.me > > wrote: > > > In case elf file interrupt vector is not supposed to be at OCRAM_S, > > > it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 > > > firmware. > > > > > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > > > > > The firmware must set stack poiner as first instruction: > > > > > > Reset_Handler: > > > ldr sp, = __stack /* set stack pointer */ > > > > > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > > > The address in the SoB has to match what is found in the "From:" > > field of the email header. Checkpatch is complaining about that, > > something I would have expected to be fixed before sending this set > > out. > > > > > --- > > > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > > > 1 file changed, 15 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/remoteproc/imx_rproc.c > > > b/drivers/remoteproc/imx_rproc.c index 3e72b6f38d4b..bebc58d0f711 > > > 100644 --- a/drivers/remoteproc/imx_rproc.c > > > +++ b/drivers/remoteproc/imx_rproc.c > > > @@ -45,6 +45,8 @@ > > > > > > #define IMX7D_RPROC_MEM_MAX 8 > > > > > > +#define IMX_BOOT_PC 0x4 > > > + > > > /** > > > * struct imx_rproc_mem - slim internal memory structure > > > * @cpu_addr: MPU virtual address of the memory region > > > @@ -85,6 +87,7 @@ struct imx_rproc { > > > const struct imx_rproc_dcfg *dcfg; > > > struct imx_rproc_mem > > > mem[IMX7D_RPROC_MEM_MAX]; struct clk *clk; > > > + void __iomem *bootreg; > > > }; > > > > > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > > > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc > > > *rproc) struct device *dev = priv->dev; > > > int ret; > > > > > > + /* write entry point to program counter */ > > > + writel(rproc->bootaddr, priv->bootreg); > > > > What happens on all the other IMX systems where this fix is not > > needed? Will they continue to work properly? > > Mathieu you are totally correct imx6/imx7 use different addresses they > boot. > > For imx7: > | On i.MX 7Dual/7Solo, the boot vector for the Cortex-M4 core is located > | at the start of the OCRAM_S (On Chip RAM - Secure) whose address is > | 0x0018_0000 from Cortex-A7. > > For imx6: > | The Boot vector for the Cortex-M4 core is located at the start of the > | TCM_L whose address is 0x007F_8000 from the Cortex-A9. This is a > | different location than on the i.MX 7Dual/7Solo > > But on imx7 0x0 is translated to 0x0018_0000 by imx_rproc_da_to_va, and > on imx7 0x0 is translated to 0x007F_8000, using imx_rproc_att_imx7d and > imx_rproc_att_imx6sx respectively. My point here is that before your patch, this driver was running on IMX platforms. How does your work impact existing platforms that are booting properly? > > I have no information about IMX8 (i have found none available > publicity), but should be the same as Cortex-M boots from 0x0. > > > > > > + > > > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > > > dcfg->src_mask, dcfg->src_start); > > > if (ret) > > > dev_err(dev, "Failed to enable M4!\n"); > > > > > > + dev_info(&rproc->dev, "Started from 0x%x\n", > > > rproc->bootaddr); + > > > return ret; > > > } > > > > > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) > > > if (ret) > > > dev_err(dev, "Failed to stop M4!\n"); > > > > > > + /* clear entry points */ > > > + writel(0, priv->bootreg); > > > + > > > return ret; > > > } > > > > > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc > > > *rproc, u64 da, int len) static const struct rproc_ops > > > imx_rproc_ops = { .start = imx_rproc_start, > > > .stop = imx_rproc_stop, > > > - .da_to_va = imx_rproc_da_to_va, > > > + .da_to_va = imx_rproc_da_to_va, > > > + .get_boot_addr = rproc_elf_get_boot_addr, > > > > How is this useful? Sure it will set rproc->bootaddr in > > rproc_fw_boot() but what good does that do when it is invariably set > > again in imx_rproc_start() ? > > > > > }; > > > > > > static int imx_rproc_addr_init(struct imx_rproc *priv, > > > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct > > > platform_device *pdev) goto err_put_rproc; > > > } > > > > > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, > > > sizeof(u32)); + > > > /* > > > * clk for M4 block including memory. Should be > > > * enabled before .start for FW transfer. > > > -- > > > 2.25.1 > > > >
On Fri, 17 Apr 2020 09:37:42 -0600 Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > On Fri, 17 Apr 2020 at 06:12, Nikita Shubin > <nikita.shubin@maquefel.me> wrote: > > > > On Tue, 14 Apr 2020 10:45:19 -0600 > > Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > > > > > Hi Nikita, > > > > > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, > > > nikita.shubin@maquefel.me wrote: > > > > In case elf file interrupt vector is not supposed to be at > > > > OCRAM_S, it is needed to write elf entry point to OCRAM_S + > > > > 0x4, to boot M4 firmware. > > > > > > > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > > > > > > > The firmware must set stack poiner as first instruction: > > > > > > > > Reset_Handler: > > > > ldr sp, = __stack /* set stack pointer */ > > > > > > > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > > > > > The address in the SoB has to match what is found in the "From:" > > > field of the email header. Checkpatch is complaining about that, > > > something I would have expected to be fixed before sending this > > > set out. > > > > > > > --- > > > > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > > > > 1 file changed, 15 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/remoteproc/imx_rproc.c > > > > b/drivers/remoteproc/imx_rproc.c index > > > > 3e72b6f38d4b..bebc58d0f711 100644 --- > > > > a/drivers/remoteproc/imx_rproc.c +++ > > > > b/drivers/remoteproc/imx_rproc.c @@ -45,6 +45,8 @@ > > > > > > > > #define IMX7D_RPROC_MEM_MAX 8 > > > > > > > > +#define IMX_BOOT_PC 0x4 > > > > + > > > > /** > > > > * struct imx_rproc_mem - slim internal memory structure > > > > * @cpu_addr: MPU virtual address of the memory region > > > > @@ -85,6 +87,7 @@ struct imx_rproc { > > > > const struct imx_rproc_dcfg *dcfg; > > > > struct imx_rproc_mem > > > > mem[IMX7D_RPROC_MEM_MAX]; struct clk > > > > *clk; > > > > + void __iomem *bootreg; > > > > }; > > > > > > > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > > > > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc > > > > *rproc) struct device *dev = priv->dev; > > > > int ret; > > > > > > > > + /* write entry point to program counter */ > > > > + writel(rproc->bootaddr, priv->bootreg); > > > > > > What happens on all the other IMX systems where this fix is not > > > needed? Will they continue to work properly? > > > > Mathieu you are totally correct imx6/imx7 use different addresses > > they boot. > > > > For imx7: > > | On i.MX 7Dual/7Solo, the boot vector for the Cortex-M4 core is > > located | at the start of the OCRAM_S (On Chip RAM - Secure) whose > > address is | 0x0018_0000 from Cortex-A7. > > > > For imx6: > > | The Boot vector for the Cortex-M4 core is located at the start of > > the | TCM_L whose address is 0x007F_8000 from the Cortex-A9. This > > is a | different location than on the i.MX 7Dual/7Solo > > > > But on imx7 0x0 is translated to 0x0018_0000 by imx_rproc_da_to_va, > > and on imx7 0x0 is translated to 0x007F_8000, using > > imx_rproc_att_imx7d and imx_rproc_att_imx6sx respectively. > > My point here is that before your patch, this driver was running on > IMX platforms. How does your work impact existing platforms that are > booting properly? Well it wasn't i am pretty sure it wasn't used at all or questions about boot process should have to be raised earlier. If we look into the first patch introduced by Oleksij Rempel https://lore.kernel.org/patchwork/cover/799614/ we can that the program he used is located at 0x0018_0000 so he didn't have any problems with Entry Point and stack pointer being at 0x0018_0000 and 0x0018_0004 respectively. But as i am trying to emphasize, IF elf is not supposed to be located starting at 0x0018_0000 it won't boot at all. Citing Oleksij: | no, currently my priority is to provide basic functionality with easy | understandable target code and dependencies. Moreover it seems not tested on IMX6 by anyone. I can limit functionality only to IMX7, until i lay hands on IMX6 with m4 co-proc - is this is what desired ? > > > > > I have no information about IMX8 (i have found none available > > publicity), but should be the same as Cortex-M boots from 0x0. > > > > > > > > > + > > > > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > > > > dcfg->src_mask, dcfg->src_start); > > > > if (ret) > > > > dev_err(dev, "Failed to enable M4!\n"); > > > > > > > > + dev_info(&rproc->dev, "Started from 0x%x\n", > > > > rproc->bootaddr); + > > > > return ret; > > > > } > > > > > > > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc > > > > *rproc) if (ret) > > > > dev_err(dev, "Failed to stop M4!\n"); > > > > > > > > + /* clear entry points */ > > > > + writel(0, priv->bootreg); > > > > + > > > > return ret; > > > > } > > > > > > > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc > > > > *rproc, u64 da, int len) static const struct rproc_ops > > > > imx_rproc_ops = { .start = imx_rproc_start, > > > > .stop = imx_rproc_stop, > > > > - .da_to_va = imx_rproc_da_to_va, > > > > + .da_to_va = imx_rproc_da_to_va, > > > > + .get_boot_addr = rproc_elf_get_boot_addr, > > > > > > How is this useful? Sure it will set rproc->bootaddr in > > > rproc_fw_boot() but what good does that do when it is invariably > > > set again in imx_rproc_start() ? > > > > > > > }; > > > > > > > > static int imx_rproc_addr_init(struct imx_rproc *priv, > > > > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct > > > > platform_device *pdev) goto err_put_rproc; > > > > } > > > > > > > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, > > > > sizeof(u32)); + > > > > /* > > > > * clk for M4 block including memory. Should be > > > > * enabled before .start for FW transfer. > > > > -- > > > > 2.25.1 > > > > > >
On Thu, 16 Apr 2020 at 23:40, <nikita.shubin@maquefel.me> wrote: > > Hi Mathieue, > > Hi Nikita, > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, nikita.shubin@maquefel.me wrote: > > In case elf file interrupt vector is not supposed to be at OCRAM_S, > it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 > firmware. > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > The firmware must set stack poiner as first instruction: > > Reset_Handler: > ldr sp, = __stack /* set stack pointer */ > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > > The address in the SoB has to match what is found in the "From:" field of the > email header. Checkpatch is complaining about that, something I would have > expected to be fixed before sending this set out. > > Noted and will be fixed. > > --- > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > 1 file changed, 15 insertions(+), 1 deletion(-) > > diff --git a/drivers/remoteproc/imx_rproc.c b/drivers/remoteproc/imx_rproc.c > index 3e72b6f38d4b..bebc58d0f711 100644 > --- a/drivers/remoteproc/imx_rproc.c > +++ b/drivers/remoteproc/imx_rproc.c > @@ -45,6 +45,8 @@ > > #define IMX7D_RPROC_MEM_MAX 8 > > +#define IMX_BOOT_PC 0x4 > + > /** > * struct imx_rproc_mem - slim internal memory structure > * @cpu_addr: MPU virtual address of the memory region > @@ -85,6 +87,7 @@ struct imx_rproc { > const struct imx_rproc_dcfg *dcfg; > struct imx_rproc_mem mem[IMX7D_RPROC_MEM_MAX]; > struct clk *clk; > + void __iomem *bootreg; > }; > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc *rproc) > struct device *dev = priv->dev; > int ret; > > + /* write entry point to program counter */ > + writel(rproc->bootaddr, priv->bootreg); > > > What happens on all the other IMX systems where this fix is not needed? Will > they continue to work properly? > > Yes, my bad, it is also needed for IMX6 (but even so i need to study this topic more carefully), > this should be applied exclusively for imx7d for now, and if will be needed someone > with imx6 hardware to test on can extend this on imx6 also. > > > > > + > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > dcfg->src_mask, dcfg->src_start); > if (ret) > dev_err(dev, "Failed to enable M4!\n"); > > + dev_info(&rproc->dev, "Started from 0x%x\n", rproc->bootaddr); > + > return ret; > } > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) > if (ret) > dev_err(dev, "Failed to stop M4!\n"); > > + /* clear entry points */ > + writel(0, priv->bootreg); > + > return ret; > } > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc *rproc, u64 da, int len) > static const struct rproc_ops imx_rproc_ops = { > .start = imx_rproc_start, > .stop = imx_rproc_stop, > - .da_to_va = imx_rproc_da_to_va, > + .da_to_va = imx_rproc_da_to_va, > + .get_boot_addr = rproc_elf_get_boot_addr, > > > How is this useful? Sure it will set rproc->bootaddr in rproc_fw_boot() but > what good does that do when it is invariably set again in imx_rproc_start() ? > > The priv->bootreg is the address where we are writing Entry Point and it is fixed, > 0x04 address is translated to 0x00180004, so don't quite understand you we > are writing rproc->bootaddr into priv->bootreg, not wiseversa. > What is your reason to set ops->get_boot_addr ? How does that help the work done in this patch? > > }; > > static int imx_rproc_addr_init(struct imx_rproc *priv, > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct platform_device *pdev) > goto err_put_rproc; > } > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, sizeof(u32)); > + > /* > * clk for M4 block including memory. Should be > * enabled before .start for FW transfer. > -- > 2.25.1 >
On Fri, 17 Apr 2020 11:01:22 -0600 Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > On Thu, 16 Apr 2020 at 23:40, <nikita.shubin@maquefel.me> wrote: > > > > Hi Mathieue, > > > > Hi Nikita, > > > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, nikita.shubin@maquefel.me > > wrote: > > > > In case elf file interrupt vector is not supposed to be at OCRAM_S, > > it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 > > firmware. > > > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > > > The firmware must set stack poiner as first instruction: > > > > Reset_Handler: > > ldr sp, = __stack /* set stack pointer */ > > > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > > > > > The address in the SoB has to match what is found in the "From:" > > field of the email header. Checkpatch is complaining about that, > > something I would have expected to be fixed before sending this set > > out. > > > > Noted and will be fixed. > > > > --- > > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > > 1 file changed, 15 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/remoteproc/imx_rproc.c > > b/drivers/remoteproc/imx_rproc.c index 3e72b6f38d4b..bebc58d0f711 > > 100644 --- a/drivers/remoteproc/imx_rproc.c > > +++ b/drivers/remoteproc/imx_rproc.c > > @@ -45,6 +45,8 @@ > > > > #define IMX7D_RPROC_MEM_MAX 8 > > > > +#define IMX_BOOT_PC 0x4 > > + > > /** > > * struct imx_rproc_mem - slim internal memory structure > > * @cpu_addr: MPU virtual address of the memory region > > @@ -85,6 +87,7 @@ struct imx_rproc { > > const struct imx_rproc_dcfg *dcfg; > > struct imx_rproc_mem mem[IMX7D_RPROC_MEM_MAX]; > > struct clk *clk; > > + void __iomem *bootreg; > > }; > > > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc > > *rproc) struct device *dev = priv->dev; > > int ret; > > > > + /* write entry point to program counter */ > > + writel(rproc->bootaddr, priv->bootreg); > > > > > > What happens on all the other IMX systems where this fix is not > > needed? Will they continue to work properly? > > > > Yes, my bad, it is also needed for IMX6 (but even so i need to > > study this topic more carefully), this should be applied > > exclusively for imx7d for now, and if will be needed someone with > > imx6 hardware to test on can extend this on imx6 also. > > > > > > > > > > + > > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > > dcfg->src_mask, dcfg->src_start); > > if (ret) > > dev_err(dev, "Failed to enable M4!\n"); > > > > + dev_info(&rproc->dev, "Started from 0x%x\n", rproc->bootaddr); > > + > > return ret; > > } > > > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) > > if (ret) > > dev_err(dev, "Failed to stop M4!\n"); > > > > + /* clear entry points */ > > + writel(0, priv->bootreg); > > + > > return ret; > > } > > > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc > > *rproc, u64 da, int len) static const struct rproc_ops > > imx_rproc_ops = { .start = imx_rproc_start, > > .stop = imx_rproc_stop, > > - .da_to_va = imx_rproc_da_to_va, > > + .da_to_va = imx_rproc_da_to_va, > > + .get_boot_addr = rproc_elf_get_boot_addr, > > > > > > How is this useful? Sure it will set rproc->bootaddr in > > rproc_fw_boot() but what good does that do when it is invariably > > set again in imx_rproc_start() ? > > > > The priv->bootreg is the address where we are writing Entry Point > > and it is fixed, 0x04 address is translated to 0x00180004, so don't > > quite understand you we are writing rproc->bootaddr into > > priv->bootreg, not wiseversa. > > > > What is your reason to set ops->get_boot_addr ? How does that help > the work done in this patch? The reason is the following : remoteproc_core.c: | rproc_fw_boot(struct rproc *rproc, const struct firmware *fw) | rproc->bootaddr = rproc_get_boot_addr(rproc, fw); remoteproc_internal.h | static inline | u32 rproc_get_boot_addr(struct rproc *rproc, const struct firmware *fw) | { | if (rproc->ops->get_boot_addr) | return rproc->ops->get_boot_addr(rproc, fw); | | return 0; | } > > > > > }; > > > > static int imx_rproc_addr_init(struct imx_rproc *priv, > > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct > > platform_device *pdev) goto err_put_rproc; > > } > > > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, > > sizeof(u32)); + > > /* > > * clk for M4 block including memory. Should be > > * enabled before .start for FW transfer. > > -- > > 2.25.1 > >
On Fri, 17 Apr 2020 at 11:27, Nikita Shubin <nikita.shubin@maquefel.me> wrote: > > On Fri, 17 Apr 2020 11:01:22 -0600 > Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > > > On Thu, 16 Apr 2020 at 23:40, <nikita.shubin@maquefel.me> wrote: > > > > > > Hi Mathieue, > > > > > > Hi Nikita, > > > > > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, nikita.shubin@maquefel.me > > > wrote: > > > > > > In case elf file interrupt vector is not supposed to be at OCRAM_S, > > > it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 > > > firmware. > > > > > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > > > > > The firmware must set stack poiner as first instruction: > > > > > > Reset_Handler: > > > ldr sp, = __stack /* set stack pointer */ > > > > > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > > > > > > > > The address in the SoB has to match what is found in the "From:" > > > field of the email header. Checkpatch is complaining about that, > > > something I would have expected to be fixed before sending this set > > > out. > > > > > > Noted and will be fixed. > > > > > > --- > > > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > > > 1 file changed, 15 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/remoteproc/imx_rproc.c > > > b/drivers/remoteproc/imx_rproc.c index 3e72b6f38d4b..bebc58d0f711 > > > 100644 --- a/drivers/remoteproc/imx_rproc.c > > > +++ b/drivers/remoteproc/imx_rproc.c > > > @@ -45,6 +45,8 @@ > > > > > > #define IMX7D_RPROC_MEM_MAX 8 > > > > > > +#define IMX_BOOT_PC 0x4 > > > + > > > /** > > > * struct imx_rproc_mem - slim internal memory structure > > > * @cpu_addr: MPU virtual address of the memory region > > > @@ -85,6 +87,7 @@ struct imx_rproc { > > > const struct imx_rproc_dcfg *dcfg; > > > struct imx_rproc_mem mem[IMX7D_RPROC_MEM_MAX]; > > > struct clk *clk; > > > + void __iomem *bootreg; > > > }; > > > > > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > > > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc > > > *rproc) struct device *dev = priv->dev; > > > int ret; > > > > > > + /* write entry point to program counter */ > > > + writel(rproc->bootaddr, priv->bootreg); > > > > > > > > > What happens on all the other IMX systems where this fix is not > > > needed? Will they continue to work properly? > > > > > > Yes, my bad, it is also needed for IMX6 (but even so i need to > > > study this topic more carefully), this should be applied > > > exclusively for imx7d for now, and if will be needed someone with > > > imx6 hardware to test on can extend this on imx6 also. > > > > > > > > > > > > > > > + > > > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > > > dcfg->src_mask, dcfg->src_start); > > > if (ret) > > > dev_err(dev, "Failed to enable M4!\n"); > > > > > > + dev_info(&rproc->dev, "Started from 0x%x\n", rproc->bootaddr); > > > + > > > return ret; > > > } > > > > > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) > > > if (ret) > > > dev_err(dev, "Failed to stop M4!\n"); > > > > > > + /* clear entry points */ > > > + writel(0, priv->bootreg); > > > + > > > return ret; > > > } > > > > > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc > > > *rproc, u64 da, int len) static const struct rproc_ops > > > imx_rproc_ops = { .start = imx_rproc_start, > > > .stop = imx_rproc_stop, > > > - .da_to_va = imx_rproc_da_to_va, > > > + .da_to_va = imx_rproc_da_to_va, > > > + .get_boot_addr = rproc_elf_get_boot_addr, > > > > > > > > > How is this useful? Sure it will set rproc->bootaddr in > > > rproc_fw_boot() but what good does that do when it is invariably > > > set again in imx_rproc_start() ? > > > > > > The priv->bootreg is the address where we are writing Entry Point > > > and it is fixed, 0x04 address is translated to 0x00180004, so don't > > > quite understand you we are writing rproc->bootaddr into > > > priv->bootreg, not wiseversa. > > > > > > > What is your reason to set ops->get_boot_addr ? How does that help > > the work done in this patch? > > The reason is the following : > > remoteproc_core.c: > | rproc_fw_boot(struct rproc *rproc, const struct firmware *fw) > | rproc->bootaddr = rproc_get_boot_addr(rproc, fw); > > remoteproc_internal.h > | static inline > | u32 rproc_get_boot_addr(struct rproc *rproc, const struct firmware > *fw) | { > | if (rproc->ops->get_boot_addr) > | return rproc->ops->get_boot_addr(rproc, fw); > | > | return 0; > | } And as I said above the value of rproc->bootaddr is set to priv->bootreg in imx_rproc_stop(). What am I missing? More over imx_rproc_ops doesn't have a ->load() function and as such rproc_alloc will set it to rproc_elf_get_boot_addr() > > > > > > > > > }; > > > > > > static int imx_rproc_addr_init(struct imx_rproc *priv, > > > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct > > > platform_device *pdev) goto err_put_rproc; > > > } > > > > > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, > > > sizeof(u32)); + > > > /* > > > * clk for M4 block including memory. Should be > > > * enabled before .start for FW transfer. > > > -- > > > 2.25.1 > > > >
On Fri, 17 Apr 2020 16:24:21 -0600 Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > On Fri, 17 Apr 2020 at 11:27, Nikita Shubin > <nikita.shubin@maquefel.me> wrote: > > > > On Fri, 17 Apr 2020 11:01:22 -0600 > > Mathieu Poirier <mathieu.poirier@linaro.org> wrote: > > > > > On Thu, 16 Apr 2020 at 23:40, <nikita.shubin@maquefel.me> wrote: > > > > > > > > Hi Mathieue, > > > > > > > > Hi Nikita, > > > > > > > > On Mon, Apr 06, 2020 at 02:33:08PM +0300, > > > > nikita.shubin@maquefel.me wrote: > > > > > > > > In case elf file interrupt vector is not supposed to be at > > > > OCRAM_S, it is needed to write elf entry point to OCRAM_S + > > > > 0x4, to boot M4 firmware. > > > > > > > > Otherwise firmware located anywhere besides OCRAM_S won't boot. > > > > > > > > The firmware must set stack poiner as first instruction: > > > > > > > > Reset_Handler: > > > > ldr sp, = __stack /* set stack pointer */ > > > > > > > > Signed-off-by: Nikita Shubin <NShubin@topcon.com> > > > > > > > > > > > > The address in the SoB has to match what is found in the "From:" > > > > field of the email header. Checkpatch is complaining about that, > > > > something I would have expected to be fixed before sending this > > > > set out. > > > > > > > > Noted and will be fixed. > > > > > > > > --- > > > > drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- > > > > 1 file changed, 15 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/remoteproc/imx_rproc.c > > > > b/drivers/remoteproc/imx_rproc.c index > > > > 3e72b6f38d4b..bebc58d0f711 100644 --- > > > > a/drivers/remoteproc/imx_rproc.c +++ > > > > b/drivers/remoteproc/imx_rproc.c @@ -45,6 +45,8 @@ > > > > > > > > #define IMX7D_RPROC_MEM_MAX 8 > > > > > > > > +#define IMX_BOOT_PC 0x4 > > > > + > > > > /** > > > > * struct imx_rproc_mem - slim internal memory structure > > > > * @cpu_addr: MPU virtual address of the memory region > > > > @@ -85,6 +87,7 @@ struct imx_rproc { > > > > const struct imx_rproc_dcfg *dcfg; > > > > struct imx_rproc_mem mem[IMX7D_RPROC_MEM_MAX]; > > > > struct clk *clk; > > > > + void __iomem *bootreg; > > > > }; > > > > > > > > static const struct imx_rproc_att imx_rproc_att_imx7d[] = { > > > > @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc > > > > *rproc) struct device *dev = priv->dev; > > > > int ret; > > > > > > > > + /* write entry point to program counter */ > > > > + writel(rproc->bootaddr, priv->bootreg); > > > > > > > > > > > > What happens on all the other IMX systems where this fix is not > > > > needed? Will they continue to work properly? > > > > > > > > Yes, my bad, it is also needed for IMX6 (but even so i need to > > > > study this topic more carefully), this should be applied > > > > exclusively for imx7d for now, and if will be needed someone > > > > with imx6 hardware to test on can extend this on imx6 also. > > > > > > > > > > > > > > > > > > > > + > > > > ret = regmap_update_bits(priv->regmap, dcfg->src_reg, > > > > dcfg->src_mask, > > > > dcfg->src_start); if (ret) > > > > dev_err(dev, "Failed to enable M4!\n"); > > > > > > > > + dev_info(&rproc->dev, "Started from 0x%x\n", > > > > rproc->bootaddr); + > > > > return ret; > > > > } > > > > > > > > @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc > > > > *rproc) if (ret) > > > > dev_err(dev, "Failed to stop M4!\n"); > > > > > > > > + /* clear entry points */ > > > > + writel(0, priv->bootreg); > > > > + > > > > return ret; > > > > } > > > > > > > > @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct > > > > rproc *rproc, u64 da, int len) static const struct rproc_ops > > > > imx_rproc_ops = { .start = imx_rproc_start, > > > > .stop = imx_rproc_stop, > > > > - .da_to_va = imx_rproc_da_to_va, > > > > + .da_to_va = imx_rproc_da_to_va, > > > > + .get_boot_addr = rproc_elf_get_boot_addr, > > > > > > > > > > > > How is this useful? Sure it will set rproc->bootaddr in > > > > rproc_fw_boot() but what good does that do when it is invariably > > > > set again in imx_rproc_start() ? > > > > > > > > The priv->bootreg is the address where we are writing Entry > > > > Point and it is fixed, 0x04 address is translated to > > > > 0x00180004, so don't quite understand you we are writing > > > > rproc->bootaddr into priv->bootreg, not wiseversa. > > > > > > > > > > What is your reason to set ops->get_boot_addr ? How does that > > > help the work done in this patch? > > > > The reason is the following : > > > > remoteproc_core.c: > > | rproc_fw_boot(struct rproc *rproc, const struct firmware *fw) > > | rproc->bootaddr = rproc_get_boot_addr(rproc, fw); > > > > remoteproc_internal.h > > | static inline > > | u32 rproc_get_boot_addr(struct rproc *rproc, const struct firmware > > *fw) | { > > | if (rproc->ops->get_boot_addr) > > | return rproc->ops->get_boot_addr(rproc, fw); > > | > > | return 0; > > | } > > And as I said above the value of rproc->bootaddr is set to > priv->bootreg in imx_rproc_stop(). What am I missing? More over > imx_rproc_ops doesn't have a ->load() function and as such rproc_alloc > will set it to rproc_elf_get_boot_addr() Yes, you are totally correct, it is not required in this patch thank you for pointing this out. > > > > > > > > > > > > > > }; > > > > > > > > static int imx_rproc_addr_init(struct imx_rproc *priv, > > > > @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct > > > > platform_device *pdev) goto err_put_rproc; > > > > } > > > > > > > > + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, > > > > sizeof(u32)); + > > > > /* > > > > * clk for M4 block including memory. Should be > > > > * enabled before .start for FW transfer. > > > > -- > > > > 2.25.1 > > > > > >
diff --git a/drivers/remoteproc/imx_rproc.c b/drivers/remoteproc/imx_rproc.c index 3e72b6f38d4b..bebc58d0f711 100644 --- a/drivers/remoteproc/imx_rproc.c +++ b/drivers/remoteproc/imx_rproc.c @@ -45,6 +45,8 @@ #define IMX7D_RPROC_MEM_MAX 8 +#define IMX_BOOT_PC 0x4 + /** * struct imx_rproc_mem - slim internal memory structure * @cpu_addr: MPU virtual address of the memory region @@ -85,6 +87,7 @@ struct imx_rproc { const struct imx_rproc_dcfg *dcfg; struct imx_rproc_mem mem[IMX7D_RPROC_MEM_MAX]; struct clk *clk; + void __iomem *bootreg; }; static const struct imx_rproc_att imx_rproc_att_imx7d[] = { @@ -162,11 +165,16 @@ static int imx_rproc_start(struct rproc *rproc) struct device *dev = priv->dev; int ret; + /* write entry point to program counter */ + writel(rproc->bootaddr, priv->bootreg); + ret = regmap_update_bits(priv->regmap, dcfg->src_reg, dcfg->src_mask, dcfg->src_start); if (ret) dev_err(dev, "Failed to enable M4!\n"); + dev_info(&rproc->dev, "Started from 0x%x\n", rproc->bootaddr); + return ret; } @@ -182,6 +190,9 @@ static int imx_rproc_stop(struct rproc *rproc) if (ret) dev_err(dev, "Failed to stop M4!\n"); + /* clear entry points */ + writel(0, priv->bootreg); + return ret; } @@ -243,7 +254,8 @@ static void *imx_rproc_da_to_va(struct rproc *rproc, u64 da, int len) static const struct rproc_ops imx_rproc_ops = { .start = imx_rproc_start, .stop = imx_rproc_stop, - .da_to_va = imx_rproc_da_to_va, + .da_to_va = imx_rproc_da_to_va, + .get_boot_addr = rproc_elf_get_boot_addr, }; static int imx_rproc_addr_init(struct imx_rproc *priv, @@ -360,6 +372,8 @@ static int imx_rproc_probe(struct platform_device *pdev) goto err_put_rproc; } + priv->bootreg = imx_rproc_da_to_va(rproc, IMX_BOOT_PC, sizeof(u32)); + /* * clk for M4 block including memory. Should be * enabled before .start for FW transfer.
In case elf file interrupt vector is not supposed to be at OCRAM_S, it is needed to write elf entry point to OCRAM_S + 0x4, to boot M4 firmware. Otherwise firmware located anywhere besides OCRAM_S won't boot. The firmware must set stack poiner as first instruction: Reset_Handler: ldr sp, = __stack /* set stack pointer */ Signed-off-by: Nikita Shubin <NShubin@topcon.com> --- drivers/remoteproc/imx_rproc.c | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-)