XMMBoot

Introduction

For both libsamsung-ipc and the Linux driver it's interesting to understand better the boot of the modem in order to come with good names for the abstraction.

High level boot process

The exact boot sequence is different on different devices:

Abstraction

TODO:

GPIOs

Devices GPIOs assignement and drivers

Hardware Linux libsamsung-ipc
Variant SOC Modem Link GPIO usage GPIO assignement device driver name
Galaxy Tab:
GT-P1000
Exynos 3310 RAM aries
Galaxy S:
GT-I9000
Exynos 3110 XMM6160 RAM aries
Nexus S:
GT-I9020
GT-I9020A
GT-I9023
Exynos 3110 RAM crespo
Galaxy SII:
GT-I9100
Exynos 4210 XMM6260 HSIC CONFIG_UMTS_MODEM_XMM6260=y galaxys2
Galaxy Note:
GT-N7000
Exynos 4210 XMM6260 HSIC CONFIG_UMTS_MODEM_XMM6260=y galaxys2
Galaxy Nexus:
GT-I9250
OMAP 4460 XMM6260 MIPI CONFIG_UMTS_MODEM_XMM6260=y
Makefile
modem_modemctl_device_xmm6260.c
maguro
Galaxy SIII:
GT-I9300
Exynos 4412 XMM6262 HSIC CONFIG_UMTS_MODEM_XMM6262=y
Makefile
modem_modemctl_device_xmm6262.c
CONFIG_SEC_MODEM_M0=y
Makefile
board-m0-modems.c
CONFIG_MACH_M0=y
gpio-midas.h
gpio-rev00-m0.h
i9300
Galaxy Note 8.0 GSM:
GT-N5100
Exynos 4412 XMM6262 HSIC CONFIG_UMTS_MODEM_XMM6262=y n5100
Galaxy Note II:
GT-N7100
Exynos 4412 XMM6262 HSIC CONFIG_UMTS_MODEM_XMM6262=y n7100
Galaxy Tab 2:
GT-P3100
GT-P5100
OMAP 4430 XMM6262 MIPI CONFIG_UMTS_MODEM_XMM6262=y piranah

GPIOs usage

TODO: make sure to mention what applies to what device

Note that we don't limit ourselves to the drivers that are in use on the devices supported by Replicant.
As Samsung wrote drivers for the modem interfaces, and that the interface is similar across many different modems, other unused drivers and their comments also gives many hints about what the GPIOs are supposed to be used for.

gpio platform data name present absent direction comments
gpio_cp_on GT-I9300 powers on the modem? in which state (PMIC?, CPU?)
* On GT-I9100 it's connected to the ON1 modem pin and ON2 is not connected.
gpio_cp_reset GT-I9300 read + write Resets the modem CPU? PMIC?:
* ''check the reset timming with C2C connection'' : Here C2C probably means chip to chip
Can also read the modem CPU? and/or PMIC? reset state?
* Reads from the GPIO and ''CP not ready, Active State low'' comment
Also used in umts_link_reconnect (todo: check firmware status)
gpio_reset_req_n GT-I9300 (define) It seems to reset the modem. On some device with an MSM SOC (modem? AP?) in the smdk4412 kernel, it resets the modem PMIC. It can also be used to keep the modem (SOC? PMIC? CPU?) in reset mode while powered on.
It could very well be the PMIC with the GT-I9300 as we have the following sequence in xmm6262_on :
    gpio_set_value(mc->gpio_reset_req_n, 0);
    gpio_set_value(mc->gpio_cp_on, 0);
    gpio_set_value(mc->gpio_cp_reset, 0);
    msleep(100);
    gpio_set_value(mc->gpio_cp_reset, 1);
    /* If XMM6262 was connected with C2C, AP wait 50ms to BB Reset*/
    msleep(50);
    gpio_set_value(mc->gpio_reset_req_n, 1);

    gpio_set_value(mc->gpio_cp_on, 1);
    udelay(60);
    gpio_set_value(mc->gpio_cp_on, 0);
    msleep(20);

SOCs probably need to have a very specific voltage ramp up sequences to boot (voltage ramps and so on are described in the SOC datasheets, TODO: check with a specific example).
If that's true it would mean that the CPU has to not be off at boot but still be in some reset mode to accept a firmware?
In that case, once everything is off, the CPU power would be turned on first, then the PMIC, then the reset line for the firmware loading (way later)
gpio_pda_active GT-I9300 Tell the modem if the SOC CPUs are sleeping/active or not?
* PDA == Application processor
* ''PDA_ACTIVE, let cp know AP sleep'' comment in status gc1-gpio.c
* PDA_ACTIVE set to 0 right after cpu_pm_enter()
* PDA_ACTIVE set to 1 right before cpu_pm_exit()
* GPIO direction is output on AP side and input on BP side, which is also confirmed by the pinout table in XDA
Also indicates when the CPU is ready to process modem stuff:
* set to 1 at the end of xmm6262_on in modem_modemctl_device_xmm6262.c
* The CPU can't process stuff if the HSIC link is in low power mode, as shown in set_hsic_lpa_states in board-m0-modems.c so it sets gpio_pda_active to 0 in these cases.
gpio_phone_active GT-I9300 Seem the modem counterpart of gpio_pda_active:
* See umts_link_reconnect in board-m0-modems.c
* See also mc_work in the unused modemctl.c driver where that GPIO is used both to signal when the modem finished booting everything and is ready, and when the modem crashes or is reset
* phone_active_irq_handler in modem_modemctl_device_xmm6262.c seem to be doing exactly the same thing.
gpio_cp_dump_int GT-I9300 Maybe used for crash dump. Reference: Presentation about exploiting XMM modems: The nonfree s-boot 4.0 bootloader can probably get the modem crashdump. (cp is the modem)
gpio_flm_uart_sel Only used for the Galaxy Nexus in libsamsung-ipc GT-I9300 Modem download mode ?
* FLM could be Firmware Load mode ?
* On several devices, that GPIO seem to be used to switch between different UART, and the PMIC seem involved too in some devices. Not sure how it switches
gpio_cp_warm_reset GT-I9300
gpio_revers_bias_clear GT-I9300
gpio_revers_bias_restore GT-I9300
gpio_sim_detect GT-I9300 (GPIO_SIM_DETECT is not set) Detect SIM card presence ?
gpio_link_enable GT-I9300 power management
gpio_link_active GT-I9300 (define) set to 0 when the (HSIC) link is in low power and to 1 when it's back, like in set_hsic_lpa_states in board-m0-modems.c , power management
gpio_link_hostwake GT-I9300 (define) power management
According to the code below in modem_link_device_hsic.c, it seems that the modem can tell the host to set the HSIC link to sleep when that GPIO is high, and to re-enable the link when it's low:
static irqreturn_t link_pm_irq_handler(int irq, void *data)
{
        int value;
        struct link_pm_data *pm_data = data;

#if defined(CONFIG_SLP)
        pm_wakeup_event(pm_data->miscdev.this_device, 0);
#endif

        if (!pm_data->link_pm_active)
                return IRQ_HANDLED;

        /* host wake up HIGH */
        /*
                resume usb runtime pm start
        */
        /* host wake up LOW */
        /*
                slave usb enumeration end,
                host can send usb packet after
                runtime pm status changes to ACTIVE
        */
        [...]
}
gpio_link_slavewake GT-I9300 (define) power management
According to the code in board-m0-modems.c, it seems to be used to wakup the modem or to make it exit low power mode by triggering a rising edge but I don't know why it checks the hostwake GPIO in the code below:
/* HSIC specific function */
void set_slave_wake(void)
{
        if (gpio_get_value(modem_link_pm_data.gpio_link_hostwake)) {
                pr_info("[MODEM_IF]Slave Wake\n");
                if (gpio_get_value(modem_link_pm_data.gpio_link_slavewake)) {
                        pr_info("[MODEM_IF]Slave Wake set _-\n");
                        gpio_direction_output(
                        modem_link_pm_data.gpio_link_slavewake, 0);
                        mdelay(10);
                }
                gpio_direction_output(
                        modem_link_pm_data.gpio_link_slavewake, 1);
        }
}
[...]
void set_hsic_lpa_states(int states)
{
        int val = gpio_get_value(umts_modem_data.gpio_cp_reset);

        mif_trace("\n");

        if (val) {
                switch (states) {
                [...]
                case STATE_HSIC_LPA_PHY_INIT:
                        gpio_set_value(umts_modem_data.gpio_pda_active, 1);
                        set_slave_wake();
                        pr_info(LOG_TAG "set hsic lpa phy init: " 
                                "slave wake-up (%d)\n",
                                gpio_get_value(
                                        modem_link_pm_data.gpio_link_slavewake)
                                );
                        break;
                }
        }
}
gpio_ap_dump_int GT-I9300 (define) Maybe used for crash dump. Reference: Presentation about exploiting XMM modems: The nonfree s-boot 4.0 bootloader can probably get the modem crashdump. Here it would indicate that the host somehow can send crashdumps to the modem somehow?
gpio_sim_io_sel GT-I9300 (define) Seem to be used for some dual modem feature which is not present in any Replicant kernels (aries, crespo, espresso10, smdk4412, tuna) for Replicant 6 (or 4.2 for aries and crespo)
gpio_cp_ctrl1 GT-I9300 (define)
gpio_cp_ctrl2 GT-I9300 (define)

Libsamsung-ipc

ioctl / function Devices
GT-I9250 GT-I9100 / GT-N7000 GT-I9300 GT-N5100 GT-N7100 GT-P3100 / GT-P5100 Galaxy S (GT-I9000) / Galaxy Tab Nexus S (GT-I902x)
libsamsung-ipc driver maguro galaxys2 i9300 n5100 n7100 piranah aries crespo
open, close, read, write
fmt/rfs
gprs
power
Yes
boot_power
status_online_wait
Yes No No
hci_power
link_control_active
link_connected_wait
link_get_hostwake_wait
No Yes No
link_control_enable Yes (ignored by Linux4[5][6]) Yes (ignored by Linux1[2][3])
psi_send xmm626_mipi_psi_send xmm626_hsic_psi_send xmm626_mipi_psi_send xmm616_psi_send
firmware_send xmm626_mipi_firmware_send xmm626_hsic_firmware_send xmm626_mipi_firmware_send xmm616_firmware_send
nv_data_send xmm626_mipi_nv_data_send xmm626_hsic_nv_data_send xmm626_mipi_nv_data_send xmm616_nv_data_send

1 https://git.replicant.us/replicant/kernel_samsung_smdk4412/tree/arch/arm/mach-exynos/board-m0-modems.c#n221

2 https://git.replicant.us/replicant/kernel_samsung_smdk4412/tree/arch/arm/mach-exynos/board-m0-modems.c#n218

3 https://git.replicant.us/replicant/kernel_samsung_smdk4412/tree/arch/arm/mach-exynos/board-m0-modems.c#n136

4 https://git.replicant.us/replicant/kernel_samsung_smdk4412/tree/arch/arm/mach-exynos/board-u1-modems.c#n153

5 https://git.replicant.us/replicant/kernel_samsung_smdk4412/tree/arch/arm/mach-exynos/board-u1-modems.c#n151

6 https://git.replicant.us/replicant/kernel_samsung_smdk4412/tree/arch/arm/mach-exynos/board-u1-modems.c#n139

libsamsung-ipc <-> kernel functions <-> gpios

libsamsung-ipc Kernel
Function using the ioctl ioctl name pointer signature GPIO used comment
xmm626_kernel_smdk4412_power IOCTL_MODEM_ON
IOCTL_MODEM_OFF
int (*modem_on)(struct modem_ctl*);

int (*modem_off)(struct modem_ctl*);
gpio_cp_on
gpio_cp_reset
gpio_reset_req_n
gpio_pda_active
gpio_phone_active
gpio_cp_dump_int
xmm626_kernel_smdk4412_boot_power IOCTL_MODEM_BOOT_ON
IOCTL_MODEM_BOOT_OFF
int (*modem_boot_on)(struct modem_ctl*);

int (*modem_boot_off)(struct modem_ctl*);
gpio_flm_uart_sel
gpio_cp_warm_reset
gpio_revers_bias_clear
gpio_revers_bias_restore
gpio_sim_detect
xmm626_kernel_smdk4412_status_online_wait IOCTL_MODEM_STATUS int phone_state; gpio_cp_on
gpio_cp_reset
gpio_pda_active
gpio_reset_req_n
gpio_phone_active
int phone_state get assigned the status computed from the various GPIO states
xmm626_kernel_smdk4412_status_online_wait only waits for the online status
xmm626_kernel_smdk4412_link_control_enable IOCTL_LINK_CONTROL_ENABLE
int (*link_ldo_enable)(bool);
gpio_link_enable on i9300:
* link_ldo_enable only returns 0 and has a comment ("Exynos HSIC V1.2 LDO was controlled by kernel")
* gpio_link_enable is set to 0 (so it's ignored)

Glossary

Terms for the modem CPU: Term for the CPU of the system on a chip running Replicant:

TODO: move in its own page and point to it

LPA: Low power mode active (Related to ULPI specs only?)

ULPI: Probably a USB PHY spec

SIM card presence detection

Do we really want to check the SIM card presence?

Would it be possible not to for privacy reasons?

Example:

TODO

Potential privacy and security issues

gpio_pda_active

From cpuidle-exynos4.c we have things like that:

    cpu_pm_enter();

#if defined(CONFIG_INTERNAL_MODEM_IF) || defined(CONFIG_SAMSUNG_PHONE_TTY)
    gpio_set_value(GPIO_PDA_ACTIVE, 0);
#endif

    if (log_en)
        pr_debug("+++lpa\n")

and:

    if (log_en)
        pr_debug("---lpa\n");
#if defined(CONFIG_INTERNAL_MODEM_IF) || defined(CONFIG_SAMSUNG_PHONE_TTY)
    gpio_set_value(GPIO_PDA_ACTIVE, 1);
#endif

    cpu_pm_exit();

Does it means that we are telling the modem about each time we go in suspend to RAM?

Devices affected or not affected:

Device Config
GT-I9300 # CONFIG_INTERNAL_MODEM_IF is not set
# CONFIG_SAMSUNG_PHONE_TTY is not set

gpio_phone_active

From ehci-s5p.c we have things like that:

#if defined(CONFIG_UMTS_MODEM_XMM6262)
    if (pdata->get_cp_active_state && !pdata->get_cp_active_state()) {
        s5p_ehci_port_control(pdev, CP_PORT, 0);
        pr_err("mif: force port%d off by cp reset\n", CP_PORT);
    }
#endif

Does it allows the modem to trigger a re-enumeration of the HSIC bus?

Devices affected or not affected:

Device Config
GT-I9300 CONFIG_UMTS_MODEM_XMM6262=y
.gpio_phone_active = GPIO_PHONE_ACTIVE

Links