Re: [PATCH v6 1/8] mmc: sdhci-of-dwcmshc: add common bulk optional clocks support

[Adrian Hunter <adrian.hunter@intel.com>]

On 24/07/2025 17:33, Diederik de Haas wrote:
> Hi Adrian,
> 
> On Wed Jul 23, 2025 at 7:33 AM CEST, Adrian Hunter wrote:
>> On 22/07/2025 21:33, Robin Murphy wrote:
>>> A bit late for a "review", but Diederik and I have just been
>>> IRC-debugging a crash on RK3568 which by inspection seems to be caused
>>> by this patch:
>>>
>>> On 2024-08-05 10:17 am, Chen Wang wrote:
>>>> From: Chen Wang <unicorn_wang@outlook.com>
>>>>
>>>> In addition to the required core clock and optional
>>>> bus clock, the soc will expand its own clocks, so
>>>> the bulk clock mechanism is abstracted.
>>>>
>>>> Note, I call the bulk clocks as "other clocks" due
>>>> to the bus clock has been called as "optional".
>>>>
>>>> Signed-off-by: Chen Wang <unicorn_wang@outlook.com>
>>>> Tested-by: Drew Fustini <drew@pdp7.com> # TH1520
>>>> Tested-by: Inochi Amaoto <inochiama@outlook.com> # Duo and Huashan Pi
>>>> ---
>>> [...]
>>>> +static int dwcmshc_get_enable_other_clks(struct device *dev,
>>>> +                     struct dwcmshc_priv *priv,
>>>> +                     int num_clks,
>>>> +                     const char * const clk_ids[])
>>>> +{
>>>> +    int err;
>>>> +
>>>> +    if (num_clks > DWCMSHC_MAX_OTHER_CLKS)
>>>> +        return -EINVAL;
>>>> +
>>>> +    for (int i = 0; i < num_clks; i++)
>>>> +        priv->other_clks[i].id = clk_ids[i];
>>>> +
>>>> +    err = devm_clk_bulk_get_optional(dev, num_clks, priv->other_clks);
>>>
>>> This leaves a pointer into "priv" in the devres list...
>>>
>>>> +    if (err) {
>>>> +        dev_err(dev, "failed to get clocks %d\n", err);
>>>> +        return err;
>>>> +    }
>>>> +
>>>> +    err = clk_bulk_prepare_enable(num_clks, priv->other_clks);
>>>> +    if (err)
>>>> +        dev_err(dev, "failed to enable clocks %d\n", err);
>>>> +
>>>> +    priv->num_other_clks = num_clks;
>>>> +
>>>> +    return err;
>>>> +}
>>>> +
>>>>   /*
>>>>    * If DMA addr spans 128MB boundary, we split the DMA transfer into two
>>>>    * so that each DMA transfer doesn't exceed the boundary.
>>> [...]
>>>> @@ -1280,9 +1300,7 @@ static int dwcmshc_probe(struct platform_device *pdev)
>>>>   err_clk:
>>>>       clk_disable_unprepare(pltfm_host->clk);
>>>>       clk_disable_unprepare(priv->bus_clk);
>>>> -    if (rk_priv)
>>>> -        clk_bulk_disable_unprepare(RK35xx_MAX_CLKS,
>>>> -                       rk_priv->rockchip_clks);
>>>> +    clk_bulk_disable_unprepare(priv->num_other_clks, priv->other_clks);
>>>>   free_pltfm:
>>>>       sdhci_pltfm_free(pdev);
>>>
>>> ...but upon, say, -EPROBE_DEFER from sdhci_setup_host() because a
>>> regulator isn't ready yet, that "priv" is freed here, so by the time the
>>> devres callbacks eventually run, that "devres->clks" pointer which used
>>> to represent "priv->other_clocks" points to who knows what, and this
>>> sort of thing happens:
>>>
>>> [   12.470827] Unable to handle kernel paging request at virtual address 002df7b378917664
>>> [   12.472104] Mem abort info:
>>> [   12.472471]   ESR = 0x0000000096000004
>>> [   12.475991]   EC = 0x25: DABT (current EL), IL = 32 bits
>>> [   12.476657]   SET = 0, FnV = 0
>>> [   12.477146]   EA = 0, S1PTW = 0
>>> [   12.477547]   FSC = 0x04: level 0 translation fault
>>> [   12.478127] Data abort info:
>>> [   12.478126] rockchip-gpio fdd60000.gpio: probed /pinctrl/gpio@fdd60000
>>> [   12.478413]   ISV = 0, ISS = 0x00000004, ISS2 = 0x00000000
>>> [   12.479826]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
>>> [   12.480418]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
>>> [   12.481282] [002df7b378917664] address between user and kernel address ranges
>>> [   12.482421] Internal error: Oops: 0000000096000004 [#1]  SMP
>>> [   12.482980] Modules linked in: sdhci_of_dwcmshc drm_dp_aux_bus gpio_rockchip(+) drm_display_helper dw_mmc_rockchip drm_client_lib sdhci_pltfm drm_dma_helper fwnode_mdio sdhci dw_mmc_pltf
>>> m libphy fixed rockchip_dfi drm_kms_helper cqhci pl330(+) phy_rockchip_naneng_combphy dw_wdt phy_rockchip_snps_pcie3 phy_rockchip_inno_usb2 dw_mmc mdio_bus dwc3 ehci_platform ohci_platform
>>> ehci_hcd drm ohci_hcd udc_core io_domain i2c_rk3x usbcore ulpi usb_common
>>> [   12.486871] CPU: 0 UID: 0 PID: 64 Comm: kworker/u16:3 Not tainted 6.16-rc7-arm64-cknow #1 PREEMPTLAZY  Debian 6.16~rc7-1
>>> [   12.487901] Hardware name: FriendlyElec NanoPi R5S (DT)
>>> [   12.488412] Workqueue: async async_run_entry_fn
>>> [   12.488879] pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>>> [   12.489539] pc : __clk_put+0x2c/0x138
>>> [   12.489913] lr : __clk_put+0x2c/0x138
>>> [   12.490281] sp : ffff800080713b10
>>> [   12.490607] x29: ffff800080713b10 x28: ffff0001f001a120 x27: 0000000000000000
>>> [   12.491302] x26: ffff0001f98e01a0 x25: 0000000000000000 x24: ffff0001f0f35408
>>> [   12.491995] x23: ffffa8da199b4b40 x22: ffff800080713bb0 x21: ffff0001f0f35010
>>> [   12.492689] x20: ffff0001f94aafd0 x19: 0a2df7b378917634 x18: 00000000ffffffff
>>> [   12.493381] x17: 3d4d455453595342 x16: 555300307075656b x15: ffff0001f4885650
>>> [   12.494075] x14: 0000000000000000 x13: ffff0001f025b810 x12: 0000000000008000
>>> [   12.494765] x11: ffffa8da1a73ef98 x10: ffffa8da1a460000 x9 : 0000000000000078
>>> [   12.495454] x8 : 0000000000000049 x7 : ffffa8da18c2fbe0 x6 : 0000000000000001
>>> [   12.496145] x5 : 0000000000000004 x4 : 000000006cb6bb63 x3 : 0000000000000000
>>> [   12.496833] x2 : 0000000000000000 x1 : ffff0001f1365ac0 x0 : 0000000000000001
>>> [   12.497524] Call trace:
>>> [   12.497776]  __clk_put+0x2c/0x138 (P)
>>> [   12.498154]  clk_put+0x18/0x30
>>> [   12.498471]  clk_bulk_put+0x40/0x68
>>> [   12.498825]  devm_clk_bulk_release+0x24/0x40
>>> [   12.499248]  release_nodes+0x64/0xa0
>>> [   12.499608]  devres_release_all+0x98/0xf8
>>> [   12.500004]  device_unbind_cleanup+0x20/0x70
>>> [   12.500426]  really_probe+0x1e8/0x3a0
>>> [   12.500793]  __driver_probe_device+0x84/0x160
>>> [   12.501225]  driver_probe_device+0x44/0x128
>>> [   12.501640]  __driver_attach_async_helper+0x5c/0x108
>>> [   12.502125]  async_run_entry_fn+0x40/0x180
>>> [   12.502535]  process_one_work+0x23c/0x640
>>> [   12.502939]  worker_thread+0x1b4/0x360
>>> [   12.503315]  kthread+0x150/0x250
>>> [   12.503646]  ret_from_fork+0x10/0x20
>>> [   12.504015] Code: aa0003f3 b140041f 540006c8 97ffd9c4 (b9403260)
>>> [   12.504598] ---[ end trace 0000000000000000 ]---
>>>
>>>
>>> TBH I'm not sure what to do as a straight revert seems impractical by
>>> now, so we hope someone else might have a good idea.
>>
>> Presumably the problem has gone away with:
>>
>> 	commit 91a001a1a0749e5d24606d46ac5dfd4433c00956
>> 	Author: Binbin Zhou <zhoubinbin@loongson.cn>
>> 	Date:   Sat Jun 7 15:39:01 2025 +0800
>>
>> 	    mmc: sdhci-of-dwcmshc: Drop the use of sdhci_pltfm_free()
>>
>> which is in next.
>>
>> In which case a separate fix is needed for stable.
> 
> Adding that patch to my 6.16-rc7 kernel indeed stopped the OOPSies.
> Thanks!

You need the other patches that it depends on, otherwise you are
just leaking the memory.  Refer:

	https://lore.kernel.org/all/cover.1749127796.git.zhoubinbin@loongson.cn/