Re: [PATCH v2] brcmfmac: Add support for BCM43596 PCIe Wi-Fi

From: Konrad Dybcio
Date: Tue Dec 06 2022 - 05:08:54 EST




On 02/12/2022 20:28, Arend Van Spriel wrote:


On 12/2/2022 4:26 PM, Arend van Spriel wrote:
On 12/2/2022 11:33 AM, Konrad Dybcio wrote:


On 1.12.2022 12:31, Arend van Spriel wrote:
On 11/28/2022 3:40 PM, Konrad Dybcio wrote:


On 26.11.2022 22:45, Linus Walleij wrote:
On Fri, Nov 25, 2022 at 1:25 PM Kalle Valo <kvalo@xxxxxxxxxx> wrote:
Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> writes:

On 25.11.2022 12:53, Kalle Valo wrote:
Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> writes:

On 21.11.2022 14:56, Linus Walleij wrote:
On Fri, Nov 18, 2022 at 5:47 PM Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:

I can think of a couple of hacky ways to force use of 43596 fw, but I
don't think any would be really upstreamable..

If it is only known to affect the Sony Xperias mentioned then
a thing such as:

if (of_machine_is_compatible("sony,xyz") ||
      of_machine_is_compatible("sony,zzz")... ) {
     // Enforce FW version
}

would be completely acceptable in my book. It hammers the
problem from the top instead of trying to figure out itsy witsy
details about firmware revisions.

Yours,
Linus Walleij

Actually, I think I came up with a better approach by pulling a page
out of Asahi folks' book - please take a look and tell me what you
think about this:

[1]
https://github.com/SoMainline/linux/commit/4b6fccc995cd79109b0dae4e4ab2e48db97695e7
[2]
https://github.com/SoMainline/linux/commit/e3ea1dc739634f734104f37fdbed046873921af7

Something in this direction works too.

The upside is that it tells all operating systems how to deal
with the firmware for this hardware.

Instead of a directory path ("brcm/brcmfmac43596-pcie") why not provide
just the chipset name ("brcmfmac43596-pcie")? IMHO it's unnecessary to
have directory names in Device Tree.

I think it's common practice to include a full $FIRMWARE_DIR-relative
path when specifying firmware in DT, though here I left out the board
name bit as that's assigned dynamically anyway. That said, if you don't
like it, I can change it.

It's just that I have understood that Device Tree is supposed to
describe hardware and to me a firmware directory "brcm/" is a software
property, not a hardware property. But this is really for the Device
Tree maintainers to decide, they know this best :)

I would personally just minimize the amount of information
put into the device tree to be exactly what is needed to find
the right firmware.

brcm,firmware-compatible = "43596";

since the code already knows how to conjure the rest of the string.

But check with Rob/Krzysztof.

Yours,
Linus Walleij

Krzysztof, Rob [added to CC] - can I have your opinions?

I tried catching up on this thread. Reading it I am not sure what the issue is, but I am happy to dive in. If you can provide a boot log with brcmfmac loaded with module parameter 'debug=0x1416' I can try and make sense of the chipid/devid confusion.

Hope this helps, thanks! https://hastebin.com/xidagekuge.yaml

It does to some extent. It is basically a 4359 revision 9:

[   25.898782] brcmfmac: brcmf_chip_recognition found AXI chip: BCM4359/9

The 4359 entry in pcie.c is applicable for revision 0 and higher (doubtful but that is in the code):

     BRCMF_FW_ENTRY(BRCM_CC_4359_CHIP_ID, 0xFFFFFFFF, 4359),

We need to change the mask above to 0x000001FF and add a new entry with mask 0xFFFFFE00. All we need is come up with a reasonable firmware filename. So can you run the strings command on the firmware you use:

$ strings fw.bin | tail -1

and let me know the output.

Actually realized you already provided a URL to the repo containing the firmware you used. So I had a look and it shows:

43596a0-roml/pcie-ag-apcs-pktctx-proptxstatus-ampduhostreorder-lpc-die3-olpc-pspretend-mfp-ltecx-clm_43xx_somc_mimo-phyflags-txpwrctrls-dpo Version: 9.75.119.15 (r691661) CRC: a6cf427b Date: Fri 2017-03-24 13:24:25 KST Ucode Ver: 1060.20542 FWID: 01-e4abc35c

However, from firmware perspective this is equivalent to 4359c0 so I would suggest the change below.

Let me know if that works.
Sorry for the late reply.

Yes, it does seem to work just fine! The kernel now looks for brcm/brcmfmac4359c-pcie.sony,kagura-row.bin as we would expect.

Could you submit this patch below to supersede my one?

Konrad

Regards,
Arend
---
diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
index cf564adc612a..b59cf0f2939c 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
@@ -55,6 +55,7 @@ BRCMF_FW_CLM_DEF(4356, "brcmfmac4356-pcie");
 BRCMF_FW_CLM_DEF(43570, "brcmfmac43570-pcie");
 BRCMF_FW_DEF(4358, "brcmfmac4358-pcie");
 BRCMF_FW_DEF(4359, "brcmfmac4359-pcie");
+BCRMF_FW_DEF(4359C, "brcmfmac4359c-pcie");
 BRCMF_FW_DEF(4364, "brcmfmac4364-pcie");
 BRCMF_FW_DEF(4365B, "brcmfmac4365b-pcie");
 BRCMF_FW_DEF(4365C, "brcmfmac4365c-pcie");
@@ -83,7 +84,8 @@ static const struct brcmf_firmware_mapping brcmf_pcie_fwnames[] = {
     BRCMF_FW_ENTRY(BRCM_CC_43569_CHIP_ID, 0xFFFFFFFF, 43570),
     BRCMF_FW_ENTRY(BRCM_CC_43570_CHIP_ID, 0xFFFFFFFF, 43570),
     BRCMF_FW_ENTRY(BRCM_CC_4358_CHIP_ID, 0xFFFFFFFF, 4358),
-    BRCMF_FW_ENTRY(BRCM_CC_4359_CHIP_ID, 0xFFFFFFFF, 4359),
+    BRCMF_FW_ENTRY(BRCM_CC_4359_CHIP_ID, 0x000001FF, 4359),
+    BRCMF_FW_ENTRY(BRCM_CC_4359_CHIP_ID, 0xFFFFFE00, 4359C),
     BRCMF_FW_ENTRY(BRCM_CC_4364_CHIP_ID, 0xFFFFFFFF, 4364),
     BRCMF_FW_ENTRY(BRCM_CC_4365_CHIP_ID, 0x0000000F, 4365B),
     BRCMF_FW_ENTRY(BRCM_CC_4365_CHIP_ID, 0xFFFFFFF0, 4365C),