
On Tue, Feb 07, 2023 at 11:38:57AM -0700, Simon Glass wrote:
Hi,
On Mon, 16 Jan 2023 at 10:28, abdellatif.elkhlifi@arm.com wrote:
From: Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
add block storage emulation for NVM XIP flash devices
Some paltforms such as Corstone-1000 need to see NVM XIP raw flash as a block storage device with read only capability.
As mentioned I didn't see this patch originally. It generally looks OK but have made some comments below.
Here NVM flash devices are devices with addressable memory (e.g: QSPI NOR flash).
The implementation is generic and can be used by different platforms.
Two drivers are provided as follows.
nvmxip-blk :
a generic block driver allowing to read from the XIP flash
nvmxip_qspi :
The driver probed with the DT and parent of the nvmxip-blk device. nvmxip_qspi can be reused by other platforms. If the platform has custom settings to apply before using the flash, then the platform can provide its own parent driver belonging to UCLASS_NVMXIP and reuse nvmxip-blk. The custom driver can be implmented like nvmxip_qspi in addition to the platform custom settings.
Platforms can use multiple NVM XIP devices at the same time by defining a DT node for each one of them.
For more details please refer to doc/develop/driver-model/nvmxip.rst
Signed-off-by: Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
MAINTAINERS | 7 ++ doc/develop/driver-model/index.rst | 1 + doc/develop/driver-model/nvmxip.rst | 70 +++++++++++ doc/device-tree-bindings/nvmxip/nvmxip.txt | 56 +++++++++ drivers/Kconfig | 2 + drivers/Makefile | 1 + drivers/block/blk-uclass.c | 1 + drivers/nvmxip/Kconfig | 19 +++ drivers/nvmxip/Makefile | 8 ++ drivers/nvmxip/nvmxip-uclass.c | 15 +++ drivers/nvmxip/nvmxip.c | 129 +++++++++++++++++++++ drivers/nvmxip/nvmxip.h | 48 ++++++++ drivers/nvmxip/nvmxip_qspi.c | 67 +++++++++++ include/dm/uclass-id.h | 1 + 14 files changed, 425 insertions(+) create mode 100644 doc/develop/driver-model/nvmxip.rst create mode 100644 doc/device-tree-bindings/nvmxip/nvmxip.txt create mode 100644 drivers/nvmxip/Kconfig create mode 100644 drivers/nvmxip/Makefile create mode 100644 drivers/nvmxip/nvmxip-uclass.c create mode 100644 drivers/nvmxip/nvmxip.c create mode 100644 drivers/nvmxip/nvmxip.h create mode 100644 drivers/nvmxip/nvmxip_qspi.c
diff --git a/MAINTAINERS b/MAINTAINERS index b2de50ccfc..539d01f68c 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1193,6 +1193,13 @@ F: cmd/nvme.c F: include/nvme.h F: doc/develop/driver-model/nvme.rst
+NVMXIP +M: Abdellatif El Khlifi abdellatif.elkhlifi@arm.com +S: Maintained +F: doc/develop/driver-model/nvmxip.rst +F: doc/device-tree-bindings/nvmxip/nvmxip.txt +F: drivers/nvmxip/
NVMEM M: Sean Anderson seanga2@gmail.com S: Maintained diff --git a/doc/develop/driver-model/index.rst b/doc/develop/driver-model/index.rst index 7366ef818c..8e12bbd936 100644 --- a/doc/develop/driver-model/index.rst +++ b/doc/develop/driver-model/index.rst @@ -20,6 +20,7 @@ subsystems livetree migration nvme
- nvmxip of-plat pci-info pmic-framework
diff --git a/doc/develop/driver-model/nvmxip.rst b/doc/develop/driver-model/nvmxip.rst new file mode 100644 index 0000000000..91b24e4e50 --- /dev/null +++ b/doc/develop/driver-model/nvmxip.rst @@ -0,0 +1,70 @@ +.. SPDX-License-Identifier: GPL-2.0+
+NVM XIP Block Storage Emulation Driver +=======================================
+Summary +-------
+Non-Volatile Memory devices with addressable memory (e.g: QSPI NOR flash) could +be used for block storage needs (e.g: parsing a GPT layout in a raw QSPI NOR flash).
+The NVMXIP class provides this functionality and can be used for any 64-bit platform.
+The NVMXIP class provides the following drivers:
nvmxip-blk :
A generic block driver allowing to read from the XIP flash.
The driver belongs to UCLASS_BLK.
The driver implemented by drivers/nvmxip/nvmxip.c
nvmxip_qspi :
The driver probed with the DT and parent of the nvmxip-blk device.
nvmxip_qspi can be reused by other platforms. If the platform
has custom settings to apply before using the flash, then the platform
can provide its own parent driver belonging to UCLASS_NVMXIP and reuse
nvmxip-blk. The custom driver can be implmented like nvmxip_qspi in
addition to the platform custom settings.
The nvmxip_qspi driver belongs to UCLASS_NVMXIP.
The driver implemented by drivers/nvmxip/nvmxip_qspi.c
- The implementation is generic and can be used by different platforms.
+Supported hardware +--------------------------------
+Any 64-bit plaform.
platform
Why not 64-bit ? Please mention that.
+Configuration +----------------------
+config NVMXIP
This option allows the emulation of a block storage device
on top of a direct access non volatile memory XIP flash devices.
This support provides the read operation.
This option provides the block storage driver nvmxip-blk which
handles the read operation. This driver is HW agnostic and can support
multiple flash devices at the same time.
+config NVMXIP_QSPI
This option allows the emulation of a block storage device on top of a QSPI XIP flash.
Any platform that needs to emulate one or multiple XIP flash devices can turn this
option on to enable the functionality. NVMXIP config is selected automatically.
Platforms that need to add custom treatments before accessing to the flash, can
write their own driver (same as nvmxip_qspi in addition to the custom settings).
+Device Tree nodes +--------------------
+Multiple XIP flash devices can be used at the same time by describing them through DT +nodes.
+Please refer to the documentation of the DT binding at:
+doc/device-tree-bindings/nvmxip/nvmxip.txt
+Contributors +------------
- Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
diff --git a/doc/device-tree-bindings/nvmxip/nvmxip.txt b/doc/device-tree-bindings/nvmxip/nvmxip.txt new file mode 100644 index 0000000000..7c4b03f66b --- /dev/null +++ b/doc/device-tree-bindings/nvmxip/nvmxip.txt @@ -0,0 +1,56 @@ +Specifying NVMXIP information for devices +======================================
+NVM XIP flash device nodes +---------------------------
+Each flash device should have its own node.
+Each node must specify the following fields:
+1)
compatible = "nvmxip,qspi";
+This allows to bind the flash device with the nvmxip_qspi driver +If a platform has its own driver, please provide your own compatible +string.
+2)
reg = <0x0 0x08000000 0x0 0x00200000>;
Do we not use regs = /bits/ 64 <...? ?
+The start address and size of the flash device. The values give here are an +example (when the cell size is 2).
+When cell size is 1, the reg field looks like this:
reg = <0x08000000 0x00200000>;
+3)
lba_shift = <9>;
+The number of bit shifts used to calculate the size in bytes of one block. +In this example the block size is 1 << 9 = 2 ^ 9 = 512 bytes
+4)
lba = <4096>;
+The number of blocks.
+Example of multiple flash devices +----------------------------------------------------
nvmxip-qspi1@08000000 {
compatible = "nvmxip,qspi";
reg = <0x0 0x08000000 0x0 0x00200000>;
lba_shift = <9>;
lba = <4096>;
};
nvmxip-qspi2@08200000 {
compatible = "nvmxip,qspi";
reg = <0x0 0x08200000 0x0 0x00100000>;
lba_shift = <9>;
lba = <2048>;
};
diff --git a/drivers/Kconfig b/drivers/Kconfig index 9101e538b0..ef321bee94 100644 --- a/drivers/Kconfig +++ b/drivers/Kconfig @@ -76,6 +76,8 @@ source "drivers/net/Kconfig"
source "drivers/nvme/Kconfig"
+source "drivers/nvmxip/Kconfig"
source "drivers/pci/Kconfig"
source "drivers/pci_endpoint/Kconfig" diff --git a/drivers/Makefile b/drivers/Makefile index 83b14ef1fd..5e7fd6dab5 100644 --- a/drivers/Makefile +++ b/drivers/Makefile @@ -89,6 +89,7 @@ obj-$(CONFIG_FWU_MDATA) += fwu-mdata/ obj-y += misc/ obj-$(CONFIG_MMC) += mmc/ obj-$(CONFIG_NVME) += nvme/ +obj-$(CONFIG_NVMXIP) += nvmxip/
Do you think this justifies its own directory? Do we expect more drivers? Another option would be drivers/mtd
Is there an equivalent Linux driver?
Yes we can have more drivers using the NVMXIP Uclass (devices that need to run board specific treatment before accessing the NVM XIP flash). Anyway, I moved the NVMXIP folder under mtd. Regarding Linux, I'm not aware of any driver doing the same thing.
obj-$(CONFIG_PCI_ENDPOINT) += pci_endpoint/ obj-y += dfu/ obj-$(CONFIG_PCH) += pch/ diff --git a/drivers/block/blk-uclass.c b/drivers/block/blk-uclass.c index c69fc4d518..e8ab576c32 100644 --- a/drivers/block/blk-uclass.c +++ b/drivers/block/blk-uclass.c @@ -28,6 +28,7 @@ static struct { { UCLASS_AHCI, "sata" }, { UCLASS_HOST, "host" }, { UCLASS_NVME, "nvme" },
{ UCLASS_NVMXIP, "nvmxip" }, { UCLASS_EFI_MEDIA, "efi" }, { UCLASS_EFI_LOADER, "efiloader" }, { UCLASS_VIRTIO, "virtio" },
diff --git a/drivers/nvmxip/Kconfig b/drivers/nvmxip/Kconfig new file mode 100644 index 0000000000..3ef7105026 --- /dev/null +++ b/drivers/nvmxip/Kconfig @@ -0,0 +1,19 @@ +# SPDX-License-Identifier: GPL-2.0+ +# +# Copyright 2023 Arm Limited and/or its affiliates open-source-office@arm.com +# Authors: +# Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
+config NVMXIP
bool "NVM XIP devices support"
select BLK
help
This option allows the emulation of a block storage device
on top of a direct access non volatile memory XIP flash devices.
This support provides the read operation.
+config NVMXIP_QSPI
bool "QSPI XIP support"
select NVMXIP
help
This option allows the emulation of a block storage device on top of a QSPI XIP flash
diff --git a/drivers/nvmxip/Makefile b/drivers/nvmxip/Makefile new file mode 100644 index 0000000000..54eacc102e --- /dev/null +++ b/drivers/nvmxip/Makefile @@ -0,0 +1,8 @@ +# SPDX-License-Identifier: GPL-2.0+ +# +# Copyright 2023 Arm Limited and/or its affiliates open-source-office@arm.com +# Authors: +# Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
+obj-y += nvmxip-uclass.o nvmxip.o +obj-$(CONFIG_NVMXIP_QSPI) += nvmxip_qspi.o diff --git a/drivers/nvmxip/nvmxip-uclass.c b/drivers/nvmxip/nvmxip-uclass.c new file mode 100644 index 0000000000..aed8e163d2 --- /dev/null +++ b/drivers/nvmxip/nvmxip-uclass.c @@ -0,0 +1,15 @@ +// SPDX-License-Identifier: GPL-2.0+ +/*
- Copyright 2023 Arm Limited and/or its affiliates open-source-office@arm.com
- Authors:
- Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
- */
+#include <common.h> +#include <dm.h>
+UCLASS_DRIVER(nvmxip) = {
.name = "nvmxip",
.id = UCLASS_NVMXIP,
+}; diff --git a/drivers/nvmxip/nvmxip.c b/drivers/nvmxip/nvmxip.c new file mode 100644 index 0000000000..c276fb147e --- /dev/null +++ b/drivers/nvmxip/nvmxip.c @@ -0,0 +1,129 @@ +// SPDX-License-Identifier: GPL-2.0+ +/*
- Copyright 2023 Arm Limited and/or its affiliates open-source-office@arm.com
- Authors:
- Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
- */
+#include <common.h> +#include <dm.h> +#include <dm/device-internal.h> +#include "nvmxip.h"
+static u32 nvmxip_bdev_max_devs;
Please put that in the uclass-private data. We don't want statics, etc. with driver model.
+static int nvmxip_mmio_rawread(const phys_addr_t address, u64 *value) +{
*value = readq(address);
return 0;
+}
+static ulong nvmxip_blk_read(struct udevice *udev, lbaint_t blknr, lbaint_t blkcnt, void *buffer) +{
struct nvmxip_blk_priv *bpriv_data = dev_get_priv(udev);
struct blk_desc *desc = dev_get_uclass_plat(udev);
drop blank line, and can you combine the two following comments:
/* size of 1 block */
/* number of the u64 words to read */
u32 qwords = (blkcnt * desc->blksz) / sizeof(u64);
/* physical address of the first block to read */
phys_addr_t blkaddr = bpriv_data->pplat_data->phys_base + blknr * desc->blksz;
u64 *virt_blkaddr;
u64 *pdst = buffer;
u32 qdata_idx;
Why is this u32? It should be uint. Try to use natural types for loops and function parameters unless there is a very good reason.
if (!pdst)
return -EINVAL;
pr_debug("[%s]: reading from blknr: %lu , blkcnt: %lu\n", udev->name, blknr, blkcnt);
virt_blkaddr = map_sysmem(blkaddr, 0);
/* assumption: the data is virtually contiguous */
[..]
+int nvmxip_init(struct udevice *udev)
Why is this function exported? It should probably be called nvmxip_post_bind() and be called by the uclass when a new device is bound?
+{
struct nvmxip_plat *plat_data = dev_get_plat(udev);
plat
struct nvmxip_priv *priv_data = dev_get_priv(udev);
priv
(please use these names throughout as that is the convention; it makes it easier for people to understand your code)
int ret;
struct udevice *bdev = NULL;
char bdev_name[NVMXIP_BLKDEV_NAME_SZ + 1] = {0};
You don't need to init it as you write to it below.
priv_data->udev = udev;
priv_data->plat_data = plat_data;
nvmxip_bdev_max_devs++;
snprintf(bdev_name, NVMXIP_BLKDEV_NAME_SZ, "nvmxip-blk#%d", nvmxip_bdev_max_devs);
ret = blk_create_devicef(udev, NVMXIP_BLKDRV_NAME, bdev_name, UCLASS_NVMXIP,
nvmxip_bdev_max_devs, NVMXIP_DEFAULT_LBA_SZ,
NVMXIP_DEFAULT_LBA_COUNT, &bdev);
if (ret) {
pr_err("[%s]: failure during creation of the block device %s, error %d\n",
udev->name, bdev_name, ret);
goto blkdev_setup_error;
}
ret = blk_probe_or_unbind(bdev);
if (ret) {
pr_err("[%s]: failure during probing the block device %s, error %d\n",
udev->name, bdev_name, ret);
goto blkdev_setup_error;
}
pr_info("[%s]: the block device %s ready for use\n", udev->name, bdev_name);
return 0;
+blkdev_setup_error:
nvmxip_bdev_max_devs--;
I'm not sure if it helps, but there is a uclass_id_count() you can use.
return ret;
+}
+static const struct blk_ops nvmxip_blk_ops = {
.read = nvmxip_blk_read,
+};
+U_BOOT_DRIVER(nvmxip_blk) = {
.name = NVMXIP_BLKDRV_NAME,
.id = UCLASS_BLK,
.probe = nvmxip_blk_probe,
.ops = &nvmxip_blk_ops,
.priv_auto = sizeof(struct nvmxip_blk_priv),
+}; diff --git a/drivers/nvmxip/nvmxip.h b/drivers/nvmxip/nvmxip.h new file mode 100644 index 0000000000..47ae3bd8ab --- /dev/null +++ b/drivers/nvmxip/nvmxip.h @@ -0,0 +1,48 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/*
- Copyright 2023 Arm Limited and/or its affiliates open-source-office@arm.com
- Authors:
- Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
- */
+#ifndef __DRIVER_NVMXIP_H__ +#define __DRIVER_NVMXIP_H__
+#include <asm/io.h> +#include <blk.h> +#include <linux/bitops.h> +#include <linux/compat.h> +#include <mapmem.h>
Please put blk and mapmem in the C file, at least. Check if you need the others here, too.
Done in patchset v2, except for blk which is needed by nvmxip.h (struct nvmxip_plat uses lbaint_t type )
Cheers, Abdellatif
+#define NVMXIP_BLKDRV_NAME "nvmxip-blk"
+#define NVMXIP_BLKDEV_NAME_SZ 20
+#define NVMXIP_DEFAULT_LBA_SHIFT 10 /* 1024 bytes per block */ +#define NVMXIP_DEFAULT_LBA_COUNT 1024 /* block count */
+#define NVMXIP_DEFAULT_LBA_SZ BIT(NVMXIP_DEFAULT_LBA_SHIFT)
This is a private header so you don't need the long names - e.g. drop the NVMXIP_
+/* NVM XIP device platform data */
I suggest using sphinx style right from the start, since this is a new file.
+struct nvmxip_plat {
phys_addr_t phys_base; /* NVM XIP device base address */
u32 lba_shift; /* block size shift count (read from device tree) */
lbaint_t lba; /* number of blocks (read from device tree) */
+};
+/* NVM XIP device private data */ +struct nvmxip_priv {
struct udevice *udev;
struct nvmxip_plat *plat_data;
Please don't store private data from devices. You can use dev_get_plat() or whatever on a stored device. It just adds confusion.
+};
+/* Private data of the block device associated with the NVM XIP device (the parent) */ +struct nvmxip_blk_priv {
struct udevice *bdev;
What is that?
struct nvmxip_plat *pplat_data; /* parent device platform data */
same here
+};
+int nvmxip_init(struct udevice *udev);
should not be exported
+#endif /* __DRIVER_NVMXIP_H__ */ diff --git a/drivers/nvmxip/nvmxip_qspi.c b/drivers/nvmxip/nvmxip_qspi.c new file mode 100644 index 0000000000..d0ce4c6633 --- /dev/null +++ b/drivers/nvmxip/nvmxip_qspi.c
Can you put the driver in a separate patch? It is a good idea to have the uclass in a patch by itself.
@@ -0,0 +1,67 @@ +// SPDX-License-Identifier: GPL-2.0+ +/*
- Copyright 2023 Arm Limited and/or its affiliates open-source-office@arm.com
- Authors:
- Abdellatif El Khlifi abdellatif.elkhlifi@arm.com
- */
+#include <common.h> +#include <dm.h> +#include <fdt_support.h> +#include "nvmxip.h"
+#include <asm/global_data.h> +DECLARE_GLOBAL_DATA_PTR;
+#define NVMXIP_QSPI_DRV_NAME "nvmxip_qspi"
+static int nvmxip_qspi_probe(struct udevice *dev) +{
pr_debug("[%s][%s]\n", __func__, dev->name);
return nvmxip_init(dev);
+}
+static int nvmxip_qspi_of_to_plat(struct udevice *dev) +{
struct nvmxip_plat *plat_data = dev_get_plat(dev);
int ret;
plat_data->phys_base = (phys_addr_t)dev_read_addr(dev);
if (plat_data->phys_base == FDT_ADDR_T_NONE) {
pr_err("[%s]: can not get base address from device tree\n", dev->name);
This is adding a lot to code size...do you want to use debug instead?
return -EINVAL;
}
ret = dev_read_u32(dev, "lba_shift", &plat_data->lba_shift);
if (ret) {
pr_err("[%s]: can not get lba_shift from device tree\n", dev->name);
return -EINVAL;
}
ret = dev_read_u32(dev, "lba", (u32 *)&plat_data->lba);
if (ret) {
pr_err("[%s]: can not get lba from device tree\n", dev->name);
return -EINVAL;
}
pr_debug("[%s]: XIP device base addr: 0x%llx , lba_shift: %d , lbas: %lu\n",
dev->name, plat_data->phys_base, plat_data->lba_shift, plat_data->lba);
return 0;
+}
+static const struct udevice_id nvmxip_qspi_ids[] = {
{ .compatible = "nvmxip,qspi" },
{ /* sentinel */ }
+};
+U_BOOT_DRIVER(nvmxip_qspi) = {
.name = NVMXIP_QSPI_DRV_NAME,
.id = UCLASS_NVMXIP,
.of_match = nvmxip_qspi_ids,
.of_to_plat = nvmxip_qspi_of_to_plat,
.priv_auto = sizeof(struct nvmxip_priv),
.plat_auto = sizeof(struct nvmxip_plat),
.probe = nvmxip_qspi_probe,
+}; diff --git a/include/dm/uclass-id.h b/include/dm/uclass-id.h index 376f741cc2..dcfe8cfe2c 100644 --- a/include/dm/uclass-id.h +++ b/include/dm/uclass-id.h @@ -88,6 +88,7 @@ enum uclass_id { UCLASS_NOP, /* No-op devices */ UCLASS_NORTHBRIDGE, /* Intel Northbridge / SDRAM controller */ UCLASS_NVME, /* NVM Express device */
UCLASS_NVMXIP, /* NVM XIP devices */ UCLASS_P2SB, /* (x86) Primary-to-Sideband Bus */ UCLASS_PANEL, /* Display panel, such as an LCD */ UCLASS_PANEL_BACKLIGHT, /* Backlight controller for panel */
-- 2.17.1
Regards, Simon