On 3/9/2024 1:51 AM, Mario Limonciello wrote:
> On 3/4/2024 00:41, Basavaraj Natikar wrote:
>> AMD MP2 STB function provides a data buffer used to log debug
>> information
>> about the system execution during S2Idle suspend/resume.
>>
>> A data buffer known as the STB (Smart Trace Buffer) is a circular buffer
>> which is a low-level log to assist in debugging by providing insights
>> into any potential hangs or stalls that may occur during the S2Idle
>> suspend/resume processes.
>>
>> The current PMC driver retrieves STB data from MP1, but there can be
>> scenarios where MP1 might hang or become unresponsive, leading to the
>> loss of critical data present in the STB buffer. This defeats the
>> purpose
>> of the STB buffer, which was originally meant to help identify system
>> failures.
>>
>> This feature creates stb_read_previous_boot debugfs allows users to
>> retrieve the STB log from MP2 specifically from the last occurrence of
>> the S2Idle suspend/resume. A userspace daemon can access STB log of last
>> S2Idle suspend/resume which can help to troubleshoot potential issues
>> related to hangs or stalls during the S2Idle suspend/resume sequence.
>>
>> Reviewed-by: Shyam Sundar S K <Shyam-sundar.S-k@xxxxxxx>
>> Signed-off-by: Basavaraj Natikar <Basavaraj.Natikar@xxxxxxx>
>> ---
>> Changes in v3:
>> - Changed an error path for deinitialization or freeing resources.
>> - Modified the macro and variable to have a more specific name.
>> - Included header files for USEC_PER_MSEC, writeq, and writel.
>> - Switched the data type of the 'len' variable from 'int' to
>> 'unsigned int'.
>> - Used one variable for return values.
>>
>> Changes in v2:
>> - Used linux/sizes.h instead of defining KB.
>> - Defined timeout containing unit as per USEC_PER_MSEC.
>> - Modified macro of AMD_C2P_MSG and AMD_P2C_MSG.
>> - Changed string stb to STB in log messages.
>> - Defined literals using Macro.
>>
>> drivers/platform/x86/amd/pmc/Kconfig | 23 +++
>> drivers/platform/x86/amd/pmc/Makefile | 1 +
>> drivers/platform/x86/amd/pmc/mp2_stb.c | 260 +++++++++++++++++++++++++
>> drivers/platform/x86/amd/pmc/pmc.c | 4 +
>> drivers/platform/x86/amd/pmc/pmc.h | 14 ++
>> 5 files changed, 302 insertions(+)
>> create mode 100644 drivers/platform/x86/amd/pmc/mp2_stb.c
>>
>> diff --git a/drivers/platform/x86/amd/pmc/Kconfig
>> b/drivers/platform/x86/amd/pmc/Kconfig
>> index 883c0a95ac0c..8177486babf9 100644
>> --- a/drivers/platform/x86/amd/pmc/Kconfig
>> +++ b/drivers/platform/x86/amd/pmc/Kconfig
>> @@ -18,3 +18,26 @@ config AMD_PMC
>> If you choose to compile this driver as a module the module
>> will be
>> called amd-pmc.
>> +
>> +config AMD_MP2_STB
>> + bool "AMD SoC MP2 STB function"
>> + depends on AMD_PMC
>
> What do you think about adding:
> default AMD_PMC
> Then this will be essentially opt out and any distros that enable
> AMD_PMC by default will also start enabling this without extra prodding.
Sure, I will add 'default AMD_PMC'
>
>> + help
>> + AMD MP2 STB function provides a data buffer used to log debug
>> + information about the system execution during S2Idle
>> suspend/resume.
>> + A data buffer known as the STB (Smart Trace Buffer) is a circular
>> + buffer which is a low-level log for the SoC which is used to
>> debug
>> + any hangs/stalls during S2Idle suspend/resume.
>> +
>> + The current PMC driver retrieves STB data from MP1, but there
>> can be
>> + scenarios where MP1 might hang or become unresponsive, leading
>> to the
>> + loss of critical data present in the STB buffer. This defeats the
>> + purpose of the STB buffer, which was originally meant to help
>> identify
>> + system failures.
>> +
>> + Creates stb_read_previous_boot debugfs to get STB, a userspace
>> daemon
>> + can access STB log of last S2Idle suspend/resume which can
>> help to
>> + debug if hangs/stalls during S2Idle suspend/resume
>> +
>> + Say Y or N here if you have a notebook powered by AMD RYZEN
>> CPU/APU
>> + and supports MP2 STB.
>
> This last sentence reads as a confusing statement. I would just drop it.
Sure I will drop.
>
>> diff --git a/drivers/platform/x86/amd/pmc/Makefile
>> b/drivers/platform/x86/amd/pmc/Makefile
>> index 4aaa29d351c9..f1d9ab19d24c 100644
>> --- a/drivers/platform/x86/amd/pmc/Makefile
>> +++ b/drivers/platform/x86/amd/pmc/Makefile
>> @@ -6,3 +6,4 @@
>> amd-pmc-objs := pmc.o pmc-quirks.o
>> obj-$(CONFIG_AMD_PMC) += amd-pmc.o
>> +amd-pmc-$(CONFIG_AMD_MP2_STB) += mp2_stb.o
>> diff --git a/drivers/platform/x86/amd/pmc/mp2_stb.c
>> b/drivers/platform/x86/amd/pmc/mp2_stb.c
>> new file mode 100644
>> index 000000000000..4acea31daaaa
>> --- /dev/null
>> +++ b/drivers/platform/x86/amd/pmc/mp2_stb.c
>> @@ -0,0 +1,260 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * AMD MP2 STB layer
>> + *
>> + * Copyright (c) 2024, Advanced Micro Devices, Inc.
>> + * All Rights Reserved.
>> + *
>> + * Author: Basavaraj Natikar <Basavaraj.Natikar@xxxxxxx>
>> + */
>> +
>> +#include <linux/debugfs.h>
>> +#include <linux/device.h>
>> +#include <linux/io.h>
>> +#include <linux/iopoll.h>
>> +#include <linux/pci.h>
>> +#include <linux/sizes.h>
>> +#include <linux/time.h>
>> +
>> +#include "pmc.h"
>> +
>> +#define VALID_MSG 0xA
>> +#define VALID_RESPONSE 2
>> +
>> +#define AMD_C2P_MSG0 0x10500
>> +#define AMD_C2P_MSG1 0x10504
>> +#define AMD_P2C_MSG0 0x10680
>> +#define AMD_P2C_MSG1 0x10684
>> +
>> +#define MP2_RESP_SLEEP_US 500
>> +#define MP2_RESP_TIMEOUT_US (1600 * USEC_PER_MSEC)
>> +
>> +#define MP2_STB_DATA_LEN_2KB 1
>> +#define MP2_STB_DATA_LEN_16KB 4
>> +
>> +static struct amd_mp2_dev *mp2;
>
> Do you really need a global variable? I would think this can all be
> children of the amd_pmc dev and you can pass pointers and references
> to that around instead.
There's no need for a global variable. I'll use a reference from |amd_pmc|instead.
>
>> +
>> +struct mp2_cmd_base {
>> + union {
>> + u32 ul;
>> + struct {
>> + u32 cmd_id : 4;
>> + u32 intr_disable : 1;
>> + u32 is_dma_used : 1;
>> + u32 rsvd : 26;
>> + } cmd;
>> + };
>> +};
>> +
>> +struct mp2_cmd_response {
>> + union {
>> + u32 resp;
>> + struct {
>> + u32 cmd_id : 4;
>> + u32 status : 4;
>> + u32 response : 4;
>> + u32 rsvd2 : 20;
>> + } response;
>> + };
>> +};
>> +
>> +struct mp2_stb_data_valid {
>> + union {
>> + u32 data_valid;
>> + struct {
>> + u32 valid : 16;
>> + u32 length : 16;
>> + } val;
>> + };
>> +};
>> +
>> +static int amd_mp2_wait_response(u8 cmd_id, u32 command_sts)
>> +{
>> + struct mp2_cmd_response cmd_resp;
>> +
>> + if (!readl_poll_timeout(mp2->mmio + AMD_P2C_MSG0, cmd_resp.resp,
>> + (cmd_resp.response.response == 0x0 &&
>> + cmd_resp.response.status == command_sts &&
>> + cmd_resp.response.cmd_id == cmd_id),
>> MP2_RESP_SLEEP_US,
>> + MP2_RESP_TIMEOUT_US))
>> + return cmd_resp.response.status;
>> +
>> + return -ETIMEDOUT;
>> +}
>> +
>> +static void amd_mp2_stb_send_cmd(u8 cmd_id, bool is_dma_used)
>> +{
>> + struct mp2_cmd_base cmd_base;
>> +
>> + cmd_base.ul = 0;
>> + cmd_base.cmd.cmd_id = cmd_id;
>> + cmd_base.cmd.intr_disable = 1;
>> + cmd_base.cmd.is_dma_used = is_dma_used;
>> +
>> + writeq(mp2->dma_addr, mp2->mmio + AMD_C2P_MSG1);
>> + writel(cmd_base.ul, mp2->mmio + AMD_C2P_MSG0);
>> +}
>> +
>> +static int amd_mp2_stb_region(unsigned int len)
>> +{
>> + struct device *dev = &mp2->pdev->dev;
>> +
>> + len *= SZ_1K;
>> +
>> + if (!mp2->stbdata) {
>> + mp2->vslbase = dmam_alloc_coherent(dev, len, &mp2->dma_addr,
>> GFP_KERNEL);
>> + if (!mp2->vslbase)
>> + return -ENOMEM;
>> +
>> + mp2->stbdata = devm_kzalloc(dev, len, GFP_KERNEL);
>> + if (!mp2->stbdata)
>> + return -ENOMEM;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int amd_mp2_process_cmd(struct file *filp)
>> +{
>> + struct device *dev = &mp2->pdev->dev;
>> + struct mp2_stb_data_valid stb_dv;
>> + int status;
>> +
>> + stb_dv.data_valid = readl(mp2->mmio + AMD_P2C_MSG1);
>> +
>> + if (stb_dv.val.valid != VALID_MSG) {
>> + dev_err(dev, "Invalid STB data\n");
>> + return -EBADMSG;
>> + }
>> +
>> + if (stb_dv.val.length != MP2_STB_DATA_LEN_2KB &&
>> + stb_dv.val.length != MP2_STB_DATA_LEN_16KB) {
>> + dev_err(dev, "Unsupported length\n");
>> + return -EMSGSIZE;
>> + }
>> +
>> + mp2->stb_len = BIT(stb_dv.val.length);
>> +
>> + status = amd_mp2_stb_region(mp2->stb_len);
>> + if (status) {
>> + dev_err(dev, "Failed to init STB region, status %d\n", status);
>> + return status;
>> + }
>> +
>> + amd_mp2_stb_send_cmd(VALID_MSG, true);
>> + status = amd_mp2_wait_response(VALID_MSG, VALID_RESPONSE);
>> + if (status == VALID_RESPONSE) {
>> + memcpy_fromio(mp2->stbdata, mp2->vslbase, SZ_1K *
>> mp2->stb_len);
>> + filp->private_data = mp2->stbdata;
>> + mp2->is_stb_data = true;
>> + } else {
>> + dev_err(dev, "Failed to start STB dump, status %d\n", status);
>> + return -EOPNOTSUPP;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int amd_mp2_stb_debugfs_open(struct inode *inode, struct file
>> *filp)
>> +{
>> + if (!mp2->is_stb_data)
>> + return amd_mp2_process_cmd(filp);
>> +
>> + filp->private_data = mp2->stbdata;
>> +
>> + return 0;
>> +}
>> +
>> +static ssize_t amd_mp2_stb_debugfs_read(struct file *filp, char
>> __user *buf, size_t size,
>> + loff_t *pos)
>> +{
>> + if (!filp->private_data)
>> + return -EINVAL;
>> +
>> + return simple_read_from_buffer(buf, size, pos,
>> filp->private_data, SZ_1K * mp2->stb_len);
>> +}
>> +
>> +static const struct file_operations amd_mp2_stb_debugfs_fops = {
>> + .owner = THIS_MODULE,
>> + .open = amd_mp2_stb_debugfs_open,
>> + .read = amd_mp2_stb_debugfs_read,
>> +};
>> +
>> +static void amd_mp2_dbgfs_register(struct amd_pmc_dev *dev)
>> +{
>> + if (!dev->dbgfs_dir)
>> + return;
> Is this right? I don't see dbgfs_dir being initialized.
Yes, that's right. dbgfs_dir is already initialized in amd_pmc i.e. before call amd_mp2_stb_init.
>
>> +
>> + debugfs_create_file("stb_read_previous_boot", 0644,
>> dev->dbgfs_dir, dev,
>> + &amd_mp2_stb_debugfs_fops);
>> +}
>> +
>> +void amd_mp2_stb_deinit(void)
>> +{
>> + struct pci_dev *pdev;
>> +
>> + if (mp2 && mp2->pdev) {
>> + pdev = mp2->pdev;
>> + pci_dev_put(pdev);
>> + if (mp2->devres_gid)
>> + devres_release_group(&pdev->dev, mp2->devres_gid);
>> + mp2 = NULL;
>> + }
>> +}
>> +
>> +void amd_mp2_stb_init(struct amd_pmc_dev *dev)
>> +{
>> + struct pci_dev *pdev;
>> + int rc;
>> +
>> + mp2 = devm_kzalloc(dev->dev, sizeof(*mp2), GFP_KERNEL);
>> + if (!mp2)
>> + goto mp2_error;
>> +
>> + pdev = pci_get_device(PCI_VENDOR_ID_AMD, 0x172c, NULL);
>
> Three things:
> 1) Could you put the definition for it in amd/pmc/pmc.h?
> 2) Is this device ID always going to be 0x172c? If not, should it be
> paired with a specific check?
>
> IE something like if the root device is
> PCI_DEVICE_ID_AMD_1AH_M20H_ROOT then look for 0x172c etc?
>
> You could make a helper that is
>
> static int amd_mp2_get_stb() to abstract it all then.
>
> This could scale better then in the future when more device IDs are
> added.
>
> 3) Does this PCI ID get used for anything else? On a matching
> platform will amd-sfh use it or anything?
PCI ID used only for MP2_STB not for anything so I will just use Macro for 0x172c and remaining
code is simple no need to abstract.
>
>> + if (!pdev)
>> + goto mp2_error;
>> +
>
> If you re-order these two calls, no extra clean up will be needed if
> the PCI device is missing.
If I re-order then need extra clean up like below
pdev = pci_get_device(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_MP2_STB, NULL);
if (!pdev)
return;
mp2 = devm_kzalloc(dev->dev, sizeof(*mp2), GFP_KERNEL);
if (!mp2) {
pci_dev_put(pdev);
So I will keep same as original order.
>
>> + mp2->pdev = pdev;
>> +
>> + mp2->devres_gid = devres_open_group(&pdev->dev, NULL, GFP_KERNEL);
>> + if (!mp2->devres_gid) {
>> + dev_err(&pdev->dev, "devres_open_group failed\n");
>> + goto mp2_error;
>> + }
>> +
>> + rc = pcim_enable_device(pdev);
>> + if (rc) {
>> + dev_err(&pdev->dev, "pcim_enable_device failed\n");
>> + goto mp2_error;
>> + }
>> +
>> + rc = pcim_iomap_regions(pdev, BIT(2), "mp2 stb");
>> + if (rc) {
>> + dev_err(&pdev->dev, "pcim_iomap_regions failed\n");
>> + goto mp2_error;
>> + }
>> +
>> + /* MP2 MMIO is mapped to BAR2 */
>> + mp2->mmio = pcim_iomap_table(pdev)[2];
>> + if (!mp2->mmio) {
>> + dev_err(&pdev->dev, "pcim_iomap_table failed\n");
>> + goto mp2_error;
>> + }
>> +
>> + pci_set_master(pdev);
>
> AFAICT you don't undo setting master (pci_clear_master) if the DMA set
> mask fails nor on the cleanup path.
I'll also add this to the cleanup path.
>
>> +
>> + rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
>> + if (rc) {
>> + dev_err(&pdev->dev, "failed to set DMA mask\n");
>> + goto mp2_error;
>> + }
>> +
>> + amd_mp2_dbgfs_register(dev);
>> + dev->mp2 = mp2;
>> +
>> + return;
>> +
>> +mp2_error:
>> + amd_mp2_stb_deinit();
>> +}
>> diff --git a/drivers/platform/x86/amd/pmc/pmc.c
>> b/drivers/platform/x86/amd/pmc/pmc.c
>> index 108e12fd580f..926257125ac1 100644
>> --- a/drivers/platform/x86/amd/pmc/pmc.c
>> +++ b/drivers/platform/x86/amd/pmc/pmc.c
>> @@ -1106,6 +1106,8 @@ static int amd_pmc_probe(struct platform_device
>> *pdev)
>> }
>> amd_pmc_dbgfs_register(dev);
>> + if (IS_ENABLED(CONFIG_AMD_MP2_STB))
>> + amd_mp2_stb_init(dev);
>> pm_report_max_hw_sleep(U64_MAX);
>> return 0;
>> @@ -1122,6 +1124,8 @@ static void amd_pmc_remove(struct
>> platform_device *pdev)
>> acpi_unregister_lps0_dev(&amd_pmc_s2idle_dev_ops);
>> amd_pmc_dbgfs_unregister(dev);
>> pci_dev_put(dev->rdev);
>> + if (IS_ENABLED(CONFIG_AMD_MP2_STB))
>> + amd_mp2_stb_deinit();
>> mutex_destroy(&dev->lock);
>> }
>> diff --git a/drivers/platform/x86/amd/pmc/pmc.h
>> b/drivers/platform/x86/amd/pmc/pmc.h
>> index 827eef65e133..a49af1fa0060 100644
>> --- a/drivers/platform/x86/amd/pmc/pmc.h
>> +++ b/drivers/platform/x86/amd/pmc/pmc.h
>> @@ -14,6 +14,17 @@
>> #include <linux/types.h>
>> #include <linux/mutex.h>
>> +struct amd_mp2_dev {
>> + void __iomem *mmio;
>> + void __iomem *vslbase;
>> + void *stbdata;
>> + void *devres_gid;
>> + struct pci_dev *pdev;
>> + dma_addr_t dma_addr;
>> + int stb_len;
>> + bool is_stb_data;
>> +};
>> +
>> struct amd_pmc_dev {
>> void __iomem *regbase;
>> void __iomem *smu_virt_addr;
>> @@ -38,10 +49,13 @@ struct amd_pmc_dev {
>> struct dentry *dbgfs_dir;
>> struct quirk_entry *quirks;
>> bool disable_8042_wakeup;
>> + struct amd_mp2_dev *mp2;
>> };
>> void amd_pmc_process_restore_quirks(struct amd_pmc_dev *dev);
>> void amd_pmc_quirks_init(struct amd_pmc_dev *dev);
>> +void amd_mp2_stb_init(struct amd_pmc_dev *dev);
>> +void amd_mp2_stb_deinit(void);
>> /* List of supported CPU ids */
>> #define AMD_CPU_ID_RV 0x15D0
>
